xref: /openbmc/linux/drivers/net/bonding/bond_main.c (revision 8795a739)
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/bonding.h>
81 #include <net/bond_3ad.h>
82 #include <net/bond_alb.h>
83 
84 #include "bonding_priv.h"
85 
86 /*---------------------------- Module parameters ----------------------------*/
87 
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 
90 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon;
94 static int updelay;
95 static int downdelay;
96 static int use_carrier	= 1;
97 static char *mode;
98 static char *primary;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
114 
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 			       "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 			       "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 			    "in milliseconds");
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 			      "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 		       "1 for active-backup, 2 for balance-xor, "
138 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 		       "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 				   "once it comes up; "
145 				   "0 for always (default), "
146 				   "1 for only if speed of primary is "
147 				   "better, "
148 				   "2 for only on active slave "
149 				   "failure");
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 			    "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
155 			    "0 for stable (default), 1 for bandwidth, "
156 			    "2 for count");
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159 
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
162 				   "0 for layer 2 (default), 1 for layer 3+4, "
163 				   "2 for layer 2+3, 3 for encap layer 2+3, "
164 				   "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 			       "0 for none (default), 1 for active, "
172 			       "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 				"the same MAC; 0 for none (default), "
178 				"1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
181 				     "by setting active flag for all slaves; "
182 				     "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 			      "link failure");
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 				    "mode; 0 for a random slave, 1 packet per "
189 				    "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 			      "the bonding driver sends learning packets to "
193 			      "each slaves peer switch. The default is 1.");
194 
195 /*----------------------------- Global variables ----------------------------*/
196 
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199 #endif
200 
201 unsigned int bond_net_id __read_mostly;
202 
203 /*-------------------------- Forward declarations ---------------------------*/
204 
205 static int bond_init(struct net_device *bond_dev);
206 static void bond_uninit(struct net_device *bond_dev);
207 static void bond_get_stats(struct net_device *bond_dev,
208 			   struct rtnl_link_stats64 *stats);
209 static void bond_slave_arr_handler(struct work_struct *work);
210 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
211 				  int mod);
212 static void bond_netdev_notify_work(struct work_struct *work);
213 
214 /*---------------------------- General routines -----------------------------*/
215 
216 const char *bond_mode_name(int mode)
217 {
218 	static const char *names[] = {
219 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 		[BOND_MODE_XOR] = "load balancing (xor)",
222 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 		[BOND_MODE_TLB] = "transmit load balancing",
225 		[BOND_MODE_ALB] = "adaptive load balancing",
226 	};
227 
228 	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
229 		return "unknown";
230 
231 	return names[mode];
232 }
233 
234 /*---------------------------------- VLAN -----------------------------------*/
235 
236 /**
237  * bond_dev_queue_xmit - Prepare skb for xmit.
238  *
239  * @bond: bond device that got this skb for tx.
240  * @skb: hw accel VLAN tagged skb to transmit
241  * @slave_dev: slave that is supposed to xmit this skbuff
242  */
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 			struct net_device *slave_dev)
245 {
246 	skb->dev = slave_dev;
247 
248 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 	skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
251 
252 	if (unlikely(netpoll_tx_running(bond->dev)))
253 		bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
254 	else
255 		dev_queue_xmit(skb);
256 }
257 
258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259  * We don't protect the slave list iteration with a lock because:
260  * a. This operation is performed in IOCTL context,
261  * b. The operation is protected by the RTNL semaphore in the 8021q code,
262  * c. Holding a lock with BH disabled while directly calling a base driver
263  *    entry point is generally a BAD idea.
264  *
265  * The design of synchronization/protection for this operation in the 8021q
266  * module is good for one or more VLAN devices over a single physical device
267  * and cannot be extended for a teaming solution like bonding, so there is a
268  * potential race condition here where a net device from the vlan group might
269  * be referenced (either by a base driver or the 8021q code) while it is being
270  * removed from the system. However, it turns out we're not making matters
271  * worse, and if it works for regular VLAN usage it will work here too.
272 */
273 
274 /**
275  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276  * @bond_dev: bonding net device that got called
277  * @vid: vlan id being added
278  */
279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 				__be16 proto, u16 vid)
281 {
282 	struct bonding *bond = netdev_priv(bond_dev);
283 	struct slave *slave, *rollback_slave;
284 	struct list_head *iter;
285 	int res;
286 
287 	bond_for_each_slave(bond, slave, iter) {
288 		res = vlan_vid_add(slave->dev, proto, vid);
289 		if (res)
290 			goto unwind;
291 	}
292 
293 	return 0;
294 
295 unwind:
296 	/* unwind to the slave that failed */
297 	bond_for_each_slave(bond, rollback_slave, iter) {
298 		if (rollback_slave == slave)
299 			break;
300 
301 		vlan_vid_del(rollback_slave->dev, proto, vid);
302 	}
303 
304 	return res;
305 }
306 
307 /**
308  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309  * @bond_dev: bonding net device that got called
310  * @vid: vlan id being removed
311  */
312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 				 __be16 proto, u16 vid)
314 {
315 	struct bonding *bond = netdev_priv(bond_dev);
316 	struct list_head *iter;
317 	struct slave *slave;
318 
319 	bond_for_each_slave(bond, slave, iter)
320 		vlan_vid_del(slave->dev, proto, vid);
321 
322 	if (bond_is_lb(bond))
323 		bond_alb_clear_vlan(bond, vid);
324 
325 	return 0;
326 }
327 
328 /*------------------------------- Link status -------------------------------*/
329 
330 /* Set the carrier state for the master according to the state of its
331  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
332  * do special 802.3ad magic.
333  *
334  * Returns zero if carrier state does not change, nonzero if it does.
335  */
336 int bond_set_carrier(struct bonding *bond)
337 {
338 	struct list_head *iter;
339 	struct slave *slave;
340 
341 	if (!bond_has_slaves(bond))
342 		goto down;
343 
344 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 		return bond_3ad_set_carrier(bond);
346 
347 	bond_for_each_slave(bond, slave, iter) {
348 		if (slave->link == BOND_LINK_UP) {
349 			if (!netif_carrier_ok(bond->dev)) {
350 				netif_carrier_on(bond->dev);
351 				return 1;
352 			}
353 			return 0;
354 		}
355 	}
356 
357 down:
358 	if (netif_carrier_ok(bond->dev)) {
359 		netif_carrier_off(bond->dev);
360 		return 1;
361 	}
362 	return 0;
363 }
364 
365 /* Get link speed and duplex from the slave's base driver
366  * using ethtool. If for some reason the call fails or the
367  * values are invalid, set speed and duplex to -1,
368  * and return. Return 1 if speed or duplex settings are
369  * UNKNOWN; 0 otherwise.
370  */
371 static int bond_update_speed_duplex(struct slave *slave)
372 {
373 	struct net_device *slave_dev = slave->dev;
374 	struct ethtool_link_ksettings ecmd;
375 	int res;
376 
377 	slave->speed = SPEED_UNKNOWN;
378 	slave->duplex = DUPLEX_UNKNOWN;
379 
380 	res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
381 	if (res < 0)
382 		return 1;
383 	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
384 		return 1;
385 	switch (ecmd.base.duplex) {
386 	case DUPLEX_FULL:
387 	case DUPLEX_HALF:
388 		break;
389 	default:
390 		return 1;
391 	}
392 
393 	slave->speed = ecmd.base.speed;
394 	slave->duplex = ecmd.base.duplex;
395 
396 	return 0;
397 }
398 
399 const char *bond_slave_link_status(s8 link)
400 {
401 	switch (link) {
402 	case BOND_LINK_UP:
403 		return "up";
404 	case BOND_LINK_FAIL:
405 		return "going down";
406 	case BOND_LINK_DOWN:
407 		return "down";
408 	case BOND_LINK_BACK:
409 		return "going back";
410 	default:
411 		return "unknown";
412 	}
413 }
414 
415 /* if <dev> supports MII link status reporting, check its link status.
416  *
417  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418  * depending upon the setting of the use_carrier parameter.
419  *
420  * Return either BMSR_LSTATUS, meaning that the link is up (or we
421  * can't tell and just pretend it is), or 0, meaning that the link is
422  * down.
423  *
424  * If reporting is non-zero, instead of faking link up, return -1 if
425  * both ETHTOOL and MII ioctls fail (meaning the device does not
426  * support them).  If use_carrier is set, return whatever it says.
427  * It'd be nice if there was a good way to tell if a driver supports
428  * netif_carrier, but there really isn't.
429  */
430 static int bond_check_dev_link(struct bonding *bond,
431 			       struct net_device *slave_dev, int reporting)
432 {
433 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 	int (*ioctl)(struct net_device *, struct ifreq *, int);
435 	struct ifreq ifr;
436 	struct mii_ioctl_data *mii;
437 
438 	if (!reporting && !netif_running(slave_dev))
439 		return 0;
440 
441 	if (bond->params.use_carrier)
442 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
443 
444 	/* Try to get link status using Ethtool first. */
445 	if (slave_dev->ethtool_ops->get_link)
446 		return slave_dev->ethtool_ops->get_link(slave_dev) ?
447 			BMSR_LSTATUS : 0;
448 
449 	/* Ethtool can't be used, fallback to MII ioctls. */
450 	ioctl = slave_ops->ndo_do_ioctl;
451 	if (ioctl) {
452 		/* TODO: set pointer to correct ioctl on a per team member
453 		 *       bases to make this more efficient. that is, once
454 		 *       we determine the correct ioctl, we will always
455 		 *       call it and not the others for that team
456 		 *       member.
457 		 */
458 
459 		/* We cannot assume that SIOCGMIIPHY will also read a
460 		 * register; not all network drivers (e.g., e100)
461 		 * support that.
462 		 */
463 
464 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 		strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
466 		mii = if_mii(&ifr);
467 		if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 			mii->reg_num = MII_BMSR;
469 			if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 				return mii->val_out & BMSR_LSTATUS;
471 		}
472 	}
473 
474 	/* If reporting, report that either there's no dev->do_ioctl,
475 	 * or both SIOCGMIIREG and get_link failed (meaning that we
476 	 * cannot report link status).  If not reporting, pretend
477 	 * we're ok.
478 	 */
479 	return reporting ? -1 : BMSR_LSTATUS;
480 }
481 
482 /*----------------------------- Multicast list ------------------------------*/
483 
484 /* Push the promiscuity flag down to appropriate slaves */
485 static int bond_set_promiscuity(struct bonding *bond, int inc)
486 {
487 	struct list_head *iter;
488 	int err = 0;
489 
490 	if (bond_uses_primary(bond)) {
491 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
492 
493 		if (curr_active)
494 			err = dev_set_promiscuity(curr_active->dev, inc);
495 	} else {
496 		struct slave *slave;
497 
498 		bond_for_each_slave(bond, slave, iter) {
499 			err = dev_set_promiscuity(slave->dev, inc);
500 			if (err)
501 				return err;
502 		}
503 	}
504 	return err;
505 }
506 
507 /* Push the allmulti flag down to all slaves */
508 static int bond_set_allmulti(struct bonding *bond, int inc)
509 {
510 	struct list_head *iter;
511 	int err = 0;
512 
513 	if (bond_uses_primary(bond)) {
514 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
515 
516 		if (curr_active)
517 			err = dev_set_allmulti(curr_active->dev, inc);
518 	} else {
519 		struct slave *slave;
520 
521 		bond_for_each_slave(bond, slave, iter) {
522 			err = dev_set_allmulti(slave->dev, inc);
523 			if (err)
524 				return err;
525 		}
526 	}
527 	return err;
528 }
529 
530 /* Retrieve the list of registered multicast addresses for the bonding
531  * device and retransmit an IGMP JOIN request to the current active
532  * slave.
533  */
534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
535 {
536 	struct bonding *bond = container_of(work, struct bonding,
537 					    mcast_work.work);
538 
539 	if (!rtnl_trylock()) {
540 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
541 		return;
542 	}
543 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
544 
545 	if (bond->igmp_retrans > 1) {
546 		bond->igmp_retrans--;
547 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
548 	}
549 	rtnl_unlock();
550 }
551 
552 /* Flush bond's hardware addresses from slave */
553 static void bond_hw_addr_flush(struct net_device *bond_dev,
554 			       struct net_device *slave_dev)
555 {
556 	struct bonding *bond = netdev_priv(bond_dev);
557 
558 	dev_uc_unsync(slave_dev, bond_dev);
559 	dev_mc_unsync(slave_dev, bond_dev);
560 
561 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 		/* del lacpdu mc addr from mc list */
563 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
564 
565 		dev_mc_del(slave_dev, lacpdu_multicast);
566 	}
567 }
568 
569 /*--------------------------- Active slave change ---------------------------*/
570 
571 /* Update the hardware address list and promisc/allmulti for the new and
572  * old active slaves (if any).  Modes that are not using primary keep all
573  * slaves up date at all times; only the modes that use primary need to call
574  * this function to swap these settings during a failover.
575  */
576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 			      struct slave *old_active)
578 {
579 	if (old_active) {
580 		if (bond->dev->flags & IFF_PROMISC)
581 			dev_set_promiscuity(old_active->dev, -1);
582 
583 		if (bond->dev->flags & IFF_ALLMULTI)
584 			dev_set_allmulti(old_active->dev, -1);
585 
586 		bond_hw_addr_flush(bond->dev, old_active->dev);
587 	}
588 
589 	if (new_active) {
590 		/* FIXME: Signal errors upstream. */
591 		if (bond->dev->flags & IFF_PROMISC)
592 			dev_set_promiscuity(new_active->dev, 1);
593 
594 		if (bond->dev->flags & IFF_ALLMULTI)
595 			dev_set_allmulti(new_active->dev, 1);
596 
597 		netif_addr_lock_bh(bond->dev);
598 		dev_uc_sync(new_active->dev, bond->dev);
599 		dev_mc_sync(new_active->dev, bond->dev);
600 		netif_addr_unlock_bh(bond->dev);
601 	}
602 }
603 
604 /**
605  * bond_set_dev_addr - clone slave's address to bond
606  * @bond_dev: bond net device
607  * @slave_dev: slave net device
608  *
609  * Should be called with RTNL held.
610  */
611 static int bond_set_dev_addr(struct net_device *bond_dev,
612 			     struct net_device *slave_dev)
613 {
614 	int err;
615 
616 	slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
617 		  bond_dev, slave_dev, slave_dev->addr_len);
618 	err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
619 	if (err)
620 		return err;
621 
622 	memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
623 	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
624 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
625 	return 0;
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[MAX_ADDR_LEN];
656 	struct sockaddr_storage ss;
657 	int rv;
658 
659 	switch (bond->params.fail_over_mac) {
660 	case BOND_FOM_ACTIVE:
661 		if (new_active) {
662 			rv = bond_set_dev_addr(bond->dev, new_active->dev);
663 			if (rv)
664 				slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
665 					  -rv);
666 		}
667 		break;
668 	case BOND_FOM_FOLLOW:
669 		/* if new_active && old_active, swap them
670 		 * if just old_active, do nothing (going to no active slave)
671 		 * if just new_active, set new_active to bond's MAC
672 		 */
673 		if (!new_active)
674 			return;
675 
676 		if (!old_active)
677 			old_active = bond_get_old_active(bond, new_active);
678 
679 		if (old_active) {
680 			bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
681 					  new_active->dev->addr_len);
682 			bond_hw_addr_copy(ss.__data,
683 					  old_active->dev->dev_addr,
684 					  old_active->dev->addr_len);
685 			ss.ss_family = new_active->dev->type;
686 		} else {
687 			bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
688 					  bond->dev->addr_len);
689 			ss.ss_family = bond->dev->type;
690 		}
691 
692 		rv = dev_set_mac_address(new_active->dev,
693 					 (struct sockaddr *)&ss, NULL);
694 		if (rv) {
695 			slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
696 				  -rv);
697 			goto out;
698 		}
699 
700 		if (!old_active)
701 			goto out;
702 
703 		bond_hw_addr_copy(ss.__data, tmp_mac,
704 				  new_active->dev->addr_len);
705 		ss.ss_family = old_active->dev->type;
706 
707 		rv = dev_set_mac_address(old_active->dev,
708 					 (struct sockaddr *)&ss, NULL);
709 		if (rv)
710 			slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
711 				  -rv);
712 out:
713 		break;
714 	default:
715 		netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
716 			   bond->params.fail_over_mac);
717 		break;
718 	}
719 
720 }
721 
722 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
723 {
724 	struct slave *prim = rtnl_dereference(bond->primary_slave);
725 	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
726 
727 	if (!prim || prim->link != BOND_LINK_UP) {
728 		if (!curr || curr->link != BOND_LINK_UP)
729 			return NULL;
730 		return curr;
731 	}
732 
733 	if (bond->force_primary) {
734 		bond->force_primary = false;
735 		return prim;
736 	}
737 
738 	if (!curr || curr->link != BOND_LINK_UP)
739 		return prim;
740 
741 	/* At this point, prim and curr are both up */
742 	switch (bond->params.primary_reselect) {
743 	case BOND_PRI_RESELECT_ALWAYS:
744 		return prim;
745 	case BOND_PRI_RESELECT_BETTER:
746 		if (prim->speed < curr->speed)
747 			return curr;
748 		if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
749 			return curr;
750 		return prim;
751 	case BOND_PRI_RESELECT_FAILURE:
752 		return curr;
753 	default:
754 		netdev_err(bond->dev, "impossible primary_reselect %d\n",
755 			   bond->params.primary_reselect);
756 		return curr;
757 	}
758 }
759 
760 /**
761  * bond_find_best_slave - select the best available slave to be the active one
762  * @bond: our bonding struct
763  */
764 static struct slave *bond_find_best_slave(struct bonding *bond)
765 {
766 	struct slave *slave, *bestslave = NULL;
767 	struct list_head *iter;
768 	int mintime = bond->params.updelay;
769 
770 	slave = bond_choose_primary_or_current(bond);
771 	if (slave)
772 		return slave;
773 
774 	bond_for_each_slave(bond, slave, iter) {
775 		if (slave->link == BOND_LINK_UP)
776 			return slave;
777 		if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
778 		    slave->delay < mintime) {
779 			mintime = slave->delay;
780 			bestslave = slave;
781 		}
782 	}
783 
784 	return bestslave;
785 }
786 
787 static bool bond_should_notify_peers(struct bonding *bond)
788 {
789 	struct slave *slave;
790 
791 	rcu_read_lock();
792 	slave = rcu_dereference(bond->curr_active_slave);
793 	rcu_read_unlock();
794 
795 	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
796 		   slave ? slave->dev->name : "NULL");
797 
798 	if (!slave || !bond->send_peer_notif ||
799 	    bond->send_peer_notif %
800 	    max(1, bond->params.peer_notif_delay) != 0 ||
801 	    !netif_carrier_ok(bond->dev) ||
802 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
803 		return false;
804 
805 	return true;
806 }
807 
808 /**
809  * change_active_interface - change the active slave into the specified one
810  * @bond: our bonding struct
811  * @new: the new slave to make the active one
812  *
813  * Set the new slave to the bond's settings and unset them on the old
814  * curr_active_slave.
815  * Setting include flags, mc-list, promiscuity, allmulti, etc.
816  *
817  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
818  * because it is apparently the best available slave we have, even though its
819  * updelay hasn't timed out yet.
820  *
821  * Caller must hold RTNL.
822  */
823 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
824 {
825 	struct slave *old_active;
826 
827 	ASSERT_RTNL();
828 
829 	old_active = rtnl_dereference(bond->curr_active_slave);
830 
831 	if (old_active == new_active)
832 		return;
833 
834 	if (new_active) {
835 		new_active->last_link_up = jiffies;
836 
837 		if (new_active->link == BOND_LINK_BACK) {
838 			if (bond_uses_primary(bond)) {
839 				slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
840 					   (bond->params.updelay - new_active->delay) * bond->params.miimon);
841 			}
842 
843 			new_active->delay = 0;
844 			bond_set_slave_link_state(new_active, BOND_LINK_UP,
845 						  BOND_SLAVE_NOTIFY_NOW);
846 
847 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
848 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
849 
850 			if (bond_is_lb(bond))
851 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
852 		} else {
853 			if (bond_uses_primary(bond)) {
854 				slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
855 			}
856 		}
857 	}
858 
859 	if (bond_uses_primary(bond))
860 		bond_hw_addr_swap(bond, new_active, old_active);
861 
862 	if (bond_is_lb(bond)) {
863 		bond_alb_handle_active_change(bond, new_active);
864 		if (old_active)
865 			bond_set_slave_inactive_flags(old_active,
866 						      BOND_SLAVE_NOTIFY_NOW);
867 		if (new_active)
868 			bond_set_slave_active_flags(new_active,
869 						    BOND_SLAVE_NOTIFY_NOW);
870 	} else {
871 		rcu_assign_pointer(bond->curr_active_slave, new_active);
872 	}
873 
874 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
875 		if (old_active)
876 			bond_set_slave_inactive_flags(old_active,
877 						      BOND_SLAVE_NOTIFY_NOW);
878 
879 		if (new_active) {
880 			bool should_notify_peers = false;
881 
882 			bond_set_slave_active_flags(new_active,
883 						    BOND_SLAVE_NOTIFY_NOW);
884 
885 			if (bond->params.fail_over_mac)
886 				bond_do_fail_over_mac(bond, new_active,
887 						      old_active);
888 
889 			if (netif_running(bond->dev)) {
890 				bond->send_peer_notif =
891 					bond->params.num_peer_notif *
892 					max(1, bond->params.peer_notif_delay);
893 				should_notify_peers =
894 					bond_should_notify_peers(bond);
895 			}
896 
897 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
898 			if (should_notify_peers) {
899 				bond->send_peer_notif--;
900 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
901 							 bond->dev);
902 			}
903 		}
904 	}
905 
906 	/* resend IGMP joins since active slave has changed or
907 	 * all were sent on curr_active_slave.
908 	 * resend only if bond is brought up with the affected
909 	 * bonding modes and the retransmission is enabled
910 	 */
911 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
912 	    ((bond_uses_primary(bond) && new_active) ||
913 	     BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
914 		bond->igmp_retrans = bond->params.resend_igmp;
915 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
916 	}
917 }
918 
919 /**
920  * bond_select_active_slave - select a new active slave, if needed
921  * @bond: our bonding struct
922  *
923  * This functions should be called when one of the following occurs:
924  * - The old curr_active_slave has been released or lost its link.
925  * - The primary_slave has got its link back.
926  * - A slave has got its link back and there's no old curr_active_slave.
927  *
928  * Caller must hold RTNL.
929  */
930 void bond_select_active_slave(struct bonding *bond)
931 {
932 	struct slave *best_slave;
933 	int rv;
934 
935 	ASSERT_RTNL();
936 
937 	best_slave = bond_find_best_slave(bond);
938 	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
939 		bond_change_active_slave(bond, best_slave);
940 		rv = bond_set_carrier(bond);
941 		if (!rv)
942 			return;
943 
944 		if (netif_carrier_ok(bond->dev))
945 			netdev_info(bond->dev, "active interface up!\n");
946 		else
947 			netdev_info(bond->dev, "now running without any active interface!\n");
948 	}
949 }
950 
951 #ifdef CONFIG_NET_POLL_CONTROLLER
952 static inline int slave_enable_netpoll(struct slave *slave)
953 {
954 	struct netpoll *np;
955 	int err = 0;
956 
957 	np = kzalloc(sizeof(*np), GFP_KERNEL);
958 	err = -ENOMEM;
959 	if (!np)
960 		goto out;
961 
962 	err = __netpoll_setup(np, slave->dev);
963 	if (err) {
964 		kfree(np);
965 		goto out;
966 	}
967 	slave->np = np;
968 out:
969 	return err;
970 }
971 static inline void slave_disable_netpoll(struct slave *slave)
972 {
973 	struct netpoll *np = slave->np;
974 
975 	if (!np)
976 		return;
977 
978 	slave->np = NULL;
979 
980 	__netpoll_free(np);
981 }
982 
983 static void bond_poll_controller(struct net_device *bond_dev)
984 {
985 	struct bonding *bond = netdev_priv(bond_dev);
986 	struct slave *slave = NULL;
987 	struct list_head *iter;
988 	struct ad_info ad_info;
989 
990 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
991 		if (bond_3ad_get_active_agg_info(bond, &ad_info))
992 			return;
993 
994 	bond_for_each_slave_rcu(bond, slave, iter) {
995 		if (!bond_slave_is_up(slave))
996 			continue;
997 
998 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
999 			struct aggregator *agg =
1000 			    SLAVE_AD_INFO(slave)->port.aggregator;
1001 
1002 			if (agg &&
1003 			    agg->aggregator_identifier != ad_info.aggregator_id)
1004 				continue;
1005 		}
1006 
1007 		netpoll_poll_dev(slave->dev);
1008 	}
1009 }
1010 
1011 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1012 {
1013 	struct bonding *bond = netdev_priv(bond_dev);
1014 	struct list_head *iter;
1015 	struct slave *slave;
1016 
1017 	bond_for_each_slave(bond, slave, iter)
1018 		if (bond_slave_is_up(slave))
1019 			slave_disable_netpoll(slave);
1020 }
1021 
1022 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1023 {
1024 	struct bonding *bond = netdev_priv(dev);
1025 	struct list_head *iter;
1026 	struct slave *slave;
1027 	int err = 0;
1028 
1029 	bond_for_each_slave(bond, slave, iter) {
1030 		err = slave_enable_netpoll(slave);
1031 		if (err) {
1032 			bond_netpoll_cleanup(dev);
1033 			break;
1034 		}
1035 	}
1036 	return err;
1037 }
1038 #else
1039 static inline int slave_enable_netpoll(struct slave *slave)
1040 {
1041 	return 0;
1042 }
1043 static inline void slave_disable_netpoll(struct slave *slave)
1044 {
1045 }
1046 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1047 {
1048 }
1049 #endif
1050 
1051 /*---------------------------------- IOCTL ----------------------------------*/
1052 
1053 static netdev_features_t bond_fix_features(struct net_device *dev,
1054 					   netdev_features_t features)
1055 {
1056 	struct bonding *bond = netdev_priv(dev);
1057 	struct list_head *iter;
1058 	netdev_features_t mask;
1059 	struct slave *slave;
1060 
1061 	mask = features;
1062 
1063 	features &= ~NETIF_F_ONE_FOR_ALL;
1064 	features |= NETIF_F_ALL_FOR_ALL;
1065 
1066 	bond_for_each_slave(bond, slave, iter) {
1067 		features = netdev_increment_features(features,
1068 						     slave->dev->features,
1069 						     mask);
1070 	}
1071 	features = netdev_add_tso_features(features, mask);
1072 
1073 	return features;
1074 }
1075 
1076 #define BOND_VLAN_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1077 				 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1078 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1079 
1080 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1081 				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1082 
1083 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1084 				 NETIF_F_ALL_TSO)
1085 
1086 static void bond_compute_features(struct bonding *bond)
1087 {
1088 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1089 					IFF_XMIT_DST_RELEASE_PERM;
1090 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1091 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
1092 	netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1093 	struct net_device *bond_dev = bond->dev;
1094 	struct list_head *iter;
1095 	struct slave *slave;
1096 	unsigned short max_hard_header_len = ETH_HLEN;
1097 	unsigned int gso_max_size = GSO_MAX_SIZE;
1098 	u16 gso_max_segs = GSO_MAX_SEGS;
1099 
1100 	if (!bond_has_slaves(bond))
1101 		goto done;
1102 	vlan_features &= NETIF_F_ALL_FOR_ALL;
1103 	mpls_features &= NETIF_F_ALL_FOR_ALL;
1104 
1105 	bond_for_each_slave(bond, slave, iter) {
1106 		vlan_features = netdev_increment_features(vlan_features,
1107 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1108 
1109 		enc_features = netdev_increment_features(enc_features,
1110 							 slave->dev->hw_enc_features,
1111 							 BOND_ENC_FEATURES);
1112 
1113 		mpls_features = netdev_increment_features(mpls_features,
1114 							  slave->dev->mpls_features,
1115 							  BOND_MPLS_FEATURES);
1116 
1117 		dst_release_flag &= slave->dev->priv_flags;
1118 		if (slave->dev->hard_header_len > max_hard_header_len)
1119 			max_hard_header_len = slave->dev->hard_header_len;
1120 
1121 		gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1122 		gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1123 	}
1124 	bond_dev->hard_header_len = max_hard_header_len;
1125 
1126 done:
1127 	bond_dev->vlan_features = vlan_features;
1128 	bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1129 				    NETIF_F_HW_VLAN_CTAG_TX |
1130 				    NETIF_F_HW_VLAN_STAG_TX |
1131 				    NETIF_F_GSO_UDP_L4;
1132 	bond_dev->mpls_features = mpls_features;
1133 	bond_dev->gso_max_segs = gso_max_segs;
1134 	netif_set_gso_max_size(bond_dev, gso_max_size);
1135 
1136 	bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1137 	if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1138 	    dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1139 		bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1140 
1141 	netdev_change_features(bond_dev);
1142 }
1143 
1144 static void bond_setup_by_slave(struct net_device *bond_dev,
1145 				struct net_device *slave_dev)
1146 {
1147 	bond_dev->header_ops	    = slave_dev->header_ops;
1148 
1149 	bond_dev->type		    = slave_dev->type;
1150 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1151 	bond_dev->addr_len	    = slave_dev->addr_len;
1152 
1153 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1154 		slave_dev->addr_len);
1155 }
1156 
1157 /* On bonding slaves other than the currently active slave, suppress
1158  * duplicates except for alb non-mcast/bcast.
1159  */
1160 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1161 					    struct slave *slave,
1162 					    struct bonding *bond)
1163 {
1164 	if (bond_is_slave_inactive(slave)) {
1165 		if (BOND_MODE(bond) == BOND_MODE_ALB &&
1166 		    skb->pkt_type != PACKET_BROADCAST &&
1167 		    skb->pkt_type != PACKET_MULTICAST)
1168 			return false;
1169 		return true;
1170 	}
1171 	return false;
1172 }
1173 
1174 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1175 {
1176 	struct sk_buff *skb = *pskb;
1177 	struct slave *slave;
1178 	struct bonding *bond;
1179 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1180 			  struct slave *);
1181 	int ret = RX_HANDLER_ANOTHER;
1182 
1183 	skb = skb_share_check(skb, GFP_ATOMIC);
1184 	if (unlikely(!skb))
1185 		return RX_HANDLER_CONSUMED;
1186 
1187 	*pskb = skb;
1188 
1189 	slave = bond_slave_get_rcu(skb->dev);
1190 	bond = slave->bond;
1191 
1192 	recv_probe = READ_ONCE(bond->recv_probe);
1193 	if (recv_probe) {
1194 		ret = recv_probe(skb, bond, slave);
1195 		if (ret == RX_HANDLER_CONSUMED) {
1196 			consume_skb(skb);
1197 			return ret;
1198 		}
1199 	}
1200 
1201 	/*
1202 	 * For packets determined by bond_should_deliver_exact_match() call to
1203 	 * be suppressed we want to make an exception for link-local packets.
1204 	 * This is necessary for e.g. LLDP daemons to be able to monitor
1205 	 * inactive slave links without being forced to bind to them
1206 	 * explicitly.
1207 	 *
1208 	 * At the same time, packets that are passed to the bonding master
1209 	 * (including link-local ones) can have their originating interface
1210 	 * determined via PACKET_ORIGDEV socket option.
1211 	 */
1212 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1213 		if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1214 			return RX_HANDLER_PASS;
1215 		return RX_HANDLER_EXACT;
1216 	}
1217 
1218 	skb->dev = bond->dev;
1219 
1220 	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1221 	    bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1222 	    skb->pkt_type == PACKET_HOST) {
1223 
1224 		if (unlikely(skb_cow_head(skb,
1225 					  skb->data - skb_mac_header(skb)))) {
1226 			kfree_skb(skb);
1227 			return RX_HANDLER_CONSUMED;
1228 		}
1229 		bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1230 				  bond->dev->addr_len);
1231 	}
1232 
1233 	return ret;
1234 }
1235 
1236 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1237 {
1238 	switch (BOND_MODE(bond)) {
1239 	case BOND_MODE_ROUNDROBIN:
1240 		return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1241 	case BOND_MODE_ACTIVEBACKUP:
1242 		return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1243 	case BOND_MODE_BROADCAST:
1244 		return NETDEV_LAG_TX_TYPE_BROADCAST;
1245 	case BOND_MODE_XOR:
1246 	case BOND_MODE_8023AD:
1247 		return NETDEV_LAG_TX_TYPE_HASH;
1248 	default:
1249 		return NETDEV_LAG_TX_TYPE_UNKNOWN;
1250 	}
1251 }
1252 
1253 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1254 					       enum netdev_lag_tx_type type)
1255 {
1256 	if (type != NETDEV_LAG_TX_TYPE_HASH)
1257 		return NETDEV_LAG_HASH_NONE;
1258 
1259 	switch (bond->params.xmit_policy) {
1260 	case BOND_XMIT_POLICY_LAYER2:
1261 		return NETDEV_LAG_HASH_L2;
1262 	case BOND_XMIT_POLICY_LAYER34:
1263 		return NETDEV_LAG_HASH_L34;
1264 	case BOND_XMIT_POLICY_LAYER23:
1265 		return NETDEV_LAG_HASH_L23;
1266 	case BOND_XMIT_POLICY_ENCAP23:
1267 		return NETDEV_LAG_HASH_E23;
1268 	case BOND_XMIT_POLICY_ENCAP34:
1269 		return NETDEV_LAG_HASH_E34;
1270 	default:
1271 		return NETDEV_LAG_HASH_UNKNOWN;
1272 	}
1273 }
1274 
1275 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1276 				      struct netlink_ext_ack *extack)
1277 {
1278 	struct netdev_lag_upper_info lag_upper_info;
1279 	enum netdev_lag_tx_type type;
1280 
1281 	type = bond_lag_tx_type(bond);
1282 	lag_upper_info.tx_type = type;
1283 	lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1284 
1285 	return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1286 					    &lag_upper_info, extack);
1287 }
1288 
1289 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1290 {
1291 	netdev_upper_dev_unlink(slave->dev, bond->dev);
1292 	slave->dev->flags &= ~IFF_SLAVE;
1293 }
1294 
1295 static struct slave *bond_alloc_slave(struct bonding *bond)
1296 {
1297 	struct slave *slave = NULL;
1298 
1299 	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1300 	if (!slave)
1301 		return NULL;
1302 
1303 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1304 		SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1305 					       GFP_KERNEL);
1306 		if (!SLAVE_AD_INFO(slave)) {
1307 			kfree(slave);
1308 			return NULL;
1309 		}
1310 	}
1311 	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1312 
1313 	return slave;
1314 }
1315 
1316 static void bond_free_slave(struct slave *slave)
1317 {
1318 	struct bonding *bond = bond_get_bond_by_slave(slave);
1319 
1320 	cancel_delayed_work_sync(&slave->notify_work);
1321 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1322 		kfree(SLAVE_AD_INFO(slave));
1323 
1324 	kfree(slave);
1325 }
1326 
1327 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1328 {
1329 	info->bond_mode = BOND_MODE(bond);
1330 	info->miimon = bond->params.miimon;
1331 	info->num_slaves = bond->slave_cnt;
1332 }
1333 
1334 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1335 {
1336 	strcpy(info->slave_name, slave->dev->name);
1337 	info->link = slave->link;
1338 	info->state = bond_slave_state(slave);
1339 	info->link_failure_count = slave->link_failure_count;
1340 }
1341 
1342 static void bond_netdev_notify_work(struct work_struct *_work)
1343 {
1344 	struct slave *slave = container_of(_work, struct slave,
1345 					   notify_work.work);
1346 
1347 	if (rtnl_trylock()) {
1348 		struct netdev_bonding_info binfo;
1349 
1350 		bond_fill_ifslave(slave, &binfo.slave);
1351 		bond_fill_ifbond(slave->bond, &binfo.master);
1352 		netdev_bonding_info_change(slave->dev, &binfo);
1353 		rtnl_unlock();
1354 	} else {
1355 		queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1356 	}
1357 }
1358 
1359 void bond_queue_slave_event(struct slave *slave)
1360 {
1361 	queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1362 }
1363 
1364 void bond_lower_state_changed(struct slave *slave)
1365 {
1366 	struct netdev_lag_lower_state_info info;
1367 
1368 	info.link_up = slave->link == BOND_LINK_UP ||
1369 		       slave->link == BOND_LINK_FAIL;
1370 	info.tx_enabled = bond_is_active_slave(slave);
1371 	netdev_lower_state_changed(slave->dev, &info);
1372 }
1373 
1374 /* enslave device <slave> to bond device <master> */
1375 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1376 		 struct netlink_ext_ack *extack)
1377 {
1378 	struct bonding *bond = netdev_priv(bond_dev);
1379 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1380 	struct slave *new_slave = NULL, *prev_slave;
1381 	struct sockaddr_storage ss;
1382 	int link_reporting;
1383 	int res = 0, i;
1384 
1385 	if (!bond->params.use_carrier &&
1386 	    slave_dev->ethtool_ops->get_link == NULL &&
1387 	    slave_ops->ndo_do_ioctl == NULL) {
1388 		slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1389 	}
1390 
1391 	/* already in-use? */
1392 	if (netdev_is_rx_handler_busy(slave_dev)) {
1393 		NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1394 		slave_err(bond_dev, slave_dev,
1395 			  "Error: Device is in use and cannot be enslaved\n");
1396 		return -EBUSY;
1397 	}
1398 
1399 	if (bond_dev == slave_dev) {
1400 		NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1401 		netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1402 		return -EPERM;
1403 	}
1404 
1405 	/* vlan challenged mutual exclusion */
1406 	/* no need to lock since we're protected by rtnl_lock */
1407 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1408 		slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1409 		if (vlan_uses_dev(bond_dev)) {
1410 			NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1411 			slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1412 			return -EPERM;
1413 		} else {
1414 			slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1415 		}
1416 	} else {
1417 		slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1418 	}
1419 
1420 	/* Old ifenslave binaries are no longer supported.  These can
1421 	 * be identified with moderate accuracy by the state of the slave:
1422 	 * the current ifenslave will set the interface down prior to
1423 	 * enslaving it; the old ifenslave will not.
1424 	 */
1425 	if (slave_dev->flags & IFF_UP) {
1426 		NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1427 		slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1428 		return -EPERM;
1429 	}
1430 
1431 	/* set bonding device ether type by slave - bonding netdevices are
1432 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1433 	 * there is a need to override some of the type dependent attribs/funcs.
1434 	 *
1435 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1436 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1437 	 */
1438 	if (!bond_has_slaves(bond)) {
1439 		if (bond_dev->type != slave_dev->type) {
1440 			slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1441 				  bond_dev->type, slave_dev->type);
1442 
1443 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1444 						       bond_dev);
1445 			res = notifier_to_errno(res);
1446 			if (res) {
1447 				slave_err(bond_dev, slave_dev, "refused to change device type\n");
1448 				return -EBUSY;
1449 			}
1450 
1451 			/* Flush unicast and multicast addresses */
1452 			dev_uc_flush(bond_dev);
1453 			dev_mc_flush(bond_dev);
1454 
1455 			if (slave_dev->type != ARPHRD_ETHER)
1456 				bond_setup_by_slave(bond_dev, slave_dev);
1457 			else {
1458 				ether_setup(bond_dev);
1459 				bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1460 			}
1461 
1462 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1463 						 bond_dev);
1464 		}
1465 	} else if (bond_dev->type != slave_dev->type) {
1466 		NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1467 		slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1468 			  slave_dev->type, bond_dev->type);
1469 		return -EINVAL;
1470 	}
1471 
1472 	if (slave_dev->type == ARPHRD_INFINIBAND &&
1473 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1474 		NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1475 		slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1476 			   slave_dev->type);
1477 		res = -EOPNOTSUPP;
1478 		goto err_undo_flags;
1479 	}
1480 
1481 	if (!slave_ops->ndo_set_mac_address ||
1482 	    slave_dev->type == ARPHRD_INFINIBAND) {
1483 		slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1484 		if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1485 		    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1486 			if (!bond_has_slaves(bond)) {
1487 				bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1488 				slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1489 			} else {
1490 				NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1491 				slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1492 				res = -EOPNOTSUPP;
1493 				goto err_undo_flags;
1494 			}
1495 		}
1496 	}
1497 
1498 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1499 
1500 	/* If this is the first slave, then we need to set the master's hardware
1501 	 * address to be the same as the slave's.
1502 	 */
1503 	if (!bond_has_slaves(bond) &&
1504 	    bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1505 		res = bond_set_dev_addr(bond->dev, slave_dev);
1506 		if (res)
1507 			goto err_undo_flags;
1508 	}
1509 
1510 	new_slave = bond_alloc_slave(bond);
1511 	if (!new_slave) {
1512 		res = -ENOMEM;
1513 		goto err_undo_flags;
1514 	}
1515 
1516 	new_slave->bond = bond;
1517 	new_slave->dev = slave_dev;
1518 	/* Set the new_slave's queue_id to be zero.  Queue ID mapping
1519 	 * is set via sysfs or module option if desired.
1520 	 */
1521 	new_slave->queue_id = 0;
1522 
1523 	/* Save slave's original mtu and then set it to match the bond */
1524 	new_slave->original_mtu = slave_dev->mtu;
1525 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1526 	if (res) {
1527 		slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1528 		goto err_free;
1529 	}
1530 
1531 	/* Save slave's original ("permanent") mac address for modes
1532 	 * that need it, and for restoring it upon release, and then
1533 	 * set it to the master's address
1534 	 */
1535 	bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1536 			  slave_dev->addr_len);
1537 
1538 	if (!bond->params.fail_over_mac ||
1539 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1540 		/* Set slave to master's mac address.  The application already
1541 		 * set the master's mac address to that of the first slave
1542 		 */
1543 		memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1544 		ss.ss_family = slave_dev->type;
1545 		res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1546 					  extack);
1547 		if (res) {
1548 			slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1549 			goto err_restore_mtu;
1550 		}
1551 	}
1552 
1553 	/* set slave flag before open to prevent IPv6 addrconf */
1554 	slave_dev->flags |= IFF_SLAVE;
1555 
1556 	/* open the slave since the application closed it */
1557 	res = dev_open(slave_dev, extack);
1558 	if (res) {
1559 		slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1560 		goto err_restore_mac;
1561 	}
1562 
1563 	slave_dev->priv_flags |= IFF_BONDING;
1564 	/* initialize slave stats */
1565 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1566 
1567 	if (bond_is_lb(bond)) {
1568 		/* bond_alb_init_slave() must be called before all other stages since
1569 		 * it might fail and we do not want to have to undo everything
1570 		 */
1571 		res = bond_alb_init_slave(bond, new_slave);
1572 		if (res)
1573 			goto err_close;
1574 	}
1575 
1576 	res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1577 	if (res) {
1578 		slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1579 		goto err_close;
1580 	}
1581 
1582 	prev_slave = bond_last_slave(bond);
1583 
1584 	new_slave->delay = 0;
1585 	new_slave->link_failure_count = 0;
1586 
1587 	if (bond_update_speed_duplex(new_slave) &&
1588 	    bond_needs_speed_duplex(bond))
1589 		new_slave->link = BOND_LINK_DOWN;
1590 
1591 	new_slave->last_rx = jiffies -
1592 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
1593 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1594 		new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1595 
1596 	if (bond->params.miimon && !bond->params.use_carrier) {
1597 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1598 
1599 		if ((link_reporting == -1) && !bond->params.arp_interval) {
1600 			/* miimon is set but a bonded network driver
1601 			 * does not support ETHTOOL/MII and
1602 			 * arp_interval is not set.  Note: if
1603 			 * use_carrier is enabled, we will never go
1604 			 * here (because netif_carrier is always
1605 			 * supported); thus, we don't need to change
1606 			 * the messages for netif_carrier.
1607 			 */
1608 			slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1609 		} else if (link_reporting == -1) {
1610 			/* unable get link status using mii/ethtool */
1611 			slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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");
1612 		}
1613 	}
1614 
1615 	/* check for initial state */
1616 	new_slave->link = BOND_LINK_NOCHANGE;
1617 	if (bond->params.miimon) {
1618 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1619 			if (bond->params.updelay) {
1620 				bond_set_slave_link_state(new_slave,
1621 							  BOND_LINK_BACK,
1622 							  BOND_SLAVE_NOTIFY_NOW);
1623 				new_slave->delay = bond->params.updelay;
1624 			} else {
1625 				bond_set_slave_link_state(new_slave,
1626 							  BOND_LINK_UP,
1627 							  BOND_SLAVE_NOTIFY_NOW);
1628 			}
1629 		} else {
1630 			bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1631 						  BOND_SLAVE_NOTIFY_NOW);
1632 		}
1633 	} else if (bond->params.arp_interval) {
1634 		bond_set_slave_link_state(new_slave,
1635 					  (netif_carrier_ok(slave_dev) ?
1636 					  BOND_LINK_UP : BOND_LINK_DOWN),
1637 					  BOND_SLAVE_NOTIFY_NOW);
1638 	} else {
1639 		bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1640 					  BOND_SLAVE_NOTIFY_NOW);
1641 	}
1642 
1643 	if (new_slave->link != BOND_LINK_DOWN)
1644 		new_slave->last_link_up = jiffies;
1645 	slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1646 		  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1647 		  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1648 
1649 	if (bond_uses_primary(bond) && bond->params.primary[0]) {
1650 		/* if there is a primary slave, remember it */
1651 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1652 			rcu_assign_pointer(bond->primary_slave, new_slave);
1653 			bond->force_primary = true;
1654 		}
1655 	}
1656 
1657 	switch (BOND_MODE(bond)) {
1658 	case BOND_MODE_ACTIVEBACKUP:
1659 		bond_set_slave_inactive_flags(new_slave,
1660 					      BOND_SLAVE_NOTIFY_NOW);
1661 		break;
1662 	case BOND_MODE_8023AD:
1663 		/* in 802.3ad mode, the internal mechanism
1664 		 * will activate the slaves in the selected
1665 		 * aggregator
1666 		 */
1667 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1668 		/* if this is the first slave */
1669 		if (!prev_slave) {
1670 			SLAVE_AD_INFO(new_slave)->id = 1;
1671 			/* Initialize AD with the number of times that the AD timer is called in 1 second
1672 			 * can be called only after the mac address of the bond is set
1673 			 */
1674 			bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1675 		} else {
1676 			SLAVE_AD_INFO(new_slave)->id =
1677 				SLAVE_AD_INFO(prev_slave)->id + 1;
1678 		}
1679 
1680 		bond_3ad_bind_slave(new_slave);
1681 		break;
1682 	case BOND_MODE_TLB:
1683 	case BOND_MODE_ALB:
1684 		bond_set_active_slave(new_slave);
1685 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1686 		break;
1687 	default:
1688 		slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1689 
1690 		/* always active in trunk mode */
1691 		bond_set_active_slave(new_slave);
1692 
1693 		/* In trunking mode there is little meaning to curr_active_slave
1694 		 * anyway (it holds no special properties of the bond device),
1695 		 * so we can change it without calling change_active_interface()
1696 		 */
1697 		if (!rcu_access_pointer(bond->curr_active_slave) &&
1698 		    new_slave->link == BOND_LINK_UP)
1699 			rcu_assign_pointer(bond->curr_active_slave, new_slave);
1700 
1701 		break;
1702 	} /* switch(bond_mode) */
1703 
1704 #ifdef CONFIG_NET_POLL_CONTROLLER
1705 	if (bond->dev->npinfo) {
1706 		if (slave_enable_netpoll(new_slave)) {
1707 			slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1708 			res = -EBUSY;
1709 			goto err_detach;
1710 		}
1711 	}
1712 #endif
1713 
1714 	if (!(bond_dev->features & NETIF_F_LRO))
1715 		dev_disable_lro(slave_dev);
1716 
1717 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1718 					 new_slave);
1719 	if (res) {
1720 		slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1721 		goto err_detach;
1722 	}
1723 
1724 	res = bond_master_upper_dev_link(bond, new_slave, extack);
1725 	if (res) {
1726 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1727 		goto err_unregister;
1728 	}
1729 
1730 	res = bond_sysfs_slave_add(new_slave);
1731 	if (res) {
1732 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1733 		goto err_upper_unlink;
1734 	}
1735 
1736 	bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1737 
1738 	/* If the mode uses primary, then the following is handled by
1739 	 * bond_change_active_slave().
1740 	 */
1741 	if (!bond_uses_primary(bond)) {
1742 		/* set promiscuity level to new slave */
1743 		if (bond_dev->flags & IFF_PROMISC) {
1744 			res = dev_set_promiscuity(slave_dev, 1);
1745 			if (res)
1746 				goto err_sysfs_del;
1747 		}
1748 
1749 		/* set allmulti level to new slave */
1750 		if (bond_dev->flags & IFF_ALLMULTI) {
1751 			res = dev_set_allmulti(slave_dev, 1);
1752 			if (res) {
1753 				if (bond_dev->flags & IFF_PROMISC)
1754 					dev_set_promiscuity(slave_dev, -1);
1755 				goto err_sysfs_del;
1756 			}
1757 		}
1758 
1759 		netif_addr_lock_bh(bond_dev);
1760 		dev_mc_sync_multiple(slave_dev, bond_dev);
1761 		dev_uc_sync_multiple(slave_dev, bond_dev);
1762 		netif_addr_unlock_bh(bond_dev);
1763 
1764 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1765 			/* add lacpdu mc addr to mc list */
1766 			u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1767 
1768 			dev_mc_add(slave_dev, lacpdu_multicast);
1769 		}
1770 	}
1771 
1772 	bond->slave_cnt++;
1773 	bond_compute_features(bond);
1774 	bond_set_carrier(bond);
1775 
1776 	if (bond_uses_primary(bond)) {
1777 		block_netpoll_tx();
1778 		bond_select_active_slave(bond);
1779 		unblock_netpoll_tx();
1780 	}
1781 
1782 	if (bond_mode_can_use_xmit_hash(bond))
1783 		bond_update_slave_arr(bond, NULL);
1784 
1785 
1786 	slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1787 		   bond_is_active_slave(new_slave) ? "an active" : "a backup",
1788 		   new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1789 
1790 	/* enslave is successful */
1791 	bond_queue_slave_event(new_slave);
1792 	return 0;
1793 
1794 /* Undo stages on error */
1795 err_sysfs_del:
1796 	bond_sysfs_slave_del(new_slave);
1797 
1798 err_upper_unlink:
1799 	bond_upper_dev_unlink(bond, new_slave);
1800 
1801 err_unregister:
1802 	netdev_rx_handler_unregister(slave_dev);
1803 
1804 err_detach:
1805 	vlan_vids_del_by_dev(slave_dev, bond_dev);
1806 	if (rcu_access_pointer(bond->primary_slave) == new_slave)
1807 		RCU_INIT_POINTER(bond->primary_slave, NULL);
1808 	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1809 		block_netpoll_tx();
1810 		bond_change_active_slave(bond, NULL);
1811 		bond_select_active_slave(bond);
1812 		unblock_netpoll_tx();
1813 	}
1814 	/* either primary_slave or curr_active_slave might've changed */
1815 	synchronize_rcu();
1816 	slave_disable_netpoll(new_slave);
1817 
1818 err_close:
1819 	slave_dev->priv_flags &= ~IFF_BONDING;
1820 	dev_close(slave_dev);
1821 
1822 err_restore_mac:
1823 	slave_dev->flags &= ~IFF_SLAVE;
1824 	if (!bond->params.fail_over_mac ||
1825 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1826 		/* XXX TODO - fom follow mode needs to change master's
1827 		 * MAC if this slave's MAC is in use by the bond, or at
1828 		 * least print a warning.
1829 		 */
1830 		bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1831 				  new_slave->dev->addr_len);
1832 		ss.ss_family = slave_dev->type;
1833 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1834 	}
1835 
1836 err_restore_mtu:
1837 	dev_set_mtu(slave_dev, new_slave->original_mtu);
1838 
1839 err_free:
1840 	bond_free_slave(new_slave);
1841 
1842 err_undo_flags:
1843 	/* Enslave of first slave has failed and we need to fix master's mac */
1844 	if (!bond_has_slaves(bond)) {
1845 		if (ether_addr_equal_64bits(bond_dev->dev_addr,
1846 					    slave_dev->dev_addr))
1847 			eth_hw_addr_random(bond_dev);
1848 		if (bond_dev->type != ARPHRD_ETHER) {
1849 			dev_close(bond_dev);
1850 			ether_setup(bond_dev);
1851 			bond_dev->flags |= IFF_MASTER;
1852 			bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1853 		}
1854 	}
1855 
1856 	return res;
1857 }
1858 
1859 /* Try to release the slave device <slave> from the bond device <master>
1860  * It is legal to access curr_active_slave without a lock because all the function
1861  * is RTNL-locked. If "all" is true it means that the function is being called
1862  * while destroying a bond interface and all slaves are being released.
1863  *
1864  * The rules for slave state should be:
1865  *   for Active/Backup:
1866  *     Active stays on all backups go down
1867  *   for Bonded connections:
1868  *     The first up interface should be left on and all others downed.
1869  */
1870 static int __bond_release_one(struct net_device *bond_dev,
1871 			      struct net_device *slave_dev,
1872 			      bool all, bool unregister)
1873 {
1874 	struct bonding *bond = netdev_priv(bond_dev);
1875 	struct slave *slave, *oldcurrent;
1876 	struct sockaddr_storage ss;
1877 	int old_flags = bond_dev->flags;
1878 	netdev_features_t old_features = bond_dev->features;
1879 
1880 	/* slave is not a slave or master is not master of this slave */
1881 	if (!(slave_dev->flags & IFF_SLAVE) ||
1882 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
1883 		slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
1884 		return -EINVAL;
1885 	}
1886 
1887 	block_netpoll_tx();
1888 
1889 	slave = bond_get_slave_by_dev(bond, slave_dev);
1890 	if (!slave) {
1891 		/* not a slave of this bond */
1892 		slave_info(bond_dev, slave_dev, "interface not enslaved\n");
1893 		unblock_netpoll_tx();
1894 		return -EINVAL;
1895 	}
1896 
1897 	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1898 
1899 	bond_sysfs_slave_del(slave);
1900 
1901 	/* recompute stats just before removing the slave */
1902 	bond_get_stats(bond->dev, &bond->bond_stats);
1903 
1904 	bond_upper_dev_unlink(bond, slave);
1905 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
1906 	 * for this slave anymore.
1907 	 */
1908 	netdev_rx_handler_unregister(slave_dev);
1909 
1910 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1911 		bond_3ad_unbind_slave(slave);
1912 
1913 	if (bond_mode_can_use_xmit_hash(bond))
1914 		bond_update_slave_arr(bond, slave);
1915 
1916 	slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
1917 		    bond_is_active_slave(slave) ? "active" : "backup");
1918 
1919 	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1920 
1921 	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1922 
1923 	if (!all && (!bond->params.fail_over_mac ||
1924 		     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1925 		if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1926 		    bond_has_slaves(bond))
1927 			slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
1928 				   slave->perm_hwaddr);
1929 	}
1930 
1931 	if (rtnl_dereference(bond->primary_slave) == slave)
1932 		RCU_INIT_POINTER(bond->primary_slave, NULL);
1933 
1934 	if (oldcurrent == slave)
1935 		bond_change_active_slave(bond, NULL);
1936 
1937 	if (bond_is_lb(bond)) {
1938 		/* Must be called only after the slave has been
1939 		 * detached from the list and the curr_active_slave
1940 		 * has been cleared (if our_slave == old_current),
1941 		 * but before a new active slave is selected.
1942 		 */
1943 		bond_alb_deinit_slave(bond, slave);
1944 	}
1945 
1946 	if (all) {
1947 		RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1948 	} else if (oldcurrent == slave) {
1949 		/* Note that we hold RTNL over this sequence, so there
1950 		 * is no concern that another slave add/remove event
1951 		 * will interfere.
1952 		 */
1953 		bond_select_active_slave(bond);
1954 	}
1955 
1956 	if (!bond_has_slaves(bond)) {
1957 		bond_set_carrier(bond);
1958 		eth_hw_addr_random(bond_dev);
1959 		bond->nest_level = SINGLE_DEPTH_NESTING;
1960 	} else {
1961 		bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1962 	}
1963 
1964 	unblock_netpoll_tx();
1965 	synchronize_rcu();
1966 	bond->slave_cnt--;
1967 
1968 	if (!bond_has_slaves(bond)) {
1969 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1970 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1971 	}
1972 
1973 	bond_compute_features(bond);
1974 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1975 	    (old_features & NETIF_F_VLAN_CHALLENGED))
1976 		slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
1977 
1978 	vlan_vids_del_by_dev(slave_dev, bond_dev);
1979 
1980 	/* If the mode uses primary, then this case was handled above by
1981 	 * bond_change_active_slave(..., NULL)
1982 	 */
1983 	if (!bond_uses_primary(bond)) {
1984 		/* unset promiscuity level from slave
1985 		 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1986 		 * of the IFF_PROMISC flag in the bond_dev, but we need the
1987 		 * value of that flag before that change, as that was the value
1988 		 * when this slave was attached, so we cache at the start of the
1989 		 * function and use it here. Same goes for ALLMULTI below
1990 		 */
1991 		if (old_flags & IFF_PROMISC)
1992 			dev_set_promiscuity(slave_dev, -1);
1993 
1994 		/* unset allmulti level from slave */
1995 		if (old_flags & IFF_ALLMULTI)
1996 			dev_set_allmulti(slave_dev, -1);
1997 
1998 		bond_hw_addr_flush(bond_dev, slave_dev);
1999 	}
2000 
2001 	slave_disable_netpoll(slave);
2002 
2003 	/* close slave before restoring its mac address */
2004 	dev_close(slave_dev);
2005 
2006 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2007 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2008 		/* restore original ("permanent") mac address */
2009 		bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2010 				  slave->dev->addr_len);
2011 		ss.ss_family = slave_dev->type;
2012 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2013 	}
2014 
2015 	if (unregister)
2016 		__dev_set_mtu(slave_dev, slave->original_mtu);
2017 	else
2018 		dev_set_mtu(slave_dev, slave->original_mtu);
2019 
2020 	slave_dev->priv_flags &= ~IFF_BONDING;
2021 
2022 	bond_free_slave(slave);
2023 
2024 	return 0;
2025 }
2026 
2027 /* A wrapper used because of ndo_del_link */
2028 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2029 {
2030 	return __bond_release_one(bond_dev, slave_dev, false, false);
2031 }
2032 
2033 /* First release a slave and then destroy the bond if no more slaves are left.
2034  * Must be under rtnl_lock when this function is called.
2035  */
2036 static int bond_release_and_destroy(struct net_device *bond_dev,
2037 				    struct net_device *slave_dev)
2038 {
2039 	struct bonding *bond = netdev_priv(bond_dev);
2040 	int ret;
2041 
2042 	ret = __bond_release_one(bond_dev, slave_dev, false, true);
2043 	if (ret == 0 && !bond_has_slaves(bond)) {
2044 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2045 		netdev_info(bond_dev, "Destroying bond\n");
2046 		bond_remove_proc_entry(bond);
2047 		unregister_netdevice(bond_dev);
2048 	}
2049 	return ret;
2050 }
2051 
2052 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2053 {
2054 	struct bonding *bond = netdev_priv(bond_dev);
2055 	bond_fill_ifbond(bond, info);
2056 }
2057 
2058 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2059 {
2060 	struct bonding *bond = netdev_priv(bond_dev);
2061 	struct list_head *iter;
2062 	int i = 0, res = -ENODEV;
2063 	struct slave *slave;
2064 
2065 	bond_for_each_slave(bond, slave, iter) {
2066 		if (i++ == (int)info->slave_id) {
2067 			res = 0;
2068 			bond_fill_ifslave(slave, info);
2069 			break;
2070 		}
2071 	}
2072 
2073 	return res;
2074 }
2075 
2076 /*-------------------------------- Monitoring -------------------------------*/
2077 
2078 /* called with rcu_read_lock() */
2079 static int bond_miimon_inspect(struct bonding *bond)
2080 {
2081 	int link_state, commit = 0;
2082 	struct list_head *iter;
2083 	struct slave *slave;
2084 	bool ignore_updelay;
2085 
2086 	ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2087 
2088 	bond_for_each_slave_rcu(bond, slave, iter) {
2089 		slave->new_link = BOND_LINK_NOCHANGE;
2090 		slave->link_new_state = slave->link;
2091 
2092 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2093 
2094 		switch (slave->link) {
2095 		case BOND_LINK_UP:
2096 			if (link_state)
2097 				continue;
2098 
2099 			bond_propose_link_state(slave, BOND_LINK_FAIL);
2100 			commit++;
2101 			slave->delay = bond->params.downdelay;
2102 			if (slave->delay) {
2103 				slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2104 					   (BOND_MODE(bond) ==
2105 					    BOND_MODE_ACTIVEBACKUP) ?
2106 					    (bond_is_active_slave(slave) ?
2107 					     "active " : "backup ") : "",
2108 					   bond->params.downdelay * bond->params.miimon);
2109 			}
2110 			/*FALLTHRU*/
2111 		case BOND_LINK_FAIL:
2112 			if (link_state) {
2113 				/* recovered before downdelay expired */
2114 				bond_propose_link_state(slave, BOND_LINK_UP);
2115 				slave->last_link_up = jiffies;
2116 				slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2117 					   (bond->params.downdelay - slave->delay) *
2118 					   bond->params.miimon);
2119 				commit++;
2120 				continue;
2121 			}
2122 
2123 			if (slave->delay <= 0) {
2124 				slave->new_link = BOND_LINK_DOWN;
2125 				commit++;
2126 				continue;
2127 			}
2128 
2129 			slave->delay--;
2130 			break;
2131 
2132 		case BOND_LINK_DOWN:
2133 			if (!link_state)
2134 				continue;
2135 
2136 			bond_propose_link_state(slave, BOND_LINK_BACK);
2137 			commit++;
2138 			slave->delay = bond->params.updelay;
2139 
2140 			if (slave->delay) {
2141 				slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2142 					   ignore_updelay ? 0 :
2143 					   bond->params.updelay *
2144 					   bond->params.miimon);
2145 			}
2146 			/*FALLTHRU*/
2147 		case BOND_LINK_BACK:
2148 			if (!link_state) {
2149 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2150 				slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2151 					   (bond->params.updelay - slave->delay) *
2152 					   bond->params.miimon);
2153 				commit++;
2154 				continue;
2155 			}
2156 
2157 			if (ignore_updelay)
2158 				slave->delay = 0;
2159 
2160 			if (slave->delay <= 0) {
2161 				slave->new_link = BOND_LINK_UP;
2162 				commit++;
2163 				ignore_updelay = false;
2164 				continue;
2165 			}
2166 
2167 			slave->delay--;
2168 			break;
2169 		}
2170 	}
2171 
2172 	return commit;
2173 }
2174 
2175 static void bond_miimon_link_change(struct bonding *bond,
2176 				    struct slave *slave,
2177 				    char link)
2178 {
2179 	switch (BOND_MODE(bond)) {
2180 	case BOND_MODE_8023AD:
2181 		bond_3ad_handle_link_change(slave, link);
2182 		break;
2183 	case BOND_MODE_TLB:
2184 	case BOND_MODE_ALB:
2185 		bond_alb_handle_link_change(bond, slave, link);
2186 		break;
2187 	case BOND_MODE_XOR:
2188 		bond_update_slave_arr(bond, NULL);
2189 		break;
2190 	}
2191 }
2192 
2193 static void bond_miimon_commit(struct bonding *bond)
2194 {
2195 	struct list_head *iter;
2196 	struct slave *slave, *primary;
2197 
2198 	bond_for_each_slave(bond, slave, iter) {
2199 		switch (slave->new_link) {
2200 		case BOND_LINK_NOCHANGE:
2201 			/* For 802.3ad mode, check current slave speed and
2202 			 * duplex again in case its port was disabled after
2203 			 * invalid speed/duplex reporting but recovered before
2204 			 * link monitoring could make a decision on the actual
2205 			 * link status
2206 			 */
2207 			if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2208 			    slave->link == BOND_LINK_UP)
2209 				bond_3ad_adapter_speed_duplex_changed(slave);
2210 			continue;
2211 
2212 		case BOND_LINK_UP:
2213 			if (bond_update_speed_duplex(slave) &&
2214 			    bond_needs_speed_duplex(bond)) {
2215 				slave->link = BOND_LINK_DOWN;
2216 				if (net_ratelimit())
2217 					slave_warn(bond->dev, slave->dev,
2218 						   "failed to get link speed/duplex\n");
2219 				continue;
2220 			}
2221 			bond_set_slave_link_state(slave, BOND_LINK_UP,
2222 						  BOND_SLAVE_NOTIFY_NOW);
2223 			slave->last_link_up = jiffies;
2224 
2225 			primary = rtnl_dereference(bond->primary_slave);
2226 			if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2227 				/* prevent it from being the active one */
2228 				bond_set_backup_slave(slave);
2229 			} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2230 				/* make it immediately active */
2231 				bond_set_active_slave(slave);
2232 			} else if (slave != primary) {
2233 				/* prevent it from being the active one */
2234 				bond_set_backup_slave(slave);
2235 			}
2236 
2237 			slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2238 				   slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2239 				   slave->duplex ? "full" : "half");
2240 
2241 			bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2242 
2243 			if (!bond->curr_active_slave || slave == primary)
2244 				goto do_failover;
2245 
2246 			continue;
2247 
2248 		case BOND_LINK_DOWN:
2249 			if (slave->link_failure_count < UINT_MAX)
2250 				slave->link_failure_count++;
2251 
2252 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2253 						  BOND_SLAVE_NOTIFY_NOW);
2254 
2255 			if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2256 			    BOND_MODE(bond) == BOND_MODE_8023AD)
2257 				bond_set_slave_inactive_flags(slave,
2258 							      BOND_SLAVE_NOTIFY_NOW);
2259 
2260 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2261 
2262 			bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2263 
2264 			if (slave == rcu_access_pointer(bond->curr_active_slave))
2265 				goto do_failover;
2266 
2267 			continue;
2268 
2269 		default:
2270 			slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2271 				  slave->new_link);
2272 			slave->new_link = BOND_LINK_NOCHANGE;
2273 
2274 			continue;
2275 		}
2276 
2277 do_failover:
2278 		block_netpoll_tx();
2279 		bond_select_active_slave(bond);
2280 		unblock_netpoll_tx();
2281 	}
2282 
2283 	bond_set_carrier(bond);
2284 }
2285 
2286 /* bond_mii_monitor
2287  *
2288  * Really a wrapper that splits the mii monitor into two phases: an
2289  * inspection, then (if inspection indicates something needs to be done)
2290  * an acquisition of appropriate locks followed by a commit phase to
2291  * implement whatever link state changes are indicated.
2292  */
2293 static void bond_mii_monitor(struct work_struct *work)
2294 {
2295 	struct bonding *bond = container_of(work, struct bonding,
2296 					    mii_work.work);
2297 	bool should_notify_peers = false;
2298 	bool commit;
2299 	unsigned long delay;
2300 	struct slave *slave;
2301 	struct list_head *iter;
2302 
2303 	delay = msecs_to_jiffies(bond->params.miimon);
2304 
2305 	if (!bond_has_slaves(bond))
2306 		goto re_arm;
2307 
2308 	rcu_read_lock();
2309 	should_notify_peers = bond_should_notify_peers(bond);
2310 	commit = !!bond_miimon_inspect(bond);
2311 	if (bond->send_peer_notif) {
2312 		rcu_read_unlock();
2313 		if (rtnl_trylock()) {
2314 			bond->send_peer_notif--;
2315 			rtnl_unlock();
2316 		}
2317 	} else {
2318 		rcu_read_unlock();
2319 	}
2320 
2321 	if (commit) {
2322 		/* Race avoidance with bond_close cancel of workqueue */
2323 		if (!rtnl_trylock()) {
2324 			delay = 1;
2325 			should_notify_peers = false;
2326 			goto re_arm;
2327 		}
2328 
2329 		bond_for_each_slave(bond, slave, iter) {
2330 			bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2331 		}
2332 		bond_miimon_commit(bond);
2333 
2334 		rtnl_unlock();	/* might sleep, hold no other locks */
2335 	}
2336 
2337 re_arm:
2338 	if (bond->params.miimon)
2339 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2340 
2341 	if (should_notify_peers) {
2342 		if (!rtnl_trylock())
2343 			return;
2344 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2345 		rtnl_unlock();
2346 	}
2347 }
2348 
2349 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2350 {
2351 	__be32 ip = *((__be32 *)data);
2352 
2353 	return ip == bond_confirm_addr(upper, 0, ip);
2354 }
2355 
2356 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2357 {
2358 	bool ret = false;
2359 
2360 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2361 		return true;
2362 
2363 	rcu_read_lock();
2364 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2365 		ret = true;
2366 	rcu_read_unlock();
2367 
2368 	return ret;
2369 }
2370 
2371 /* We go to the (large) trouble of VLAN tagging ARP frames because
2372  * switches in VLAN mode (especially if ports are configured as
2373  * "native" to a VLAN) might not pass non-tagged frames.
2374  */
2375 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2376 			  __be32 src_ip, struct bond_vlan_tag *tags)
2377 {
2378 	struct sk_buff *skb;
2379 	struct bond_vlan_tag *outer_tag = tags;
2380 	struct net_device *slave_dev = slave->dev;
2381 	struct net_device *bond_dev = slave->bond->dev;
2382 
2383 	slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2384 		  arp_op, &dest_ip, &src_ip);
2385 
2386 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2387 			 NULL, slave_dev->dev_addr, NULL);
2388 
2389 	if (!skb) {
2390 		net_err_ratelimited("ARP packet allocation failed\n");
2391 		return;
2392 	}
2393 
2394 	if (!tags || tags->vlan_proto == VLAN_N_VID)
2395 		goto xmit;
2396 
2397 	tags++;
2398 
2399 	/* Go through all the tags backwards and add them to the packet */
2400 	while (tags->vlan_proto != VLAN_N_VID) {
2401 		if (!tags->vlan_id) {
2402 			tags++;
2403 			continue;
2404 		}
2405 
2406 		slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2407 			  ntohs(outer_tag->vlan_proto), tags->vlan_id);
2408 		skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2409 						tags->vlan_id);
2410 		if (!skb) {
2411 			net_err_ratelimited("failed to insert inner VLAN tag\n");
2412 			return;
2413 		}
2414 
2415 		tags++;
2416 	}
2417 	/* Set the outer tag */
2418 	if (outer_tag->vlan_id) {
2419 		slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2420 			  ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2421 		__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2422 				       outer_tag->vlan_id);
2423 	}
2424 
2425 xmit:
2426 	arp_xmit(skb);
2427 }
2428 
2429 /* Validate the device path between the @start_dev and the @end_dev.
2430  * The path is valid if the @end_dev is reachable through device
2431  * stacking.
2432  * When the path is validated, collect any vlan information in the
2433  * path.
2434  */
2435 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2436 					      struct net_device *end_dev,
2437 					      int level)
2438 {
2439 	struct bond_vlan_tag *tags;
2440 	struct net_device *upper;
2441 	struct list_head  *iter;
2442 
2443 	if (start_dev == end_dev) {
2444 		tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2445 		if (!tags)
2446 			return ERR_PTR(-ENOMEM);
2447 		tags[level].vlan_proto = VLAN_N_VID;
2448 		return tags;
2449 	}
2450 
2451 	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2452 		tags = bond_verify_device_path(upper, end_dev, level + 1);
2453 		if (IS_ERR_OR_NULL(tags)) {
2454 			if (IS_ERR(tags))
2455 				return tags;
2456 			continue;
2457 		}
2458 		if (is_vlan_dev(upper)) {
2459 			tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2460 			tags[level].vlan_id = vlan_dev_vlan_id(upper);
2461 		}
2462 
2463 		return tags;
2464 	}
2465 
2466 	return NULL;
2467 }
2468 
2469 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2470 {
2471 	struct rtable *rt;
2472 	struct bond_vlan_tag *tags;
2473 	__be32 *targets = bond->params.arp_targets, addr;
2474 	int i;
2475 
2476 	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2477 		slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2478 			  __func__, &targets[i]);
2479 		tags = NULL;
2480 
2481 		/* Find out through which dev should the packet go */
2482 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2483 				     RTO_ONLINK, 0);
2484 		if (IS_ERR(rt)) {
2485 			/* there's no route to target - try to send arp
2486 			 * probe to generate any traffic (arp_validate=0)
2487 			 */
2488 			if (bond->params.arp_validate)
2489 				net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2490 						     bond->dev->name,
2491 						     &targets[i]);
2492 			bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2493 				      0, tags);
2494 			continue;
2495 		}
2496 
2497 		/* bond device itself */
2498 		if (rt->dst.dev == bond->dev)
2499 			goto found;
2500 
2501 		rcu_read_lock();
2502 		tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2503 		rcu_read_unlock();
2504 
2505 		if (!IS_ERR_OR_NULL(tags))
2506 			goto found;
2507 
2508 		/* Not our device - skip */
2509 		slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2510 			   &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2511 
2512 		ip_rt_put(rt);
2513 		continue;
2514 
2515 found:
2516 		addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2517 		ip_rt_put(rt);
2518 		bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2519 		kfree(tags);
2520 	}
2521 }
2522 
2523 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2524 {
2525 	int i;
2526 
2527 	if (!sip || !bond_has_this_ip(bond, tip)) {
2528 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2529 			   __func__, &sip, &tip);
2530 		return;
2531 	}
2532 
2533 	i = bond_get_targets_ip(bond->params.arp_targets, sip);
2534 	if (i == -1) {
2535 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2536 			   __func__, &sip);
2537 		return;
2538 	}
2539 	slave->last_rx = jiffies;
2540 	slave->target_last_arp_rx[i] = jiffies;
2541 }
2542 
2543 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2544 		 struct slave *slave)
2545 {
2546 	struct arphdr *arp = (struct arphdr *)skb->data;
2547 	struct slave *curr_active_slave, *curr_arp_slave;
2548 	unsigned char *arp_ptr;
2549 	__be32 sip, tip;
2550 	int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2551 	unsigned int alen;
2552 
2553 	if (!slave_do_arp_validate(bond, slave)) {
2554 		if ((slave_do_arp_validate_only(bond) && is_arp) ||
2555 		    !slave_do_arp_validate_only(bond))
2556 			slave->last_rx = jiffies;
2557 		return RX_HANDLER_ANOTHER;
2558 	} else if (!is_arp) {
2559 		return RX_HANDLER_ANOTHER;
2560 	}
2561 
2562 	alen = arp_hdr_len(bond->dev);
2563 
2564 	slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2565 		   __func__, skb->dev->name);
2566 
2567 	if (alen > skb_headlen(skb)) {
2568 		arp = kmalloc(alen, GFP_ATOMIC);
2569 		if (!arp)
2570 			goto out_unlock;
2571 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
2572 			goto out_unlock;
2573 	}
2574 
2575 	if (arp->ar_hln != bond->dev->addr_len ||
2576 	    skb->pkt_type == PACKET_OTHERHOST ||
2577 	    skb->pkt_type == PACKET_LOOPBACK ||
2578 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
2579 	    arp->ar_pro != htons(ETH_P_IP) ||
2580 	    arp->ar_pln != 4)
2581 		goto out_unlock;
2582 
2583 	arp_ptr = (unsigned char *)(arp + 1);
2584 	arp_ptr += bond->dev->addr_len;
2585 	memcpy(&sip, arp_ptr, 4);
2586 	arp_ptr += 4 + bond->dev->addr_len;
2587 	memcpy(&tip, arp_ptr, 4);
2588 
2589 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2590 		  __func__, slave->dev->name, bond_slave_state(slave),
2591 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2592 		  &sip, &tip);
2593 
2594 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
2595 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2596 
2597 	/* We 'trust' the received ARP enough to validate it if:
2598 	 *
2599 	 * (a) the slave receiving the ARP is active (which includes the
2600 	 * current ARP slave, if any), or
2601 	 *
2602 	 * (b) the receiving slave isn't active, but there is a currently
2603 	 * active slave and it received valid arp reply(s) after it became
2604 	 * the currently active slave, or
2605 	 *
2606 	 * (c) there is an ARP slave that sent an ARP during the prior ARP
2607 	 * interval, and we receive an ARP reply on any slave.  We accept
2608 	 * these because switch FDB update delays may deliver the ARP
2609 	 * reply to a slave other than the sender of the ARP request.
2610 	 *
2611 	 * Note: for (b), backup slaves are receiving the broadcast ARP
2612 	 * request, not a reply.  This request passes from the sending
2613 	 * slave through the L2 switch(es) to the receiving slave.  Since
2614 	 * this is checking the request, sip/tip are swapped for
2615 	 * validation.
2616 	 *
2617 	 * This is done to avoid endless looping when we can't reach the
2618 	 * arp_ip_target and fool ourselves with our own arp requests.
2619 	 */
2620 	if (bond_is_active_slave(slave))
2621 		bond_validate_arp(bond, slave, sip, tip);
2622 	else if (curr_active_slave &&
2623 		 time_after(slave_last_rx(bond, curr_active_slave),
2624 			    curr_active_slave->last_link_up))
2625 		bond_validate_arp(bond, slave, tip, sip);
2626 	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2627 		 bond_time_in_interval(bond,
2628 				       dev_trans_start(curr_arp_slave->dev), 1))
2629 		bond_validate_arp(bond, slave, sip, tip);
2630 
2631 out_unlock:
2632 	if (arp != (struct arphdr *)skb->data)
2633 		kfree(arp);
2634 	return RX_HANDLER_ANOTHER;
2635 }
2636 
2637 /* function to verify if we're in the arp_interval timeslice, returns true if
2638  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2639  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2640  */
2641 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2642 				  int mod)
2643 {
2644 	int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2645 
2646 	return time_in_range(jiffies,
2647 			     last_act - delta_in_ticks,
2648 			     last_act + mod * delta_in_ticks + delta_in_ticks/2);
2649 }
2650 
2651 /* This function is called regularly to monitor each slave's link
2652  * ensuring that traffic is being sent and received when arp monitoring
2653  * is used in load-balancing mode. if the adapter has been dormant, then an
2654  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2655  * arp monitoring in active backup mode.
2656  */
2657 static void bond_loadbalance_arp_mon(struct bonding *bond)
2658 {
2659 	struct slave *slave, *oldcurrent;
2660 	struct list_head *iter;
2661 	int do_failover = 0, slave_state_changed = 0;
2662 
2663 	if (!bond_has_slaves(bond))
2664 		goto re_arm;
2665 
2666 	rcu_read_lock();
2667 
2668 	oldcurrent = rcu_dereference(bond->curr_active_slave);
2669 	/* see if any of the previous devices are up now (i.e. they have
2670 	 * xmt and rcv traffic). the curr_active_slave does not come into
2671 	 * the picture unless it is null. also, slave->last_link_up is not
2672 	 * needed here because we send an arp on each slave and give a slave
2673 	 * as long as it needs to get the tx/rx within the delta.
2674 	 * TODO: what about up/down delay in arp mode? it wasn't here before
2675 	 *       so it can wait
2676 	 */
2677 	bond_for_each_slave_rcu(bond, slave, iter) {
2678 		unsigned long trans_start = dev_trans_start(slave->dev);
2679 
2680 		slave->new_link = BOND_LINK_NOCHANGE;
2681 
2682 		if (slave->link != BOND_LINK_UP) {
2683 			if (bond_time_in_interval(bond, trans_start, 1) &&
2684 			    bond_time_in_interval(bond, slave->last_rx, 1)) {
2685 
2686 				slave->new_link = BOND_LINK_UP;
2687 				slave_state_changed = 1;
2688 
2689 				/* primary_slave has no meaning in round-robin
2690 				 * mode. the window of a slave being up and
2691 				 * curr_active_slave being null after enslaving
2692 				 * is closed.
2693 				 */
2694 				if (!oldcurrent) {
2695 					slave_info(bond->dev, slave->dev, "link status definitely up\n");
2696 					do_failover = 1;
2697 				} else {
2698 					slave_info(bond->dev, slave->dev, "interface is now up\n");
2699 				}
2700 			}
2701 		} else {
2702 			/* slave->link == BOND_LINK_UP */
2703 
2704 			/* not all switches will respond to an arp request
2705 			 * when the source ip is 0, so don't take the link down
2706 			 * if we don't know our ip yet
2707 			 */
2708 			if (!bond_time_in_interval(bond, trans_start, 2) ||
2709 			    !bond_time_in_interval(bond, slave->last_rx, 2)) {
2710 
2711 				slave->new_link = BOND_LINK_DOWN;
2712 				slave_state_changed = 1;
2713 
2714 				if (slave->link_failure_count < UINT_MAX)
2715 					slave->link_failure_count++;
2716 
2717 				slave_info(bond->dev, slave->dev, "interface is now down\n");
2718 
2719 				if (slave == oldcurrent)
2720 					do_failover = 1;
2721 			}
2722 		}
2723 
2724 		/* note: if switch is in round-robin mode, all links
2725 		 * must tx arp to ensure all links rx an arp - otherwise
2726 		 * links may oscillate or not come up at all; if switch is
2727 		 * in something like xor mode, there is nothing we can
2728 		 * do - all replies will be rx'ed on same link causing slaves
2729 		 * to be unstable during low/no traffic periods
2730 		 */
2731 		if (bond_slave_is_up(slave))
2732 			bond_arp_send_all(bond, slave);
2733 	}
2734 
2735 	rcu_read_unlock();
2736 
2737 	if (do_failover || slave_state_changed) {
2738 		if (!rtnl_trylock())
2739 			goto re_arm;
2740 
2741 		bond_for_each_slave(bond, slave, iter) {
2742 			if (slave->new_link != BOND_LINK_NOCHANGE)
2743 				slave->link = slave->new_link;
2744 		}
2745 
2746 		if (slave_state_changed) {
2747 			bond_slave_state_change(bond);
2748 			if (BOND_MODE(bond) == BOND_MODE_XOR)
2749 				bond_update_slave_arr(bond, NULL);
2750 		}
2751 		if (do_failover) {
2752 			block_netpoll_tx();
2753 			bond_select_active_slave(bond);
2754 			unblock_netpoll_tx();
2755 		}
2756 		rtnl_unlock();
2757 	}
2758 
2759 re_arm:
2760 	if (bond->params.arp_interval)
2761 		queue_delayed_work(bond->wq, &bond->arp_work,
2762 				   msecs_to_jiffies(bond->params.arp_interval));
2763 }
2764 
2765 /* Called to inspect slaves for active-backup mode ARP monitor link state
2766  * changes.  Sets new_link in slaves to specify what action should take
2767  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2768  * to link states must be committed.
2769  *
2770  * Called with rcu_read_lock held.
2771  */
2772 static int bond_ab_arp_inspect(struct bonding *bond)
2773 {
2774 	unsigned long trans_start, last_rx;
2775 	struct list_head *iter;
2776 	struct slave *slave;
2777 	int commit = 0;
2778 
2779 	bond_for_each_slave_rcu(bond, slave, iter) {
2780 		slave->new_link = BOND_LINK_NOCHANGE;
2781 		last_rx = slave_last_rx(bond, slave);
2782 
2783 		if (slave->link != BOND_LINK_UP) {
2784 			if (bond_time_in_interval(bond, last_rx, 1)) {
2785 				slave->new_link = BOND_LINK_UP;
2786 				commit++;
2787 			}
2788 			continue;
2789 		}
2790 
2791 		/* Give slaves 2*delta after being enslaved or made
2792 		 * active.  This avoids bouncing, as the last receive
2793 		 * times need a full ARP monitor cycle to be updated.
2794 		 */
2795 		if (bond_time_in_interval(bond, slave->last_link_up, 2))
2796 			continue;
2797 
2798 		/* Backup slave is down if:
2799 		 * - No current_arp_slave AND
2800 		 * - more than 3*delta since last receive AND
2801 		 * - the bond has an IP address
2802 		 *
2803 		 * Note: a non-null current_arp_slave indicates
2804 		 * the curr_active_slave went down and we are
2805 		 * searching for a new one; under this condition
2806 		 * we only take the curr_active_slave down - this
2807 		 * gives each slave a chance to tx/rx traffic
2808 		 * before being taken out
2809 		 */
2810 		if (!bond_is_active_slave(slave) &&
2811 		    !rcu_access_pointer(bond->current_arp_slave) &&
2812 		    !bond_time_in_interval(bond, last_rx, 3)) {
2813 			slave->new_link = BOND_LINK_DOWN;
2814 			commit++;
2815 		}
2816 
2817 		/* Active slave is down if:
2818 		 * - more than 2*delta since transmitting OR
2819 		 * - (more than 2*delta since receive AND
2820 		 *    the bond has an IP address)
2821 		 */
2822 		trans_start = dev_trans_start(slave->dev);
2823 		if (bond_is_active_slave(slave) &&
2824 		    (!bond_time_in_interval(bond, trans_start, 2) ||
2825 		     !bond_time_in_interval(bond, last_rx, 2))) {
2826 			slave->new_link = BOND_LINK_DOWN;
2827 			commit++;
2828 		}
2829 	}
2830 
2831 	return commit;
2832 }
2833 
2834 /* Called to commit link state changes noted by inspection step of
2835  * active-backup mode ARP monitor.
2836  *
2837  * Called with RTNL hold.
2838  */
2839 static void bond_ab_arp_commit(struct bonding *bond)
2840 {
2841 	unsigned long trans_start;
2842 	struct list_head *iter;
2843 	struct slave *slave;
2844 
2845 	bond_for_each_slave(bond, slave, iter) {
2846 		switch (slave->new_link) {
2847 		case BOND_LINK_NOCHANGE:
2848 			continue;
2849 
2850 		case BOND_LINK_UP:
2851 			trans_start = dev_trans_start(slave->dev);
2852 			if (rtnl_dereference(bond->curr_active_slave) != slave ||
2853 			    (!rtnl_dereference(bond->curr_active_slave) &&
2854 			     bond_time_in_interval(bond, trans_start, 1))) {
2855 				struct slave *current_arp_slave;
2856 
2857 				current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2858 				bond_set_slave_link_state(slave, BOND_LINK_UP,
2859 							  BOND_SLAVE_NOTIFY_NOW);
2860 				if (current_arp_slave) {
2861 					bond_set_slave_inactive_flags(
2862 						current_arp_slave,
2863 						BOND_SLAVE_NOTIFY_NOW);
2864 					RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2865 				}
2866 
2867 				slave_info(bond->dev, slave->dev, "link status definitely up\n");
2868 
2869 				if (!rtnl_dereference(bond->curr_active_slave) ||
2870 				    slave == rtnl_dereference(bond->primary_slave))
2871 					goto do_failover;
2872 
2873 			}
2874 
2875 			continue;
2876 
2877 		case BOND_LINK_DOWN:
2878 			if (slave->link_failure_count < UINT_MAX)
2879 				slave->link_failure_count++;
2880 
2881 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2882 						  BOND_SLAVE_NOTIFY_NOW);
2883 			bond_set_slave_inactive_flags(slave,
2884 						      BOND_SLAVE_NOTIFY_NOW);
2885 
2886 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2887 
2888 			if (slave == rtnl_dereference(bond->curr_active_slave)) {
2889 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2890 				goto do_failover;
2891 			}
2892 
2893 			continue;
2894 
2895 		default:
2896 			slave_err(bond->dev, slave->dev, "impossible: new_link %d on slave\n",
2897 				  slave->new_link);
2898 			continue;
2899 		}
2900 
2901 do_failover:
2902 		block_netpoll_tx();
2903 		bond_select_active_slave(bond);
2904 		unblock_netpoll_tx();
2905 	}
2906 
2907 	bond_set_carrier(bond);
2908 }
2909 
2910 /* Send ARP probes for active-backup mode ARP monitor.
2911  *
2912  * Called with rcu_read_lock held.
2913  */
2914 static bool bond_ab_arp_probe(struct bonding *bond)
2915 {
2916 	struct slave *slave, *before = NULL, *new_slave = NULL,
2917 		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2918 		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2919 	struct list_head *iter;
2920 	bool found = false;
2921 	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2922 
2923 	if (curr_arp_slave && curr_active_slave)
2924 		netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2925 			    curr_arp_slave->dev->name,
2926 			    curr_active_slave->dev->name);
2927 
2928 	if (curr_active_slave) {
2929 		bond_arp_send_all(bond, curr_active_slave);
2930 		return should_notify_rtnl;
2931 	}
2932 
2933 	/* if we don't have a curr_active_slave, search for the next available
2934 	 * backup slave from the current_arp_slave and make it the candidate
2935 	 * for becoming the curr_active_slave
2936 	 */
2937 
2938 	if (!curr_arp_slave) {
2939 		curr_arp_slave = bond_first_slave_rcu(bond);
2940 		if (!curr_arp_slave)
2941 			return should_notify_rtnl;
2942 	}
2943 
2944 	bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2945 
2946 	bond_for_each_slave_rcu(bond, slave, iter) {
2947 		if (!found && !before && bond_slave_is_up(slave))
2948 			before = slave;
2949 
2950 		if (found && !new_slave && bond_slave_is_up(slave))
2951 			new_slave = slave;
2952 		/* if the link state is up at this point, we
2953 		 * mark it down - this can happen if we have
2954 		 * simultaneous link failures and
2955 		 * reselect_active_interface doesn't make this
2956 		 * one the current slave so it is still marked
2957 		 * up when it is actually down
2958 		 */
2959 		if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2960 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2961 						  BOND_SLAVE_NOTIFY_LATER);
2962 			if (slave->link_failure_count < UINT_MAX)
2963 				slave->link_failure_count++;
2964 
2965 			bond_set_slave_inactive_flags(slave,
2966 						      BOND_SLAVE_NOTIFY_LATER);
2967 
2968 			slave_info(bond->dev, slave->dev, "backup interface is now down\n");
2969 		}
2970 		if (slave == curr_arp_slave)
2971 			found = true;
2972 	}
2973 
2974 	if (!new_slave && before)
2975 		new_slave = before;
2976 
2977 	if (!new_slave)
2978 		goto check_state;
2979 
2980 	bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2981 				  BOND_SLAVE_NOTIFY_LATER);
2982 	bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2983 	bond_arp_send_all(bond, new_slave);
2984 	new_slave->last_link_up = jiffies;
2985 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
2986 
2987 check_state:
2988 	bond_for_each_slave_rcu(bond, slave, iter) {
2989 		if (slave->should_notify || slave->should_notify_link) {
2990 			should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2991 			break;
2992 		}
2993 	}
2994 	return should_notify_rtnl;
2995 }
2996 
2997 static void bond_activebackup_arp_mon(struct bonding *bond)
2998 {
2999 	bool should_notify_peers = false;
3000 	bool should_notify_rtnl = false;
3001 	int delta_in_ticks;
3002 
3003 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3004 
3005 	if (!bond_has_slaves(bond))
3006 		goto re_arm;
3007 
3008 	rcu_read_lock();
3009 
3010 	should_notify_peers = bond_should_notify_peers(bond);
3011 
3012 	if (bond_ab_arp_inspect(bond)) {
3013 		rcu_read_unlock();
3014 
3015 		/* Race avoidance with bond_close flush of workqueue */
3016 		if (!rtnl_trylock()) {
3017 			delta_in_ticks = 1;
3018 			should_notify_peers = false;
3019 			goto re_arm;
3020 		}
3021 
3022 		bond_ab_arp_commit(bond);
3023 
3024 		rtnl_unlock();
3025 		rcu_read_lock();
3026 	}
3027 
3028 	should_notify_rtnl = bond_ab_arp_probe(bond);
3029 	rcu_read_unlock();
3030 
3031 re_arm:
3032 	if (bond->params.arp_interval)
3033 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3034 
3035 	if (should_notify_peers || should_notify_rtnl) {
3036 		if (!rtnl_trylock())
3037 			return;
3038 
3039 		if (should_notify_peers)
3040 			call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3041 						 bond->dev);
3042 		if (should_notify_rtnl) {
3043 			bond_slave_state_notify(bond);
3044 			bond_slave_link_notify(bond);
3045 		}
3046 
3047 		rtnl_unlock();
3048 	}
3049 }
3050 
3051 static void bond_arp_monitor(struct work_struct *work)
3052 {
3053 	struct bonding *bond = container_of(work, struct bonding,
3054 					    arp_work.work);
3055 
3056 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3057 		bond_activebackup_arp_mon(bond);
3058 	else
3059 		bond_loadbalance_arp_mon(bond);
3060 }
3061 
3062 /*-------------------------- netdev event handling --------------------------*/
3063 
3064 /* Change device name */
3065 static int bond_event_changename(struct bonding *bond)
3066 {
3067 	bond_remove_proc_entry(bond);
3068 	bond_create_proc_entry(bond);
3069 
3070 	bond_debug_reregister(bond);
3071 
3072 	return NOTIFY_DONE;
3073 }
3074 
3075 static int bond_master_netdev_event(unsigned long event,
3076 				    struct net_device *bond_dev)
3077 {
3078 	struct bonding *event_bond = netdev_priv(bond_dev);
3079 
3080 	netdev_dbg(bond_dev, "%s called\n", __func__);
3081 
3082 	switch (event) {
3083 	case NETDEV_CHANGENAME:
3084 		return bond_event_changename(event_bond);
3085 	case NETDEV_UNREGISTER:
3086 		bond_remove_proc_entry(event_bond);
3087 		break;
3088 	case NETDEV_REGISTER:
3089 		bond_create_proc_entry(event_bond);
3090 		break;
3091 	default:
3092 		break;
3093 	}
3094 
3095 	return NOTIFY_DONE;
3096 }
3097 
3098 static int bond_slave_netdev_event(unsigned long event,
3099 				   struct net_device *slave_dev)
3100 {
3101 	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3102 	struct bonding *bond;
3103 	struct net_device *bond_dev;
3104 
3105 	/* A netdev event can be generated while enslaving a device
3106 	 * before netdev_rx_handler_register is called in which case
3107 	 * slave will be NULL
3108 	 */
3109 	if (!slave) {
3110 		netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3111 		return NOTIFY_DONE;
3112 	}
3113 
3114 	bond_dev = slave->bond->dev;
3115 	bond = slave->bond;
3116 	primary = rtnl_dereference(bond->primary_slave);
3117 
3118 	slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3119 
3120 	switch (event) {
3121 	case NETDEV_UNREGISTER:
3122 		if (bond_dev->type != ARPHRD_ETHER)
3123 			bond_release_and_destroy(bond_dev, slave_dev);
3124 		else
3125 			__bond_release_one(bond_dev, slave_dev, false, true);
3126 		break;
3127 	case NETDEV_UP:
3128 	case NETDEV_CHANGE:
3129 		/* For 802.3ad mode only:
3130 		 * Getting invalid Speed/Duplex values here will put slave
3131 		 * in weird state. Mark it as link-fail if the link was
3132 		 * previously up or link-down if it hasn't yet come up, and
3133 		 * let link-monitoring (miimon) set it right when correct
3134 		 * speeds/duplex are available.
3135 		 */
3136 		if (bond_update_speed_duplex(slave) &&
3137 		    BOND_MODE(bond) == BOND_MODE_8023AD) {
3138 			if (slave->last_link_up)
3139 				slave->link = BOND_LINK_FAIL;
3140 			else
3141 				slave->link = BOND_LINK_DOWN;
3142 		}
3143 
3144 		if (BOND_MODE(bond) == BOND_MODE_8023AD)
3145 			bond_3ad_adapter_speed_duplex_changed(slave);
3146 		/* Fallthrough */
3147 	case NETDEV_DOWN:
3148 		/* Refresh slave-array if applicable!
3149 		 * If the setup does not use miimon or arpmon (mode-specific!),
3150 		 * then these events will not cause the slave-array to be
3151 		 * refreshed. This will cause xmit to use a slave that is not
3152 		 * usable. Avoid such situation by refeshing the array at these
3153 		 * events. If these (miimon/arpmon) parameters are configured
3154 		 * then array gets refreshed twice and that should be fine!
3155 		 */
3156 		if (bond_mode_can_use_xmit_hash(bond))
3157 			bond_update_slave_arr(bond, NULL);
3158 		break;
3159 	case NETDEV_CHANGEMTU:
3160 		/* TODO: Should slaves be allowed to
3161 		 * independently alter their MTU?  For
3162 		 * an active-backup bond, slaves need
3163 		 * not be the same type of device, so
3164 		 * MTUs may vary.  For other modes,
3165 		 * slaves arguably should have the
3166 		 * same MTUs. To do this, we'd need to
3167 		 * take over the slave's change_mtu
3168 		 * function for the duration of their
3169 		 * servitude.
3170 		 */
3171 		break;
3172 	case NETDEV_CHANGENAME:
3173 		/* we don't care if we don't have primary set */
3174 		if (!bond_uses_primary(bond) ||
3175 		    !bond->params.primary[0])
3176 			break;
3177 
3178 		if (slave == primary) {
3179 			/* slave's name changed - he's no longer primary */
3180 			RCU_INIT_POINTER(bond->primary_slave, NULL);
3181 		} else if (!strcmp(slave_dev->name, bond->params.primary)) {
3182 			/* we have a new primary slave */
3183 			rcu_assign_pointer(bond->primary_slave, slave);
3184 		} else { /* we didn't change primary - exit */
3185 			break;
3186 		}
3187 
3188 		netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3189 			    primary ? slave_dev->name : "none");
3190 
3191 		block_netpoll_tx();
3192 		bond_select_active_slave(bond);
3193 		unblock_netpoll_tx();
3194 		break;
3195 	case NETDEV_FEAT_CHANGE:
3196 		bond_compute_features(bond);
3197 		break;
3198 	case NETDEV_RESEND_IGMP:
3199 		/* Propagate to master device */
3200 		call_netdevice_notifiers(event, slave->bond->dev);
3201 		break;
3202 	default:
3203 		break;
3204 	}
3205 
3206 	return NOTIFY_DONE;
3207 }
3208 
3209 /* bond_netdev_event: handle netdev notifier chain events.
3210  *
3211  * This function receives events for the netdev chain.  The caller (an
3212  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3213  * locks for us to safely manipulate the slave devices (RTNL lock,
3214  * dev_probe_lock).
3215  */
3216 static int bond_netdev_event(struct notifier_block *this,
3217 			     unsigned long event, void *ptr)
3218 {
3219 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3220 
3221 	netdev_dbg(event_dev, "%s received %s\n",
3222 		   __func__, netdev_cmd_to_name(event));
3223 
3224 	if (!(event_dev->priv_flags & IFF_BONDING))
3225 		return NOTIFY_DONE;
3226 
3227 	if (event_dev->flags & IFF_MASTER) {
3228 		int ret;
3229 
3230 		ret = bond_master_netdev_event(event, event_dev);
3231 		if (ret != NOTIFY_DONE)
3232 			return ret;
3233 	}
3234 
3235 	if (event_dev->flags & IFF_SLAVE)
3236 		return bond_slave_netdev_event(event, event_dev);
3237 
3238 	return NOTIFY_DONE;
3239 }
3240 
3241 static struct notifier_block bond_netdev_notifier = {
3242 	.notifier_call = bond_netdev_event,
3243 };
3244 
3245 /*---------------------------- Hashing Policies -----------------------------*/
3246 
3247 /* L2 hash helper */
3248 static inline u32 bond_eth_hash(struct sk_buff *skb)
3249 {
3250 	struct ethhdr *ep, hdr_tmp;
3251 
3252 	ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3253 	if (ep)
3254 		return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3255 	return 0;
3256 }
3257 
3258 /* Extract the appropriate headers based on bond's xmit policy */
3259 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3260 			      struct flow_keys *fk)
3261 {
3262 	const struct ipv6hdr *iph6;
3263 	const struct iphdr *iph;
3264 	int noff, proto = -1;
3265 
3266 	if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3267 		return skb_flow_dissect_flow_keys(skb, fk, 0);
3268 
3269 	fk->ports.ports = 0;
3270 	noff = skb_network_offset(skb);
3271 	if (skb->protocol == htons(ETH_P_IP)) {
3272 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3273 			return false;
3274 		iph = ip_hdr(skb);
3275 		iph_to_flow_copy_v4addrs(fk, iph);
3276 		noff += iph->ihl << 2;
3277 		if (!ip_is_fragment(iph))
3278 			proto = iph->protocol;
3279 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
3280 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3281 			return false;
3282 		iph6 = ipv6_hdr(skb);
3283 		iph_to_flow_copy_v6addrs(fk, iph6);
3284 		noff += sizeof(*iph6);
3285 		proto = iph6->nexthdr;
3286 	} else {
3287 		return false;
3288 	}
3289 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3290 		fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3291 
3292 	return true;
3293 }
3294 
3295 /**
3296  * bond_xmit_hash - generate a hash value based on the xmit policy
3297  * @bond: bonding device
3298  * @skb: buffer to use for headers
3299  *
3300  * This function will extract the necessary headers from the skb buffer and use
3301  * them to generate a hash based on the xmit_policy set in the bonding device
3302  */
3303 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3304 {
3305 	struct flow_keys flow;
3306 	u32 hash;
3307 
3308 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3309 	    skb->l4_hash)
3310 		return skb->hash;
3311 
3312 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3313 	    !bond_flow_dissect(bond, skb, &flow))
3314 		return bond_eth_hash(skb);
3315 
3316 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3317 	    bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3318 		hash = bond_eth_hash(skb);
3319 	else
3320 		hash = (__force u32)flow.ports.ports;
3321 	hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3322 		(__force u32)flow_get_u32_src(&flow);
3323 	hash ^= (hash >> 16);
3324 	hash ^= (hash >> 8);
3325 
3326 	return hash >> 1;
3327 }
3328 
3329 /*-------------------------- Device entry points ----------------------------*/
3330 
3331 void bond_work_init_all(struct bonding *bond)
3332 {
3333 	INIT_DELAYED_WORK(&bond->mcast_work,
3334 			  bond_resend_igmp_join_requests_delayed);
3335 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3336 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3337 	INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3338 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3339 	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3340 }
3341 
3342 static void bond_work_cancel_all(struct bonding *bond)
3343 {
3344 	cancel_delayed_work_sync(&bond->mii_work);
3345 	cancel_delayed_work_sync(&bond->arp_work);
3346 	cancel_delayed_work_sync(&bond->alb_work);
3347 	cancel_delayed_work_sync(&bond->ad_work);
3348 	cancel_delayed_work_sync(&bond->mcast_work);
3349 	cancel_delayed_work_sync(&bond->slave_arr_work);
3350 }
3351 
3352 static int bond_open(struct net_device *bond_dev)
3353 {
3354 	struct bonding *bond = netdev_priv(bond_dev);
3355 	struct list_head *iter;
3356 	struct slave *slave;
3357 
3358 	/* reset slave->backup and slave->inactive */
3359 	if (bond_has_slaves(bond)) {
3360 		bond_for_each_slave(bond, slave, iter) {
3361 			if (bond_uses_primary(bond) &&
3362 			    slave != rcu_access_pointer(bond->curr_active_slave)) {
3363 				bond_set_slave_inactive_flags(slave,
3364 							      BOND_SLAVE_NOTIFY_NOW);
3365 			} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3366 				bond_set_slave_active_flags(slave,
3367 							    BOND_SLAVE_NOTIFY_NOW);
3368 			}
3369 		}
3370 	}
3371 
3372 	if (bond_is_lb(bond)) {
3373 		/* bond_alb_initialize must be called before the timer
3374 		 * is started.
3375 		 */
3376 		if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3377 			return -ENOMEM;
3378 		if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3379 			queue_delayed_work(bond->wq, &bond->alb_work, 0);
3380 	}
3381 
3382 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
3383 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
3384 
3385 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3386 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
3387 		bond->recv_probe = bond_arp_rcv;
3388 	}
3389 
3390 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3391 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
3392 		/* register to receive LACPDUs */
3393 		bond->recv_probe = bond_3ad_lacpdu_recv;
3394 		bond_3ad_initiate_agg_selection(bond, 1);
3395 	}
3396 
3397 	if (bond_mode_can_use_xmit_hash(bond))
3398 		bond_update_slave_arr(bond, NULL);
3399 
3400 	return 0;
3401 }
3402 
3403 static int bond_close(struct net_device *bond_dev)
3404 {
3405 	struct bonding *bond = netdev_priv(bond_dev);
3406 
3407 	bond_work_cancel_all(bond);
3408 	bond->send_peer_notif = 0;
3409 	if (bond_is_lb(bond))
3410 		bond_alb_deinitialize(bond);
3411 	bond->recv_probe = NULL;
3412 
3413 	return 0;
3414 }
3415 
3416 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3417  * that some drivers can provide 32bit values only.
3418  */
3419 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3420 			    const struct rtnl_link_stats64 *_new,
3421 			    const struct rtnl_link_stats64 *_old)
3422 {
3423 	const u64 *new = (const u64 *)_new;
3424 	const u64 *old = (const u64 *)_old;
3425 	u64 *res = (u64 *)_res;
3426 	int i;
3427 
3428 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3429 		u64 nv = new[i];
3430 		u64 ov = old[i];
3431 		s64 delta = nv - ov;
3432 
3433 		/* detects if this particular field is 32bit only */
3434 		if (((nv | ov) >> 32) == 0)
3435 			delta = (s64)(s32)((u32)nv - (u32)ov);
3436 
3437 		/* filter anomalies, some drivers reset their stats
3438 		 * at down/up events.
3439 		 */
3440 		if (delta > 0)
3441 			res[i] += delta;
3442 	}
3443 }
3444 
3445 static int bond_get_nest_level(struct net_device *bond_dev)
3446 {
3447 	struct bonding *bond = netdev_priv(bond_dev);
3448 
3449 	return bond->nest_level;
3450 }
3451 
3452 static void bond_get_stats(struct net_device *bond_dev,
3453 			   struct rtnl_link_stats64 *stats)
3454 {
3455 	struct bonding *bond = netdev_priv(bond_dev);
3456 	struct rtnl_link_stats64 temp;
3457 	struct list_head *iter;
3458 	struct slave *slave;
3459 
3460 	spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3461 	memcpy(stats, &bond->bond_stats, sizeof(*stats));
3462 
3463 	rcu_read_lock();
3464 	bond_for_each_slave_rcu(bond, slave, iter) {
3465 		const struct rtnl_link_stats64 *new =
3466 			dev_get_stats(slave->dev, &temp);
3467 
3468 		bond_fold_stats(stats, new, &slave->slave_stats);
3469 
3470 		/* save off the slave stats for the next run */
3471 		memcpy(&slave->slave_stats, new, sizeof(*new));
3472 	}
3473 	rcu_read_unlock();
3474 
3475 	memcpy(&bond->bond_stats, stats, sizeof(*stats));
3476 	spin_unlock(&bond->stats_lock);
3477 }
3478 
3479 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3480 {
3481 	struct bonding *bond = netdev_priv(bond_dev);
3482 	struct net_device *slave_dev = NULL;
3483 	struct ifbond k_binfo;
3484 	struct ifbond __user *u_binfo = NULL;
3485 	struct ifslave k_sinfo;
3486 	struct ifslave __user *u_sinfo = NULL;
3487 	struct mii_ioctl_data *mii = NULL;
3488 	struct bond_opt_value newval;
3489 	struct net *net;
3490 	int res = 0;
3491 
3492 	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3493 
3494 	switch (cmd) {
3495 	case SIOCGMIIPHY:
3496 		mii = if_mii(ifr);
3497 		if (!mii)
3498 			return -EINVAL;
3499 
3500 		mii->phy_id = 0;
3501 		/* Fall Through */
3502 	case SIOCGMIIREG:
3503 		/* We do this again just in case we were called by SIOCGMIIREG
3504 		 * instead of SIOCGMIIPHY.
3505 		 */
3506 		mii = if_mii(ifr);
3507 		if (!mii)
3508 			return -EINVAL;
3509 
3510 		if (mii->reg_num == 1) {
3511 			mii->val_out = 0;
3512 			if (netif_carrier_ok(bond->dev))
3513 				mii->val_out = BMSR_LSTATUS;
3514 		}
3515 
3516 		return 0;
3517 	case BOND_INFO_QUERY_OLD:
3518 	case SIOCBONDINFOQUERY:
3519 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
3520 
3521 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3522 			return -EFAULT;
3523 
3524 		bond_info_query(bond_dev, &k_binfo);
3525 		if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3526 			return -EFAULT;
3527 
3528 		return 0;
3529 	case BOND_SLAVE_INFO_QUERY_OLD:
3530 	case SIOCBONDSLAVEINFOQUERY:
3531 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3532 
3533 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3534 			return -EFAULT;
3535 
3536 		res = bond_slave_info_query(bond_dev, &k_sinfo);
3537 		if (res == 0 &&
3538 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3539 			return -EFAULT;
3540 
3541 		return res;
3542 	default:
3543 		break;
3544 	}
3545 
3546 	net = dev_net(bond_dev);
3547 
3548 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3549 		return -EPERM;
3550 
3551 	slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3552 
3553 	slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3554 
3555 	if (!slave_dev)
3556 		return -ENODEV;
3557 
3558 	switch (cmd) {
3559 	case BOND_ENSLAVE_OLD:
3560 	case SIOCBONDENSLAVE:
3561 		res = bond_enslave(bond_dev, slave_dev, NULL);
3562 		break;
3563 	case BOND_RELEASE_OLD:
3564 	case SIOCBONDRELEASE:
3565 		res = bond_release(bond_dev, slave_dev);
3566 		break;
3567 	case BOND_SETHWADDR_OLD:
3568 	case SIOCBONDSETHWADDR:
3569 		res = bond_set_dev_addr(bond_dev, slave_dev);
3570 		break;
3571 	case BOND_CHANGE_ACTIVE_OLD:
3572 	case SIOCBONDCHANGEACTIVE:
3573 		bond_opt_initstr(&newval, slave_dev->name);
3574 		res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3575 					    &newval);
3576 		break;
3577 	default:
3578 		res = -EOPNOTSUPP;
3579 	}
3580 
3581 	return res;
3582 }
3583 
3584 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3585 {
3586 	struct bonding *bond = netdev_priv(bond_dev);
3587 
3588 	if (change & IFF_PROMISC)
3589 		bond_set_promiscuity(bond,
3590 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
3591 
3592 	if (change & IFF_ALLMULTI)
3593 		bond_set_allmulti(bond,
3594 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3595 }
3596 
3597 static void bond_set_rx_mode(struct net_device *bond_dev)
3598 {
3599 	struct bonding *bond = netdev_priv(bond_dev);
3600 	struct list_head *iter;
3601 	struct slave *slave;
3602 
3603 	rcu_read_lock();
3604 	if (bond_uses_primary(bond)) {
3605 		slave = rcu_dereference(bond->curr_active_slave);
3606 		if (slave) {
3607 			dev_uc_sync(slave->dev, bond_dev);
3608 			dev_mc_sync(slave->dev, bond_dev);
3609 		}
3610 	} else {
3611 		bond_for_each_slave_rcu(bond, slave, iter) {
3612 			dev_uc_sync_multiple(slave->dev, bond_dev);
3613 			dev_mc_sync_multiple(slave->dev, bond_dev);
3614 		}
3615 	}
3616 	rcu_read_unlock();
3617 }
3618 
3619 static int bond_neigh_init(struct neighbour *n)
3620 {
3621 	struct bonding *bond = netdev_priv(n->dev);
3622 	const struct net_device_ops *slave_ops;
3623 	struct neigh_parms parms;
3624 	struct slave *slave;
3625 	int ret;
3626 
3627 	slave = bond_first_slave(bond);
3628 	if (!slave)
3629 		return 0;
3630 	slave_ops = slave->dev->netdev_ops;
3631 	if (!slave_ops->ndo_neigh_setup)
3632 		return 0;
3633 
3634 	parms.neigh_setup = NULL;
3635 	parms.neigh_cleanup = NULL;
3636 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3637 	if (ret)
3638 		return ret;
3639 
3640 	/* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3641 	 * after the last slave has been detached.  Assumes that all slaves
3642 	 * utilize the same neigh_cleanup (true at this writing as only user
3643 	 * is ipoib).
3644 	 */
3645 	n->parms->neigh_cleanup = parms.neigh_cleanup;
3646 
3647 	if (!parms.neigh_setup)
3648 		return 0;
3649 
3650 	return parms.neigh_setup(n);
3651 }
3652 
3653 /* The bonding ndo_neigh_setup is called at init time beofre any
3654  * slave exists. So we must declare proxy setup function which will
3655  * be used at run time to resolve the actual slave neigh param setup.
3656  *
3657  * It's also called by master devices (such as vlans) to setup their
3658  * underlying devices. In that case - do nothing, we're already set up from
3659  * our init.
3660  */
3661 static int bond_neigh_setup(struct net_device *dev,
3662 			    struct neigh_parms *parms)
3663 {
3664 	/* modify only our neigh_parms */
3665 	if (parms->dev == dev)
3666 		parms->neigh_setup = bond_neigh_init;
3667 
3668 	return 0;
3669 }
3670 
3671 /* Change the MTU of all of a master's slaves to match the master */
3672 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3673 {
3674 	struct bonding *bond = netdev_priv(bond_dev);
3675 	struct slave *slave, *rollback_slave;
3676 	struct list_head *iter;
3677 	int res = 0;
3678 
3679 	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3680 
3681 	bond_for_each_slave(bond, slave, iter) {
3682 		slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3683 			   slave, slave->dev->netdev_ops->ndo_change_mtu);
3684 
3685 		res = dev_set_mtu(slave->dev, new_mtu);
3686 
3687 		if (res) {
3688 			/* If we failed to set the slave's mtu to the new value
3689 			 * we must abort the operation even in ACTIVE_BACKUP
3690 			 * mode, because if we allow the backup slaves to have
3691 			 * different mtu values than the active slave we'll
3692 			 * need to change their mtu when doing a failover. That
3693 			 * means changing their mtu from timer context, which
3694 			 * is probably not a good idea.
3695 			 */
3696 			slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3697 				  res, new_mtu);
3698 			goto unwind;
3699 		}
3700 	}
3701 
3702 	bond_dev->mtu = new_mtu;
3703 
3704 	return 0;
3705 
3706 unwind:
3707 	/* unwind from head to the slave that failed */
3708 	bond_for_each_slave(bond, rollback_slave, iter) {
3709 		int tmp_res;
3710 
3711 		if (rollback_slave == slave)
3712 			break;
3713 
3714 		tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3715 		if (tmp_res)
3716 			slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3717 				  tmp_res);
3718 	}
3719 
3720 	return res;
3721 }
3722 
3723 /* Change HW address
3724  *
3725  * Note that many devices must be down to change the HW address, and
3726  * downing the master releases all slaves.  We can make bonds full of
3727  * bonding devices to test this, however.
3728  */
3729 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3730 {
3731 	struct bonding *bond = netdev_priv(bond_dev);
3732 	struct slave *slave, *rollback_slave;
3733 	struct sockaddr_storage *ss = addr, tmp_ss;
3734 	struct list_head *iter;
3735 	int res = 0;
3736 
3737 	if (BOND_MODE(bond) == BOND_MODE_ALB)
3738 		return bond_alb_set_mac_address(bond_dev, addr);
3739 
3740 
3741 	netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
3742 
3743 	/* If fail_over_mac is enabled, do nothing and return success.
3744 	 * Returning an error causes ifenslave to fail.
3745 	 */
3746 	if (bond->params.fail_over_mac &&
3747 	    BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3748 		return 0;
3749 
3750 	if (!is_valid_ether_addr(ss->__data))
3751 		return -EADDRNOTAVAIL;
3752 
3753 	bond_for_each_slave(bond, slave, iter) {
3754 		slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
3755 			  __func__, slave);
3756 		res = dev_set_mac_address(slave->dev, addr, NULL);
3757 		if (res) {
3758 			/* TODO: consider downing the slave
3759 			 * and retry ?
3760 			 * User should expect communications
3761 			 * breakage anyway until ARP finish
3762 			 * updating, so...
3763 			 */
3764 			slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
3765 				  __func__, res);
3766 			goto unwind;
3767 		}
3768 	}
3769 
3770 	/* success */
3771 	memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3772 	return 0;
3773 
3774 unwind:
3775 	memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3776 	tmp_ss.ss_family = bond_dev->type;
3777 
3778 	/* unwind from head to the slave that failed */
3779 	bond_for_each_slave(bond, rollback_slave, iter) {
3780 		int tmp_res;
3781 
3782 		if (rollback_slave == slave)
3783 			break;
3784 
3785 		tmp_res = dev_set_mac_address(rollback_slave->dev,
3786 					      (struct sockaddr *)&tmp_ss, NULL);
3787 		if (tmp_res) {
3788 			slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
3789 				   __func__, tmp_res);
3790 		}
3791 	}
3792 
3793 	return res;
3794 }
3795 
3796 /**
3797  * bond_xmit_slave_id - transmit skb through slave with slave_id
3798  * @bond: bonding device that is transmitting
3799  * @skb: buffer to transmit
3800  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3801  *
3802  * This function tries to transmit through slave with slave_id but in case
3803  * it fails, it tries to find the first available slave for transmission.
3804  * The skb is consumed in all cases, thus the function is void.
3805  */
3806 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3807 {
3808 	struct list_head *iter;
3809 	struct slave *slave;
3810 	int i = slave_id;
3811 
3812 	/* Here we start from the slave with slave_id */
3813 	bond_for_each_slave_rcu(bond, slave, iter) {
3814 		if (--i < 0) {
3815 			if (bond_slave_can_tx(slave)) {
3816 				bond_dev_queue_xmit(bond, skb, slave->dev);
3817 				return;
3818 			}
3819 		}
3820 	}
3821 
3822 	/* Here we start from the first slave up to slave_id */
3823 	i = slave_id;
3824 	bond_for_each_slave_rcu(bond, slave, iter) {
3825 		if (--i < 0)
3826 			break;
3827 		if (bond_slave_can_tx(slave)) {
3828 			bond_dev_queue_xmit(bond, skb, slave->dev);
3829 			return;
3830 		}
3831 	}
3832 	/* no slave that can tx has been found */
3833 	bond_tx_drop(bond->dev, skb);
3834 }
3835 
3836 /**
3837  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3838  * @bond: bonding device to use
3839  *
3840  * Based on the value of the bonding device's packets_per_slave parameter
3841  * this function generates a slave id, which is usually used as the next
3842  * slave to transmit through.
3843  */
3844 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3845 {
3846 	u32 slave_id;
3847 	struct reciprocal_value reciprocal_packets_per_slave;
3848 	int packets_per_slave = bond->params.packets_per_slave;
3849 
3850 	switch (packets_per_slave) {
3851 	case 0:
3852 		slave_id = prandom_u32();
3853 		break;
3854 	case 1:
3855 		slave_id = bond->rr_tx_counter;
3856 		break;
3857 	default:
3858 		reciprocal_packets_per_slave =
3859 			bond->params.reciprocal_packets_per_slave;
3860 		slave_id = reciprocal_divide(bond->rr_tx_counter,
3861 					     reciprocal_packets_per_slave);
3862 		break;
3863 	}
3864 	bond->rr_tx_counter++;
3865 
3866 	return slave_id;
3867 }
3868 
3869 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3870 					struct net_device *bond_dev)
3871 {
3872 	struct bonding *bond = netdev_priv(bond_dev);
3873 	struct slave *slave;
3874 	int slave_cnt;
3875 	u32 slave_id;
3876 
3877 	/* Start with the curr_active_slave that joined the bond as the
3878 	 * default for sending IGMP traffic.  For failover purposes one
3879 	 * needs to maintain some consistency for the interface that will
3880 	 * send the join/membership reports.  The curr_active_slave found
3881 	 * will send all of this type of traffic.
3882 	 */
3883 	if (skb->protocol == htons(ETH_P_IP)) {
3884 		int noff = skb_network_offset(skb);
3885 		struct iphdr *iph;
3886 
3887 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3888 			goto non_igmp;
3889 
3890 		iph = ip_hdr(skb);
3891 		if (iph->protocol == IPPROTO_IGMP) {
3892 			slave = rcu_dereference(bond->curr_active_slave);
3893 			if (slave)
3894 				bond_dev_queue_xmit(bond, skb, slave->dev);
3895 			else
3896 				bond_xmit_slave_id(bond, skb, 0);
3897 			return NETDEV_TX_OK;
3898 		}
3899 	}
3900 
3901 non_igmp:
3902 	slave_cnt = READ_ONCE(bond->slave_cnt);
3903 	if (likely(slave_cnt)) {
3904 		slave_id = bond_rr_gen_slave_id(bond);
3905 		bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3906 	} else {
3907 		bond_tx_drop(bond_dev, skb);
3908 	}
3909 	return NETDEV_TX_OK;
3910 }
3911 
3912 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3913  * the bond has a usable interface.
3914  */
3915 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3916 					  struct net_device *bond_dev)
3917 {
3918 	struct bonding *bond = netdev_priv(bond_dev);
3919 	struct slave *slave;
3920 
3921 	slave = rcu_dereference(bond->curr_active_slave);
3922 	if (slave)
3923 		bond_dev_queue_xmit(bond, skb, slave->dev);
3924 	else
3925 		bond_tx_drop(bond_dev, skb);
3926 
3927 	return NETDEV_TX_OK;
3928 }
3929 
3930 /* Use this to update slave_array when (a) it's not appropriate to update
3931  * slave_array right away (note that update_slave_array() may sleep)
3932  * and / or (b) RTNL is not held.
3933  */
3934 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3935 {
3936 	queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3937 }
3938 
3939 /* Slave array work handler. Holds only RTNL */
3940 static void bond_slave_arr_handler(struct work_struct *work)
3941 {
3942 	struct bonding *bond = container_of(work, struct bonding,
3943 					    slave_arr_work.work);
3944 	int ret;
3945 
3946 	if (!rtnl_trylock())
3947 		goto err;
3948 
3949 	ret = bond_update_slave_arr(bond, NULL);
3950 	rtnl_unlock();
3951 	if (ret) {
3952 		pr_warn_ratelimited("Failed to update slave array from WT\n");
3953 		goto err;
3954 	}
3955 	return;
3956 
3957 err:
3958 	bond_slave_arr_work_rearm(bond, 1);
3959 }
3960 
3961 /* Build the usable slaves array in control path for modes that use xmit-hash
3962  * to determine the slave interface -
3963  * (a) BOND_MODE_8023AD
3964  * (b) BOND_MODE_XOR
3965  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3966  *
3967  * The caller is expected to hold RTNL only and NO other lock!
3968  */
3969 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3970 {
3971 	struct slave *slave;
3972 	struct list_head *iter;
3973 	struct bond_up_slave *new_arr, *old_arr;
3974 	int agg_id = 0;
3975 	int ret = 0;
3976 
3977 #ifdef CONFIG_LOCKDEP
3978 	WARN_ON(lockdep_is_held(&bond->mode_lock));
3979 #endif
3980 
3981 	new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3982 			  GFP_KERNEL);
3983 	if (!new_arr) {
3984 		ret = -ENOMEM;
3985 		pr_err("Failed to build slave-array.\n");
3986 		goto out;
3987 	}
3988 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3989 		struct ad_info ad_info;
3990 
3991 		if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3992 			pr_debug("bond_3ad_get_active_agg_info failed\n");
3993 			kfree_rcu(new_arr, rcu);
3994 			/* No active aggragator means it's not safe to use
3995 			 * the previous array.
3996 			 */
3997 			old_arr = rtnl_dereference(bond->slave_arr);
3998 			if (old_arr) {
3999 				RCU_INIT_POINTER(bond->slave_arr, NULL);
4000 				kfree_rcu(old_arr, rcu);
4001 			}
4002 			goto out;
4003 		}
4004 		agg_id = ad_info.aggregator_id;
4005 	}
4006 	bond_for_each_slave(bond, slave, iter) {
4007 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4008 			struct aggregator *agg;
4009 
4010 			agg = SLAVE_AD_INFO(slave)->port.aggregator;
4011 			if (!agg || agg->aggregator_identifier != agg_id)
4012 				continue;
4013 		}
4014 		if (!bond_slave_can_tx(slave))
4015 			continue;
4016 		if (skipslave == slave)
4017 			continue;
4018 
4019 		slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4020 			  new_arr->count);
4021 
4022 		new_arr->arr[new_arr->count++] = slave;
4023 	}
4024 
4025 	old_arr = rtnl_dereference(bond->slave_arr);
4026 	rcu_assign_pointer(bond->slave_arr, new_arr);
4027 	if (old_arr)
4028 		kfree_rcu(old_arr, rcu);
4029 out:
4030 	if (ret != 0 && skipslave) {
4031 		int idx;
4032 
4033 		/* Rare situation where caller has asked to skip a specific
4034 		 * slave but allocation failed (most likely!). BTW this is
4035 		 * only possible when the call is initiated from
4036 		 * __bond_release_one(). In this situation; overwrite the
4037 		 * skipslave entry in the array with the last entry from the
4038 		 * array to avoid a situation where the xmit path may choose
4039 		 * this to-be-skipped slave to send a packet out.
4040 		 */
4041 		old_arr = rtnl_dereference(bond->slave_arr);
4042 		for (idx = 0; idx < old_arr->count; idx++) {
4043 			if (skipslave == old_arr->arr[idx]) {
4044 				old_arr->arr[idx] =
4045 				    old_arr->arr[old_arr->count-1];
4046 				old_arr->count--;
4047 				break;
4048 			}
4049 		}
4050 	}
4051 	return ret;
4052 }
4053 
4054 /* Use this Xmit function for 3AD as well as XOR modes. The current
4055  * usable slave array is formed in the control path. The xmit function
4056  * just calculates hash and sends the packet out.
4057  */
4058 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4059 				     struct net_device *dev)
4060 {
4061 	struct bonding *bond = netdev_priv(dev);
4062 	struct slave *slave;
4063 	struct bond_up_slave *slaves;
4064 	unsigned int count;
4065 
4066 	slaves = rcu_dereference(bond->slave_arr);
4067 	count = slaves ? READ_ONCE(slaves->count) : 0;
4068 	if (likely(count)) {
4069 		slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4070 		bond_dev_queue_xmit(bond, skb, slave->dev);
4071 	} else {
4072 		bond_tx_drop(dev, skb);
4073 	}
4074 
4075 	return NETDEV_TX_OK;
4076 }
4077 
4078 /* in broadcast mode, we send everything to all usable interfaces. */
4079 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4080 				       struct net_device *bond_dev)
4081 {
4082 	struct bonding *bond = netdev_priv(bond_dev);
4083 	struct slave *slave = NULL;
4084 	struct list_head *iter;
4085 
4086 	bond_for_each_slave_rcu(bond, slave, iter) {
4087 		if (bond_is_last_slave(bond, slave))
4088 			break;
4089 		if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4090 			struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4091 
4092 			if (!skb2) {
4093 				net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4094 						    bond_dev->name, __func__);
4095 				continue;
4096 			}
4097 			bond_dev_queue_xmit(bond, skb2, slave->dev);
4098 		}
4099 	}
4100 	if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4101 		bond_dev_queue_xmit(bond, skb, slave->dev);
4102 	else
4103 		bond_tx_drop(bond_dev, skb);
4104 
4105 	return NETDEV_TX_OK;
4106 }
4107 
4108 /*------------------------- Device initialization ---------------------------*/
4109 
4110 /* Lookup the slave that corresponds to a qid */
4111 static inline int bond_slave_override(struct bonding *bond,
4112 				      struct sk_buff *skb)
4113 {
4114 	struct slave *slave = NULL;
4115 	struct list_head *iter;
4116 
4117 	if (!skb_rx_queue_recorded(skb))
4118 		return 1;
4119 
4120 	/* Find out if any slaves have the same mapping as this skb. */
4121 	bond_for_each_slave_rcu(bond, slave, iter) {
4122 		if (slave->queue_id == skb_get_queue_mapping(skb)) {
4123 			if (bond_slave_is_up(slave) &&
4124 			    slave->link == BOND_LINK_UP) {
4125 				bond_dev_queue_xmit(bond, skb, slave->dev);
4126 				return 0;
4127 			}
4128 			/* If the slave isn't UP, use default transmit policy. */
4129 			break;
4130 		}
4131 	}
4132 
4133 	return 1;
4134 }
4135 
4136 
4137 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4138 			     struct net_device *sb_dev)
4139 {
4140 	/* This helper function exists to help dev_pick_tx get the correct
4141 	 * destination queue.  Using a helper function skips a call to
4142 	 * skb_tx_hash and will put the skbs in the queue we expect on their
4143 	 * way down to the bonding driver.
4144 	 */
4145 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4146 
4147 	/* Save the original txq to restore before passing to the driver */
4148 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4149 
4150 	if (unlikely(txq >= dev->real_num_tx_queues)) {
4151 		do {
4152 			txq -= dev->real_num_tx_queues;
4153 		} while (txq >= dev->real_num_tx_queues);
4154 	}
4155 	return txq;
4156 }
4157 
4158 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4159 {
4160 	struct bonding *bond = netdev_priv(dev);
4161 
4162 	if (bond_should_override_tx_queue(bond) &&
4163 	    !bond_slave_override(bond, skb))
4164 		return NETDEV_TX_OK;
4165 
4166 	switch (BOND_MODE(bond)) {
4167 	case BOND_MODE_ROUNDROBIN:
4168 		return bond_xmit_roundrobin(skb, dev);
4169 	case BOND_MODE_ACTIVEBACKUP:
4170 		return bond_xmit_activebackup(skb, dev);
4171 	case BOND_MODE_8023AD:
4172 	case BOND_MODE_XOR:
4173 		return bond_3ad_xor_xmit(skb, dev);
4174 	case BOND_MODE_BROADCAST:
4175 		return bond_xmit_broadcast(skb, dev);
4176 	case BOND_MODE_ALB:
4177 		return bond_alb_xmit(skb, dev);
4178 	case BOND_MODE_TLB:
4179 		return bond_tlb_xmit(skb, dev);
4180 	default:
4181 		/* Should never happen, mode already checked */
4182 		netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4183 		WARN_ON_ONCE(1);
4184 		bond_tx_drop(dev, skb);
4185 		return NETDEV_TX_OK;
4186 	}
4187 }
4188 
4189 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4190 {
4191 	struct bonding *bond = netdev_priv(dev);
4192 	netdev_tx_t ret = NETDEV_TX_OK;
4193 
4194 	/* If we risk deadlock from transmitting this in the
4195 	 * netpoll path, tell netpoll to queue the frame for later tx
4196 	 */
4197 	if (unlikely(is_netpoll_tx_blocked(dev)))
4198 		return NETDEV_TX_BUSY;
4199 
4200 	rcu_read_lock();
4201 	if (bond_has_slaves(bond))
4202 		ret = __bond_start_xmit(skb, dev);
4203 	else
4204 		bond_tx_drop(dev, skb);
4205 	rcu_read_unlock();
4206 
4207 	return ret;
4208 }
4209 
4210 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4211 					   struct ethtool_link_ksettings *cmd)
4212 {
4213 	struct bonding *bond = netdev_priv(bond_dev);
4214 	unsigned long speed = 0;
4215 	struct list_head *iter;
4216 	struct slave *slave;
4217 
4218 	cmd->base.duplex = DUPLEX_UNKNOWN;
4219 	cmd->base.port = PORT_OTHER;
4220 
4221 	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4222 	 * do not need to check mode.  Though link speed might not represent
4223 	 * the true receive or transmit bandwidth (not all modes are symmetric)
4224 	 * this is an accurate maximum.
4225 	 */
4226 	bond_for_each_slave(bond, slave, iter) {
4227 		if (bond_slave_can_tx(slave)) {
4228 			if (slave->speed != SPEED_UNKNOWN)
4229 				speed += slave->speed;
4230 			if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4231 			    slave->duplex != DUPLEX_UNKNOWN)
4232 				cmd->base.duplex = slave->duplex;
4233 		}
4234 	}
4235 	cmd->base.speed = speed ? : SPEED_UNKNOWN;
4236 
4237 	return 0;
4238 }
4239 
4240 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4241 				     struct ethtool_drvinfo *drvinfo)
4242 {
4243 	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4244 	strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4245 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4246 		 BOND_ABI_VERSION);
4247 }
4248 
4249 static const struct ethtool_ops bond_ethtool_ops = {
4250 	.get_drvinfo		= bond_ethtool_get_drvinfo,
4251 	.get_link		= ethtool_op_get_link,
4252 	.get_link_ksettings	= bond_ethtool_get_link_ksettings,
4253 };
4254 
4255 static const struct net_device_ops bond_netdev_ops = {
4256 	.ndo_init		= bond_init,
4257 	.ndo_uninit		= bond_uninit,
4258 	.ndo_open		= bond_open,
4259 	.ndo_stop		= bond_close,
4260 	.ndo_start_xmit		= bond_start_xmit,
4261 	.ndo_select_queue	= bond_select_queue,
4262 	.ndo_get_stats64	= bond_get_stats,
4263 	.ndo_do_ioctl		= bond_do_ioctl,
4264 	.ndo_change_rx_flags	= bond_change_rx_flags,
4265 	.ndo_set_rx_mode	= bond_set_rx_mode,
4266 	.ndo_change_mtu		= bond_change_mtu,
4267 	.ndo_set_mac_address	= bond_set_mac_address,
4268 	.ndo_neigh_setup	= bond_neigh_setup,
4269 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
4270 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
4271 	.ndo_get_lock_subclass  = bond_get_nest_level,
4272 #ifdef CONFIG_NET_POLL_CONTROLLER
4273 	.ndo_netpoll_setup	= bond_netpoll_setup,
4274 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
4275 	.ndo_poll_controller	= bond_poll_controller,
4276 #endif
4277 	.ndo_add_slave		= bond_enslave,
4278 	.ndo_del_slave		= bond_release,
4279 	.ndo_fix_features	= bond_fix_features,
4280 	.ndo_features_check	= passthru_features_check,
4281 };
4282 
4283 static const struct device_type bond_type = {
4284 	.name = "bond",
4285 };
4286 
4287 static void bond_destructor(struct net_device *bond_dev)
4288 {
4289 	struct bonding *bond = netdev_priv(bond_dev);
4290 	if (bond->wq)
4291 		destroy_workqueue(bond->wq);
4292 }
4293 
4294 void bond_setup(struct net_device *bond_dev)
4295 {
4296 	struct bonding *bond = netdev_priv(bond_dev);
4297 
4298 	spin_lock_init(&bond->mode_lock);
4299 	spin_lock_init(&bond->stats_lock);
4300 	bond->params = bonding_defaults;
4301 
4302 	/* Initialize pointers */
4303 	bond->dev = bond_dev;
4304 
4305 	/* Initialize the device entry points */
4306 	ether_setup(bond_dev);
4307 	bond_dev->max_mtu = ETH_MAX_MTU;
4308 	bond_dev->netdev_ops = &bond_netdev_ops;
4309 	bond_dev->ethtool_ops = &bond_ethtool_ops;
4310 
4311 	bond_dev->needs_free_netdev = true;
4312 	bond_dev->priv_destructor = bond_destructor;
4313 
4314 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4315 
4316 	/* Initialize the device options */
4317 	bond_dev->flags |= IFF_MASTER;
4318 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4319 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4320 
4321 	/* don't acquire bond device's netif_tx_lock when transmitting */
4322 	bond_dev->features |= NETIF_F_LLTX;
4323 
4324 	/* By default, we declare the bond to be fully
4325 	 * VLAN hardware accelerated capable. Special
4326 	 * care is taken in the various xmit functions
4327 	 * when there are slaves that are not hw accel
4328 	 * capable
4329 	 */
4330 
4331 	/* Don't allow bond devices to change network namespaces. */
4332 	bond_dev->features |= NETIF_F_NETNS_LOCAL;
4333 
4334 	bond_dev->hw_features = BOND_VLAN_FEATURES |
4335 				NETIF_F_HW_VLAN_CTAG_RX |
4336 				NETIF_F_HW_VLAN_CTAG_FILTER;
4337 
4338 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4339 	bond_dev->features |= bond_dev->hw_features;
4340 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4341 }
4342 
4343 /* Destroy a bonding device.
4344  * Must be under rtnl_lock when this function is called.
4345  */
4346 static void bond_uninit(struct net_device *bond_dev)
4347 {
4348 	struct bonding *bond = netdev_priv(bond_dev);
4349 	struct list_head *iter;
4350 	struct slave *slave;
4351 	struct bond_up_slave *arr;
4352 
4353 	bond_netpoll_cleanup(bond_dev);
4354 
4355 	/* Release the bonded slaves */
4356 	bond_for_each_slave(bond, slave, iter)
4357 		__bond_release_one(bond_dev, slave->dev, true, true);
4358 	netdev_info(bond_dev, "Released all slaves\n");
4359 
4360 	arr = rtnl_dereference(bond->slave_arr);
4361 	if (arr) {
4362 		RCU_INIT_POINTER(bond->slave_arr, NULL);
4363 		kfree_rcu(arr, rcu);
4364 	}
4365 
4366 	list_del(&bond->bond_list);
4367 
4368 	bond_debug_unregister(bond);
4369 }
4370 
4371 /*------------------------- Module initialization ---------------------------*/
4372 
4373 static int bond_check_params(struct bond_params *params)
4374 {
4375 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4376 	struct bond_opt_value newval;
4377 	const struct bond_opt_value *valptr;
4378 	int arp_all_targets_value = 0;
4379 	u16 ad_actor_sys_prio = 0;
4380 	u16 ad_user_port_key = 0;
4381 	__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4382 	int arp_ip_count;
4383 	int bond_mode	= BOND_MODE_ROUNDROBIN;
4384 	int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4385 	int lacp_fast = 0;
4386 	int tlb_dynamic_lb;
4387 
4388 	/* Convert string parameters. */
4389 	if (mode) {
4390 		bond_opt_initstr(&newval, mode);
4391 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4392 		if (!valptr) {
4393 			pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4394 			return -EINVAL;
4395 		}
4396 		bond_mode = valptr->value;
4397 	}
4398 
4399 	if (xmit_hash_policy) {
4400 		if (bond_mode == BOND_MODE_ROUNDROBIN ||
4401 		    bond_mode == BOND_MODE_ACTIVEBACKUP ||
4402 		    bond_mode == BOND_MODE_BROADCAST) {
4403 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4404 				bond_mode_name(bond_mode));
4405 		} else {
4406 			bond_opt_initstr(&newval, xmit_hash_policy);
4407 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4408 						&newval);
4409 			if (!valptr) {
4410 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4411 				       xmit_hash_policy);
4412 				return -EINVAL;
4413 			}
4414 			xmit_hashtype = valptr->value;
4415 		}
4416 	}
4417 
4418 	if (lacp_rate) {
4419 		if (bond_mode != BOND_MODE_8023AD) {
4420 			pr_info("lacp_rate param is irrelevant in mode %s\n",
4421 				bond_mode_name(bond_mode));
4422 		} else {
4423 			bond_opt_initstr(&newval, lacp_rate);
4424 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4425 						&newval);
4426 			if (!valptr) {
4427 				pr_err("Error: Invalid lacp rate \"%s\"\n",
4428 				       lacp_rate);
4429 				return -EINVAL;
4430 			}
4431 			lacp_fast = valptr->value;
4432 		}
4433 	}
4434 
4435 	if (ad_select) {
4436 		bond_opt_initstr(&newval, ad_select);
4437 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4438 					&newval);
4439 		if (!valptr) {
4440 			pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4441 			return -EINVAL;
4442 		}
4443 		params->ad_select = valptr->value;
4444 		if (bond_mode != BOND_MODE_8023AD)
4445 			pr_warn("ad_select param only affects 802.3ad mode\n");
4446 	} else {
4447 		params->ad_select = BOND_AD_STABLE;
4448 	}
4449 
4450 	if (max_bonds < 0) {
4451 		pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4452 			max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4453 		max_bonds = BOND_DEFAULT_MAX_BONDS;
4454 	}
4455 
4456 	if (miimon < 0) {
4457 		pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4458 			miimon, INT_MAX);
4459 		miimon = 0;
4460 	}
4461 
4462 	if (updelay < 0) {
4463 		pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4464 			updelay, INT_MAX);
4465 		updelay = 0;
4466 	}
4467 
4468 	if (downdelay < 0) {
4469 		pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4470 			downdelay, INT_MAX);
4471 		downdelay = 0;
4472 	}
4473 
4474 	if ((use_carrier != 0) && (use_carrier != 1)) {
4475 		pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4476 			use_carrier);
4477 		use_carrier = 1;
4478 	}
4479 
4480 	if (num_peer_notif < 0 || num_peer_notif > 255) {
4481 		pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4482 			num_peer_notif);
4483 		num_peer_notif = 1;
4484 	}
4485 
4486 	/* reset values for 802.3ad/TLB/ALB */
4487 	if (!bond_mode_uses_arp(bond_mode)) {
4488 		if (!miimon) {
4489 			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");
4490 			pr_warn("Forcing miimon to 100msec\n");
4491 			miimon = BOND_DEFAULT_MIIMON;
4492 		}
4493 	}
4494 
4495 	if (tx_queues < 1 || tx_queues > 255) {
4496 		pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4497 			tx_queues, BOND_DEFAULT_TX_QUEUES);
4498 		tx_queues = BOND_DEFAULT_TX_QUEUES;
4499 	}
4500 
4501 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4502 		pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4503 			all_slaves_active);
4504 		all_slaves_active = 0;
4505 	}
4506 
4507 	if (resend_igmp < 0 || resend_igmp > 255) {
4508 		pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4509 			resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4510 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4511 	}
4512 
4513 	bond_opt_initval(&newval, packets_per_slave);
4514 	if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4515 		pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4516 			packets_per_slave, USHRT_MAX);
4517 		packets_per_slave = 1;
4518 	}
4519 
4520 	if (bond_mode == BOND_MODE_ALB) {
4521 		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",
4522 			  updelay);
4523 	}
4524 
4525 	if (!miimon) {
4526 		if (updelay || downdelay) {
4527 			/* just warn the user the up/down delay will have
4528 			 * no effect since miimon is zero...
4529 			 */
4530 			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",
4531 				updelay, downdelay);
4532 		}
4533 	} else {
4534 		/* don't allow arp monitoring */
4535 		if (arp_interval) {
4536 			pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4537 				miimon, arp_interval);
4538 			arp_interval = 0;
4539 		}
4540 
4541 		if ((updelay % miimon) != 0) {
4542 			pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4543 				updelay, miimon, (updelay / miimon) * miimon);
4544 		}
4545 
4546 		updelay /= miimon;
4547 
4548 		if ((downdelay % miimon) != 0) {
4549 			pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4550 				downdelay, miimon,
4551 				(downdelay / miimon) * miimon);
4552 		}
4553 
4554 		downdelay /= miimon;
4555 	}
4556 
4557 	if (arp_interval < 0) {
4558 		pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4559 			arp_interval, INT_MAX);
4560 		arp_interval = 0;
4561 	}
4562 
4563 	for (arp_ip_count = 0, i = 0;
4564 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4565 		__be32 ip;
4566 
4567 		/* not a complete check, but good enough to catch mistakes */
4568 		if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4569 		    !bond_is_ip_target_ok(ip)) {
4570 			pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4571 				arp_ip_target[i]);
4572 			arp_interval = 0;
4573 		} else {
4574 			if (bond_get_targets_ip(arp_target, ip) == -1)
4575 				arp_target[arp_ip_count++] = ip;
4576 			else
4577 				pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4578 					&ip);
4579 		}
4580 	}
4581 
4582 	if (arp_interval && !arp_ip_count) {
4583 		/* don't allow arping if no arp_ip_target given... */
4584 		pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4585 			arp_interval);
4586 		arp_interval = 0;
4587 	}
4588 
4589 	if (arp_validate) {
4590 		if (!arp_interval) {
4591 			pr_err("arp_validate requires arp_interval\n");
4592 			return -EINVAL;
4593 		}
4594 
4595 		bond_opt_initstr(&newval, arp_validate);
4596 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4597 					&newval);
4598 		if (!valptr) {
4599 			pr_err("Error: invalid arp_validate \"%s\"\n",
4600 			       arp_validate);
4601 			return -EINVAL;
4602 		}
4603 		arp_validate_value = valptr->value;
4604 	} else {
4605 		arp_validate_value = 0;
4606 	}
4607 
4608 	if (arp_all_targets) {
4609 		bond_opt_initstr(&newval, arp_all_targets);
4610 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4611 					&newval);
4612 		if (!valptr) {
4613 			pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4614 			       arp_all_targets);
4615 			arp_all_targets_value = 0;
4616 		} else {
4617 			arp_all_targets_value = valptr->value;
4618 		}
4619 	}
4620 
4621 	if (miimon) {
4622 		pr_info("MII link monitoring set to %d ms\n", miimon);
4623 	} else if (arp_interval) {
4624 		valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4625 					  arp_validate_value);
4626 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4627 			arp_interval, valptr->string, arp_ip_count);
4628 
4629 		for (i = 0; i < arp_ip_count; i++)
4630 			pr_cont(" %s", arp_ip_target[i]);
4631 
4632 		pr_cont("\n");
4633 
4634 	} else if (max_bonds) {
4635 		/* miimon and arp_interval not set, we need one so things
4636 		 * work as expected, see bonding.txt for details
4637 		 */
4638 		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");
4639 	}
4640 
4641 	if (primary && !bond_mode_uses_primary(bond_mode)) {
4642 		/* currently, using a primary only makes sense
4643 		 * in active backup, TLB or ALB modes
4644 		 */
4645 		pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4646 			primary, bond_mode_name(bond_mode));
4647 		primary = NULL;
4648 	}
4649 
4650 	if (primary && primary_reselect) {
4651 		bond_opt_initstr(&newval, primary_reselect);
4652 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4653 					&newval);
4654 		if (!valptr) {
4655 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
4656 			       primary_reselect);
4657 			return -EINVAL;
4658 		}
4659 		primary_reselect_value = valptr->value;
4660 	} else {
4661 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4662 	}
4663 
4664 	if (fail_over_mac) {
4665 		bond_opt_initstr(&newval, fail_over_mac);
4666 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4667 					&newval);
4668 		if (!valptr) {
4669 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
4670 			       fail_over_mac);
4671 			return -EINVAL;
4672 		}
4673 		fail_over_mac_value = valptr->value;
4674 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4675 			pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4676 	} else {
4677 		fail_over_mac_value = BOND_FOM_NONE;
4678 	}
4679 
4680 	bond_opt_initstr(&newval, "default");
4681 	valptr = bond_opt_parse(
4682 			bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4683 				     &newval);
4684 	if (!valptr) {
4685 		pr_err("Error: No ad_actor_sys_prio default value");
4686 		return -EINVAL;
4687 	}
4688 	ad_actor_sys_prio = valptr->value;
4689 
4690 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4691 				&newval);
4692 	if (!valptr) {
4693 		pr_err("Error: No ad_user_port_key default value");
4694 		return -EINVAL;
4695 	}
4696 	ad_user_port_key = valptr->value;
4697 
4698 	bond_opt_initstr(&newval, "default");
4699 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4700 	if (!valptr) {
4701 		pr_err("Error: No tlb_dynamic_lb default value");
4702 		return -EINVAL;
4703 	}
4704 	tlb_dynamic_lb = valptr->value;
4705 
4706 	if (lp_interval == 0) {
4707 		pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4708 			INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4709 		lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4710 	}
4711 
4712 	/* fill params struct with the proper values */
4713 	params->mode = bond_mode;
4714 	params->xmit_policy = xmit_hashtype;
4715 	params->miimon = miimon;
4716 	params->num_peer_notif = num_peer_notif;
4717 	params->arp_interval = arp_interval;
4718 	params->arp_validate = arp_validate_value;
4719 	params->arp_all_targets = arp_all_targets_value;
4720 	params->updelay = updelay;
4721 	params->downdelay = downdelay;
4722 	params->peer_notif_delay = 0;
4723 	params->use_carrier = use_carrier;
4724 	params->lacp_fast = lacp_fast;
4725 	params->primary[0] = 0;
4726 	params->primary_reselect = primary_reselect_value;
4727 	params->fail_over_mac = fail_over_mac_value;
4728 	params->tx_queues = tx_queues;
4729 	params->all_slaves_active = all_slaves_active;
4730 	params->resend_igmp = resend_igmp;
4731 	params->min_links = min_links;
4732 	params->lp_interval = lp_interval;
4733 	params->packets_per_slave = packets_per_slave;
4734 	params->tlb_dynamic_lb = tlb_dynamic_lb;
4735 	params->ad_actor_sys_prio = ad_actor_sys_prio;
4736 	eth_zero_addr(params->ad_actor_system);
4737 	params->ad_user_port_key = ad_user_port_key;
4738 	if (packets_per_slave > 0) {
4739 		params->reciprocal_packets_per_slave =
4740 			reciprocal_value(packets_per_slave);
4741 	} else {
4742 		/* reciprocal_packets_per_slave is unused if
4743 		 * packets_per_slave is 0 or 1, just initialize it
4744 		 */
4745 		params->reciprocal_packets_per_slave =
4746 			(struct reciprocal_value) { 0 };
4747 	}
4748 
4749 	if (primary) {
4750 		strncpy(params->primary, primary, IFNAMSIZ);
4751 		params->primary[IFNAMSIZ - 1] = 0;
4752 	}
4753 
4754 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4755 
4756 	return 0;
4757 }
4758 
4759 /* Called from registration process */
4760 static int bond_init(struct net_device *bond_dev)
4761 {
4762 	struct bonding *bond = netdev_priv(bond_dev);
4763 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4764 
4765 	netdev_dbg(bond_dev, "Begin bond_init\n");
4766 
4767 	bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4768 	if (!bond->wq)
4769 		return -ENOMEM;
4770 
4771 	bond->nest_level = SINGLE_DEPTH_NESTING;
4772 	netdev_lockdep_set_classes(bond_dev);
4773 
4774 	list_add_tail(&bond->bond_list, &bn->dev_list);
4775 
4776 	bond_prepare_sysfs_group(bond);
4777 
4778 	bond_debug_register(bond);
4779 
4780 	/* Ensure valid dev_addr */
4781 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
4782 	    bond_dev->addr_assign_type == NET_ADDR_PERM)
4783 		eth_hw_addr_random(bond_dev);
4784 
4785 	return 0;
4786 }
4787 
4788 unsigned int bond_get_num_tx_queues(void)
4789 {
4790 	return tx_queues;
4791 }
4792 
4793 /* Create a new bond based on the specified name and bonding parameters.
4794  * If name is NULL, obtain a suitable "bond%d" name for us.
4795  * Caller must NOT hold rtnl_lock; we need to release it here before we
4796  * set up our sysfs entries.
4797  */
4798 int bond_create(struct net *net, const char *name)
4799 {
4800 	struct net_device *bond_dev;
4801 	struct bonding *bond;
4802 	struct alb_bond_info *bond_info;
4803 	int res;
4804 
4805 	rtnl_lock();
4806 
4807 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4808 				   name ? name : "bond%d", NET_NAME_UNKNOWN,
4809 				   bond_setup, tx_queues);
4810 	if (!bond_dev) {
4811 		pr_err("%s: eek! can't alloc netdev!\n", name);
4812 		rtnl_unlock();
4813 		return -ENOMEM;
4814 	}
4815 
4816 	/*
4817 	 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4818 	 * It is set to 0 by default which is wrong.
4819 	 */
4820 	bond = netdev_priv(bond_dev);
4821 	bond_info = &(BOND_ALB_INFO(bond));
4822 	bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4823 
4824 	dev_net_set(bond_dev, net);
4825 	bond_dev->rtnl_link_ops = &bond_link_ops;
4826 
4827 	res = register_netdevice(bond_dev);
4828 
4829 	netif_carrier_off(bond_dev);
4830 
4831 	bond_work_init_all(bond);
4832 
4833 	rtnl_unlock();
4834 	if (res < 0)
4835 		free_netdev(bond_dev);
4836 	return res;
4837 }
4838 
4839 static int __net_init bond_net_init(struct net *net)
4840 {
4841 	struct bond_net *bn = net_generic(net, bond_net_id);
4842 
4843 	bn->net = net;
4844 	INIT_LIST_HEAD(&bn->dev_list);
4845 
4846 	bond_create_proc_dir(bn);
4847 	bond_create_sysfs(bn);
4848 
4849 	return 0;
4850 }
4851 
4852 static void __net_exit bond_net_exit(struct net *net)
4853 {
4854 	struct bond_net *bn = net_generic(net, bond_net_id);
4855 	struct bonding *bond, *tmp_bond;
4856 	LIST_HEAD(list);
4857 
4858 	bond_destroy_sysfs(bn);
4859 
4860 	/* Kill off any bonds created after unregistering bond rtnl ops */
4861 	rtnl_lock();
4862 	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4863 		unregister_netdevice_queue(bond->dev, &list);
4864 	unregister_netdevice_many(&list);
4865 	rtnl_unlock();
4866 
4867 	bond_destroy_proc_dir(bn);
4868 }
4869 
4870 static struct pernet_operations bond_net_ops = {
4871 	.init = bond_net_init,
4872 	.exit = bond_net_exit,
4873 	.id   = &bond_net_id,
4874 	.size = sizeof(struct bond_net),
4875 };
4876 
4877 static int __init bonding_init(void)
4878 {
4879 	int i;
4880 	int res;
4881 
4882 	pr_info("%s", bond_version);
4883 
4884 	res = bond_check_params(&bonding_defaults);
4885 	if (res)
4886 		goto out;
4887 
4888 	res = register_pernet_subsys(&bond_net_ops);
4889 	if (res)
4890 		goto out;
4891 
4892 	res = bond_netlink_init();
4893 	if (res)
4894 		goto err_link;
4895 
4896 	bond_create_debugfs();
4897 
4898 	for (i = 0; i < max_bonds; i++) {
4899 		res = bond_create(&init_net, NULL);
4900 		if (res)
4901 			goto err;
4902 	}
4903 
4904 	register_netdevice_notifier(&bond_netdev_notifier);
4905 out:
4906 	return res;
4907 err:
4908 	bond_destroy_debugfs();
4909 	bond_netlink_fini();
4910 err_link:
4911 	unregister_pernet_subsys(&bond_net_ops);
4912 	goto out;
4913 
4914 }
4915 
4916 static void __exit bonding_exit(void)
4917 {
4918 	unregister_netdevice_notifier(&bond_netdev_notifier);
4919 
4920 	bond_destroy_debugfs();
4921 
4922 	bond_netlink_fini();
4923 	unregister_pernet_subsys(&bond_net_ops);
4924 
4925 #ifdef CONFIG_NET_POLL_CONTROLLER
4926 	/* Make sure we don't have an imbalance on our netpoll blocking */
4927 	WARN_ON(atomic_read(&netpoll_block_tx));
4928 #endif
4929 }
4930 
4931 module_init(bonding_init);
4932 module_exit(bonding_exit);
4933 MODULE_LICENSE("GPL");
4934 MODULE_VERSION(DRV_VERSION);
4935 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4936 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4937