xref: /openbmc/linux/drivers/net/bonding/bond_main.c (revision c51d39010a1bccc9c1294e2d7c00005aefeb2b5c)
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *	Cisco 5500
11  *	Sun Trunking (Solaris)
12  *	Alteon AceDirector Trunks
13  *	Linux Bonding
14  *	and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *	will be assigned at this time.  The hw mac address will come from
20  *	the first slave bonded to the channel.  All slaves will then use
21  *	this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *	will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *	a: be used as initial mac address
29  *	b: if a hw mac address already is there, eth0's hw mac address
30  *	   will then be set from bond0.
31  *
32  */
33 
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
41 #include <linux/in.h>
42 #include <net/ip.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <linux/io.h>
55 #include <asm/dma.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
63 #include <net/sock.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
67 #include <net/arp.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
84 
85 #include "bonding_priv.h"
86 
87 /*---------------------------- Module parameters ----------------------------*/
88 
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 
91 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon;
95 static int updelay;
96 static int downdelay;
97 static int use_carrier	= 1;
98 static char *mode;
99 static char *primary;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 			       "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 			       "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 			    "in milliseconds");
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 			      "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 		       "1 for active-backup, 2 for balance-xor, "
139 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 		       "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 				   "once it comes up; "
146 				   "0 for always (default), "
147 				   "1 for only if speed of primary is "
148 				   "better, "
149 				   "2 for only on active slave "
150 				   "failure");
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 			    "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 			    "0 for stable (default), 1 for bandwidth, "
157 			    "2 for count");
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 				   "0 for layer 2 (default), 1 for layer 3+4, "
164 				   "2 for layer 2+3, 3 for encap layer 2+3, "
165 				   "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 			       "0 for none (default), 1 for active, "
173 			       "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 				"the same MAC; 0 for none (default), "
179 				"1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 				     "by setting active flag for all slaves; "
183 				     "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 			      "link failure");
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 				    "mode; 0 for a random slave, 1 packet per "
190 				    "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 			      "the bonding driver sends learning packets to "
194 			      "each slaves peer switch. The default is 1.");
195 
196 /*----------------------------- Global variables ----------------------------*/
197 
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200 #endif
201 
202 unsigned int bond_net_id __read_mostly;
203 
204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205 static int arp_ip_count;
206 static int bond_mode	= BOND_MODE_ROUNDROBIN;
207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208 static int lacp_fast;
209 
210 /*-------------------------- Forward declarations ---------------------------*/
211 
212 static int bond_init(struct net_device *bond_dev);
213 static void bond_uninit(struct net_device *bond_dev);
214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 						struct rtnl_link_stats64 *stats);
216 static void bond_slave_arr_handler(struct work_struct *work);
217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
218 				  int mod);
219 
220 /*---------------------------- General routines -----------------------------*/
221 
222 const char *bond_mode_name(int mode)
223 {
224 	static const char *names[] = {
225 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
226 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
227 		[BOND_MODE_XOR] = "load balancing (xor)",
228 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
229 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
230 		[BOND_MODE_TLB] = "transmit load balancing",
231 		[BOND_MODE_ALB] = "adaptive load balancing",
232 	};
233 
234 	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
235 		return "unknown";
236 
237 	return names[mode];
238 }
239 
240 /*---------------------------------- VLAN -----------------------------------*/
241 
242 /**
243  * bond_dev_queue_xmit - Prepare skb for xmit.
244  *
245  * @bond: bond device that got this skb for tx.
246  * @skb: hw accel VLAN tagged skb to transmit
247  * @slave_dev: slave that is supposed to xmit this skbuff
248  */
249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
250 			struct net_device *slave_dev)
251 {
252 	skb->dev = slave_dev;
253 
254 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
255 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
256 	skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
257 
258 	if (unlikely(netpoll_tx_running(bond->dev)))
259 		bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
260 	else
261 		dev_queue_xmit(skb);
262 }
263 
264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
265  * We don't protect the slave list iteration with a lock because:
266  * a. This operation is performed in IOCTL context,
267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
268  * c. Holding a lock with BH disabled while directly calling a base driver
269  *    entry point is generally a BAD idea.
270  *
271  * The design of synchronization/protection for this operation in the 8021q
272  * module is good for one or more VLAN devices over a single physical device
273  * and cannot be extended for a teaming solution like bonding, so there is a
274  * potential race condition here where a net device from the vlan group might
275  * be referenced (either by a base driver or the 8021q code) while it is being
276  * removed from the system. However, it turns out we're not making matters
277  * worse, and if it works for regular VLAN usage it will work here too.
278 */
279 
280 /**
281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
282  * @bond_dev: bonding net device that got called
283  * @vid: vlan id being added
284  */
285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
286 				__be16 proto, u16 vid)
287 {
288 	struct bonding *bond = netdev_priv(bond_dev);
289 	struct slave *slave, *rollback_slave;
290 	struct list_head *iter;
291 	int res;
292 
293 	bond_for_each_slave(bond, slave, iter) {
294 		res = vlan_vid_add(slave->dev, proto, vid);
295 		if (res)
296 			goto unwind;
297 	}
298 
299 	return 0;
300 
301 unwind:
302 	/* unwind to the slave that failed */
303 	bond_for_each_slave(bond, rollback_slave, iter) {
304 		if (rollback_slave == slave)
305 			break;
306 
307 		vlan_vid_del(rollback_slave->dev, proto, vid);
308 	}
309 
310 	return res;
311 }
312 
313 /**
314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
315  * @bond_dev: bonding net device that got called
316  * @vid: vlan id being removed
317  */
318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
319 				 __be16 proto, u16 vid)
320 {
321 	struct bonding *bond = netdev_priv(bond_dev);
322 	struct list_head *iter;
323 	struct slave *slave;
324 
325 	bond_for_each_slave(bond, slave, iter)
326 		vlan_vid_del(slave->dev, proto, vid);
327 
328 	if (bond_is_lb(bond))
329 		bond_alb_clear_vlan(bond, vid);
330 
331 	return 0;
332 }
333 
334 /*------------------------------- Link status -------------------------------*/
335 
336 /* Set the carrier state for the master according to the state of its
337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
338  * do special 802.3ad magic.
339  *
340  * Returns zero if carrier state does not change, nonzero if it does.
341  */
342 int bond_set_carrier(struct bonding *bond)
343 {
344 	struct list_head *iter;
345 	struct slave *slave;
346 
347 	if (!bond_has_slaves(bond))
348 		goto down;
349 
350 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
351 		return bond_3ad_set_carrier(bond);
352 
353 	bond_for_each_slave(bond, slave, iter) {
354 		if (slave->link == BOND_LINK_UP) {
355 			if (!netif_carrier_ok(bond->dev)) {
356 				netif_carrier_on(bond->dev);
357 				return 1;
358 			}
359 			return 0;
360 		}
361 	}
362 
363 down:
364 	if (netif_carrier_ok(bond->dev)) {
365 		netif_carrier_off(bond->dev);
366 		return 1;
367 	}
368 	return 0;
369 }
370 
371 /* Get link speed and duplex from the slave's base driver
372  * using ethtool. If for some reason the call fails or the
373  * values are invalid, set speed and duplex to -1,
374  * and return.
375  */
376 static void bond_update_speed_duplex(struct slave *slave)
377 {
378 	struct net_device *slave_dev = slave->dev;
379 	struct ethtool_link_ksettings ecmd;
380 	int res;
381 
382 	slave->speed = SPEED_UNKNOWN;
383 	slave->duplex = DUPLEX_UNKNOWN;
384 
385 	res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
386 	if (res < 0)
387 		return;
388 
389 	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
390 		return;
391 
392 	switch (ecmd.base.duplex) {
393 	case DUPLEX_FULL:
394 	case DUPLEX_HALF:
395 		break;
396 	default:
397 		return;
398 	}
399 
400 	slave->speed = ecmd.base.speed;
401 	slave->duplex = ecmd.base.duplex;
402 
403 	return;
404 }
405 
406 const char *bond_slave_link_status(s8 link)
407 {
408 	switch (link) {
409 	case BOND_LINK_UP:
410 		return "up";
411 	case BOND_LINK_FAIL:
412 		return "going down";
413 	case BOND_LINK_DOWN:
414 		return "down";
415 	case BOND_LINK_BACK:
416 		return "going back";
417 	default:
418 		return "unknown";
419 	}
420 }
421 
422 /* if <dev> supports MII link status reporting, check its link status.
423  *
424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425  * depending upon the setting of the use_carrier parameter.
426  *
427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
428  * can't tell and just pretend it is), or 0, meaning that the link is
429  * down.
430  *
431  * If reporting is non-zero, instead of faking link up, return -1 if
432  * both ETHTOOL and MII ioctls fail (meaning the device does not
433  * support them).  If use_carrier is set, return whatever it says.
434  * It'd be nice if there was a good way to tell if a driver supports
435  * netif_carrier, but there really isn't.
436  */
437 static int bond_check_dev_link(struct bonding *bond,
438 			       struct net_device *slave_dev, int reporting)
439 {
440 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 	int (*ioctl)(struct net_device *, struct ifreq *, int);
442 	struct ifreq ifr;
443 	struct mii_ioctl_data *mii;
444 
445 	if (!reporting && !netif_running(slave_dev))
446 		return 0;
447 
448 	if (bond->params.use_carrier)
449 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
450 
451 	/* Try to get link status using Ethtool first. */
452 	if (slave_dev->ethtool_ops->get_link)
453 		return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 			BMSR_LSTATUS : 0;
455 
456 	/* Ethtool can't be used, fallback to MII ioctls. */
457 	ioctl = slave_ops->ndo_do_ioctl;
458 	if (ioctl) {
459 		/* TODO: set pointer to correct ioctl on a per team member
460 		 *       bases to make this more efficient. that is, once
461 		 *       we determine the correct ioctl, we will always
462 		 *       call it and not the others for that team
463 		 *       member.
464 		 */
465 
466 		/* We cannot assume that SIOCGMIIPHY will also read a
467 		 * register; not all network drivers (e.g., e100)
468 		 * support that.
469 		 */
470 
471 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 		strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
473 		mii = if_mii(&ifr);
474 		if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 			mii->reg_num = MII_BMSR;
476 			if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 				return mii->val_out & BMSR_LSTATUS;
478 		}
479 	}
480 
481 	/* If reporting, report that either there's no dev->do_ioctl,
482 	 * or both SIOCGMIIREG and get_link failed (meaning that we
483 	 * cannot report link status).  If not reporting, pretend
484 	 * we're ok.
485 	 */
486 	return reporting ? -1 : BMSR_LSTATUS;
487 }
488 
489 /*----------------------------- Multicast list ------------------------------*/
490 
491 /* Push the promiscuity flag down to appropriate slaves */
492 static int bond_set_promiscuity(struct bonding *bond, int inc)
493 {
494 	struct list_head *iter;
495 	int err = 0;
496 
497 	if (bond_uses_primary(bond)) {
498 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
499 
500 		if (curr_active)
501 			err = dev_set_promiscuity(curr_active->dev, inc);
502 	} else {
503 		struct slave *slave;
504 
505 		bond_for_each_slave(bond, slave, iter) {
506 			err = dev_set_promiscuity(slave->dev, inc);
507 			if (err)
508 				return err;
509 		}
510 	}
511 	return err;
512 }
513 
514 /* Push the allmulti flag down to all slaves */
515 static int bond_set_allmulti(struct bonding *bond, int inc)
516 {
517 	struct list_head *iter;
518 	int err = 0;
519 
520 	if (bond_uses_primary(bond)) {
521 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
522 
523 		if (curr_active)
524 			err = dev_set_allmulti(curr_active->dev, inc);
525 	} else {
526 		struct slave *slave;
527 
528 		bond_for_each_slave(bond, slave, iter) {
529 			err = dev_set_allmulti(slave->dev, inc);
530 			if (err)
531 				return err;
532 		}
533 	}
534 	return err;
535 }
536 
537 /* Retrieve the list of registered multicast addresses for the bonding
538  * device and retransmit an IGMP JOIN request to the current active
539  * slave.
540  */
541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
542 {
543 	struct bonding *bond = container_of(work, struct bonding,
544 					    mcast_work.work);
545 
546 	if (!rtnl_trylock()) {
547 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
548 		return;
549 	}
550 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
551 
552 	if (bond->igmp_retrans > 1) {
553 		bond->igmp_retrans--;
554 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
555 	}
556 	rtnl_unlock();
557 }
558 
559 /* Flush bond's hardware addresses from slave */
560 static void bond_hw_addr_flush(struct net_device *bond_dev,
561 			       struct net_device *slave_dev)
562 {
563 	struct bonding *bond = netdev_priv(bond_dev);
564 
565 	dev_uc_unsync(slave_dev, bond_dev);
566 	dev_mc_unsync(slave_dev, bond_dev);
567 
568 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
569 		/* del lacpdu mc addr from mc list */
570 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
571 
572 		dev_mc_del(slave_dev, lacpdu_multicast);
573 	}
574 }
575 
576 /*--------------------------- Active slave change ---------------------------*/
577 
578 /* Update the hardware address list and promisc/allmulti for the new and
579  * old active slaves (if any).  Modes that are not using primary keep all
580  * slaves up date at all times; only the modes that use primary need to call
581  * this function to swap these settings during a failover.
582  */
583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 			      struct slave *old_active)
585 {
586 	if (old_active) {
587 		if (bond->dev->flags & IFF_PROMISC)
588 			dev_set_promiscuity(old_active->dev, -1);
589 
590 		if (bond->dev->flags & IFF_ALLMULTI)
591 			dev_set_allmulti(old_active->dev, -1);
592 
593 		bond_hw_addr_flush(bond->dev, old_active->dev);
594 	}
595 
596 	if (new_active) {
597 		/* FIXME: Signal errors upstream. */
598 		if (bond->dev->flags & IFF_PROMISC)
599 			dev_set_promiscuity(new_active->dev, 1);
600 
601 		if (bond->dev->flags & IFF_ALLMULTI)
602 			dev_set_allmulti(new_active->dev, 1);
603 
604 		netif_addr_lock_bh(bond->dev);
605 		dev_uc_sync(new_active->dev, bond->dev);
606 		dev_mc_sync(new_active->dev, bond->dev);
607 		netif_addr_unlock_bh(bond->dev);
608 	}
609 }
610 
611 /**
612  * bond_set_dev_addr - clone slave's address to bond
613  * @bond_dev: bond net device
614  * @slave_dev: slave net device
615  *
616  * Should be called with RTNL held.
617  */
618 static void bond_set_dev_addr(struct net_device *bond_dev,
619 			      struct net_device *slave_dev)
620 {
621 	netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
622 		   bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
623 	memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
626 }
627 
628 static struct slave *bond_get_old_active(struct bonding *bond,
629 					 struct slave *new_active)
630 {
631 	struct slave *slave;
632 	struct list_head *iter;
633 
634 	bond_for_each_slave(bond, slave, iter) {
635 		if (slave == new_active)
636 			continue;
637 
638 		if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
639 			return slave;
640 	}
641 
642 	return NULL;
643 }
644 
645 /* bond_do_fail_over_mac
646  *
647  * Perform special MAC address swapping for fail_over_mac settings
648  *
649  * Called with RTNL
650  */
651 static void bond_do_fail_over_mac(struct bonding *bond,
652 				  struct slave *new_active,
653 				  struct slave *old_active)
654 {
655 	u8 tmp_mac[ETH_ALEN];
656 	struct sockaddr saddr;
657 	int rv;
658 
659 	switch (bond->params.fail_over_mac) {
660 	case BOND_FOM_ACTIVE:
661 		if (new_active)
662 			bond_set_dev_addr(bond->dev, new_active->dev);
663 		break;
664 	case BOND_FOM_FOLLOW:
665 		/* if new_active && old_active, swap them
666 		 * if just old_active, do nothing (going to no active slave)
667 		 * if just new_active, set new_active to bond's MAC
668 		 */
669 		if (!new_active)
670 			return;
671 
672 		if (!old_active)
673 			old_active = bond_get_old_active(bond, new_active);
674 
675 		if (old_active) {
676 			ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
677 			ether_addr_copy(saddr.sa_data,
678 					old_active->dev->dev_addr);
679 			saddr.sa_family = new_active->dev->type;
680 		} else {
681 			ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
682 			saddr.sa_family = bond->dev->type;
683 		}
684 
685 		rv = dev_set_mac_address(new_active->dev, &saddr);
686 		if (rv) {
687 			netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
688 				   -rv, new_active->dev->name);
689 			goto out;
690 		}
691 
692 		if (!old_active)
693 			goto out;
694 
695 		ether_addr_copy(saddr.sa_data, tmp_mac);
696 		saddr.sa_family = old_active->dev->type;
697 
698 		rv = dev_set_mac_address(old_active->dev, &saddr);
699 		if (rv)
700 			netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 				   -rv, new_active->dev->name);
702 out:
703 		break;
704 	default:
705 		netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 			   bond->params.fail_over_mac);
707 		break;
708 	}
709 
710 }
711 
712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
713 {
714 	struct slave *prim = rtnl_dereference(bond->primary_slave);
715 	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
716 
717 	if (!prim || prim->link != BOND_LINK_UP) {
718 		if (!curr || curr->link != BOND_LINK_UP)
719 			return NULL;
720 		return curr;
721 	}
722 
723 	if (bond->force_primary) {
724 		bond->force_primary = false;
725 		return prim;
726 	}
727 
728 	if (!curr || curr->link != BOND_LINK_UP)
729 		return prim;
730 
731 	/* At this point, prim and curr are both up */
732 	switch (bond->params.primary_reselect) {
733 	case BOND_PRI_RESELECT_ALWAYS:
734 		return prim;
735 	case BOND_PRI_RESELECT_BETTER:
736 		if (prim->speed < curr->speed)
737 			return curr;
738 		if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
739 			return curr;
740 		return prim;
741 	case BOND_PRI_RESELECT_FAILURE:
742 		return curr;
743 	default:
744 		netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 			   bond->params.primary_reselect);
746 		return curr;
747 	}
748 }
749 
750 /**
751  * bond_find_best_slave - select the best available slave to be the active one
752  * @bond: our bonding struct
753  */
754 static struct slave *bond_find_best_slave(struct bonding *bond)
755 {
756 	struct slave *slave, *bestslave = NULL;
757 	struct list_head *iter;
758 	int mintime = bond->params.updelay;
759 
760 	slave = bond_choose_primary_or_current(bond);
761 	if (slave)
762 		return slave;
763 
764 	bond_for_each_slave(bond, slave, iter) {
765 		if (slave->link == BOND_LINK_UP)
766 			return slave;
767 		if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 		    slave->delay < mintime) {
769 			mintime = slave->delay;
770 			bestslave = slave;
771 		}
772 	}
773 
774 	return bestslave;
775 }
776 
777 static bool bond_should_notify_peers(struct bonding *bond)
778 {
779 	struct slave *slave;
780 
781 	rcu_read_lock();
782 	slave = rcu_dereference(bond->curr_active_slave);
783 	rcu_read_unlock();
784 
785 	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 		   slave ? slave->dev->name : "NULL");
787 
788 	if (!slave || !bond->send_peer_notif ||
789 	    !netif_carrier_ok(bond->dev) ||
790 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
791 		return false;
792 
793 	return true;
794 }
795 
796 /**
797  * change_active_interface - change the active slave into the specified one
798  * @bond: our bonding struct
799  * @new: the new slave to make the active one
800  *
801  * Set the new slave to the bond's settings and unset them on the old
802  * curr_active_slave.
803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
804  *
805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806  * because it is apparently the best available slave we have, even though its
807  * updelay hasn't timed out yet.
808  *
809  * Caller must hold RTNL.
810  */
811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
812 {
813 	struct slave *old_active;
814 
815 	ASSERT_RTNL();
816 
817 	old_active = rtnl_dereference(bond->curr_active_slave);
818 
819 	if (old_active == new_active)
820 		return;
821 
822 	if (new_active) {
823 		new_active->last_link_up = jiffies;
824 
825 		if (new_active->link == BOND_LINK_BACK) {
826 			if (bond_uses_primary(bond)) {
827 				netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 					    new_active->dev->name,
829 					    (bond->params.updelay - new_active->delay) * bond->params.miimon);
830 			}
831 
832 			new_active->delay = 0;
833 			bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 						  BOND_SLAVE_NOTIFY_NOW);
835 
836 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
838 
839 			if (bond_is_lb(bond))
840 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
841 		} else {
842 			if (bond_uses_primary(bond)) {
843 				netdev_info(bond->dev, "making interface %s the new active one\n",
844 					    new_active->dev->name);
845 			}
846 		}
847 	}
848 
849 	if (bond_uses_primary(bond))
850 		bond_hw_addr_swap(bond, new_active, old_active);
851 
852 	if (bond_is_lb(bond)) {
853 		bond_alb_handle_active_change(bond, new_active);
854 		if (old_active)
855 			bond_set_slave_inactive_flags(old_active,
856 						      BOND_SLAVE_NOTIFY_NOW);
857 		if (new_active)
858 			bond_set_slave_active_flags(new_active,
859 						    BOND_SLAVE_NOTIFY_NOW);
860 	} else {
861 		rcu_assign_pointer(bond->curr_active_slave, new_active);
862 	}
863 
864 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
865 		if (old_active)
866 			bond_set_slave_inactive_flags(old_active,
867 						      BOND_SLAVE_NOTIFY_NOW);
868 
869 		if (new_active) {
870 			bool should_notify_peers = false;
871 
872 			bond_set_slave_active_flags(new_active,
873 						    BOND_SLAVE_NOTIFY_NOW);
874 
875 			if (bond->params.fail_over_mac)
876 				bond_do_fail_over_mac(bond, new_active,
877 						      old_active);
878 
879 			if (netif_running(bond->dev)) {
880 				bond->send_peer_notif =
881 					bond->params.num_peer_notif;
882 				should_notify_peers =
883 					bond_should_notify_peers(bond);
884 			}
885 
886 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 			if (should_notify_peers)
888 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
889 							 bond->dev);
890 		}
891 	}
892 
893 	/* resend IGMP joins since active slave has changed or
894 	 * all were sent on curr_active_slave.
895 	 * resend only if bond is brought up with the affected
896 	 * bonding modes and the retransmission is enabled
897 	 */
898 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 	    ((bond_uses_primary(bond) && new_active) ||
900 	     BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 		bond->igmp_retrans = bond->params.resend_igmp;
902 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
903 	}
904 }
905 
906 /**
907  * bond_select_active_slave - select a new active slave, if needed
908  * @bond: our bonding struct
909  *
910  * This functions should be called when one of the following occurs:
911  * - The old curr_active_slave has been released or lost its link.
912  * - The primary_slave has got its link back.
913  * - A slave has got its link back and there's no old curr_active_slave.
914  *
915  * Caller must hold RTNL.
916  */
917 void bond_select_active_slave(struct bonding *bond)
918 {
919 	struct slave *best_slave;
920 	int rv;
921 
922 	ASSERT_RTNL();
923 
924 	best_slave = bond_find_best_slave(bond);
925 	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 		bond_change_active_slave(bond, best_slave);
927 		rv = bond_set_carrier(bond);
928 		if (!rv)
929 			return;
930 
931 		if (netif_carrier_ok(bond->dev))
932 			netdev_info(bond->dev, "first active interface up!\n");
933 		else
934 			netdev_info(bond->dev, "now running without any active interface!\n");
935 	}
936 }
937 
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave *slave)
940 {
941 	struct netpoll *np;
942 	int err = 0;
943 
944 	np = kzalloc(sizeof(*np), GFP_KERNEL);
945 	err = -ENOMEM;
946 	if (!np)
947 		goto out;
948 
949 	err = __netpoll_setup(np, slave->dev);
950 	if (err) {
951 		kfree(np);
952 		goto out;
953 	}
954 	slave->np = np;
955 out:
956 	return err;
957 }
958 static inline void slave_disable_netpoll(struct slave *slave)
959 {
960 	struct netpoll *np = slave->np;
961 
962 	if (!np)
963 		return;
964 
965 	slave->np = NULL;
966 	__netpoll_free_async(np);
967 }
968 
969 static void bond_poll_controller(struct net_device *bond_dev)
970 {
971 	struct bonding *bond = netdev_priv(bond_dev);
972 	struct slave *slave = NULL;
973 	struct list_head *iter;
974 	struct ad_info ad_info;
975 	struct netpoll_info *ni;
976 	const struct net_device_ops *ops;
977 
978 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
979 		if (bond_3ad_get_active_agg_info(bond, &ad_info))
980 			return;
981 
982 	bond_for_each_slave_rcu(bond, slave, iter) {
983 		ops = slave->dev->netdev_ops;
984 		if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
985 			continue;
986 
987 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
988 			struct aggregator *agg =
989 			    SLAVE_AD_INFO(slave)->port.aggregator;
990 
991 			if (agg &&
992 			    agg->aggregator_identifier != ad_info.aggregator_id)
993 				continue;
994 		}
995 
996 		ni = rcu_dereference_bh(slave->dev->npinfo);
997 		if (down_trylock(&ni->dev_lock))
998 			continue;
999 		ops->ndo_poll_controller(slave->dev);
1000 		up(&ni->dev_lock);
1001 	}
1002 }
1003 
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 {
1006 	struct bonding *bond = netdev_priv(bond_dev);
1007 	struct list_head *iter;
1008 	struct slave *slave;
1009 
1010 	bond_for_each_slave(bond, slave, iter)
1011 		if (bond_slave_is_up(slave))
1012 			slave_disable_netpoll(slave);
1013 }
1014 
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 {
1017 	struct bonding *bond = netdev_priv(dev);
1018 	struct list_head *iter;
1019 	struct slave *slave;
1020 	int err = 0;
1021 
1022 	bond_for_each_slave(bond, slave, iter) {
1023 		err = slave_enable_netpoll(slave);
1024 		if (err) {
1025 			bond_netpoll_cleanup(dev);
1026 			break;
1027 		}
1028 	}
1029 	return err;
1030 }
1031 #else
1032 static inline int slave_enable_netpoll(struct slave *slave)
1033 {
1034 	return 0;
1035 }
1036 static inline void slave_disable_netpoll(struct slave *slave)
1037 {
1038 }
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 {
1041 }
1042 #endif
1043 
1044 /*---------------------------------- IOCTL ----------------------------------*/
1045 
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047 					   netdev_features_t features)
1048 {
1049 	struct bonding *bond = netdev_priv(dev);
1050 	struct list_head *iter;
1051 	netdev_features_t mask;
1052 	struct slave *slave;
1053 
1054 	mask = features;
1055 
1056 	features &= ~NETIF_F_ONE_FOR_ALL;
1057 	features |= NETIF_F_ALL_FOR_ALL;
1058 
1059 	bond_for_each_slave(bond, slave, iter) {
1060 		features = netdev_increment_features(features,
1061 						     slave->dev->features,
1062 						     mask);
1063 	}
1064 	features = netdev_add_tso_features(features, mask);
1065 
1066 	return features;
1067 }
1068 
1069 #define BOND_VLAN_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1070 				 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1072 
1073 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1074 				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075 
1076 static void bond_compute_features(struct bonding *bond)
1077 {
1078 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079 					IFF_XMIT_DST_RELEASE_PERM;
1080 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
1082 	struct net_device *bond_dev = bond->dev;
1083 	struct list_head *iter;
1084 	struct slave *slave;
1085 	unsigned short max_hard_header_len = ETH_HLEN;
1086 	unsigned int gso_max_size = GSO_MAX_SIZE;
1087 	u16 gso_max_segs = GSO_MAX_SEGS;
1088 
1089 	if (!bond_has_slaves(bond))
1090 		goto done;
1091 	vlan_features &= NETIF_F_ALL_FOR_ALL;
1092 
1093 	bond_for_each_slave(bond, slave, iter) {
1094 		vlan_features = netdev_increment_features(vlan_features,
1095 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096 
1097 		enc_features = netdev_increment_features(enc_features,
1098 							 slave->dev->hw_enc_features,
1099 							 BOND_ENC_FEATURES);
1100 		dst_release_flag &= slave->dev->priv_flags;
1101 		if (slave->dev->hard_header_len > max_hard_header_len)
1102 			max_hard_header_len = slave->dev->hard_header_len;
1103 
1104 		gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105 		gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106 	}
1107 
1108 done:
1109 	bond_dev->vlan_features = vlan_features;
1110 	bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 	bond_dev->hard_header_len = max_hard_header_len;
1112 	bond_dev->gso_max_segs = gso_max_segs;
1113 	netif_set_gso_max_size(bond_dev, gso_max_size);
1114 
1115 	bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116 	if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117 	    dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118 		bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119 
1120 	netdev_change_features(bond_dev);
1121 }
1122 
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124 				struct net_device *slave_dev)
1125 {
1126 	bond_dev->header_ops	    = slave_dev->header_ops;
1127 
1128 	bond_dev->type		    = slave_dev->type;
1129 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1130 	bond_dev->addr_len	    = slave_dev->addr_len;
1131 
1132 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133 		slave_dev->addr_len);
1134 }
1135 
1136 /* On bonding slaves other than the currently active slave, suppress
1137  * duplicates except for alb non-mcast/bcast.
1138  */
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140 					    struct slave *slave,
1141 					    struct bonding *bond)
1142 {
1143 	if (bond_is_slave_inactive(slave)) {
1144 		if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145 		    skb->pkt_type != PACKET_BROADCAST &&
1146 		    skb->pkt_type != PACKET_MULTICAST)
1147 			return false;
1148 		return true;
1149 	}
1150 	return false;
1151 }
1152 
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 {
1155 	struct sk_buff *skb = *pskb;
1156 	struct slave *slave;
1157 	struct bonding *bond;
1158 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159 			  struct slave *);
1160 	int ret = RX_HANDLER_ANOTHER;
1161 
1162 	skb = skb_share_check(skb, GFP_ATOMIC);
1163 	if (unlikely(!skb))
1164 		return RX_HANDLER_CONSUMED;
1165 
1166 	*pskb = skb;
1167 
1168 	slave = bond_slave_get_rcu(skb->dev);
1169 	bond = slave->bond;
1170 
1171 	recv_probe = ACCESS_ONCE(bond->recv_probe);
1172 	if (recv_probe) {
1173 		ret = recv_probe(skb, bond, slave);
1174 		if (ret == RX_HANDLER_CONSUMED) {
1175 			consume_skb(skb);
1176 			return ret;
1177 		}
1178 	}
1179 
1180 	if (bond_should_deliver_exact_match(skb, slave, bond))
1181 		return RX_HANDLER_EXACT;
1182 
1183 	skb->dev = bond->dev;
1184 
1185 	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186 	    bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187 	    skb->pkt_type == PACKET_HOST) {
1188 
1189 		if (unlikely(skb_cow_head(skb,
1190 					  skb->data - skb_mac_header(skb)))) {
1191 			kfree_skb(skb);
1192 			return RX_HANDLER_CONSUMED;
1193 		}
1194 		ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1195 	}
1196 
1197 	return ret;
1198 }
1199 
1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1201 {
1202 	switch (BOND_MODE(bond)) {
1203 	case BOND_MODE_ROUNDROBIN:
1204 		return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205 	case BOND_MODE_ACTIVEBACKUP:
1206 		return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207 	case BOND_MODE_BROADCAST:
1208 		return NETDEV_LAG_TX_TYPE_BROADCAST;
1209 	case BOND_MODE_XOR:
1210 	case BOND_MODE_8023AD:
1211 		return NETDEV_LAG_TX_TYPE_HASH;
1212 	default:
1213 		return NETDEV_LAG_TX_TYPE_UNKNOWN;
1214 	}
1215 }
1216 
1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1218 {
1219 	struct netdev_lag_upper_info lag_upper_info;
1220 	int err;
1221 
1222 	lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223 	err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1224 					   &lag_upper_info);
1225 	if (err)
1226 		return err;
1227 	rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1228 	return 0;
1229 }
1230 
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232 {
1233 	netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 	slave->dev->flags &= ~IFF_SLAVE;
1235 	rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1236 }
1237 
1238 static struct slave *bond_alloc_slave(struct bonding *bond)
1239 {
1240 	struct slave *slave = NULL;
1241 
1242 	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1243 	if (!slave)
1244 		return NULL;
1245 
1246 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247 		SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1248 					       GFP_KERNEL);
1249 		if (!SLAVE_AD_INFO(slave)) {
1250 			kfree(slave);
1251 			return NULL;
1252 		}
1253 	}
1254 	return slave;
1255 }
1256 
1257 static void bond_free_slave(struct slave *slave)
1258 {
1259 	struct bonding *bond = bond_get_bond_by_slave(slave);
1260 
1261 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262 		kfree(SLAVE_AD_INFO(slave));
1263 
1264 	kfree(slave);
1265 }
1266 
1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1268 {
1269 	info->bond_mode = BOND_MODE(bond);
1270 	info->miimon = bond->params.miimon;
1271 	info->num_slaves = bond->slave_cnt;
1272 }
1273 
1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1275 {
1276 	strcpy(info->slave_name, slave->dev->name);
1277 	info->link = slave->link;
1278 	info->state = bond_slave_state(slave);
1279 	info->link_failure_count = slave->link_failure_count;
1280 }
1281 
1282 static void bond_netdev_notify(struct net_device *dev,
1283 			       struct netdev_bonding_info *info)
1284 {
1285 	rtnl_lock();
1286 	netdev_bonding_info_change(dev, info);
1287 	rtnl_unlock();
1288 }
1289 
1290 static void bond_netdev_notify_work(struct work_struct *_work)
1291 {
1292 	struct netdev_notify_work *w =
1293 		container_of(_work, struct netdev_notify_work, work.work);
1294 
1295 	bond_netdev_notify(w->dev, &w->bonding_info);
1296 	dev_put(w->dev);
1297 	kfree(w);
1298 }
1299 
1300 void bond_queue_slave_event(struct slave *slave)
1301 {
1302 	struct bonding *bond = slave->bond;
1303 	struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1304 
1305 	if (!nnw)
1306 		return;
1307 
1308 	dev_hold(slave->dev);
1309 	nnw->dev = slave->dev;
1310 	bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311 	bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312 	INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1313 
1314 	queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1315 }
1316 
1317 void bond_lower_state_changed(struct slave *slave)
1318 {
1319 	struct netdev_lag_lower_state_info info;
1320 
1321 	info.link_up = slave->link == BOND_LINK_UP ||
1322 		       slave->link == BOND_LINK_FAIL;
1323 	info.tx_enabled = bond_is_active_slave(slave);
1324 	netdev_lower_state_changed(slave->dev, &info);
1325 }
1326 
1327 /* enslave device <slave> to bond device <master> */
1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1329 {
1330 	struct bonding *bond = netdev_priv(bond_dev);
1331 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 	struct slave *new_slave = NULL, *prev_slave;
1333 	struct sockaddr addr;
1334 	int link_reporting;
1335 	int res = 0, i;
1336 
1337 	if (!bond->params.use_carrier &&
1338 	    slave_dev->ethtool_ops->get_link == NULL &&
1339 	    slave_ops->ndo_do_ioctl == NULL) {
1340 		netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341 			    slave_dev->name);
1342 	}
1343 
1344 	/* already in-use? */
1345 	if (netdev_is_rx_handler_busy(slave_dev)) {
1346 		netdev_err(bond_dev,
1347 			   "Error: Device is in use and cannot be enslaved\n");
1348 		return -EBUSY;
1349 	}
1350 
1351 	if (bond_dev == slave_dev) {
1352 		netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1353 		return -EPERM;
1354 	}
1355 
1356 	/* vlan challenged mutual exclusion */
1357 	/* no need to lock since we're protected by rtnl_lock */
1358 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1359 		netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1360 			   slave_dev->name);
1361 		if (vlan_uses_dev(bond_dev)) {
1362 			netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1363 				   slave_dev->name, bond_dev->name);
1364 			return -EPERM;
1365 		} else {
1366 			netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1367 				    slave_dev->name, slave_dev->name,
1368 				    bond_dev->name);
1369 		}
1370 	} else {
1371 		netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1372 			   slave_dev->name);
1373 	}
1374 
1375 	/* Old ifenslave binaries are no longer supported.  These can
1376 	 * be identified with moderate accuracy by the state of the slave:
1377 	 * the current ifenslave will set the interface down prior to
1378 	 * enslaving it; the old ifenslave will not.
1379 	 */
1380 	if (slave_dev->flags & IFF_UP) {
1381 		netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1382 			   slave_dev->name);
1383 		return -EPERM;
1384 	}
1385 
1386 	/* set bonding device ether type by slave - bonding netdevices are
1387 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1388 	 * there is a need to override some of the type dependent attribs/funcs.
1389 	 *
1390 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1391 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1392 	 */
1393 	if (!bond_has_slaves(bond)) {
1394 		if (bond_dev->type != slave_dev->type) {
1395 			netdev_dbg(bond_dev, "change device type from %d to %d\n",
1396 				   bond_dev->type, slave_dev->type);
1397 
1398 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1399 						       bond_dev);
1400 			res = notifier_to_errno(res);
1401 			if (res) {
1402 				netdev_err(bond_dev, "refused to change device type\n");
1403 				return -EBUSY;
1404 			}
1405 
1406 			/* Flush unicast and multicast addresses */
1407 			dev_uc_flush(bond_dev);
1408 			dev_mc_flush(bond_dev);
1409 
1410 			if (slave_dev->type != ARPHRD_ETHER)
1411 				bond_setup_by_slave(bond_dev, slave_dev);
1412 			else {
1413 				ether_setup(bond_dev);
1414 				bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1415 			}
1416 
1417 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1418 						 bond_dev);
1419 		}
1420 	} else if (bond_dev->type != slave_dev->type) {
1421 		netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1422 			   slave_dev->name, slave_dev->type, bond_dev->type);
1423 		return -EINVAL;
1424 	}
1425 
1426 	if (slave_dev->type == ARPHRD_INFINIBAND &&
1427 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1428 		netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1429 			    slave_dev->type);
1430 		res = -EOPNOTSUPP;
1431 		goto err_undo_flags;
1432 	}
1433 
1434 	if (!slave_ops->ndo_set_mac_address ||
1435 	    slave_dev->type == ARPHRD_INFINIBAND) {
1436 		netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1437 		if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1438 		    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1439 			if (!bond_has_slaves(bond)) {
1440 				bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1441 				netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1442 			} else {
1443 				netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1444 				res = -EOPNOTSUPP;
1445 				goto err_undo_flags;
1446 			}
1447 		}
1448 	}
1449 
1450 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1451 
1452 	/* If this is the first slave, then we need to set the master's hardware
1453 	 * address to be the same as the slave's.
1454 	 */
1455 	if (!bond_has_slaves(bond) &&
1456 	    bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1457 		bond_set_dev_addr(bond->dev, slave_dev);
1458 
1459 	new_slave = bond_alloc_slave(bond);
1460 	if (!new_slave) {
1461 		res = -ENOMEM;
1462 		goto err_undo_flags;
1463 	}
1464 
1465 	new_slave->bond = bond;
1466 	new_slave->dev = slave_dev;
1467 	/* Set the new_slave's queue_id to be zero.  Queue ID mapping
1468 	 * is set via sysfs or module option if desired.
1469 	 */
1470 	new_slave->queue_id = 0;
1471 
1472 	/* Save slave's original mtu and then set it to match the bond */
1473 	new_slave->original_mtu = slave_dev->mtu;
1474 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1475 	if (res) {
1476 		netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1477 		goto err_free;
1478 	}
1479 
1480 	/* Save slave's original ("permanent") mac address for modes
1481 	 * that need it, and for restoring it upon release, and then
1482 	 * set it to the master's address
1483 	 */
1484 	ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1485 
1486 	if (!bond->params.fail_over_mac ||
1487 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1488 		/* Set slave to master's mac address.  The application already
1489 		 * set the master's mac address to that of the first slave
1490 		 */
1491 		memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1492 		addr.sa_family = slave_dev->type;
1493 		res = dev_set_mac_address(slave_dev, &addr);
1494 		if (res) {
1495 			netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1496 			goto err_restore_mtu;
1497 		}
1498 	}
1499 
1500 	/* set slave flag before open to prevent IPv6 addrconf */
1501 	slave_dev->flags |= IFF_SLAVE;
1502 
1503 	/* open the slave since the application closed it */
1504 	res = dev_open(slave_dev);
1505 	if (res) {
1506 		netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1507 		goto err_restore_mac;
1508 	}
1509 
1510 	slave_dev->priv_flags |= IFF_BONDING;
1511 	/* initialize slave stats */
1512 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1513 
1514 	if (bond_is_lb(bond)) {
1515 		/* bond_alb_init_slave() must be called before all other stages since
1516 		 * it might fail and we do not want to have to undo everything
1517 		 */
1518 		res = bond_alb_init_slave(bond, new_slave);
1519 		if (res)
1520 			goto err_close;
1521 	}
1522 
1523 	/* If the mode uses primary, then the following is handled by
1524 	 * bond_change_active_slave().
1525 	 */
1526 	if (!bond_uses_primary(bond)) {
1527 		/* set promiscuity level to new slave */
1528 		if (bond_dev->flags & IFF_PROMISC) {
1529 			res = dev_set_promiscuity(slave_dev, 1);
1530 			if (res)
1531 				goto err_close;
1532 		}
1533 
1534 		/* set allmulti level to new slave */
1535 		if (bond_dev->flags & IFF_ALLMULTI) {
1536 			res = dev_set_allmulti(slave_dev, 1);
1537 			if (res)
1538 				goto err_close;
1539 		}
1540 
1541 		netif_addr_lock_bh(bond_dev);
1542 
1543 		dev_mc_sync_multiple(slave_dev, bond_dev);
1544 		dev_uc_sync_multiple(slave_dev, bond_dev);
1545 
1546 		netif_addr_unlock_bh(bond_dev);
1547 	}
1548 
1549 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1550 		/* add lacpdu mc addr to mc list */
1551 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1552 
1553 		dev_mc_add(slave_dev, lacpdu_multicast);
1554 	}
1555 
1556 	res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1557 	if (res) {
1558 		netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1559 			   slave_dev->name);
1560 		goto err_close;
1561 	}
1562 
1563 	prev_slave = bond_last_slave(bond);
1564 
1565 	new_slave->delay = 0;
1566 	new_slave->link_failure_count = 0;
1567 
1568 	bond_update_speed_duplex(new_slave);
1569 
1570 	new_slave->last_rx = jiffies -
1571 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
1572 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1573 		new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1574 
1575 	if (bond->params.miimon && !bond->params.use_carrier) {
1576 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1577 
1578 		if ((link_reporting == -1) && !bond->params.arp_interval) {
1579 			/* miimon is set but a bonded network driver
1580 			 * does not support ETHTOOL/MII and
1581 			 * arp_interval is not set.  Note: if
1582 			 * use_carrier is enabled, we will never go
1583 			 * here (because netif_carrier is always
1584 			 * supported); thus, we don't need to change
1585 			 * the messages for netif_carrier.
1586 			 */
1587 			netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1588 				    slave_dev->name);
1589 		} else if (link_reporting == -1) {
1590 			/* unable get link status using mii/ethtool */
1591 			netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1592 				    slave_dev->name);
1593 		}
1594 	}
1595 
1596 	/* check for initial state */
1597 	new_slave->link = BOND_LINK_NOCHANGE;
1598 	if (bond->params.miimon) {
1599 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1600 			if (bond->params.updelay) {
1601 				bond_set_slave_link_state(new_slave,
1602 							  BOND_LINK_BACK,
1603 							  BOND_SLAVE_NOTIFY_NOW);
1604 				new_slave->delay = bond->params.updelay;
1605 			} else {
1606 				bond_set_slave_link_state(new_slave,
1607 							  BOND_LINK_UP,
1608 							  BOND_SLAVE_NOTIFY_NOW);
1609 			}
1610 		} else {
1611 			bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1612 						  BOND_SLAVE_NOTIFY_NOW);
1613 		}
1614 	} else if (bond->params.arp_interval) {
1615 		bond_set_slave_link_state(new_slave,
1616 					  (netif_carrier_ok(slave_dev) ?
1617 					  BOND_LINK_UP : BOND_LINK_DOWN),
1618 					  BOND_SLAVE_NOTIFY_NOW);
1619 	} else {
1620 		bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1621 					  BOND_SLAVE_NOTIFY_NOW);
1622 	}
1623 
1624 	if (new_slave->link != BOND_LINK_DOWN)
1625 		new_slave->last_link_up = jiffies;
1626 	netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1627 		   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1628 		   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1629 
1630 	if (bond_uses_primary(bond) && bond->params.primary[0]) {
1631 		/* if there is a primary slave, remember it */
1632 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1633 			rcu_assign_pointer(bond->primary_slave, new_slave);
1634 			bond->force_primary = true;
1635 		}
1636 	}
1637 
1638 	switch (BOND_MODE(bond)) {
1639 	case BOND_MODE_ACTIVEBACKUP:
1640 		bond_set_slave_inactive_flags(new_slave,
1641 					      BOND_SLAVE_NOTIFY_NOW);
1642 		break;
1643 	case BOND_MODE_8023AD:
1644 		/* in 802.3ad mode, the internal mechanism
1645 		 * will activate the slaves in the selected
1646 		 * aggregator
1647 		 */
1648 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1649 		/* if this is the first slave */
1650 		if (!prev_slave) {
1651 			SLAVE_AD_INFO(new_slave)->id = 1;
1652 			/* Initialize AD with the number of times that the AD timer is called in 1 second
1653 			 * can be called only after the mac address of the bond is set
1654 			 */
1655 			bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1656 		} else {
1657 			SLAVE_AD_INFO(new_slave)->id =
1658 				SLAVE_AD_INFO(prev_slave)->id + 1;
1659 		}
1660 
1661 		bond_3ad_bind_slave(new_slave);
1662 		break;
1663 	case BOND_MODE_TLB:
1664 	case BOND_MODE_ALB:
1665 		bond_set_active_slave(new_slave);
1666 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1667 		break;
1668 	default:
1669 		netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1670 
1671 		/* always active in trunk mode */
1672 		bond_set_active_slave(new_slave);
1673 
1674 		/* In trunking mode there is little meaning to curr_active_slave
1675 		 * anyway (it holds no special properties of the bond device),
1676 		 * so we can change it without calling change_active_interface()
1677 		 */
1678 		if (!rcu_access_pointer(bond->curr_active_slave) &&
1679 		    new_slave->link == BOND_LINK_UP)
1680 			rcu_assign_pointer(bond->curr_active_slave, new_slave);
1681 
1682 		break;
1683 	} /* switch(bond_mode) */
1684 
1685 #ifdef CONFIG_NET_POLL_CONTROLLER
1686 	slave_dev->npinfo = bond->dev->npinfo;
1687 	if (slave_dev->npinfo) {
1688 		if (slave_enable_netpoll(new_slave)) {
1689 			netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1690 			res = -EBUSY;
1691 			goto err_detach;
1692 		}
1693 	}
1694 #endif
1695 
1696 	if (!(bond_dev->features & NETIF_F_LRO))
1697 		dev_disable_lro(slave_dev);
1698 
1699 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1700 					 new_slave);
1701 	if (res) {
1702 		netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1703 		goto err_detach;
1704 	}
1705 
1706 	res = bond_master_upper_dev_link(bond, new_slave);
1707 	if (res) {
1708 		netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1709 		goto err_unregister;
1710 	}
1711 
1712 	res = bond_sysfs_slave_add(new_slave);
1713 	if (res) {
1714 		netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1715 		goto err_upper_unlink;
1716 	}
1717 
1718 	bond->slave_cnt++;
1719 	bond_compute_features(bond);
1720 	bond_set_carrier(bond);
1721 
1722 	if (bond_uses_primary(bond)) {
1723 		block_netpoll_tx();
1724 		bond_select_active_slave(bond);
1725 		unblock_netpoll_tx();
1726 	}
1727 
1728 	if (bond_mode_uses_xmit_hash(bond))
1729 		bond_update_slave_arr(bond, NULL);
1730 
1731 	netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1732 		    slave_dev->name,
1733 		    bond_is_active_slave(new_slave) ? "an active" : "a backup",
1734 		    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1735 
1736 	/* enslave is successful */
1737 	bond_queue_slave_event(new_slave);
1738 	return 0;
1739 
1740 /* Undo stages on error */
1741 err_upper_unlink:
1742 	bond_upper_dev_unlink(bond, new_slave);
1743 
1744 err_unregister:
1745 	netdev_rx_handler_unregister(slave_dev);
1746 
1747 err_detach:
1748 	if (!bond_uses_primary(bond))
1749 		bond_hw_addr_flush(bond_dev, slave_dev);
1750 
1751 	vlan_vids_del_by_dev(slave_dev, bond_dev);
1752 	if (rcu_access_pointer(bond->primary_slave) == new_slave)
1753 		RCU_INIT_POINTER(bond->primary_slave, NULL);
1754 	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1755 		block_netpoll_tx();
1756 		bond_change_active_slave(bond, NULL);
1757 		bond_select_active_slave(bond);
1758 		unblock_netpoll_tx();
1759 	}
1760 	/* either primary_slave or curr_active_slave might've changed */
1761 	synchronize_rcu();
1762 	slave_disable_netpoll(new_slave);
1763 
1764 err_close:
1765 	slave_dev->priv_flags &= ~IFF_BONDING;
1766 	dev_close(slave_dev);
1767 
1768 err_restore_mac:
1769 	slave_dev->flags &= ~IFF_SLAVE;
1770 	if (!bond->params.fail_over_mac ||
1771 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1772 		/* XXX TODO - fom follow mode needs to change master's
1773 		 * MAC if this slave's MAC is in use by the bond, or at
1774 		 * least print a warning.
1775 		 */
1776 		ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1777 		addr.sa_family = slave_dev->type;
1778 		dev_set_mac_address(slave_dev, &addr);
1779 	}
1780 
1781 err_restore_mtu:
1782 	dev_set_mtu(slave_dev, new_slave->original_mtu);
1783 
1784 err_free:
1785 	bond_free_slave(new_slave);
1786 
1787 err_undo_flags:
1788 	/* Enslave of first slave has failed and we need to fix master's mac */
1789 	if (!bond_has_slaves(bond)) {
1790 		if (ether_addr_equal_64bits(bond_dev->dev_addr,
1791 					    slave_dev->dev_addr))
1792 			eth_hw_addr_random(bond_dev);
1793 		if (bond_dev->type != ARPHRD_ETHER) {
1794 			dev_close(bond_dev);
1795 			ether_setup(bond_dev);
1796 			bond_dev->flags |= IFF_MASTER;
1797 			bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1798 		}
1799 	}
1800 
1801 	return res;
1802 }
1803 
1804 /* Try to release the slave device <slave> from the bond device <master>
1805  * It is legal to access curr_active_slave without a lock because all the function
1806  * is RTNL-locked. If "all" is true it means that the function is being called
1807  * while destroying a bond interface and all slaves are being released.
1808  *
1809  * The rules for slave state should be:
1810  *   for Active/Backup:
1811  *     Active stays on all backups go down
1812  *   for Bonded connections:
1813  *     The first up interface should be left on and all others downed.
1814  */
1815 static int __bond_release_one(struct net_device *bond_dev,
1816 			      struct net_device *slave_dev,
1817 			      bool all)
1818 {
1819 	struct bonding *bond = netdev_priv(bond_dev);
1820 	struct slave *slave, *oldcurrent;
1821 	struct sockaddr addr;
1822 	int old_flags = bond_dev->flags;
1823 	netdev_features_t old_features = bond_dev->features;
1824 
1825 	/* slave is not a slave or master is not master of this slave */
1826 	if (!(slave_dev->flags & IFF_SLAVE) ||
1827 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
1828 		netdev_dbg(bond_dev, "cannot release %s\n",
1829 			   slave_dev->name);
1830 		return -EINVAL;
1831 	}
1832 
1833 	block_netpoll_tx();
1834 
1835 	slave = bond_get_slave_by_dev(bond, slave_dev);
1836 	if (!slave) {
1837 		/* not a slave of this bond */
1838 		netdev_info(bond_dev, "%s not enslaved\n",
1839 			    slave_dev->name);
1840 		unblock_netpoll_tx();
1841 		return -EINVAL;
1842 	}
1843 
1844 	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1845 
1846 	bond_sysfs_slave_del(slave);
1847 
1848 	/* recompute stats just before removing the slave */
1849 	bond_get_stats(bond->dev, &bond->bond_stats);
1850 
1851 	bond_upper_dev_unlink(bond, slave);
1852 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
1853 	 * for this slave anymore.
1854 	 */
1855 	netdev_rx_handler_unregister(slave_dev);
1856 
1857 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1858 		bond_3ad_unbind_slave(slave);
1859 
1860 	if (bond_mode_uses_xmit_hash(bond))
1861 		bond_update_slave_arr(bond, slave);
1862 
1863 	netdev_info(bond_dev, "Releasing %s interface %s\n",
1864 		    bond_is_active_slave(slave) ? "active" : "backup",
1865 		    slave_dev->name);
1866 
1867 	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1868 
1869 	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1870 
1871 	if (!all && (!bond->params.fail_over_mac ||
1872 		     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1873 		if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1874 		    bond_has_slaves(bond))
1875 			netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1876 				    slave_dev->name, slave->perm_hwaddr,
1877 				    bond_dev->name, slave_dev->name);
1878 	}
1879 
1880 	if (rtnl_dereference(bond->primary_slave) == slave)
1881 		RCU_INIT_POINTER(bond->primary_slave, NULL);
1882 
1883 	if (oldcurrent == slave)
1884 		bond_change_active_slave(bond, NULL);
1885 
1886 	if (bond_is_lb(bond)) {
1887 		/* Must be called only after the slave has been
1888 		 * detached from the list and the curr_active_slave
1889 		 * has been cleared (if our_slave == old_current),
1890 		 * but before a new active slave is selected.
1891 		 */
1892 		bond_alb_deinit_slave(bond, slave);
1893 	}
1894 
1895 	if (all) {
1896 		RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1897 	} else if (oldcurrent == slave) {
1898 		/* Note that we hold RTNL over this sequence, so there
1899 		 * is no concern that another slave add/remove event
1900 		 * will interfere.
1901 		 */
1902 		bond_select_active_slave(bond);
1903 	}
1904 
1905 	if (!bond_has_slaves(bond)) {
1906 		bond_set_carrier(bond);
1907 		eth_hw_addr_random(bond_dev);
1908 	}
1909 
1910 	unblock_netpoll_tx();
1911 	synchronize_rcu();
1912 	bond->slave_cnt--;
1913 
1914 	if (!bond_has_slaves(bond)) {
1915 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1916 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1917 	}
1918 
1919 	bond_compute_features(bond);
1920 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1921 	    (old_features & NETIF_F_VLAN_CHALLENGED))
1922 		netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1923 			    slave_dev->name, bond_dev->name);
1924 
1925 	vlan_vids_del_by_dev(slave_dev, bond_dev);
1926 
1927 	/* If the mode uses primary, then this case was handled above by
1928 	 * bond_change_active_slave(..., NULL)
1929 	 */
1930 	if (!bond_uses_primary(bond)) {
1931 		/* unset promiscuity level from slave
1932 		 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1933 		 * of the IFF_PROMISC flag in the bond_dev, but we need the
1934 		 * value of that flag before that change, as that was the value
1935 		 * when this slave was attached, so we cache at the start of the
1936 		 * function and use it here. Same goes for ALLMULTI below
1937 		 */
1938 		if (old_flags & IFF_PROMISC)
1939 			dev_set_promiscuity(slave_dev, -1);
1940 
1941 		/* unset allmulti level from slave */
1942 		if (old_flags & IFF_ALLMULTI)
1943 			dev_set_allmulti(slave_dev, -1);
1944 
1945 		bond_hw_addr_flush(bond_dev, slave_dev);
1946 	}
1947 
1948 	slave_disable_netpoll(slave);
1949 
1950 	/* close slave before restoring its mac address */
1951 	dev_close(slave_dev);
1952 
1953 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1954 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1955 		/* restore original ("permanent") mac address */
1956 		ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1957 		addr.sa_family = slave_dev->type;
1958 		dev_set_mac_address(slave_dev, &addr);
1959 	}
1960 
1961 	dev_set_mtu(slave_dev, slave->original_mtu);
1962 
1963 	slave_dev->priv_flags &= ~IFF_BONDING;
1964 
1965 	bond_free_slave(slave);
1966 
1967 	return 0;
1968 }
1969 
1970 /* A wrapper used because of ndo_del_link */
1971 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1972 {
1973 	return __bond_release_one(bond_dev, slave_dev, false);
1974 }
1975 
1976 /* First release a slave and then destroy the bond if no more slaves are left.
1977  * Must be under rtnl_lock when this function is called.
1978  */
1979 static int  bond_release_and_destroy(struct net_device *bond_dev,
1980 				     struct net_device *slave_dev)
1981 {
1982 	struct bonding *bond = netdev_priv(bond_dev);
1983 	int ret;
1984 
1985 	ret = bond_release(bond_dev, slave_dev);
1986 	if (ret == 0 && !bond_has_slaves(bond)) {
1987 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1988 		netdev_info(bond_dev, "Destroying bond %s\n",
1989 			    bond_dev->name);
1990 		bond_remove_proc_entry(bond);
1991 		unregister_netdevice(bond_dev);
1992 	}
1993 	return ret;
1994 }
1995 
1996 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1997 {
1998 	struct bonding *bond = netdev_priv(bond_dev);
1999 	bond_fill_ifbond(bond, info);
2000 	return 0;
2001 }
2002 
2003 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2004 {
2005 	struct bonding *bond = netdev_priv(bond_dev);
2006 	struct list_head *iter;
2007 	int i = 0, res = -ENODEV;
2008 	struct slave *slave;
2009 
2010 	bond_for_each_slave(bond, slave, iter) {
2011 		if (i++ == (int)info->slave_id) {
2012 			res = 0;
2013 			bond_fill_ifslave(slave, info);
2014 			break;
2015 		}
2016 	}
2017 
2018 	return res;
2019 }
2020 
2021 /*-------------------------------- Monitoring -------------------------------*/
2022 
2023 /* called with rcu_read_lock() */
2024 static int bond_miimon_inspect(struct bonding *bond)
2025 {
2026 	int link_state, commit = 0;
2027 	struct list_head *iter;
2028 	struct slave *slave;
2029 	bool ignore_updelay;
2030 
2031 	ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2032 
2033 	bond_for_each_slave_rcu(bond, slave, iter) {
2034 		slave->new_link = BOND_LINK_NOCHANGE;
2035 
2036 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2037 
2038 		switch (slave->link) {
2039 		case BOND_LINK_UP:
2040 			if (link_state)
2041 				continue;
2042 
2043 			bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2044 						  BOND_SLAVE_NOTIFY_LATER);
2045 			slave->delay = bond->params.downdelay;
2046 			if (slave->delay) {
2047 				netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2048 					    (BOND_MODE(bond) ==
2049 					     BOND_MODE_ACTIVEBACKUP) ?
2050 					     (bond_is_active_slave(slave) ?
2051 					      "active " : "backup ") : "",
2052 					    slave->dev->name,
2053 					    bond->params.downdelay * bond->params.miimon);
2054 			}
2055 			/*FALLTHRU*/
2056 		case BOND_LINK_FAIL:
2057 			if (link_state) {
2058 				/* recovered before downdelay expired */
2059 				bond_set_slave_link_state(slave, BOND_LINK_UP,
2060 							  BOND_SLAVE_NOTIFY_LATER);
2061 				slave->last_link_up = jiffies;
2062 				netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2063 					    (bond->params.downdelay - slave->delay) *
2064 					    bond->params.miimon,
2065 					    slave->dev->name);
2066 				continue;
2067 			}
2068 
2069 			if (slave->delay <= 0) {
2070 				slave->new_link = BOND_LINK_DOWN;
2071 				commit++;
2072 				continue;
2073 			}
2074 
2075 			slave->delay--;
2076 			break;
2077 
2078 		case BOND_LINK_DOWN:
2079 			if (!link_state)
2080 				continue;
2081 
2082 			bond_set_slave_link_state(slave, BOND_LINK_BACK,
2083 						  BOND_SLAVE_NOTIFY_LATER);
2084 			slave->delay = bond->params.updelay;
2085 
2086 			if (slave->delay) {
2087 				netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2088 					    slave->dev->name,
2089 					    ignore_updelay ? 0 :
2090 					    bond->params.updelay *
2091 					    bond->params.miimon);
2092 			}
2093 			/*FALLTHRU*/
2094 		case BOND_LINK_BACK:
2095 			if (!link_state) {
2096 				bond_set_slave_link_state(slave,
2097 							  BOND_LINK_DOWN,
2098 							  BOND_SLAVE_NOTIFY_LATER);
2099 				netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2100 					    (bond->params.updelay - slave->delay) *
2101 					    bond->params.miimon,
2102 					    slave->dev->name);
2103 
2104 				continue;
2105 			}
2106 
2107 			if (ignore_updelay)
2108 				slave->delay = 0;
2109 
2110 			if (slave->delay <= 0) {
2111 				slave->new_link = BOND_LINK_UP;
2112 				commit++;
2113 				ignore_updelay = false;
2114 				continue;
2115 			}
2116 
2117 			slave->delay--;
2118 			break;
2119 		}
2120 	}
2121 
2122 	return commit;
2123 }
2124 
2125 static void bond_miimon_commit(struct bonding *bond)
2126 {
2127 	struct list_head *iter;
2128 	struct slave *slave, *primary;
2129 
2130 	bond_for_each_slave(bond, slave, iter) {
2131 		switch (slave->new_link) {
2132 		case BOND_LINK_NOCHANGE:
2133 			continue;
2134 
2135 		case BOND_LINK_UP:
2136 			bond_update_speed_duplex(slave);
2137 			bond_set_slave_link_state(slave, BOND_LINK_UP,
2138 						  BOND_SLAVE_NOTIFY_NOW);
2139 			slave->last_link_up = jiffies;
2140 
2141 			primary = rtnl_dereference(bond->primary_slave);
2142 			if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2143 				/* prevent it from being the active one */
2144 				bond_set_backup_slave(slave);
2145 			} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2146 				/* make it immediately active */
2147 				bond_set_active_slave(slave);
2148 			} else if (slave != primary) {
2149 				/* prevent it from being the active one */
2150 				bond_set_backup_slave(slave);
2151 			}
2152 
2153 			netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2154 				    slave->dev->name,
2155 				    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2156 				    slave->duplex ? "full" : "half");
2157 
2158 			/* notify ad that the link status has changed */
2159 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
2160 				bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2161 
2162 			if (bond_is_lb(bond))
2163 				bond_alb_handle_link_change(bond, slave,
2164 							    BOND_LINK_UP);
2165 
2166 			if (BOND_MODE(bond) == BOND_MODE_XOR)
2167 				bond_update_slave_arr(bond, NULL);
2168 
2169 			if (!bond->curr_active_slave || slave == primary)
2170 				goto do_failover;
2171 
2172 			continue;
2173 
2174 		case BOND_LINK_DOWN:
2175 			if (slave->link_failure_count < UINT_MAX)
2176 				slave->link_failure_count++;
2177 
2178 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2179 						  BOND_SLAVE_NOTIFY_NOW);
2180 
2181 			if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2182 			    BOND_MODE(bond) == BOND_MODE_8023AD)
2183 				bond_set_slave_inactive_flags(slave,
2184 							      BOND_SLAVE_NOTIFY_NOW);
2185 
2186 			netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2187 				    slave->dev->name);
2188 
2189 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
2190 				bond_3ad_handle_link_change(slave,
2191 							    BOND_LINK_DOWN);
2192 
2193 			if (bond_is_lb(bond))
2194 				bond_alb_handle_link_change(bond, slave,
2195 							    BOND_LINK_DOWN);
2196 
2197 			if (BOND_MODE(bond) == BOND_MODE_XOR)
2198 				bond_update_slave_arr(bond, NULL);
2199 
2200 			if (slave == rcu_access_pointer(bond->curr_active_slave))
2201 				goto do_failover;
2202 
2203 			continue;
2204 
2205 		default:
2206 			netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2207 				   slave->new_link, slave->dev->name);
2208 			slave->new_link = BOND_LINK_NOCHANGE;
2209 
2210 			continue;
2211 		}
2212 
2213 do_failover:
2214 		block_netpoll_tx();
2215 		bond_select_active_slave(bond);
2216 		unblock_netpoll_tx();
2217 	}
2218 
2219 	bond_set_carrier(bond);
2220 }
2221 
2222 /* bond_mii_monitor
2223  *
2224  * Really a wrapper that splits the mii monitor into two phases: an
2225  * inspection, then (if inspection indicates something needs to be done)
2226  * an acquisition of appropriate locks followed by a commit phase to
2227  * implement whatever link state changes are indicated.
2228  */
2229 static void bond_mii_monitor(struct work_struct *work)
2230 {
2231 	struct bonding *bond = container_of(work, struct bonding,
2232 					    mii_work.work);
2233 	bool should_notify_peers = false;
2234 	unsigned long delay;
2235 
2236 	delay = msecs_to_jiffies(bond->params.miimon);
2237 
2238 	if (!bond_has_slaves(bond))
2239 		goto re_arm;
2240 
2241 	rcu_read_lock();
2242 
2243 	should_notify_peers = bond_should_notify_peers(bond);
2244 
2245 	if (bond_miimon_inspect(bond)) {
2246 		rcu_read_unlock();
2247 
2248 		/* Race avoidance with bond_close cancel of workqueue */
2249 		if (!rtnl_trylock()) {
2250 			delay = 1;
2251 			should_notify_peers = false;
2252 			goto re_arm;
2253 		}
2254 
2255 		bond_miimon_commit(bond);
2256 
2257 		rtnl_unlock();	/* might sleep, hold no other locks */
2258 	} else
2259 		rcu_read_unlock();
2260 
2261 re_arm:
2262 	if (bond->params.miimon)
2263 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2264 
2265 	if (should_notify_peers) {
2266 		if (!rtnl_trylock())
2267 			return;
2268 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2269 		rtnl_unlock();
2270 	}
2271 }
2272 
2273 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2274 {
2275 	__be32 ip = *((__be32 *)data);
2276 
2277 	return ip == bond_confirm_addr(upper, 0, ip);
2278 }
2279 
2280 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2281 {
2282 	bool ret = false;
2283 
2284 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2285 		return true;
2286 
2287 	rcu_read_lock();
2288 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2289 		ret = true;
2290 	rcu_read_unlock();
2291 
2292 	return ret;
2293 }
2294 
2295 /* We go to the (large) trouble of VLAN tagging ARP frames because
2296  * switches in VLAN mode (especially if ports are configured as
2297  * "native" to a VLAN) might not pass non-tagged frames.
2298  */
2299 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2300 			  __be32 dest_ip, __be32 src_ip,
2301 			  struct bond_vlan_tag *tags)
2302 {
2303 	struct sk_buff *skb;
2304 	struct bond_vlan_tag *outer_tag = tags;
2305 
2306 	netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2307 		   arp_op, slave_dev->name, &dest_ip, &src_ip);
2308 
2309 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2310 			 NULL, slave_dev->dev_addr, NULL);
2311 
2312 	if (!skb) {
2313 		net_err_ratelimited("ARP packet allocation failed\n");
2314 		return;
2315 	}
2316 
2317 	if (!tags || tags->vlan_proto == VLAN_N_VID)
2318 		goto xmit;
2319 
2320 	tags++;
2321 
2322 	/* Go through all the tags backwards and add them to the packet */
2323 	while (tags->vlan_proto != VLAN_N_VID) {
2324 		if (!tags->vlan_id) {
2325 			tags++;
2326 			continue;
2327 		}
2328 
2329 		netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2330 			   ntohs(outer_tag->vlan_proto), tags->vlan_id);
2331 		skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2332 						tags->vlan_id);
2333 		if (!skb) {
2334 			net_err_ratelimited("failed to insert inner VLAN tag\n");
2335 			return;
2336 		}
2337 
2338 		tags++;
2339 	}
2340 	/* Set the outer tag */
2341 	if (outer_tag->vlan_id) {
2342 		netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2343 			   ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2344 		__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2345 				       outer_tag->vlan_id);
2346 	}
2347 
2348 xmit:
2349 	arp_xmit(skb);
2350 }
2351 
2352 /* Validate the device path between the @start_dev and the @end_dev.
2353  * The path is valid if the @end_dev is reachable through device
2354  * stacking.
2355  * When the path is validated, collect any vlan information in the
2356  * path.
2357  */
2358 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2359 					      struct net_device *end_dev,
2360 					      int level)
2361 {
2362 	struct bond_vlan_tag *tags;
2363 	struct net_device *upper;
2364 	struct list_head  *iter;
2365 
2366 	if (start_dev == end_dev) {
2367 		tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2368 		if (!tags)
2369 			return ERR_PTR(-ENOMEM);
2370 		tags[level].vlan_proto = VLAN_N_VID;
2371 		return tags;
2372 	}
2373 
2374 	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2375 		tags = bond_verify_device_path(upper, end_dev, level + 1);
2376 		if (IS_ERR_OR_NULL(tags)) {
2377 			if (IS_ERR(tags))
2378 				return tags;
2379 			continue;
2380 		}
2381 		if (is_vlan_dev(upper)) {
2382 			tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2383 			tags[level].vlan_id = vlan_dev_vlan_id(upper);
2384 		}
2385 
2386 		return tags;
2387 	}
2388 
2389 	return NULL;
2390 }
2391 
2392 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2393 {
2394 	struct rtable *rt;
2395 	struct bond_vlan_tag *tags;
2396 	__be32 *targets = bond->params.arp_targets, addr;
2397 	int i;
2398 
2399 	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2400 		netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2401 		tags = NULL;
2402 
2403 		/* Find out through which dev should the packet go */
2404 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2405 				     RTO_ONLINK, 0);
2406 		if (IS_ERR(rt)) {
2407 			/* there's no route to target - try to send arp
2408 			 * probe to generate any traffic (arp_validate=0)
2409 			 */
2410 			if (bond->params.arp_validate)
2411 				net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2412 						     bond->dev->name,
2413 						     &targets[i]);
2414 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2415 				      0, tags);
2416 			continue;
2417 		}
2418 
2419 		/* bond device itself */
2420 		if (rt->dst.dev == bond->dev)
2421 			goto found;
2422 
2423 		rcu_read_lock();
2424 		tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2425 		rcu_read_unlock();
2426 
2427 		if (!IS_ERR_OR_NULL(tags))
2428 			goto found;
2429 
2430 		/* Not our device - skip */
2431 		netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2432 			   &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2433 
2434 		ip_rt_put(rt);
2435 		continue;
2436 
2437 found:
2438 		addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2439 		ip_rt_put(rt);
2440 		bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2441 			      addr, tags);
2442 		kfree(tags);
2443 	}
2444 }
2445 
2446 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2447 {
2448 	int i;
2449 
2450 	if (!sip || !bond_has_this_ip(bond, tip)) {
2451 		netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2452 			   &sip, &tip);
2453 		return;
2454 	}
2455 
2456 	i = bond_get_targets_ip(bond->params.arp_targets, sip);
2457 	if (i == -1) {
2458 		netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2459 			   &sip);
2460 		return;
2461 	}
2462 	slave->last_rx = jiffies;
2463 	slave->target_last_arp_rx[i] = jiffies;
2464 }
2465 
2466 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2467 		 struct slave *slave)
2468 {
2469 	struct arphdr *arp = (struct arphdr *)skb->data;
2470 	struct slave *curr_active_slave, *curr_arp_slave;
2471 	unsigned char *arp_ptr;
2472 	__be32 sip, tip;
2473 	int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2474 
2475 	if (!slave_do_arp_validate(bond, slave)) {
2476 		if ((slave_do_arp_validate_only(bond) && is_arp) ||
2477 		    !slave_do_arp_validate_only(bond))
2478 			slave->last_rx = jiffies;
2479 		return RX_HANDLER_ANOTHER;
2480 	} else if (!is_arp) {
2481 		return RX_HANDLER_ANOTHER;
2482 	}
2483 
2484 	alen = arp_hdr_len(bond->dev);
2485 
2486 	netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2487 		   skb->dev->name);
2488 
2489 	if (alen > skb_headlen(skb)) {
2490 		arp = kmalloc(alen, GFP_ATOMIC);
2491 		if (!arp)
2492 			goto out_unlock;
2493 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
2494 			goto out_unlock;
2495 	}
2496 
2497 	if (arp->ar_hln != bond->dev->addr_len ||
2498 	    skb->pkt_type == PACKET_OTHERHOST ||
2499 	    skb->pkt_type == PACKET_LOOPBACK ||
2500 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
2501 	    arp->ar_pro != htons(ETH_P_IP) ||
2502 	    arp->ar_pln != 4)
2503 		goto out_unlock;
2504 
2505 	arp_ptr = (unsigned char *)(arp + 1);
2506 	arp_ptr += bond->dev->addr_len;
2507 	memcpy(&sip, arp_ptr, 4);
2508 	arp_ptr += 4 + bond->dev->addr_len;
2509 	memcpy(&tip, arp_ptr, 4);
2510 
2511 	netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2512 		   slave->dev->name, bond_slave_state(slave),
2513 		     bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2514 		     &sip, &tip);
2515 
2516 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
2517 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2518 
2519 	/* We 'trust' the received ARP enough to validate it if:
2520 	 *
2521 	 * (a) the slave receiving the ARP is active (which includes the
2522 	 * current ARP slave, if any), or
2523 	 *
2524 	 * (b) the receiving slave isn't active, but there is a currently
2525 	 * active slave and it received valid arp reply(s) after it became
2526 	 * the currently active slave, or
2527 	 *
2528 	 * (c) there is an ARP slave that sent an ARP during the prior ARP
2529 	 * interval, and we receive an ARP reply on any slave.  We accept
2530 	 * these because switch FDB update delays may deliver the ARP
2531 	 * reply to a slave other than the sender of the ARP request.
2532 	 *
2533 	 * Note: for (b), backup slaves are receiving the broadcast ARP
2534 	 * request, not a reply.  This request passes from the sending
2535 	 * slave through the L2 switch(es) to the receiving slave.  Since
2536 	 * this is checking the request, sip/tip are swapped for
2537 	 * validation.
2538 	 *
2539 	 * This is done to avoid endless looping when we can't reach the
2540 	 * arp_ip_target and fool ourselves with our own arp requests.
2541 	 */
2542 	if (bond_is_active_slave(slave))
2543 		bond_validate_arp(bond, slave, sip, tip);
2544 	else if (curr_active_slave &&
2545 		 time_after(slave_last_rx(bond, curr_active_slave),
2546 			    curr_active_slave->last_link_up))
2547 		bond_validate_arp(bond, slave, tip, sip);
2548 	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2549 		 bond_time_in_interval(bond,
2550 				       dev_trans_start(curr_arp_slave->dev), 1))
2551 		bond_validate_arp(bond, slave, sip, tip);
2552 
2553 out_unlock:
2554 	if (arp != (struct arphdr *)skb->data)
2555 		kfree(arp);
2556 	return RX_HANDLER_ANOTHER;
2557 }
2558 
2559 /* function to verify if we're in the arp_interval timeslice, returns true if
2560  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2561  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2562  */
2563 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2564 				  int mod)
2565 {
2566 	int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2567 
2568 	return time_in_range(jiffies,
2569 			     last_act - delta_in_ticks,
2570 			     last_act + mod * delta_in_ticks + delta_in_ticks/2);
2571 }
2572 
2573 /* This function is called regularly to monitor each slave's link
2574  * ensuring that traffic is being sent and received when arp monitoring
2575  * is used in load-balancing mode. if the adapter has been dormant, then an
2576  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2577  * arp monitoring in active backup mode.
2578  */
2579 static void bond_loadbalance_arp_mon(struct work_struct *work)
2580 {
2581 	struct bonding *bond = container_of(work, struct bonding,
2582 					    arp_work.work);
2583 	struct slave *slave, *oldcurrent;
2584 	struct list_head *iter;
2585 	int do_failover = 0, slave_state_changed = 0;
2586 
2587 	if (!bond_has_slaves(bond))
2588 		goto re_arm;
2589 
2590 	rcu_read_lock();
2591 
2592 	oldcurrent = rcu_dereference(bond->curr_active_slave);
2593 	/* see if any of the previous devices are up now (i.e. they have
2594 	 * xmt and rcv traffic). the curr_active_slave does not come into
2595 	 * the picture unless it is null. also, slave->last_link_up is not
2596 	 * needed here because we send an arp on each slave and give a slave
2597 	 * as long as it needs to get the tx/rx within the delta.
2598 	 * TODO: what about up/down delay in arp mode? it wasn't here before
2599 	 *       so it can wait
2600 	 */
2601 	bond_for_each_slave_rcu(bond, slave, iter) {
2602 		unsigned long trans_start = dev_trans_start(slave->dev);
2603 
2604 		if (slave->link != BOND_LINK_UP) {
2605 			if (bond_time_in_interval(bond, trans_start, 1) &&
2606 			    bond_time_in_interval(bond, slave->last_rx, 1)) {
2607 
2608 				slave->link  = BOND_LINK_UP;
2609 				slave_state_changed = 1;
2610 
2611 				/* primary_slave has no meaning in round-robin
2612 				 * mode. the window of a slave being up and
2613 				 * curr_active_slave being null after enslaving
2614 				 * is closed.
2615 				 */
2616 				if (!oldcurrent) {
2617 					netdev_info(bond->dev, "link status definitely up for interface %s\n",
2618 						    slave->dev->name);
2619 					do_failover = 1;
2620 				} else {
2621 					netdev_info(bond->dev, "interface %s is now up\n",
2622 						    slave->dev->name);
2623 				}
2624 			}
2625 		} else {
2626 			/* slave->link == BOND_LINK_UP */
2627 
2628 			/* not all switches will respond to an arp request
2629 			 * when the source ip is 0, so don't take the link down
2630 			 * if we don't know our ip yet
2631 			 */
2632 			if (!bond_time_in_interval(bond, trans_start, 2) ||
2633 			    !bond_time_in_interval(bond, slave->last_rx, 2)) {
2634 
2635 				slave->link  = BOND_LINK_DOWN;
2636 				slave_state_changed = 1;
2637 
2638 				if (slave->link_failure_count < UINT_MAX)
2639 					slave->link_failure_count++;
2640 
2641 				netdev_info(bond->dev, "interface %s is now down\n",
2642 					    slave->dev->name);
2643 
2644 				if (slave == oldcurrent)
2645 					do_failover = 1;
2646 			}
2647 		}
2648 
2649 		/* note: if switch is in round-robin mode, all links
2650 		 * must tx arp to ensure all links rx an arp - otherwise
2651 		 * links may oscillate or not come up at all; if switch is
2652 		 * in something like xor mode, there is nothing we can
2653 		 * do - all replies will be rx'ed on same link causing slaves
2654 		 * to be unstable during low/no traffic periods
2655 		 */
2656 		if (bond_slave_is_up(slave))
2657 			bond_arp_send_all(bond, slave);
2658 	}
2659 
2660 	rcu_read_unlock();
2661 
2662 	if (do_failover || slave_state_changed) {
2663 		if (!rtnl_trylock())
2664 			goto re_arm;
2665 
2666 		if (slave_state_changed) {
2667 			bond_slave_state_change(bond);
2668 			if (BOND_MODE(bond) == BOND_MODE_XOR)
2669 				bond_update_slave_arr(bond, NULL);
2670 		}
2671 		if (do_failover) {
2672 			block_netpoll_tx();
2673 			bond_select_active_slave(bond);
2674 			unblock_netpoll_tx();
2675 		}
2676 		rtnl_unlock();
2677 	}
2678 
2679 re_arm:
2680 	if (bond->params.arp_interval)
2681 		queue_delayed_work(bond->wq, &bond->arp_work,
2682 				   msecs_to_jiffies(bond->params.arp_interval));
2683 }
2684 
2685 /* Called to inspect slaves for active-backup mode ARP monitor link state
2686  * changes.  Sets new_link in slaves to specify what action should take
2687  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2688  * to link states must be committed.
2689  *
2690  * Called with rcu_read_lock held.
2691  */
2692 static int bond_ab_arp_inspect(struct bonding *bond)
2693 {
2694 	unsigned long trans_start, last_rx;
2695 	struct list_head *iter;
2696 	struct slave *slave;
2697 	int commit = 0;
2698 
2699 	bond_for_each_slave_rcu(bond, slave, iter) {
2700 		slave->new_link = BOND_LINK_NOCHANGE;
2701 		last_rx = slave_last_rx(bond, slave);
2702 
2703 		if (slave->link != BOND_LINK_UP) {
2704 			if (bond_time_in_interval(bond, last_rx, 1)) {
2705 				slave->new_link = BOND_LINK_UP;
2706 				commit++;
2707 			}
2708 			continue;
2709 		}
2710 
2711 		/* Give slaves 2*delta after being enslaved or made
2712 		 * active.  This avoids bouncing, as the last receive
2713 		 * times need a full ARP monitor cycle to be updated.
2714 		 */
2715 		if (bond_time_in_interval(bond, slave->last_link_up, 2))
2716 			continue;
2717 
2718 		/* Backup slave is down if:
2719 		 * - No current_arp_slave AND
2720 		 * - more than 3*delta since last receive AND
2721 		 * - the bond has an IP address
2722 		 *
2723 		 * Note: a non-null current_arp_slave indicates
2724 		 * the curr_active_slave went down and we are
2725 		 * searching for a new one; under this condition
2726 		 * we only take the curr_active_slave down - this
2727 		 * gives each slave a chance to tx/rx traffic
2728 		 * before being taken out
2729 		 */
2730 		if (!bond_is_active_slave(slave) &&
2731 		    !rcu_access_pointer(bond->current_arp_slave) &&
2732 		    !bond_time_in_interval(bond, last_rx, 3)) {
2733 			slave->new_link = BOND_LINK_DOWN;
2734 			commit++;
2735 		}
2736 
2737 		/* Active slave is down if:
2738 		 * - more than 2*delta since transmitting OR
2739 		 * - (more than 2*delta since receive AND
2740 		 *    the bond has an IP address)
2741 		 */
2742 		trans_start = dev_trans_start(slave->dev);
2743 		if (bond_is_active_slave(slave) &&
2744 		    (!bond_time_in_interval(bond, trans_start, 2) ||
2745 		     !bond_time_in_interval(bond, last_rx, 2))) {
2746 			slave->new_link = BOND_LINK_DOWN;
2747 			commit++;
2748 		}
2749 	}
2750 
2751 	return commit;
2752 }
2753 
2754 /* Called to commit link state changes noted by inspection step of
2755  * active-backup mode ARP monitor.
2756  *
2757  * Called with RTNL hold.
2758  */
2759 static void bond_ab_arp_commit(struct bonding *bond)
2760 {
2761 	unsigned long trans_start;
2762 	struct list_head *iter;
2763 	struct slave *slave;
2764 
2765 	bond_for_each_slave(bond, slave, iter) {
2766 		switch (slave->new_link) {
2767 		case BOND_LINK_NOCHANGE:
2768 			continue;
2769 
2770 		case BOND_LINK_UP:
2771 			trans_start = dev_trans_start(slave->dev);
2772 			if (rtnl_dereference(bond->curr_active_slave) != slave ||
2773 			    (!rtnl_dereference(bond->curr_active_slave) &&
2774 			     bond_time_in_interval(bond, trans_start, 1))) {
2775 				struct slave *current_arp_slave;
2776 
2777 				current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2778 				bond_set_slave_link_state(slave, BOND_LINK_UP,
2779 							  BOND_SLAVE_NOTIFY_NOW);
2780 				if (current_arp_slave) {
2781 					bond_set_slave_inactive_flags(
2782 						current_arp_slave,
2783 						BOND_SLAVE_NOTIFY_NOW);
2784 					RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2785 				}
2786 
2787 				netdev_info(bond->dev, "link status definitely up for interface %s\n",
2788 					    slave->dev->name);
2789 
2790 				if (!rtnl_dereference(bond->curr_active_slave) ||
2791 				    slave == rtnl_dereference(bond->primary_slave))
2792 					goto do_failover;
2793 
2794 			}
2795 
2796 			continue;
2797 
2798 		case BOND_LINK_DOWN:
2799 			if (slave->link_failure_count < UINT_MAX)
2800 				slave->link_failure_count++;
2801 
2802 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2803 						  BOND_SLAVE_NOTIFY_NOW);
2804 			bond_set_slave_inactive_flags(slave,
2805 						      BOND_SLAVE_NOTIFY_NOW);
2806 
2807 			netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2808 				    slave->dev->name);
2809 
2810 			if (slave == rtnl_dereference(bond->curr_active_slave)) {
2811 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2812 				goto do_failover;
2813 			}
2814 
2815 			continue;
2816 
2817 		default:
2818 			netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2819 				   slave->new_link, slave->dev->name);
2820 			continue;
2821 		}
2822 
2823 do_failover:
2824 		block_netpoll_tx();
2825 		bond_select_active_slave(bond);
2826 		unblock_netpoll_tx();
2827 	}
2828 
2829 	bond_set_carrier(bond);
2830 }
2831 
2832 /* Send ARP probes for active-backup mode ARP monitor.
2833  *
2834  * Called with rcu_read_lock held.
2835  */
2836 static bool bond_ab_arp_probe(struct bonding *bond)
2837 {
2838 	struct slave *slave, *before = NULL, *new_slave = NULL,
2839 		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2840 		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2841 	struct list_head *iter;
2842 	bool found = false;
2843 	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2844 
2845 	if (curr_arp_slave && curr_active_slave)
2846 		netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2847 			    curr_arp_slave->dev->name,
2848 			    curr_active_slave->dev->name);
2849 
2850 	if (curr_active_slave) {
2851 		bond_arp_send_all(bond, curr_active_slave);
2852 		return should_notify_rtnl;
2853 	}
2854 
2855 	/* if we don't have a curr_active_slave, search for the next available
2856 	 * backup slave from the current_arp_slave and make it the candidate
2857 	 * for becoming the curr_active_slave
2858 	 */
2859 
2860 	if (!curr_arp_slave) {
2861 		curr_arp_slave = bond_first_slave_rcu(bond);
2862 		if (!curr_arp_slave)
2863 			return should_notify_rtnl;
2864 	}
2865 
2866 	bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2867 
2868 	bond_for_each_slave_rcu(bond, slave, iter) {
2869 		if (!found && !before && bond_slave_is_up(slave))
2870 			before = slave;
2871 
2872 		if (found && !new_slave && bond_slave_is_up(slave))
2873 			new_slave = slave;
2874 		/* if the link state is up at this point, we
2875 		 * mark it down - this can happen if we have
2876 		 * simultaneous link failures and
2877 		 * reselect_active_interface doesn't make this
2878 		 * one the current slave so it is still marked
2879 		 * up when it is actually down
2880 		 */
2881 		if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2882 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2883 						  BOND_SLAVE_NOTIFY_LATER);
2884 			if (slave->link_failure_count < UINT_MAX)
2885 				slave->link_failure_count++;
2886 
2887 			bond_set_slave_inactive_flags(slave,
2888 						      BOND_SLAVE_NOTIFY_LATER);
2889 
2890 			netdev_info(bond->dev, "backup interface %s is now down\n",
2891 				    slave->dev->name);
2892 		}
2893 		if (slave == curr_arp_slave)
2894 			found = true;
2895 	}
2896 
2897 	if (!new_slave && before)
2898 		new_slave = before;
2899 
2900 	if (!new_slave)
2901 		goto check_state;
2902 
2903 	bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2904 				  BOND_SLAVE_NOTIFY_LATER);
2905 	bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2906 	bond_arp_send_all(bond, new_slave);
2907 	new_slave->last_link_up = jiffies;
2908 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
2909 
2910 check_state:
2911 	bond_for_each_slave_rcu(bond, slave, iter) {
2912 		if (slave->should_notify || slave->should_notify_link) {
2913 			should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2914 			break;
2915 		}
2916 	}
2917 	return should_notify_rtnl;
2918 }
2919 
2920 static void bond_activebackup_arp_mon(struct work_struct *work)
2921 {
2922 	struct bonding *bond = container_of(work, struct bonding,
2923 					    arp_work.work);
2924 	bool should_notify_peers = false;
2925 	bool should_notify_rtnl = false;
2926 	int delta_in_ticks;
2927 
2928 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2929 
2930 	if (!bond_has_slaves(bond))
2931 		goto re_arm;
2932 
2933 	rcu_read_lock();
2934 
2935 	should_notify_peers = bond_should_notify_peers(bond);
2936 
2937 	if (bond_ab_arp_inspect(bond)) {
2938 		rcu_read_unlock();
2939 
2940 		/* Race avoidance with bond_close flush of workqueue */
2941 		if (!rtnl_trylock()) {
2942 			delta_in_ticks = 1;
2943 			should_notify_peers = false;
2944 			goto re_arm;
2945 		}
2946 
2947 		bond_ab_arp_commit(bond);
2948 
2949 		rtnl_unlock();
2950 		rcu_read_lock();
2951 	}
2952 
2953 	should_notify_rtnl = bond_ab_arp_probe(bond);
2954 	rcu_read_unlock();
2955 
2956 re_arm:
2957 	if (bond->params.arp_interval)
2958 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2959 
2960 	if (should_notify_peers || should_notify_rtnl) {
2961 		if (!rtnl_trylock())
2962 			return;
2963 
2964 		if (should_notify_peers)
2965 			call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2966 						 bond->dev);
2967 		if (should_notify_rtnl) {
2968 			bond_slave_state_notify(bond);
2969 			bond_slave_link_notify(bond);
2970 		}
2971 
2972 		rtnl_unlock();
2973 	}
2974 }
2975 
2976 /*-------------------------- netdev event handling --------------------------*/
2977 
2978 /* Change device name */
2979 static int bond_event_changename(struct bonding *bond)
2980 {
2981 	bond_remove_proc_entry(bond);
2982 	bond_create_proc_entry(bond);
2983 
2984 	bond_debug_reregister(bond);
2985 
2986 	return NOTIFY_DONE;
2987 }
2988 
2989 static int bond_master_netdev_event(unsigned long event,
2990 				    struct net_device *bond_dev)
2991 {
2992 	struct bonding *event_bond = netdev_priv(bond_dev);
2993 
2994 	switch (event) {
2995 	case NETDEV_CHANGENAME:
2996 		return bond_event_changename(event_bond);
2997 	case NETDEV_UNREGISTER:
2998 		bond_remove_proc_entry(event_bond);
2999 		break;
3000 	case NETDEV_REGISTER:
3001 		bond_create_proc_entry(event_bond);
3002 		break;
3003 	case NETDEV_NOTIFY_PEERS:
3004 		if (event_bond->send_peer_notif)
3005 			event_bond->send_peer_notif--;
3006 		break;
3007 	default:
3008 		break;
3009 	}
3010 
3011 	return NOTIFY_DONE;
3012 }
3013 
3014 static int bond_slave_netdev_event(unsigned long event,
3015 				   struct net_device *slave_dev)
3016 {
3017 	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3018 	struct bonding *bond;
3019 	struct net_device *bond_dev;
3020 
3021 	/* A netdev event can be generated while enslaving a device
3022 	 * before netdev_rx_handler_register is called in which case
3023 	 * slave will be NULL
3024 	 */
3025 	if (!slave)
3026 		return NOTIFY_DONE;
3027 	bond_dev = slave->bond->dev;
3028 	bond = slave->bond;
3029 	primary = rtnl_dereference(bond->primary_slave);
3030 
3031 	switch (event) {
3032 	case NETDEV_UNREGISTER:
3033 		if (bond_dev->type != ARPHRD_ETHER)
3034 			bond_release_and_destroy(bond_dev, slave_dev);
3035 		else
3036 			bond_release(bond_dev, slave_dev);
3037 		break;
3038 	case NETDEV_UP:
3039 	case NETDEV_CHANGE:
3040 		bond_update_speed_duplex(slave);
3041 		if (BOND_MODE(bond) == BOND_MODE_8023AD)
3042 			bond_3ad_adapter_speed_duplex_changed(slave);
3043 		/* Fallthrough */
3044 	case NETDEV_DOWN:
3045 		/* Refresh slave-array if applicable!
3046 		 * If the setup does not use miimon or arpmon (mode-specific!),
3047 		 * then these events will not cause the slave-array to be
3048 		 * refreshed. This will cause xmit to use a slave that is not
3049 		 * usable. Avoid such situation by refeshing the array at these
3050 		 * events. If these (miimon/arpmon) parameters are configured
3051 		 * then array gets refreshed twice and that should be fine!
3052 		 */
3053 		if (bond_mode_uses_xmit_hash(bond))
3054 			bond_update_slave_arr(bond, NULL);
3055 		break;
3056 	case NETDEV_CHANGEMTU:
3057 		/* TODO: Should slaves be allowed to
3058 		 * independently alter their MTU?  For
3059 		 * an active-backup bond, slaves need
3060 		 * not be the same type of device, so
3061 		 * MTUs may vary.  For other modes,
3062 		 * slaves arguably should have the
3063 		 * same MTUs. To do this, we'd need to
3064 		 * take over the slave's change_mtu
3065 		 * function for the duration of their
3066 		 * servitude.
3067 		 */
3068 		break;
3069 	case NETDEV_CHANGENAME:
3070 		/* we don't care if we don't have primary set */
3071 		if (!bond_uses_primary(bond) ||
3072 		    !bond->params.primary[0])
3073 			break;
3074 
3075 		if (slave == primary) {
3076 			/* slave's name changed - he's no longer primary */
3077 			RCU_INIT_POINTER(bond->primary_slave, NULL);
3078 		} else if (!strcmp(slave_dev->name, bond->params.primary)) {
3079 			/* we have a new primary slave */
3080 			rcu_assign_pointer(bond->primary_slave, slave);
3081 		} else { /* we didn't change primary - exit */
3082 			break;
3083 		}
3084 
3085 		netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3086 			    primary ? slave_dev->name : "none");
3087 
3088 		block_netpoll_tx();
3089 		bond_select_active_slave(bond);
3090 		unblock_netpoll_tx();
3091 		break;
3092 	case NETDEV_FEAT_CHANGE:
3093 		bond_compute_features(bond);
3094 		break;
3095 	case NETDEV_RESEND_IGMP:
3096 		/* Propagate to master device */
3097 		call_netdevice_notifiers(event, slave->bond->dev);
3098 		break;
3099 	default:
3100 		break;
3101 	}
3102 
3103 	return NOTIFY_DONE;
3104 }
3105 
3106 /* bond_netdev_event: handle netdev notifier chain events.
3107  *
3108  * This function receives events for the netdev chain.  The caller (an
3109  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3110  * locks for us to safely manipulate the slave devices (RTNL lock,
3111  * dev_probe_lock).
3112  */
3113 static int bond_netdev_event(struct notifier_block *this,
3114 			     unsigned long event, void *ptr)
3115 {
3116 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3117 
3118 	netdev_dbg(event_dev, "event: %lx\n", event);
3119 
3120 	if (!(event_dev->priv_flags & IFF_BONDING))
3121 		return NOTIFY_DONE;
3122 
3123 	if (event_dev->flags & IFF_MASTER) {
3124 		netdev_dbg(event_dev, "IFF_MASTER\n");
3125 		return bond_master_netdev_event(event, event_dev);
3126 	}
3127 
3128 	if (event_dev->flags & IFF_SLAVE) {
3129 		netdev_dbg(event_dev, "IFF_SLAVE\n");
3130 		return bond_slave_netdev_event(event, event_dev);
3131 	}
3132 
3133 	return NOTIFY_DONE;
3134 }
3135 
3136 static struct notifier_block bond_netdev_notifier = {
3137 	.notifier_call = bond_netdev_event,
3138 };
3139 
3140 /*---------------------------- Hashing Policies -----------------------------*/
3141 
3142 /* L2 hash helper */
3143 static inline u32 bond_eth_hash(struct sk_buff *skb)
3144 {
3145 	struct ethhdr *ep, hdr_tmp;
3146 
3147 	ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3148 	if (ep)
3149 		return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3150 	return 0;
3151 }
3152 
3153 /* Extract the appropriate headers based on bond's xmit policy */
3154 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3155 			      struct flow_keys *fk)
3156 {
3157 	const struct ipv6hdr *iph6;
3158 	const struct iphdr *iph;
3159 	int noff, proto = -1;
3160 
3161 	if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3162 		return skb_flow_dissect_flow_keys(skb, fk, 0);
3163 
3164 	fk->ports.ports = 0;
3165 	noff = skb_network_offset(skb);
3166 	if (skb->protocol == htons(ETH_P_IP)) {
3167 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3168 			return false;
3169 		iph = ip_hdr(skb);
3170 		iph_to_flow_copy_v4addrs(fk, iph);
3171 		noff += iph->ihl << 2;
3172 		if (!ip_is_fragment(iph))
3173 			proto = iph->protocol;
3174 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
3175 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3176 			return false;
3177 		iph6 = ipv6_hdr(skb);
3178 		iph_to_flow_copy_v6addrs(fk, iph6);
3179 		noff += sizeof(*iph6);
3180 		proto = iph6->nexthdr;
3181 	} else {
3182 		return false;
3183 	}
3184 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3185 		fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3186 
3187 	return true;
3188 }
3189 
3190 /**
3191  * bond_xmit_hash - generate a hash value based on the xmit policy
3192  * @bond: bonding device
3193  * @skb: buffer to use for headers
3194  *
3195  * This function will extract the necessary headers from the skb buffer and use
3196  * them to generate a hash based on the xmit_policy set in the bonding device
3197  */
3198 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3199 {
3200 	struct flow_keys flow;
3201 	u32 hash;
3202 
3203 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3204 	    skb->l4_hash)
3205 		return skb->hash;
3206 
3207 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3208 	    !bond_flow_dissect(bond, skb, &flow))
3209 		return bond_eth_hash(skb);
3210 
3211 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3212 	    bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3213 		hash = bond_eth_hash(skb);
3214 	else
3215 		hash = (__force u32)flow.ports.ports;
3216 	hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3217 		(__force u32)flow_get_u32_src(&flow);
3218 	hash ^= (hash >> 16);
3219 	hash ^= (hash >> 8);
3220 
3221 	return hash;
3222 }
3223 
3224 /*-------------------------- Device entry points ----------------------------*/
3225 
3226 static void bond_work_init_all(struct bonding *bond)
3227 {
3228 	INIT_DELAYED_WORK(&bond->mcast_work,
3229 			  bond_resend_igmp_join_requests_delayed);
3230 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3231 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3232 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3233 		INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3234 	else
3235 		INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3236 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3237 	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3238 }
3239 
3240 static void bond_work_cancel_all(struct bonding *bond)
3241 {
3242 	cancel_delayed_work_sync(&bond->mii_work);
3243 	cancel_delayed_work_sync(&bond->arp_work);
3244 	cancel_delayed_work_sync(&bond->alb_work);
3245 	cancel_delayed_work_sync(&bond->ad_work);
3246 	cancel_delayed_work_sync(&bond->mcast_work);
3247 	cancel_delayed_work_sync(&bond->slave_arr_work);
3248 }
3249 
3250 static int bond_open(struct net_device *bond_dev)
3251 {
3252 	struct bonding *bond = netdev_priv(bond_dev);
3253 	struct list_head *iter;
3254 	struct slave *slave;
3255 
3256 	/* reset slave->backup and slave->inactive */
3257 	if (bond_has_slaves(bond)) {
3258 		bond_for_each_slave(bond, slave, iter) {
3259 			if (bond_uses_primary(bond) &&
3260 			    slave != rcu_access_pointer(bond->curr_active_slave)) {
3261 				bond_set_slave_inactive_flags(slave,
3262 							      BOND_SLAVE_NOTIFY_NOW);
3263 			} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3264 				bond_set_slave_active_flags(slave,
3265 							    BOND_SLAVE_NOTIFY_NOW);
3266 			}
3267 		}
3268 	}
3269 
3270 	bond_work_init_all(bond);
3271 
3272 	if (bond_is_lb(bond)) {
3273 		/* bond_alb_initialize must be called before the timer
3274 		 * is started.
3275 		 */
3276 		if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3277 			return -ENOMEM;
3278 		if (bond->params.tlb_dynamic_lb)
3279 			queue_delayed_work(bond->wq, &bond->alb_work, 0);
3280 	}
3281 
3282 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
3283 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
3284 
3285 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3286 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
3287 		bond->recv_probe = bond_arp_rcv;
3288 	}
3289 
3290 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3291 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
3292 		/* register to receive LACPDUs */
3293 		bond->recv_probe = bond_3ad_lacpdu_recv;
3294 		bond_3ad_initiate_agg_selection(bond, 1);
3295 	}
3296 
3297 	if (bond_mode_uses_xmit_hash(bond))
3298 		bond_update_slave_arr(bond, NULL);
3299 
3300 	return 0;
3301 }
3302 
3303 static int bond_close(struct net_device *bond_dev)
3304 {
3305 	struct bonding *bond = netdev_priv(bond_dev);
3306 
3307 	bond_work_cancel_all(bond);
3308 	bond->send_peer_notif = 0;
3309 	if (bond_is_lb(bond))
3310 		bond_alb_deinitialize(bond);
3311 	bond->recv_probe = NULL;
3312 
3313 	return 0;
3314 }
3315 
3316 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3317  * that some drivers can provide 32bit values only.
3318  */
3319 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3320 			    const struct rtnl_link_stats64 *_new,
3321 			    const struct rtnl_link_stats64 *_old)
3322 {
3323 	const u64 *new = (const u64 *)_new;
3324 	const u64 *old = (const u64 *)_old;
3325 	u64 *res = (u64 *)_res;
3326 	int i;
3327 
3328 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3329 		u64 nv = new[i];
3330 		u64 ov = old[i];
3331 
3332 		/* detects if this particular field is 32bit only */
3333 		if (((nv | ov) >> 32) == 0)
3334 			res[i] += (u32)nv - (u32)ov;
3335 		else
3336 			res[i] += nv - ov;
3337 	}
3338 }
3339 
3340 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3341 						struct rtnl_link_stats64 *stats)
3342 {
3343 	struct bonding *bond = netdev_priv(bond_dev);
3344 	struct rtnl_link_stats64 temp;
3345 	struct list_head *iter;
3346 	struct slave *slave;
3347 
3348 	spin_lock(&bond->stats_lock);
3349 	memcpy(stats, &bond->bond_stats, sizeof(*stats));
3350 
3351 	rcu_read_lock();
3352 	bond_for_each_slave_rcu(bond, slave, iter) {
3353 		const struct rtnl_link_stats64 *new =
3354 			dev_get_stats(slave->dev, &temp);
3355 
3356 		bond_fold_stats(stats, new, &slave->slave_stats);
3357 
3358 		/* save off the slave stats for the next run */
3359 		memcpy(&slave->slave_stats, new, sizeof(*new));
3360 	}
3361 	rcu_read_unlock();
3362 
3363 	memcpy(&bond->bond_stats, stats, sizeof(*stats));
3364 	spin_unlock(&bond->stats_lock);
3365 
3366 	return stats;
3367 }
3368 
3369 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3370 {
3371 	struct bonding *bond = netdev_priv(bond_dev);
3372 	struct net_device *slave_dev = NULL;
3373 	struct ifbond k_binfo;
3374 	struct ifbond __user *u_binfo = NULL;
3375 	struct ifslave k_sinfo;
3376 	struct ifslave __user *u_sinfo = NULL;
3377 	struct mii_ioctl_data *mii = NULL;
3378 	struct bond_opt_value newval;
3379 	struct net *net;
3380 	int res = 0;
3381 
3382 	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3383 
3384 	switch (cmd) {
3385 	case SIOCGMIIPHY:
3386 		mii = if_mii(ifr);
3387 		if (!mii)
3388 			return -EINVAL;
3389 
3390 		mii->phy_id = 0;
3391 		/* Fall Through */
3392 	case SIOCGMIIREG:
3393 		/* We do this again just in case we were called by SIOCGMIIREG
3394 		 * instead of SIOCGMIIPHY.
3395 		 */
3396 		mii = if_mii(ifr);
3397 		if (!mii)
3398 			return -EINVAL;
3399 
3400 		if (mii->reg_num == 1) {
3401 			mii->val_out = 0;
3402 			if (netif_carrier_ok(bond->dev))
3403 				mii->val_out = BMSR_LSTATUS;
3404 		}
3405 
3406 		return 0;
3407 	case BOND_INFO_QUERY_OLD:
3408 	case SIOCBONDINFOQUERY:
3409 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
3410 
3411 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3412 			return -EFAULT;
3413 
3414 		res = bond_info_query(bond_dev, &k_binfo);
3415 		if (res == 0 &&
3416 		    copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3417 			return -EFAULT;
3418 
3419 		return res;
3420 	case BOND_SLAVE_INFO_QUERY_OLD:
3421 	case SIOCBONDSLAVEINFOQUERY:
3422 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3423 
3424 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3425 			return -EFAULT;
3426 
3427 		res = bond_slave_info_query(bond_dev, &k_sinfo);
3428 		if (res == 0 &&
3429 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3430 			return -EFAULT;
3431 
3432 		return res;
3433 	default:
3434 		break;
3435 	}
3436 
3437 	net = dev_net(bond_dev);
3438 
3439 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3440 		return -EPERM;
3441 
3442 	slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3443 
3444 	netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3445 
3446 	if (!slave_dev)
3447 		return -ENODEV;
3448 
3449 	netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3450 	switch (cmd) {
3451 	case BOND_ENSLAVE_OLD:
3452 	case SIOCBONDENSLAVE:
3453 		res = bond_enslave(bond_dev, slave_dev);
3454 		break;
3455 	case BOND_RELEASE_OLD:
3456 	case SIOCBONDRELEASE:
3457 		res = bond_release(bond_dev, slave_dev);
3458 		break;
3459 	case BOND_SETHWADDR_OLD:
3460 	case SIOCBONDSETHWADDR:
3461 		bond_set_dev_addr(bond_dev, slave_dev);
3462 		res = 0;
3463 		break;
3464 	case BOND_CHANGE_ACTIVE_OLD:
3465 	case SIOCBONDCHANGEACTIVE:
3466 		bond_opt_initstr(&newval, slave_dev->name);
3467 		res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3468 		break;
3469 	default:
3470 		res = -EOPNOTSUPP;
3471 	}
3472 
3473 	return res;
3474 }
3475 
3476 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3477 {
3478 	struct bonding *bond = netdev_priv(bond_dev);
3479 
3480 	if (change & IFF_PROMISC)
3481 		bond_set_promiscuity(bond,
3482 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
3483 
3484 	if (change & IFF_ALLMULTI)
3485 		bond_set_allmulti(bond,
3486 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3487 }
3488 
3489 static void bond_set_rx_mode(struct net_device *bond_dev)
3490 {
3491 	struct bonding *bond = netdev_priv(bond_dev);
3492 	struct list_head *iter;
3493 	struct slave *slave;
3494 
3495 	rcu_read_lock();
3496 	if (bond_uses_primary(bond)) {
3497 		slave = rcu_dereference(bond->curr_active_slave);
3498 		if (slave) {
3499 			dev_uc_sync(slave->dev, bond_dev);
3500 			dev_mc_sync(slave->dev, bond_dev);
3501 		}
3502 	} else {
3503 		bond_for_each_slave_rcu(bond, slave, iter) {
3504 			dev_uc_sync_multiple(slave->dev, bond_dev);
3505 			dev_mc_sync_multiple(slave->dev, bond_dev);
3506 		}
3507 	}
3508 	rcu_read_unlock();
3509 }
3510 
3511 static int bond_neigh_init(struct neighbour *n)
3512 {
3513 	struct bonding *bond = netdev_priv(n->dev);
3514 	const struct net_device_ops *slave_ops;
3515 	struct neigh_parms parms;
3516 	struct slave *slave;
3517 	int ret;
3518 
3519 	slave = bond_first_slave(bond);
3520 	if (!slave)
3521 		return 0;
3522 	slave_ops = slave->dev->netdev_ops;
3523 	if (!slave_ops->ndo_neigh_setup)
3524 		return 0;
3525 
3526 	parms.neigh_setup = NULL;
3527 	parms.neigh_cleanup = NULL;
3528 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3529 	if (ret)
3530 		return ret;
3531 
3532 	/* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3533 	 * after the last slave has been detached.  Assumes that all slaves
3534 	 * utilize the same neigh_cleanup (true at this writing as only user
3535 	 * is ipoib).
3536 	 */
3537 	n->parms->neigh_cleanup = parms.neigh_cleanup;
3538 
3539 	if (!parms.neigh_setup)
3540 		return 0;
3541 
3542 	return parms.neigh_setup(n);
3543 }
3544 
3545 /* The bonding ndo_neigh_setup is called at init time beofre any
3546  * slave exists. So we must declare proxy setup function which will
3547  * be used at run time to resolve the actual slave neigh param setup.
3548  *
3549  * It's also called by master devices (such as vlans) to setup their
3550  * underlying devices. In that case - do nothing, we're already set up from
3551  * our init.
3552  */
3553 static int bond_neigh_setup(struct net_device *dev,
3554 			    struct neigh_parms *parms)
3555 {
3556 	/* modify only our neigh_parms */
3557 	if (parms->dev == dev)
3558 		parms->neigh_setup = bond_neigh_init;
3559 
3560 	return 0;
3561 }
3562 
3563 /* Change the MTU of all of a master's slaves to match the master */
3564 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3565 {
3566 	struct bonding *bond = netdev_priv(bond_dev);
3567 	struct slave *slave, *rollback_slave;
3568 	struct list_head *iter;
3569 	int res = 0;
3570 
3571 	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3572 
3573 	bond_for_each_slave(bond, slave, iter) {
3574 		netdev_dbg(bond_dev, "s %p c_m %p\n",
3575 			   slave, slave->dev->netdev_ops->ndo_change_mtu);
3576 
3577 		res = dev_set_mtu(slave->dev, new_mtu);
3578 
3579 		if (res) {
3580 			/* If we failed to set the slave's mtu to the new value
3581 			 * we must abort the operation even in ACTIVE_BACKUP
3582 			 * mode, because if we allow the backup slaves to have
3583 			 * different mtu values than the active slave we'll
3584 			 * need to change their mtu when doing a failover. That
3585 			 * means changing their mtu from timer context, which
3586 			 * is probably not a good idea.
3587 			 */
3588 			netdev_dbg(bond_dev, "err %d %s\n", res,
3589 				   slave->dev->name);
3590 			goto unwind;
3591 		}
3592 	}
3593 
3594 	bond_dev->mtu = new_mtu;
3595 
3596 	return 0;
3597 
3598 unwind:
3599 	/* unwind from head to the slave that failed */
3600 	bond_for_each_slave(bond, rollback_slave, iter) {
3601 		int tmp_res;
3602 
3603 		if (rollback_slave == slave)
3604 			break;
3605 
3606 		tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3607 		if (tmp_res) {
3608 			netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3609 				   tmp_res, rollback_slave->dev->name);
3610 		}
3611 	}
3612 
3613 	return res;
3614 }
3615 
3616 /* Change HW address
3617  *
3618  * Note that many devices must be down to change the HW address, and
3619  * downing the master releases all slaves.  We can make bonds full of
3620  * bonding devices to test this, however.
3621  */
3622 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3623 {
3624 	struct bonding *bond = netdev_priv(bond_dev);
3625 	struct slave *slave, *rollback_slave;
3626 	struct sockaddr *sa = addr, tmp_sa;
3627 	struct list_head *iter;
3628 	int res = 0;
3629 
3630 	if (BOND_MODE(bond) == BOND_MODE_ALB)
3631 		return bond_alb_set_mac_address(bond_dev, addr);
3632 
3633 
3634 	netdev_dbg(bond_dev, "bond=%p\n", bond);
3635 
3636 	/* If fail_over_mac is enabled, do nothing and return success.
3637 	 * Returning an error causes ifenslave to fail.
3638 	 */
3639 	if (bond->params.fail_over_mac &&
3640 	    BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3641 		return 0;
3642 
3643 	if (!is_valid_ether_addr(sa->sa_data))
3644 		return -EADDRNOTAVAIL;
3645 
3646 	bond_for_each_slave(bond, slave, iter) {
3647 		netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3648 		res = dev_set_mac_address(slave->dev, addr);
3649 		if (res) {
3650 			/* TODO: consider downing the slave
3651 			 * and retry ?
3652 			 * User should expect communications
3653 			 * breakage anyway until ARP finish
3654 			 * updating, so...
3655 			 */
3656 			netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3657 			goto unwind;
3658 		}
3659 	}
3660 
3661 	/* success */
3662 	memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3663 	return 0;
3664 
3665 unwind:
3666 	memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3667 	tmp_sa.sa_family = bond_dev->type;
3668 
3669 	/* unwind from head to the slave that failed */
3670 	bond_for_each_slave(bond, rollback_slave, iter) {
3671 		int tmp_res;
3672 
3673 		if (rollback_slave == slave)
3674 			break;
3675 
3676 		tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3677 		if (tmp_res) {
3678 			netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3679 				   tmp_res, rollback_slave->dev->name);
3680 		}
3681 	}
3682 
3683 	return res;
3684 }
3685 
3686 /**
3687  * bond_xmit_slave_id - transmit skb through slave with slave_id
3688  * @bond: bonding device that is transmitting
3689  * @skb: buffer to transmit
3690  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3691  *
3692  * This function tries to transmit through slave with slave_id but in case
3693  * it fails, it tries to find the first available slave for transmission.
3694  * The skb is consumed in all cases, thus the function is void.
3695  */
3696 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3697 {
3698 	struct list_head *iter;
3699 	struct slave *slave;
3700 	int i = slave_id;
3701 
3702 	/* Here we start from the slave with slave_id */
3703 	bond_for_each_slave_rcu(bond, slave, iter) {
3704 		if (--i < 0) {
3705 			if (bond_slave_can_tx(slave)) {
3706 				bond_dev_queue_xmit(bond, skb, slave->dev);
3707 				return;
3708 			}
3709 		}
3710 	}
3711 
3712 	/* Here we start from the first slave up to slave_id */
3713 	i = slave_id;
3714 	bond_for_each_slave_rcu(bond, slave, iter) {
3715 		if (--i < 0)
3716 			break;
3717 		if (bond_slave_can_tx(slave)) {
3718 			bond_dev_queue_xmit(bond, skb, slave->dev);
3719 			return;
3720 		}
3721 	}
3722 	/* no slave that can tx has been found */
3723 	bond_tx_drop(bond->dev, skb);
3724 }
3725 
3726 /**
3727  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3728  * @bond: bonding device to use
3729  *
3730  * Based on the value of the bonding device's packets_per_slave parameter
3731  * this function generates a slave id, which is usually used as the next
3732  * slave to transmit through.
3733  */
3734 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3735 {
3736 	u32 slave_id;
3737 	struct reciprocal_value reciprocal_packets_per_slave;
3738 	int packets_per_slave = bond->params.packets_per_slave;
3739 
3740 	switch (packets_per_slave) {
3741 	case 0:
3742 		slave_id = prandom_u32();
3743 		break;
3744 	case 1:
3745 		slave_id = bond->rr_tx_counter;
3746 		break;
3747 	default:
3748 		reciprocal_packets_per_slave =
3749 			bond->params.reciprocal_packets_per_slave;
3750 		slave_id = reciprocal_divide(bond->rr_tx_counter,
3751 					     reciprocal_packets_per_slave);
3752 		break;
3753 	}
3754 	bond->rr_tx_counter++;
3755 
3756 	return slave_id;
3757 }
3758 
3759 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3760 {
3761 	struct bonding *bond = netdev_priv(bond_dev);
3762 	struct iphdr *iph = ip_hdr(skb);
3763 	struct slave *slave;
3764 	u32 slave_id;
3765 
3766 	/* Start with the curr_active_slave that joined the bond as the
3767 	 * default for sending IGMP traffic.  For failover purposes one
3768 	 * needs to maintain some consistency for the interface that will
3769 	 * send the join/membership reports.  The curr_active_slave found
3770 	 * will send all of this type of traffic.
3771 	 */
3772 	if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3773 		slave = rcu_dereference(bond->curr_active_slave);
3774 		if (slave)
3775 			bond_dev_queue_xmit(bond, skb, slave->dev);
3776 		else
3777 			bond_xmit_slave_id(bond, skb, 0);
3778 	} else {
3779 		int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3780 
3781 		if (likely(slave_cnt)) {
3782 			slave_id = bond_rr_gen_slave_id(bond);
3783 			bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3784 		} else {
3785 			bond_tx_drop(bond_dev, skb);
3786 		}
3787 	}
3788 
3789 	return NETDEV_TX_OK;
3790 }
3791 
3792 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3793  * the bond has a usable interface.
3794  */
3795 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3796 {
3797 	struct bonding *bond = netdev_priv(bond_dev);
3798 	struct slave *slave;
3799 
3800 	slave = rcu_dereference(bond->curr_active_slave);
3801 	if (slave)
3802 		bond_dev_queue_xmit(bond, skb, slave->dev);
3803 	else
3804 		bond_tx_drop(bond_dev, skb);
3805 
3806 	return NETDEV_TX_OK;
3807 }
3808 
3809 /* Use this to update slave_array when (a) it's not appropriate to update
3810  * slave_array right away (note that update_slave_array() may sleep)
3811  * and / or (b) RTNL is not held.
3812  */
3813 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3814 {
3815 	queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3816 }
3817 
3818 /* Slave array work handler. Holds only RTNL */
3819 static void bond_slave_arr_handler(struct work_struct *work)
3820 {
3821 	struct bonding *bond = container_of(work, struct bonding,
3822 					    slave_arr_work.work);
3823 	int ret;
3824 
3825 	if (!rtnl_trylock())
3826 		goto err;
3827 
3828 	ret = bond_update_slave_arr(bond, NULL);
3829 	rtnl_unlock();
3830 	if (ret) {
3831 		pr_warn_ratelimited("Failed to update slave array from WT\n");
3832 		goto err;
3833 	}
3834 	return;
3835 
3836 err:
3837 	bond_slave_arr_work_rearm(bond, 1);
3838 }
3839 
3840 /* Build the usable slaves array in control path for modes that use xmit-hash
3841  * to determine the slave interface -
3842  * (a) BOND_MODE_8023AD
3843  * (b) BOND_MODE_XOR
3844  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3845  *
3846  * The caller is expected to hold RTNL only and NO other lock!
3847  */
3848 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3849 {
3850 	struct slave *slave;
3851 	struct list_head *iter;
3852 	struct bond_up_slave *new_arr, *old_arr;
3853 	int agg_id = 0;
3854 	int ret = 0;
3855 
3856 #ifdef CONFIG_LOCKDEP
3857 	WARN_ON(lockdep_is_held(&bond->mode_lock));
3858 #endif
3859 
3860 	new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3861 			  GFP_KERNEL);
3862 	if (!new_arr) {
3863 		ret = -ENOMEM;
3864 		pr_err("Failed to build slave-array.\n");
3865 		goto out;
3866 	}
3867 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3868 		struct ad_info ad_info;
3869 
3870 		if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3871 			pr_debug("bond_3ad_get_active_agg_info failed\n");
3872 			kfree_rcu(new_arr, rcu);
3873 			/* No active aggragator means it's not safe to use
3874 			 * the previous array.
3875 			 */
3876 			old_arr = rtnl_dereference(bond->slave_arr);
3877 			if (old_arr) {
3878 				RCU_INIT_POINTER(bond->slave_arr, NULL);
3879 				kfree_rcu(old_arr, rcu);
3880 			}
3881 			goto out;
3882 		}
3883 		agg_id = ad_info.aggregator_id;
3884 	}
3885 	bond_for_each_slave(bond, slave, iter) {
3886 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3887 			struct aggregator *agg;
3888 
3889 			agg = SLAVE_AD_INFO(slave)->port.aggregator;
3890 			if (!agg || agg->aggregator_identifier != agg_id)
3891 				continue;
3892 		}
3893 		if (!bond_slave_can_tx(slave))
3894 			continue;
3895 		if (skipslave == slave)
3896 			continue;
3897 		new_arr->arr[new_arr->count++] = slave;
3898 	}
3899 
3900 	old_arr = rtnl_dereference(bond->slave_arr);
3901 	rcu_assign_pointer(bond->slave_arr, new_arr);
3902 	if (old_arr)
3903 		kfree_rcu(old_arr, rcu);
3904 out:
3905 	if (ret != 0 && skipslave) {
3906 		int idx;
3907 
3908 		/* Rare situation where caller has asked to skip a specific
3909 		 * slave but allocation failed (most likely!). BTW this is
3910 		 * only possible when the call is initiated from
3911 		 * __bond_release_one(). In this situation; overwrite the
3912 		 * skipslave entry in the array with the last entry from the
3913 		 * array to avoid a situation where the xmit path may choose
3914 		 * this to-be-skipped slave to send a packet out.
3915 		 */
3916 		old_arr = rtnl_dereference(bond->slave_arr);
3917 		for (idx = 0; idx < old_arr->count; idx++) {
3918 			if (skipslave == old_arr->arr[idx]) {
3919 				old_arr->arr[idx] =
3920 				    old_arr->arr[old_arr->count-1];
3921 				old_arr->count--;
3922 				break;
3923 			}
3924 		}
3925 	}
3926 	return ret;
3927 }
3928 
3929 /* Use this Xmit function for 3AD as well as XOR modes. The current
3930  * usable slave array is formed in the control path. The xmit function
3931  * just calculates hash and sends the packet out.
3932  */
3933 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3934 {
3935 	struct bonding *bond = netdev_priv(dev);
3936 	struct slave *slave;
3937 	struct bond_up_slave *slaves;
3938 	unsigned int count;
3939 
3940 	slaves = rcu_dereference(bond->slave_arr);
3941 	count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3942 	if (likely(count)) {
3943 		slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3944 		bond_dev_queue_xmit(bond, skb, slave->dev);
3945 	} else {
3946 		bond_tx_drop(dev, skb);
3947 	}
3948 
3949 	return NETDEV_TX_OK;
3950 }
3951 
3952 /* in broadcast mode, we send everything to all usable interfaces. */
3953 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3954 {
3955 	struct bonding *bond = netdev_priv(bond_dev);
3956 	struct slave *slave = NULL;
3957 	struct list_head *iter;
3958 
3959 	bond_for_each_slave_rcu(bond, slave, iter) {
3960 		if (bond_is_last_slave(bond, slave))
3961 			break;
3962 		if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3963 			struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3964 
3965 			if (!skb2) {
3966 				net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3967 						    bond_dev->name, __func__);
3968 				continue;
3969 			}
3970 			bond_dev_queue_xmit(bond, skb2, slave->dev);
3971 		}
3972 	}
3973 	if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3974 		bond_dev_queue_xmit(bond, skb, slave->dev);
3975 	else
3976 		bond_tx_drop(bond_dev, skb);
3977 
3978 	return NETDEV_TX_OK;
3979 }
3980 
3981 /*------------------------- Device initialization ---------------------------*/
3982 
3983 /* Lookup the slave that corresponds to a qid */
3984 static inline int bond_slave_override(struct bonding *bond,
3985 				      struct sk_buff *skb)
3986 {
3987 	struct slave *slave = NULL;
3988 	struct list_head *iter;
3989 
3990 	if (!skb->queue_mapping)
3991 		return 1;
3992 
3993 	/* Find out if any slaves have the same mapping as this skb. */
3994 	bond_for_each_slave_rcu(bond, slave, iter) {
3995 		if (slave->queue_id == skb->queue_mapping) {
3996 			if (bond_slave_is_up(slave) &&
3997 			    slave->link == BOND_LINK_UP) {
3998 				bond_dev_queue_xmit(bond, skb, slave->dev);
3999 				return 0;
4000 			}
4001 			/* If the slave isn't UP, use default transmit policy. */
4002 			break;
4003 		}
4004 	}
4005 
4006 	return 1;
4007 }
4008 
4009 
4010 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4011 			     void *accel_priv, select_queue_fallback_t fallback)
4012 {
4013 	/* This helper function exists to help dev_pick_tx get the correct
4014 	 * destination queue.  Using a helper function skips a call to
4015 	 * skb_tx_hash and will put the skbs in the queue we expect on their
4016 	 * way down to the bonding driver.
4017 	 */
4018 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4019 
4020 	/* Save the original txq to restore before passing to the driver */
4021 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4022 
4023 	if (unlikely(txq >= dev->real_num_tx_queues)) {
4024 		do {
4025 			txq -= dev->real_num_tx_queues;
4026 		} while (txq >= dev->real_num_tx_queues);
4027 	}
4028 	return txq;
4029 }
4030 
4031 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4032 {
4033 	struct bonding *bond = netdev_priv(dev);
4034 
4035 	if (bond_should_override_tx_queue(bond) &&
4036 	    !bond_slave_override(bond, skb))
4037 		return NETDEV_TX_OK;
4038 
4039 	switch (BOND_MODE(bond)) {
4040 	case BOND_MODE_ROUNDROBIN:
4041 		return bond_xmit_roundrobin(skb, dev);
4042 	case BOND_MODE_ACTIVEBACKUP:
4043 		return bond_xmit_activebackup(skb, dev);
4044 	case BOND_MODE_8023AD:
4045 	case BOND_MODE_XOR:
4046 		return bond_3ad_xor_xmit(skb, dev);
4047 	case BOND_MODE_BROADCAST:
4048 		return bond_xmit_broadcast(skb, dev);
4049 	case BOND_MODE_ALB:
4050 		return bond_alb_xmit(skb, dev);
4051 	case BOND_MODE_TLB:
4052 		return bond_tlb_xmit(skb, dev);
4053 	default:
4054 		/* Should never happen, mode already checked */
4055 		netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4056 		WARN_ON_ONCE(1);
4057 		bond_tx_drop(dev, skb);
4058 		return NETDEV_TX_OK;
4059 	}
4060 }
4061 
4062 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4063 {
4064 	struct bonding *bond = netdev_priv(dev);
4065 	netdev_tx_t ret = NETDEV_TX_OK;
4066 
4067 	/* If we risk deadlock from transmitting this in the
4068 	 * netpoll path, tell netpoll to queue the frame for later tx
4069 	 */
4070 	if (unlikely(is_netpoll_tx_blocked(dev)))
4071 		return NETDEV_TX_BUSY;
4072 
4073 	rcu_read_lock();
4074 	if (bond_has_slaves(bond))
4075 		ret = __bond_start_xmit(skb, dev);
4076 	else
4077 		bond_tx_drop(dev, skb);
4078 	rcu_read_unlock();
4079 
4080 	return ret;
4081 }
4082 
4083 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4084 					   struct ethtool_link_ksettings *cmd)
4085 {
4086 	struct bonding *bond = netdev_priv(bond_dev);
4087 	unsigned long speed = 0;
4088 	struct list_head *iter;
4089 	struct slave *slave;
4090 
4091 	cmd->base.duplex = DUPLEX_UNKNOWN;
4092 	cmd->base.port = PORT_OTHER;
4093 
4094 	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4095 	 * do not need to check mode.  Though link speed might not represent
4096 	 * the true receive or transmit bandwidth (not all modes are symmetric)
4097 	 * this is an accurate maximum.
4098 	 */
4099 	bond_for_each_slave(bond, slave, iter) {
4100 		if (bond_slave_can_tx(slave)) {
4101 			if (slave->speed != SPEED_UNKNOWN)
4102 				speed += slave->speed;
4103 			if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4104 			    slave->duplex != DUPLEX_UNKNOWN)
4105 				cmd->base.duplex = slave->duplex;
4106 		}
4107 	}
4108 	cmd->base.speed = speed ? : SPEED_UNKNOWN;
4109 
4110 	return 0;
4111 }
4112 
4113 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4114 				     struct ethtool_drvinfo *drvinfo)
4115 {
4116 	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4117 	strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4118 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4119 		 BOND_ABI_VERSION);
4120 }
4121 
4122 static const struct ethtool_ops bond_ethtool_ops = {
4123 	.get_drvinfo		= bond_ethtool_get_drvinfo,
4124 	.get_link		= ethtool_op_get_link,
4125 	.get_link_ksettings	= bond_ethtool_get_link_ksettings,
4126 };
4127 
4128 static const struct net_device_ops bond_netdev_ops = {
4129 	.ndo_init		= bond_init,
4130 	.ndo_uninit		= bond_uninit,
4131 	.ndo_open		= bond_open,
4132 	.ndo_stop		= bond_close,
4133 	.ndo_start_xmit		= bond_start_xmit,
4134 	.ndo_select_queue	= bond_select_queue,
4135 	.ndo_get_stats64	= bond_get_stats,
4136 	.ndo_do_ioctl		= bond_do_ioctl,
4137 	.ndo_change_rx_flags	= bond_change_rx_flags,
4138 	.ndo_set_rx_mode	= bond_set_rx_mode,
4139 	.ndo_change_mtu		= bond_change_mtu,
4140 	.ndo_set_mac_address	= bond_set_mac_address,
4141 	.ndo_neigh_setup	= bond_neigh_setup,
4142 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
4143 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
4144 #ifdef CONFIG_NET_POLL_CONTROLLER
4145 	.ndo_netpoll_setup	= bond_netpoll_setup,
4146 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
4147 	.ndo_poll_controller	= bond_poll_controller,
4148 #endif
4149 	.ndo_add_slave		= bond_enslave,
4150 	.ndo_del_slave		= bond_release,
4151 	.ndo_fix_features	= bond_fix_features,
4152 	.ndo_neigh_construct	= netdev_default_l2upper_neigh_construct,
4153 	.ndo_neigh_destroy	= netdev_default_l2upper_neigh_destroy,
4154 	.ndo_bridge_setlink	= switchdev_port_bridge_setlink,
4155 	.ndo_bridge_getlink	= switchdev_port_bridge_getlink,
4156 	.ndo_bridge_dellink	= switchdev_port_bridge_dellink,
4157 	.ndo_fdb_add		= switchdev_port_fdb_add,
4158 	.ndo_fdb_del		= switchdev_port_fdb_del,
4159 	.ndo_fdb_dump		= switchdev_port_fdb_dump,
4160 	.ndo_features_check	= passthru_features_check,
4161 };
4162 
4163 static const struct device_type bond_type = {
4164 	.name = "bond",
4165 };
4166 
4167 static void bond_destructor(struct net_device *bond_dev)
4168 {
4169 	struct bonding *bond = netdev_priv(bond_dev);
4170 	if (bond->wq)
4171 		destroy_workqueue(bond->wq);
4172 	free_netdev(bond_dev);
4173 }
4174 
4175 void bond_setup(struct net_device *bond_dev)
4176 {
4177 	struct bonding *bond = netdev_priv(bond_dev);
4178 
4179 	spin_lock_init(&bond->mode_lock);
4180 	spin_lock_init(&bond->stats_lock);
4181 	bond->params = bonding_defaults;
4182 
4183 	/* Initialize pointers */
4184 	bond->dev = bond_dev;
4185 
4186 	/* Initialize the device entry points */
4187 	ether_setup(bond_dev);
4188 	bond_dev->netdev_ops = &bond_netdev_ops;
4189 	bond_dev->ethtool_ops = &bond_ethtool_ops;
4190 
4191 	bond_dev->destructor = bond_destructor;
4192 
4193 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4194 
4195 	/* Initialize the device options */
4196 	bond_dev->flags |= IFF_MASTER;
4197 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4198 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4199 
4200 	/* don't acquire bond device's netif_tx_lock when transmitting */
4201 	bond_dev->features |= NETIF_F_LLTX;
4202 
4203 	/* By default, we declare the bond to be fully
4204 	 * VLAN hardware accelerated capable. Special
4205 	 * care is taken in the various xmit functions
4206 	 * when there are slaves that are not hw accel
4207 	 * capable
4208 	 */
4209 
4210 	/* Don't allow bond devices to change network namespaces. */
4211 	bond_dev->features |= NETIF_F_NETNS_LOCAL;
4212 
4213 	bond_dev->hw_features = BOND_VLAN_FEATURES |
4214 				NETIF_F_HW_VLAN_CTAG_TX |
4215 				NETIF_F_HW_VLAN_CTAG_RX |
4216 				NETIF_F_HW_VLAN_CTAG_FILTER;
4217 
4218 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4219 	bond_dev->features |= bond_dev->hw_features;
4220 }
4221 
4222 /* Destroy a bonding device.
4223  * Must be under rtnl_lock when this function is called.
4224  */
4225 static void bond_uninit(struct net_device *bond_dev)
4226 {
4227 	struct bonding *bond = netdev_priv(bond_dev);
4228 	struct list_head *iter;
4229 	struct slave *slave;
4230 	struct bond_up_slave *arr;
4231 
4232 	bond_netpoll_cleanup(bond_dev);
4233 
4234 	/* Release the bonded slaves */
4235 	bond_for_each_slave(bond, slave, iter)
4236 		__bond_release_one(bond_dev, slave->dev, true);
4237 	netdev_info(bond_dev, "Released all slaves\n");
4238 
4239 	arr = rtnl_dereference(bond->slave_arr);
4240 	if (arr) {
4241 		RCU_INIT_POINTER(bond->slave_arr, NULL);
4242 		kfree_rcu(arr, rcu);
4243 	}
4244 
4245 	list_del(&bond->bond_list);
4246 
4247 	bond_debug_unregister(bond);
4248 }
4249 
4250 /*------------------------- Module initialization ---------------------------*/
4251 
4252 static int bond_check_params(struct bond_params *params)
4253 {
4254 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4255 	struct bond_opt_value newval;
4256 	const struct bond_opt_value *valptr;
4257 	int arp_all_targets_value;
4258 	u16 ad_actor_sys_prio = 0;
4259 	u16 ad_user_port_key = 0;
4260 
4261 	/* Convert string parameters. */
4262 	if (mode) {
4263 		bond_opt_initstr(&newval, mode);
4264 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4265 		if (!valptr) {
4266 			pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4267 			return -EINVAL;
4268 		}
4269 		bond_mode = valptr->value;
4270 	}
4271 
4272 	if (xmit_hash_policy) {
4273 		if ((bond_mode != BOND_MODE_XOR) &&
4274 		    (bond_mode != BOND_MODE_8023AD) &&
4275 		    (bond_mode != BOND_MODE_TLB)) {
4276 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4277 				bond_mode_name(bond_mode));
4278 		} else {
4279 			bond_opt_initstr(&newval, xmit_hash_policy);
4280 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4281 						&newval);
4282 			if (!valptr) {
4283 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4284 				       xmit_hash_policy);
4285 				return -EINVAL;
4286 			}
4287 			xmit_hashtype = valptr->value;
4288 		}
4289 	}
4290 
4291 	if (lacp_rate) {
4292 		if (bond_mode != BOND_MODE_8023AD) {
4293 			pr_info("lacp_rate param is irrelevant in mode %s\n",
4294 				bond_mode_name(bond_mode));
4295 		} else {
4296 			bond_opt_initstr(&newval, lacp_rate);
4297 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4298 						&newval);
4299 			if (!valptr) {
4300 				pr_err("Error: Invalid lacp rate \"%s\"\n",
4301 				       lacp_rate);
4302 				return -EINVAL;
4303 			}
4304 			lacp_fast = valptr->value;
4305 		}
4306 	}
4307 
4308 	if (ad_select) {
4309 		bond_opt_initstr(&newval, ad_select);
4310 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4311 					&newval);
4312 		if (!valptr) {
4313 			pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4314 			return -EINVAL;
4315 		}
4316 		params->ad_select = valptr->value;
4317 		if (bond_mode != BOND_MODE_8023AD)
4318 			pr_warn("ad_select param only affects 802.3ad mode\n");
4319 	} else {
4320 		params->ad_select = BOND_AD_STABLE;
4321 	}
4322 
4323 	if (max_bonds < 0) {
4324 		pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4325 			max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4326 		max_bonds = BOND_DEFAULT_MAX_BONDS;
4327 	}
4328 
4329 	if (miimon < 0) {
4330 		pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4331 			miimon, INT_MAX);
4332 		miimon = 0;
4333 	}
4334 
4335 	if (updelay < 0) {
4336 		pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4337 			updelay, INT_MAX);
4338 		updelay = 0;
4339 	}
4340 
4341 	if (downdelay < 0) {
4342 		pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4343 			downdelay, INT_MAX);
4344 		downdelay = 0;
4345 	}
4346 
4347 	if ((use_carrier != 0) && (use_carrier != 1)) {
4348 		pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4349 			use_carrier);
4350 		use_carrier = 1;
4351 	}
4352 
4353 	if (num_peer_notif < 0 || num_peer_notif > 255) {
4354 		pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4355 			num_peer_notif);
4356 		num_peer_notif = 1;
4357 	}
4358 
4359 	/* reset values for 802.3ad/TLB/ALB */
4360 	if (!bond_mode_uses_arp(bond_mode)) {
4361 		if (!miimon) {
4362 			pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4363 			pr_warn("Forcing miimon to 100msec\n");
4364 			miimon = BOND_DEFAULT_MIIMON;
4365 		}
4366 	}
4367 
4368 	if (tx_queues < 1 || tx_queues > 255) {
4369 		pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4370 			tx_queues, BOND_DEFAULT_TX_QUEUES);
4371 		tx_queues = BOND_DEFAULT_TX_QUEUES;
4372 	}
4373 
4374 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4375 		pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4376 			all_slaves_active);
4377 		all_slaves_active = 0;
4378 	}
4379 
4380 	if (resend_igmp < 0 || resend_igmp > 255) {
4381 		pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4382 			resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4383 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4384 	}
4385 
4386 	bond_opt_initval(&newval, packets_per_slave);
4387 	if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4388 		pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4389 			packets_per_slave, USHRT_MAX);
4390 		packets_per_slave = 1;
4391 	}
4392 
4393 	if (bond_mode == BOND_MODE_ALB) {
4394 		pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4395 			  updelay);
4396 	}
4397 
4398 	if (!miimon) {
4399 		if (updelay || downdelay) {
4400 			/* just warn the user the up/down delay will have
4401 			 * no effect since miimon is zero...
4402 			 */
4403 			pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4404 				updelay, downdelay);
4405 		}
4406 	} else {
4407 		/* don't allow arp monitoring */
4408 		if (arp_interval) {
4409 			pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4410 				miimon, arp_interval);
4411 			arp_interval = 0;
4412 		}
4413 
4414 		if ((updelay % miimon) != 0) {
4415 			pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4416 				updelay, miimon, (updelay / miimon) * miimon);
4417 		}
4418 
4419 		updelay /= miimon;
4420 
4421 		if ((downdelay % miimon) != 0) {
4422 			pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4423 				downdelay, miimon,
4424 				(downdelay / miimon) * miimon);
4425 		}
4426 
4427 		downdelay /= miimon;
4428 	}
4429 
4430 	if (arp_interval < 0) {
4431 		pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4432 			arp_interval, INT_MAX);
4433 		arp_interval = 0;
4434 	}
4435 
4436 	for (arp_ip_count = 0, i = 0;
4437 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4438 		__be32 ip;
4439 
4440 		/* not a complete check, but good enough to catch mistakes */
4441 		if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4442 		    !bond_is_ip_target_ok(ip)) {
4443 			pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4444 				arp_ip_target[i]);
4445 			arp_interval = 0;
4446 		} else {
4447 			if (bond_get_targets_ip(arp_target, ip) == -1)
4448 				arp_target[arp_ip_count++] = ip;
4449 			else
4450 				pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4451 					&ip);
4452 		}
4453 	}
4454 
4455 	if (arp_interval && !arp_ip_count) {
4456 		/* don't allow arping if no arp_ip_target given... */
4457 		pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4458 			arp_interval);
4459 		arp_interval = 0;
4460 	}
4461 
4462 	if (arp_validate) {
4463 		if (!arp_interval) {
4464 			pr_err("arp_validate requires arp_interval\n");
4465 			return -EINVAL;
4466 		}
4467 
4468 		bond_opt_initstr(&newval, arp_validate);
4469 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4470 					&newval);
4471 		if (!valptr) {
4472 			pr_err("Error: invalid arp_validate \"%s\"\n",
4473 			       arp_validate);
4474 			return -EINVAL;
4475 		}
4476 		arp_validate_value = valptr->value;
4477 	} else {
4478 		arp_validate_value = 0;
4479 	}
4480 
4481 	arp_all_targets_value = 0;
4482 	if (arp_all_targets) {
4483 		bond_opt_initstr(&newval, arp_all_targets);
4484 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4485 					&newval);
4486 		if (!valptr) {
4487 			pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4488 			       arp_all_targets);
4489 			arp_all_targets_value = 0;
4490 		} else {
4491 			arp_all_targets_value = valptr->value;
4492 		}
4493 	}
4494 
4495 	if (miimon) {
4496 		pr_info("MII link monitoring set to %d ms\n", miimon);
4497 	} else if (arp_interval) {
4498 		valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4499 					  arp_validate_value);
4500 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4501 			arp_interval, valptr->string, arp_ip_count);
4502 
4503 		for (i = 0; i < arp_ip_count; i++)
4504 			pr_cont(" %s", arp_ip_target[i]);
4505 
4506 		pr_cont("\n");
4507 
4508 	} else if (max_bonds) {
4509 		/* miimon and arp_interval not set, we need one so things
4510 		 * work as expected, see bonding.txt for details
4511 		 */
4512 		pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
4513 	}
4514 
4515 	if (primary && !bond_mode_uses_primary(bond_mode)) {
4516 		/* currently, using a primary only makes sense
4517 		 * in active backup, TLB or ALB modes
4518 		 */
4519 		pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4520 			primary, bond_mode_name(bond_mode));
4521 		primary = NULL;
4522 	}
4523 
4524 	if (primary && primary_reselect) {
4525 		bond_opt_initstr(&newval, primary_reselect);
4526 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4527 					&newval);
4528 		if (!valptr) {
4529 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
4530 			       primary_reselect);
4531 			return -EINVAL;
4532 		}
4533 		primary_reselect_value = valptr->value;
4534 	} else {
4535 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4536 	}
4537 
4538 	if (fail_over_mac) {
4539 		bond_opt_initstr(&newval, fail_over_mac);
4540 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4541 					&newval);
4542 		if (!valptr) {
4543 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
4544 			       fail_over_mac);
4545 			return -EINVAL;
4546 		}
4547 		fail_over_mac_value = valptr->value;
4548 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4549 			pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4550 	} else {
4551 		fail_over_mac_value = BOND_FOM_NONE;
4552 	}
4553 
4554 	bond_opt_initstr(&newval, "default");
4555 	valptr = bond_opt_parse(
4556 			bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4557 				     &newval);
4558 	if (!valptr) {
4559 		pr_err("Error: No ad_actor_sys_prio default value");
4560 		return -EINVAL;
4561 	}
4562 	ad_actor_sys_prio = valptr->value;
4563 
4564 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4565 				&newval);
4566 	if (!valptr) {
4567 		pr_err("Error: No ad_user_port_key default value");
4568 		return -EINVAL;
4569 	}
4570 	ad_user_port_key = valptr->value;
4571 
4572 	if (lp_interval == 0) {
4573 		pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4574 			INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4575 		lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4576 	}
4577 
4578 	/* fill params struct with the proper values */
4579 	params->mode = bond_mode;
4580 	params->xmit_policy = xmit_hashtype;
4581 	params->miimon = miimon;
4582 	params->num_peer_notif = num_peer_notif;
4583 	params->arp_interval = arp_interval;
4584 	params->arp_validate = arp_validate_value;
4585 	params->arp_all_targets = arp_all_targets_value;
4586 	params->updelay = updelay;
4587 	params->downdelay = downdelay;
4588 	params->use_carrier = use_carrier;
4589 	params->lacp_fast = lacp_fast;
4590 	params->primary[0] = 0;
4591 	params->primary_reselect = primary_reselect_value;
4592 	params->fail_over_mac = fail_over_mac_value;
4593 	params->tx_queues = tx_queues;
4594 	params->all_slaves_active = all_slaves_active;
4595 	params->resend_igmp = resend_igmp;
4596 	params->min_links = min_links;
4597 	params->lp_interval = lp_interval;
4598 	params->packets_per_slave = packets_per_slave;
4599 	params->tlb_dynamic_lb = 1; /* Default value */
4600 	params->ad_actor_sys_prio = ad_actor_sys_prio;
4601 	eth_zero_addr(params->ad_actor_system);
4602 	params->ad_user_port_key = ad_user_port_key;
4603 	if (packets_per_slave > 0) {
4604 		params->reciprocal_packets_per_slave =
4605 			reciprocal_value(packets_per_slave);
4606 	} else {
4607 		/* reciprocal_packets_per_slave is unused if
4608 		 * packets_per_slave is 0 or 1, just initialize it
4609 		 */
4610 		params->reciprocal_packets_per_slave =
4611 			(struct reciprocal_value) { 0 };
4612 	}
4613 
4614 	if (primary) {
4615 		strncpy(params->primary, primary, IFNAMSIZ);
4616 		params->primary[IFNAMSIZ - 1] = 0;
4617 	}
4618 
4619 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4620 
4621 	return 0;
4622 }
4623 
4624 /* Called from registration process */
4625 static int bond_init(struct net_device *bond_dev)
4626 {
4627 	struct bonding *bond = netdev_priv(bond_dev);
4628 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4629 
4630 	netdev_dbg(bond_dev, "Begin bond_init\n");
4631 
4632 	bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4633 	if (!bond->wq)
4634 		return -ENOMEM;
4635 
4636 	netdev_lockdep_set_classes(bond_dev);
4637 
4638 	list_add_tail(&bond->bond_list, &bn->dev_list);
4639 
4640 	bond_prepare_sysfs_group(bond);
4641 
4642 	bond_debug_register(bond);
4643 
4644 	/* Ensure valid dev_addr */
4645 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
4646 	    bond_dev->addr_assign_type == NET_ADDR_PERM)
4647 		eth_hw_addr_random(bond_dev);
4648 
4649 	return 0;
4650 }
4651 
4652 unsigned int bond_get_num_tx_queues(void)
4653 {
4654 	return tx_queues;
4655 }
4656 
4657 /* Create a new bond based on the specified name and bonding parameters.
4658  * If name is NULL, obtain a suitable "bond%d" name for us.
4659  * Caller must NOT hold rtnl_lock; we need to release it here before we
4660  * set up our sysfs entries.
4661  */
4662 int bond_create(struct net *net, const char *name)
4663 {
4664 	struct net_device *bond_dev;
4665 	struct bonding *bond;
4666 	struct alb_bond_info *bond_info;
4667 	int res;
4668 
4669 	rtnl_lock();
4670 
4671 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4672 				   name ? name : "bond%d", NET_NAME_UNKNOWN,
4673 				   bond_setup, tx_queues);
4674 	if (!bond_dev) {
4675 		pr_err("%s: eek! can't alloc netdev!\n", name);
4676 		rtnl_unlock();
4677 		return -ENOMEM;
4678 	}
4679 
4680 	/*
4681 	 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4682 	 * It is set to 0 by default which is wrong.
4683 	 */
4684 	bond = netdev_priv(bond_dev);
4685 	bond_info = &(BOND_ALB_INFO(bond));
4686 	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4687 
4688 	dev_net_set(bond_dev, net);
4689 	bond_dev->rtnl_link_ops = &bond_link_ops;
4690 
4691 	res = register_netdevice(bond_dev);
4692 
4693 	netif_carrier_off(bond_dev);
4694 
4695 	rtnl_unlock();
4696 	if (res < 0)
4697 		bond_destructor(bond_dev);
4698 	return res;
4699 }
4700 
4701 static int __net_init bond_net_init(struct net *net)
4702 {
4703 	struct bond_net *bn = net_generic(net, bond_net_id);
4704 
4705 	bn->net = net;
4706 	INIT_LIST_HEAD(&bn->dev_list);
4707 
4708 	bond_create_proc_dir(bn);
4709 	bond_create_sysfs(bn);
4710 
4711 	return 0;
4712 }
4713 
4714 static void __net_exit bond_net_exit(struct net *net)
4715 {
4716 	struct bond_net *bn = net_generic(net, bond_net_id);
4717 	struct bonding *bond, *tmp_bond;
4718 	LIST_HEAD(list);
4719 
4720 	bond_destroy_sysfs(bn);
4721 
4722 	/* Kill off any bonds created after unregistering bond rtnl ops */
4723 	rtnl_lock();
4724 	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4725 		unregister_netdevice_queue(bond->dev, &list);
4726 	unregister_netdevice_many(&list);
4727 	rtnl_unlock();
4728 
4729 	bond_destroy_proc_dir(bn);
4730 }
4731 
4732 static struct pernet_operations bond_net_ops = {
4733 	.init = bond_net_init,
4734 	.exit = bond_net_exit,
4735 	.id   = &bond_net_id,
4736 	.size = sizeof(struct bond_net),
4737 };
4738 
4739 static int __init bonding_init(void)
4740 {
4741 	int i;
4742 	int res;
4743 
4744 	pr_info("%s", bond_version);
4745 
4746 	res = bond_check_params(&bonding_defaults);
4747 	if (res)
4748 		goto out;
4749 
4750 	res = register_pernet_subsys(&bond_net_ops);
4751 	if (res)
4752 		goto out;
4753 
4754 	res = bond_netlink_init();
4755 	if (res)
4756 		goto err_link;
4757 
4758 	bond_create_debugfs();
4759 
4760 	for (i = 0; i < max_bonds; i++) {
4761 		res = bond_create(&init_net, NULL);
4762 		if (res)
4763 			goto err;
4764 	}
4765 
4766 	register_netdevice_notifier(&bond_netdev_notifier);
4767 out:
4768 	return res;
4769 err:
4770 	bond_destroy_debugfs();
4771 	bond_netlink_fini();
4772 err_link:
4773 	unregister_pernet_subsys(&bond_net_ops);
4774 	goto out;
4775 
4776 }
4777 
4778 static void __exit bonding_exit(void)
4779 {
4780 	unregister_netdevice_notifier(&bond_netdev_notifier);
4781 
4782 	bond_destroy_debugfs();
4783 
4784 	bond_netlink_fini();
4785 	unregister_pernet_subsys(&bond_net_ops);
4786 
4787 #ifdef CONFIG_NET_POLL_CONTROLLER
4788 	/* Make sure we don't have an imbalance on our netpoll blocking */
4789 	WARN_ON(atomic_read(&netpoll_block_tx));
4790 #endif
4791 }
4792 
4793 module_init(bonding_init);
4794 module_exit(bonding_exit);
4795 MODULE_LICENSE("GPL");
4796 MODULE_VERSION(DRV_VERSION);
4797 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4798 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4799