xref: /linux/drivers/net/bonding/bond_main.c (revision 235f0da3274690f540aa53fccf77d433e344e4b8)
1 // SPDX-License-Identifier: GPL-1.0+
2 /*
3  * originally based on the dummy device.
4  *
5  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
6  * Based on dummy.c, and eql.c devices.
7  *
8  * bonding.c: an Ethernet Bonding driver
9  *
10  * This is useful to talk to a Cisco EtherChannel compatible equipment:
11  *	Cisco 5500
12  *	Sun Trunking (Solaris)
13  *	Alteon AceDirector Trunks
14  *	Linux Bonding
15  *	and probably many L2 switches ...
16  *
17  * How it works:
18  *    ifconfig bond0 ipaddress netmask up
19  *      will setup a network device, with an ip address.  No mac address
20  *	will be assigned at this time.  The hw mac address will come from
21  *	the first slave bonded to the channel.  All slaves will then use
22  *	this hw mac address.
23  *
24  *    ifconfig bond0 down
25  *         will release all slaves, marking them as down.
26  *
27  *    ifenslave bond0 eth0
28  *	will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
29  *	a: be used as initial mac address
30  *	b: if a hw mac address already is there, eth0's hw mac address
31  *	   will then be set from bond0.
32  *
33  */
34 
35 #include <linux/kernel.h>
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/fcntl.h>
39 #include <linux/filter.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/icmp.h>
47 #include <linux/icmpv6.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <linux/io.h>
59 #include <asm/dma.h>
60 #include <linux/uaccess.h>
61 #include <linux/errno.h>
62 #include <linux/netdevice.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/smp.h>
70 #include <linux/if_ether.h>
71 #include <net/arp.h>
72 #include <linux/mii.h>
73 #include <linux/ethtool.h>
74 #include <linux/if_vlan.h>
75 #include <linux/if_bonding.h>
76 #include <linux/phy.h>
77 #include <linux/jiffies.h>
78 #include <linux/preempt.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include <net/pkt_sched.h>
83 #include <linux/rculist.h>
84 #include <net/flow_dissector.h>
85 #include <net/xfrm.h>
86 #include <net/bonding.h>
87 #include <net/bond_3ad.h>
88 #include <net/bond_alb.h>
89 #if IS_ENABLED(CONFIG_TLS_DEVICE)
90 #include <net/tls.h>
91 #endif
92 #include <net/ip6_route.h>
93 #include <net/xdp.h>
94 
95 #include "bonding_priv.h"
96 
97 /*---------------------------- Module parameters ----------------------------*/
98 
99 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
100 
101 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
102 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
103 static int num_peer_notif = 1;
104 static int miimon;
105 static int updelay;
106 static int downdelay;
107 static int use_carrier	= 1;
108 static char *mode;
109 static char *primary;
110 static char *primary_reselect;
111 static char *lacp_rate;
112 static int min_links;
113 static char *ad_select;
114 static char *xmit_hash_policy;
115 static int arp_interval;
116 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
117 static char *arp_validate;
118 static char *arp_all_targets;
119 static char *fail_over_mac;
120 static int all_slaves_active;
121 static struct bond_params bonding_defaults;
122 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
123 static int packets_per_slave = 1;
124 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
125 
126 module_param(max_bonds, int, 0);
127 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
128 module_param(tx_queues, int, 0);
129 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
130 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
131 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
132 			       "failover event (alias of num_unsol_na)");
133 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
134 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
135 			       "failover event (alias of num_grat_arp)");
136 module_param(miimon, int, 0);
137 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
138 module_param(updelay, int, 0);
139 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
140 module_param(downdelay, int, 0);
141 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
142 			    "in milliseconds");
143 module_param(use_carrier, int, 0);
144 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
145 			      "0 for off, 1 for on (default)");
146 module_param(mode, charp, 0);
147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
148 		       "1 for active-backup, 2 for balance-xor, "
149 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
150 		       "6 for balance-alb");
151 module_param(primary, charp, 0);
152 MODULE_PARM_DESC(primary, "Primary network device to use");
153 module_param(primary_reselect, charp, 0);
154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
155 				   "once it comes up; "
156 				   "0 for always (default), "
157 				   "1 for only if speed of primary is "
158 				   "better, "
159 				   "2 for only on active slave "
160 				   "failure");
161 module_param(lacp_rate, charp, 0);
162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163 			    "0 for slow, 1 for fast");
164 module_param(ad_select, charp, 0);
165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166 			    "0 for stable (default), 1 for bandwidth, "
167 			    "2 for count");
168 module_param(min_links, int, 0);
169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
170 
171 module_param(xmit_hash_policy, charp, 0);
172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
173 				   "0 for layer 2 (default), 1 for layer 3+4, "
174 				   "2 for layer 2+3, 3 for encap layer 2+3, "
175 				   "4 for encap layer 3+4, 5 for vlan+srcmac");
176 module_param(arp_interval, int, 0);
177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178 module_param_array(arp_ip_target, charp, NULL, 0);
179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180 module_param(arp_validate, charp, 0);
181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
182 			       "0 for none (default), 1 for active, "
183 			       "2 for backup, 3 for all");
184 module_param(arp_all_targets, charp, 0);
185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186 module_param(fail_over_mac, charp, 0);
187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
188 				"the same MAC; 0 for none (default), "
189 				"1 for active, 2 for follow");
190 module_param(all_slaves_active, int, 0);
191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
192 				     "by setting active flag for all slaves; "
193 				     "0 for never (default), 1 for always.");
194 module_param(resend_igmp, int, 0);
195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
196 			      "link failure");
197 module_param(packets_per_slave, int, 0);
198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
199 				    "mode; 0 for a random slave, 1 packet per "
200 				    "slave (default), >1 packets per slave.");
201 module_param(lp_interval, uint, 0);
202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
203 			      "the bonding driver sends learning packets to "
204 			      "each slaves peer switch. The default is 1.");
205 
206 /*----------------------------- Global variables ----------------------------*/
207 
208 #ifdef CONFIG_NET_POLL_CONTROLLER
209 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210 #endif
211 
212 unsigned int bond_net_id __read_mostly;
213 
214 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
215 	{
216 		.key_id = FLOW_DISSECTOR_KEY_CONTROL,
217 		.offset = offsetof(struct flow_keys, control),
218 	},
219 	{
220 		.key_id = FLOW_DISSECTOR_KEY_BASIC,
221 		.offset = offsetof(struct flow_keys, basic),
222 	},
223 	{
224 		.key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
225 		.offset = offsetof(struct flow_keys, addrs.v4addrs),
226 	},
227 	{
228 		.key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
229 		.offset = offsetof(struct flow_keys, addrs.v6addrs),
230 	},
231 	{
232 		.key_id = FLOW_DISSECTOR_KEY_TIPC,
233 		.offset = offsetof(struct flow_keys, addrs.tipckey),
234 	},
235 	{
236 		.key_id = FLOW_DISSECTOR_KEY_PORTS,
237 		.offset = offsetof(struct flow_keys, ports),
238 	},
239 	{
240 		.key_id = FLOW_DISSECTOR_KEY_ICMP,
241 		.offset = offsetof(struct flow_keys, icmp),
242 	},
243 	{
244 		.key_id = FLOW_DISSECTOR_KEY_VLAN,
245 		.offset = offsetof(struct flow_keys, vlan),
246 	},
247 	{
248 		.key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
249 		.offset = offsetof(struct flow_keys, tags),
250 	},
251 	{
252 		.key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
253 		.offset = offsetof(struct flow_keys, keyid),
254 	},
255 };
256 
257 static struct flow_dissector flow_keys_bonding __read_mostly;
258 
259 /*-------------------------- Forward declarations ---------------------------*/
260 
261 static int bond_init(struct net_device *bond_dev);
262 static void bond_uninit(struct net_device *bond_dev);
263 static void bond_get_stats(struct net_device *bond_dev,
264 			   struct rtnl_link_stats64 *stats);
265 static void bond_slave_arr_handler(struct work_struct *work);
266 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
267 				  int mod);
268 static void bond_netdev_notify_work(struct work_struct *work);
269 
270 /*---------------------------- General routines -----------------------------*/
271 
272 const char *bond_mode_name(int mode)
273 {
274 	static const char *names[] = {
275 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
276 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
277 		[BOND_MODE_XOR] = "load balancing (xor)",
278 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
279 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
280 		[BOND_MODE_TLB] = "transmit load balancing",
281 		[BOND_MODE_ALB] = "adaptive load balancing",
282 	};
283 
284 	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
285 		return "unknown";
286 
287 	return names[mode];
288 }
289 
290 /**
291  * bond_dev_queue_xmit - Prepare skb for xmit.
292  *
293  * @bond: bond device that got this skb for tx.
294  * @skb: hw accel VLAN tagged skb to transmit
295  * @slave_dev: slave that is supposed to xmit this skbuff
296  */
297 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
298 			struct net_device *slave_dev)
299 {
300 	skb->dev = slave_dev;
301 
302 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
303 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
304 	skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
305 
306 	if (unlikely(netpoll_tx_running(bond->dev)))
307 		return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
308 
309 	return dev_queue_xmit(skb);
310 }
311 
312 static bool bond_sk_check(struct bonding *bond)
313 {
314 	switch (BOND_MODE(bond)) {
315 	case BOND_MODE_8023AD:
316 	case BOND_MODE_XOR:
317 		if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
318 			return true;
319 		fallthrough;
320 	default:
321 		return false;
322 	}
323 }
324 
325 static bool bond_xdp_check(struct bonding *bond)
326 {
327 	switch (BOND_MODE(bond)) {
328 	case BOND_MODE_ROUNDROBIN:
329 	case BOND_MODE_ACTIVEBACKUP:
330 		return true;
331 	case BOND_MODE_8023AD:
332 	case BOND_MODE_XOR:
333 		/* vlan+srcmac is not supported with XDP as in most cases the 802.1q
334 		 * payload is not in the packet due to hardware offload.
335 		 */
336 		if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
337 			return true;
338 		fallthrough;
339 	default:
340 		return false;
341 	}
342 }
343 
344 /*---------------------------------- VLAN -----------------------------------*/
345 
346 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
347  * We don't protect the slave list iteration with a lock because:
348  * a. This operation is performed in IOCTL context,
349  * b. The operation is protected by the RTNL semaphore in the 8021q code,
350  * c. Holding a lock with BH disabled while directly calling a base driver
351  *    entry point is generally a BAD idea.
352  *
353  * The design of synchronization/protection for this operation in the 8021q
354  * module is good for one or more VLAN devices over a single physical device
355  * and cannot be extended for a teaming solution like bonding, so there is a
356  * potential race condition here where a net device from the vlan group might
357  * be referenced (either by a base driver or the 8021q code) while it is being
358  * removed from the system. However, it turns out we're not making matters
359  * worse, and if it works for regular VLAN usage it will work here too.
360 */
361 
362 /**
363  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
364  * @bond_dev: bonding net device that got called
365  * @proto: network protocol ID
366  * @vid: vlan id being added
367  */
368 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
369 				__be16 proto, u16 vid)
370 {
371 	struct bonding *bond = netdev_priv(bond_dev);
372 	struct slave *slave, *rollback_slave;
373 	struct list_head *iter;
374 	int res;
375 
376 	bond_for_each_slave(bond, slave, iter) {
377 		res = vlan_vid_add(slave->dev, proto, vid);
378 		if (res)
379 			goto unwind;
380 	}
381 
382 	return 0;
383 
384 unwind:
385 	/* unwind to the slave that failed */
386 	bond_for_each_slave(bond, rollback_slave, iter) {
387 		if (rollback_slave == slave)
388 			break;
389 
390 		vlan_vid_del(rollback_slave->dev, proto, vid);
391 	}
392 
393 	return res;
394 }
395 
396 /**
397  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
398  * @bond_dev: bonding net device that got called
399  * @proto: network protocol ID
400  * @vid: vlan id being removed
401  */
402 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
403 				 __be16 proto, u16 vid)
404 {
405 	struct bonding *bond = netdev_priv(bond_dev);
406 	struct list_head *iter;
407 	struct slave *slave;
408 
409 	bond_for_each_slave(bond, slave, iter)
410 		vlan_vid_del(slave->dev, proto, vid);
411 
412 	if (bond_is_lb(bond))
413 		bond_alb_clear_vlan(bond, vid);
414 
415 	return 0;
416 }
417 
418 /*---------------------------------- XFRM -----------------------------------*/
419 
420 #ifdef CONFIG_XFRM_OFFLOAD
421 /**
422  * bond_ipsec_add_sa - program device with a security association
423  * @xs: pointer to transformer state struct
424  * @extack: extack point to fill failure reason
425  **/
426 static int bond_ipsec_add_sa(struct xfrm_state *xs,
427 			     struct netlink_ext_ack *extack)
428 {
429 	struct net_device *bond_dev = xs->xso.dev;
430 	struct net_device *real_dev;
431 	netdevice_tracker tracker;
432 	struct bond_ipsec *ipsec;
433 	struct bonding *bond;
434 	struct slave *slave;
435 	int err;
436 
437 	if (!bond_dev)
438 		return -EINVAL;
439 
440 	rcu_read_lock();
441 	bond = netdev_priv(bond_dev);
442 	slave = rcu_dereference(bond->curr_active_slave);
443 	real_dev = slave ? slave->dev : NULL;
444 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
445 	rcu_read_unlock();
446 	if (!real_dev) {
447 		err = -ENODEV;
448 		goto out;
449 	}
450 
451 	if (!real_dev->xfrmdev_ops ||
452 	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
453 	    netif_is_bond_master(real_dev)) {
454 		NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
455 		err = -EINVAL;
456 		goto out;
457 	}
458 
459 	ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL);
460 	if (!ipsec) {
461 		err = -ENOMEM;
462 		goto out;
463 	}
464 
465 	xs->xso.real_dev = real_dev;
466 	err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
467 	if (!err) {
468 		ipsec->xs = xs;
469 		INIT_LIST_HEAD(&ipsec->list);
470 		mutex_lock(&bond->ipsec_lock);
471 		list_add(&ipsec->list, &bond->ipsec_list);
472 		mutex_unlock(&bond->ipsec_lock);
473 	} else {
474 		kfree(ipsec);
475 	}
476 out:
477 	netdev_put(real_dev, &tracker);
478 	return err;
479 }
480 
481 static void bond_ipsec_add_sa_all(struct bonding *bond)
482 {
483 	struct net_device *bond_dev = bond->dev;
484 	struct net_device *real_dev;
485 	struct bond_ipsec *ipsec;
486 	struct slave *slave;
487 
488 	slave = rtnl_dereference(bond->curr_active_slave);
489 	real_dev = slave ? slave->dev : NULL;
490 	if (!real_dev)
491 		return;
492 
493 	mutex_lock(&bond->ipsec_lock);
494 	if (!real_dev->xfrmdev_ops ||
495 	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
496 	    netif_is_bond_master(real_dev)) {
497 		if (!list_empty(&bond->ipsec_list))
498 			slave_warn(bond_dev, real_dev,
499 				   "%s: no slave xdo_dev_state_add\n",
500 				   __func__);
501 		goto out;
502 	}
503 
504 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
505 		/* If new state is added before ipsec_lock acquired */
506 		if (ipsec->xs->xso.real_dev == real_dev)
507 			continue;
508 
509 		ipsec->xs->xso.real_dev = real_dev;
510 		if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
511 			slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__);
512 			ipsec->xs->xso.real_dev = NULL;
513 		}
514 	}
515 out:
516 	mutex_unlock(&bond->ipsec_lock);
517 }
518 
519 /**
520  * bond_ipsec_del_sa - clear out this specific SA
521  * @xs: pointer to transformer state struct
522  **/
523 static void bond_ipsec_del_sa(struct xfrm_state *xs)
524 {
525 	struct net_device *bond_dev = xs->xso.dev;
526 	struct net_device *real_dev;
527 	netdevice_tracker tracker;
528 	struct bond_ipsec *ipsec;
529 	struct bonding *bond;
530 	struct slave *slave;
531 
532 	if (!bond_dev)
533 		return;
534 
535 	rcu_read_lock();
536 	bond = netdev_priv(bond_dev);
537 	slave = rcu_dereference(bond->curr_active_slave);
538 	real_dev = slave ? slave->dev : NULL;
539 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
540 	rcu_read_unlock();
541 
542 	if (!slave)
543 		goto out;
544 
545 	if (!xs->xso.real_dev)
546 		goto out;
547 
548 	WARN_ON(xs->xso.real_dev != real_dev);
549 
550 	if (!real_dev->xfrmdev_ops ||
551 	    !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
552 	    netif_is_bond_master(real_dev)) {
553 		slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__);
554 		goto out;
555 	}
556 
557 	real_dev->xfrmdev_ops->xdo_dev_state_delete(xs);
558 out:
559 	netdev_put(real_dev, &tracker);
560 	mutex_lock(&bond->ipsec_lock);
561 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
562 		if (ipsec->xs == xs) {
563 			list_del(&ipsec->list);
564 			kfree(ipsec);
565 			break;
566 		}
567 	}
568 	mutex_unlock(&bond->ipsec_lock);
569 }
570 
571 static void bond_ipsec_del_sa_all(struct bonding *bond)
572 {
573 	struct net_device *bond_dev = bond->dev;
574 	struct net_device *real_dev;
575 	struct bond_ipsec *ipsec;
576 	struct slave *slave;
577 
578 	slave = rtnl_dereference(bond->curr_active_slave);
579 	real_dev = slave ? slave->dev : NULL;
580 	if (!real_dev)
581 		return;
582 
583 	mutex_lock(&bond->ipsec_lock);
584 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
585 		if (!ipsec->xs->xso.real_dev)
586 			continue;
587 
588 		if (!real_dev->xfrmdev_ops ||
589 		    !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
590 		    netif_is_bond_master(real_dev)) {
591 			slave_warn(bond_dev, real_dev,
592 				   "%s: no slave xdo_dev_state_delete\n",
593 				   __func__);
594 		} else {
595 			real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
596 			if (real_dev->xfrmdev_ops->xdo_dev_state_free)
597 				real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs);
598 		}
599 	}
600 	mutex_unlock(&bond->ipsec_lock);
601 }
602 
603 static void bond_ipsec_free_sa(struct xfrm_state *xs)
604 {
605 	struct net_device *bond_dev = xs->xso.dev;
606 	struct net_device *real_dev;
607 	netdevice_tracker tracker;
608 	struct bonding *bond;
609 	struct slave *slave;
610 
611 	if (!bond_dev)
612 		return;
613 
614 	rcu_read_lock();
615 	bond = netdev_priv(bond_dev);
616 	slave = rcu_dereference(bond->curr_active_slave);
617 	real_dev = slave ? slave->dev : NULL;
618 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
619 	rcu_read_unlock();
620 
621 	if (!slave)
622 		goto out;
623 
624 	if (!xs->xso.real_dev)
625 		goto out;
626 
627 	WARN_ON(xs->xso.real_dev != real_dev);
628 
629 	if (real_dev && real_dev->xfrmdev_ops &&
630 	    real_dev->xfrmdev_ops->xdo_dev_state_free)
631 		real_dev->xfrmdev_ops->xdo_dev_state_free(xs);
632 out:
633 	netdev_put(real_dev, &tracker);
634 }
635 
636 /**
637  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
638  * @skb: current data packet
639  * @xs: pointer to transformer state struct
640  **/
641 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
642 {
643 	struct net_device *bond_dev = xs->xso.dev;
644 	struct net_device *real_dev;
645 	struct slave *curr_active;
646 	struct bonding *bond;
647 	bool ok = false;
648 
649 	bond = netdev_priv(bond_dev);
650 	rcu_read_lock();
651 	curr_active = rcu_dereference(bond->curr_active_slave);
652 	if (!curr_active)
653 		goto out;
654 	real_dev = curr_active->dev;
655 
656 	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
657 		goto out;
658 
659 	if (!xs->xso.real_dev)
660 		goto out;
661 
662 	if (!real_dev->xfrmdev_ops ||
663 	    !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
664 	    netif_is_bond_master(real_dev))
665 		goto out;
666 
667 	ok = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
668 out:
669 	rcu_read_unlock();
670 	return ok;
671 }
672 
673 static const struct xfrmdev_ops bond_xfrmdev_ops = {
674 	.xdo_dev_state_add = bond_ipsec_add_sa,
675 	.xdo_dev_state_delete = bond_ipsec_del_sa,
676 	.xdo_dev_state_free = bond_ipsec_free_sa,
677 	.xdo_dev_offload_ok = bond_ipsec_offload_ok,
678 };
679 #endif /* CONFIG_XFRM_OFFLOAD */
680 
681 /*------------------------------- Link status -------------------------------*/
682 
683 /* Set the carrier state for the master according to the state of its
684  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
685  * do special 802.3ad magic.
686  *
687  * Returns zero if carrier state does not change, nonzero if it does.
688  */
689 int bond_set_carrier(struct bonding *bond)
690 {
691 	struct list_head *iter;
692 	struct slave *slave;
693 
694 	if (!bond_has_slaves(bond))
695 		goto down;
696 
697 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
698 		return bond_3ad_set_carrier(bond);
699 
700 	bond_for_each_slave(bond, slave, iter) {
701 		if (slave->link == BOND_LINK_UP) {
702 			if (!netif_carrier_ok(bond->dev)) {
703 				netif_carrier_on(bond->dev);
704 				return 1;
705 			}
706 			return 0;
707 		}
708 	}
709 
710 down:
711 	if (netif_carrier_ok(bond->dev)) {
712 		netif_carrier_off(bond->dev);
713 		return 1;
714 	}
715 	return 0;
716 }
717 
718 /* Get link speed and duplex from the slave's base driver
719  * using ethtool. If for some reason the call fails or the
720  * values are invalid, set speed and duplex to -1,
721  * and return. Return 1 if speed or duplex settings are
722  * UNKNOWN; 0 otherwise.
723  */
724 static int bond_update_speed_duplex(struct slave *slave)
725 {
726 	struct net_device *slave_dev = slave->dev;
727 	struct ethtool_link_ksettings ecmd;
728 	int res;
729 
730 	slave->speed = SPEED_UNKNOWN;
731 	slave->duplex = DUPLEX_UNKNOWN;
732 
733 	res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
734 	if (res < 0)
735 		return 1;
736 	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
737 		return 1;
738 	switch (ecmd.base.duplex) {
739 	case DUPLEX_FULL:
740 	case DUPLEX_HALF:
741 		break;
742 	default:
743 		return 1;
744 	}
745 
746 	slave->speed = ecmd.base.speed;
747 	slave->duplex = ecmd.base.duplex;
748 
749 	return 0;
750 }
751 
752 const char *bond_slave_link_status(s8 link)
753 {
754 	switch (link) {
755 	case BOND_LINK_UP:
756 		return "up";
757 	case BOND_LINK_FAIL:
758 		return "going down";
759 	case BOND_LINK_DOWN:
760 		return "down";
761 	case BOND_LINK_BACK:
762 		return "going back";
763 	default:
764 		return "unknown";
765 	}
766 }
767 
768 /* if <dev> supports MII link status reporting, check its link status.
769  *
770  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
771  * depending upon the setting of the use_carrier parameter.
772  *
773  * Return either BMSR_LSTATUS, meaning that the link is up (or we
774  * can't tell and just pretend it is), or 0, meaning that the link is
775  * down.
776  *
777  * If reporting is non-zero, instead of faking link up, return -1 if
778  * both ETHTOOL and MII ioctls fail (meaning the device does not
779  * support them).  If use_carrier is set, return whatever it says.
780  * It'd be nice if there was a good way to tell if a driver supports
781  * netif_carrier, but there really isn't.
782  */
783 static int bond_check_dev_link(struct bonding *bond,
784 			       struct net_device *slave_dev, int reporting)
785 {
786 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
787 	int (*ioctl)(struct net_device *, struct ifreq *, int);
788 	struct ifreq ifr;
789 	struct mii_ioctl_data *mii;
790 
791 	if (!reporting && !netif_running(slave_dev))
792 		return 0;
793 
794 	if (bond->params.use_carrier)
795 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
796 
797 	/* Try to get link status using Ethtool first. */
798 	if (slave_dev->ethtool_ops->get_link)
799 		return slave_dev->ethtool_ops->get_link(slave_dev) ?
800 			BMSR_LSTATUS : 0;
801 
802 	/* Ethtool can't be used, fallback to MII ioctls. */
803 	ioctl = slave_ops->ndo_eth_ioctl;
804 	if (ioctl) {
805 		/* TODO: set pointer to correct ioctl on a per team member
806 		 *       bases to make this more efficient. that is, once
807 		 *       we determine the correct ioctl, we will always
808 		 *       call it and not the others for that team
809 		 *       member.
810 		 */
811 
812 		/* We cannot assume that SIOCGMIIPHY will also read a
813 		 * register; not all network drivers (e.g., e100)
814 		 * support that.
815 		 */
816 
817 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
818 		strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
819 		mii = if_mii(&ifr);
820 		if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
821 			mii->reg_num = MII_BMSR;
822 			if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
823 				return mii->val_out & BMSR_LSTATUS;
824 		}
825 	}
826 
827 	/* If reporting, report that either there's no ndo_eth_ioctl,
828 	 * or both SIOCGMIIREG and get_link failed (meaning that we
829 	 * cannot report link status).  If not reporting, pretend
830 	 * we're ok.
831 	 */
832 	return reporting ? -1 : BMSR_LSTATUS;
833 }
834 
835 /*----------------------------- Multicast list ------------------------------*/
836 
837 /* Push the promiscuity flag down to appropriate slaves */
838 static int bond_set_promiscuity(struct bonding *bond, int inc)
839 {
840 	struct list_head *iter;
841 	int err = 0;
842 
843 	if (bond_uses_primary(bond)) {
844 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
845 
846 		if (curr_active)
847 			err = dev_set_promiscuity(curr_active->dev, inc);
848 	} else {
849 		struct slave *slave;
850 
851 		bond_for_each_slave(bond, slave, iter) {
852 			err = dev_set_promiscuity(slave->dev, inc);
853 			if (err)
854 				return err;
855 		}
856 	}
857 	return err;
858 }
859 
860 /* Push the allmulti flag down to all slaves */
861 static int bond_set_allmulti(struct bonding *bond, int inc)
862 {
863 	struct list_head *iter;
864 	int err = 0;
865 
866 	if (bond_uses_primary(bond)) {
867 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
868 
869 		if (curr_active)
870 			err = dev_set_allmulti(curr_active->dev, inc);
871 	} else {
872 		struct slave *slave;
873 
874 		bond_for_each_slave(bond, slave, iter) {
875 			err = dev_set_allmulti(slave->dev, inc);
876 			if (err)
877 				return err;
878 		}
879 	}
880 	return err;
881 }
882 
883 /* Retrieve the list of registered multicast addresses for the bonding
884  * device and retransmit an IGMP JOIN request to the current active
885  * slave.
886  */
887 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
888 {
889 	struct bonding *bond = container_of(work, struct bonding,
890 					    mcast_work.work);
891 
892 	if (!rtnl_trylock()) {
893 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
894 		return;
895 	}
896 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
897 
898 	if (bond->igmp_retrans > 1) {
899 		bond->igmp_retrans--;
900 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
901 	}
902 	rtnl_unlock();
903 }
904 
905 /* Flush bond's hardware addresses from slave */
906 static void bond_hw_addr_flush(struct net_device *bond_dev,
907 			       struct net_device *slave_dev)
908 {
909 	struct bonding *bond = netdev_priv(bond_dev);
910 
911 	dev_uc_unsync(slave_dev, bond_dev);
912 	dev_mc_unsync(slave_dev, bond_dev);
913 
914 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
915 		dev_mc_del(slave_dev, lacpdu_mcast_addr);
916 }
917 
918 /*--------------------------- Active slave change ---------------------------*/
919 
920 /* Update the hardware address list and promisc/allmulti for the new and
921  * old active slaves (if any).  Modes that are not using primary keep all
922  * slaves up date at all times; only the modes that use primary need to call
923  * this function to swap these settings during a failover.
924  */
925 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
926 			      struct slave *old_active)
927 {
928 	if (old_active) {
929 		if (bond->dev->flags & IFF_PROMISC)
930 			dev_set_promiscuity(old_active->dev, -1);
931 
932 		if (bond->dev->flags & IFF_ALLMULTI)
933 			dev_set_allmulti(old_active->dev, -1);
934 
935 		if (bond->dev->flags & IFF_UP)
936 			bond_hw_addr_flush(bond->dev, old_active->dev);
937 	}
938 
939 	if (new_active) {
940 		/* FIXME: Signal errors upstream. */
941 		if (bond->dev->flags & IFF_PROMISC)
942 			dev_set_promiscuity(new_active->dev, 1);
943 
944 		if (bond->dev->flags & IFF_ALLMULTI)
945 			dev_set_allmulti(new_active->dev, 1);
946 
947 		if (bond->dev->flags & IFF_UP) {
948 			netif_addr_lock_bh(bond->dev);
949 			dev_uc_sync(new_active->dev, bond->dev);
950 			dev_mc_sync(new_active->dev, bond->dev);
951 			netif_addr_unlock_bh(bond->dev);
952 		}
953 	}
954 }
955 
956 /**
957  * bond_set_dev_addr - clone slave's address to bond
958  * @bond_dev: bond net device
959  * @slave_dev: slave net device
960  *
961  * Should be called with RTNL held.
962  */
963 static int bond_set_dev_addr(struct net_device *bond_dev,
964 			     struct net_device *slave_dev)
965 {
966 	int err;
967 
968 	slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
969 		  bond_dev, slave_dev, slave_dev->addr_len);
970 	err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
971 	if (err)
972 		return err;
973 
974 	__dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
975 	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
976 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
977 	return 0;
978 }
979 
980 static struct slave *bond_get_old_active(struct bonding *bond,
981 					 struct slave *new_active)
982 {
983 	struct slave *slave;
984 	struct list_head *iter;
985 
986 	bond_for_each_slave(bond, slave, iter) {
987 		if (slave == new_active)
988 			continue;
989 
990 		if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
991 			return slave;
992 	}
993 
994 	return NULL;
995 }
996 
997 /* bond_do_fail_over_mac
998  *
999  * Perform special MAC address swapping for fail_over_mac settings
1000  *
1001  * Called with RTNL
1002  */
1003 static void bond_do_fail_over_mac(struct bonding *bond,
1004 				  struct slave *new_active,
1005 				  struct slave *old_active)
1006 {
1007 	u8 tmp_mac[MAX_ADDR_LEN];
1008 	struct sockaddr_storage ss;
1009 	int rv;
1010 
1011 	switch (bond->params.fail_over_mac) {
1012 	case BOND_FOM_ACTIVE:
1013 		if (new_active) {
1014 			rv = bond_set_dev_addr(bond->dev, new_active->dev);
1015 			if (rv)
1016 				slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
1017 					  -rv);
1018 		}
1019 		break;
1020 	case BOND_FOM_FOLLOW:
1021 		/* if new_active && old_active, swap them
1022 		 * if just old_active, do nothing (going to no active slave)
1023 		 * if just new_active, set new_active to bond's MAC
1024 		 */
1025 		if (!new_active)
1026 			return;
1027 
1028 		if (!old_active)
1029 			old_active = bond_get_old_active(bond, new_active);
1030 
1031 		if (old_active) {
1032 			bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
1033 					  new_active->dev->addr_len);
1034 			bond_hw_addr_copy(ss.__data,
1035 					  old_active->dev->dev_addr,
1036 					  old_active->dev->addr_len);
1037 			ss.ss_family = new_active->dev->type;
1038 		} else {
1039 			bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1040 					  bond->dev->addr_len);
1041 			ss.ss_family = bond->dev->type;
1042 		}
1043 
1044 		rv = dev_set_mac_address(new_active->dev,
1045 					 (struct sockaddr *)&ss, NULL);
1046 		if (rv) {
1047 			slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1048 				  -rv);
1049 			goto out;
1050 		}
1051 
1052 		if (!old_active)
1053 			goto out;
1054 
1055 		bond_hw_addr_copy(ss.__data, tmp_mac,
1056 				  new_active->dev->addr_len);
1057 		ss.ss_family = old_active->dev->type;
1058 
1059 		rv = dev_set_mac_address(old_active->dev,
1060 					 (struct sockaddr *)&ss, NULL);
1061 		if (rv)
1062 			slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1063 				  -rv);
1064 out:
1065 		break;
1066 	default:
1067 		netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1068 			   bond->params.fail_over_mac);
1069 		break;
1070 	}
1071 
1072 }
1073 
1074 /**
1075  * bond_choose_primary_or_current - select the primary or high priority slave
1076  * @bond: our bonding struct
1077  *
1078  * - Check if there is a primary link. If the primary link was set and is up,
1079  *   go on and do link reselection.
1080  *
1081  * - If primary link is not set or down, find the highest priority link.
1082  *   If the highest priority link is not current slave, set it as primary
1083  *   link and do link reselection.
1084  */
1085 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1086 {
1087 	struct slave *prim = rtnl_dereference(bond->primary_slave);
1088 	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1089 	struct slave *slave, *hprio = NULL;
1090 	struct list_head *iter;
1091 
1092 	if (!prim || prim->link != BOND_LINK_UP) {
1093 		bond_for_each_slave(bond, slave, iter) {
1094 			if (slave->link == BOND_LINK_UP) {
1095 				hprio = hprio ?: slave;
1096 				if (slave->prio > hprio->prio)
1097 					hprio = slave;
1098 			}
1099 		}
1100 
1101 		if (hprio && hprio != curr) {
1102 			prim = hprio;
1103 			goto link_reselect;
1104 		}
1105 
1106 		if (!curr || curr->link != BOND_LINK_UP)
1107 			return NULL;
1108 		return curr;
1109 	}
1110 
1111 	if (bond->force_primary) {
1112 		bond->force_primary = false;
1113 		return prim;
1114 	}
1115 
1116 link_reselect:
1117 	if (!curr || curr->link != BOND_LINK_UP)
1118 		return prim;
1119 
1120 	/* At this point, prim and curr are both up */
1121 	switch (bond->params.primary_reselect) {
1122 	case BOND_PRI_RESELECT_ALWAYS:
1123 		return prim;
1124 	case BOND_PRI_RESELECT_BETTER:
1125 		if (prim->speed < curr->speed)
1126 			return curr;
1127 		if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1128 			return curr;
1129 		return prim;
1130 	case BOND_PRI_RESELECT_FAILURE:
1131 		return curr;
1132 	default:
1133 		netdev_err(bond->dev, "impossible primary_reselect %d\n",
1134 			   bond->params.primary_reselect);
1135 		return curr;
1136 	}
1137 }
1138 
1139 /**
1140  * bond_find_best_slave - select the best available slave to be the active one
1141  * @bond: our bonding struct
1142  */
1143 static struct slave *bond_find_best_slave(struct bonding *bond)
1144 {
1145 	struct slave *slave, *bestslave = NULL;
1146 	struct list_head *iter;
1147 	int mintime = bond->params.updelay;
1148 
1149 	slave = bond_choose_primary_or_current(bond);
1150 	if (slave)
1151 		return slave;
1152 
1153 	bond_for_each_slave(bond, slave, iter) {
1154 		if (slave->link == BOND_LINK_UP)
1155 			return slave;
1156 		if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1157 		    slave->delay < mintime) {
1158 			mintime = slave->delay;
1159 			bestslave = slave;
1160 		}
1161 	}
1162 
1163 	return bestslave;
1164 }
1165 
1166 /* must be called in RCU critical section or with RTNL held */
1167 static bool bond_should_notify_peers(struct bonding *bond)
1168 {
1169 	struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
1170 
1171 	if (!slave || !bond->send_peer_notif ||
1172 	    bond->send_peer_notif %
1173 	    max(1, bond->params.peer_notif_delay) != 0 ||
1174 	    !netif_carrier_ok(bond->dev) ||
1175 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1176 		return false;
1177 
1178 	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1179 		   slave ? slave->dev->name : "NULL");
1180 
1181 	return true;
1182 }
1183 
1184 /**
1185  * bond_change_active_slave - change the active slave into the specified one
1186  * @bond: our bonding struct
1187  * @new_active: the new slave to make the active one
1188  *
1189  * Set the new slave to the bond's settings and unset them on the old
1190  * curr_active_slave.
1191  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1192  *
1193  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1194  * because it is apparently the best available slave we have, even though its
1195  * updelay hasn't timed out yet.
1196  *
1197  * Caller must hold RTNL.
1198  */
1199 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1200 {
1201 	struct slave *old_active;
1202 
1203 	ASSERT_RTNL();
1204 
1205 	old_active = rtnl_dereference(bond->curr_active_slave);
1206 
1207 	if (old_active == new_active)
1208 		return;
1209 
1210 #ifdef CONFIG_XFRM_OFFLOAD
1211 	bond_ipsec_del_sa_all(bond);
1212 #endif /* CONFIG_XFRM_OFFLOAD */
1213 
1214 	if (new_active) {
1215 		new_active->last_link_up = jiffies;
1216 
1217 		if (new_active->link == BOND_LINK_BACK) {
1218 			if (bond_uses_primary(bond)) {
1219 				slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1220 					   (bond->params.updelay - new_active->delay) * bond->params.miimon);
1221 			}
1222 
1223 			new_active->delay = 0;
1224 			bond_set_slave_link_state(new_active, BOND_LINK_UP,
1225 						  BOND_SLAVE_NOTIFY_NOW);
1226 
1227 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
1228 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1229 
1230 			if (bond_is_lb(bond))
1231 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1232 		} else {
1233 			if (bond_uses_primary(bond))
1234 				slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1235 		}
1236 	}
1237 
1238 	if (bond_uses_primary(bond))
1239 		bond_hw_addr_swap(bond, new_active, old_active);
1240 
1241 	if (bond_is_lb(bond)) {
1242 		bond_alb_handle_active_change(bond, new_active);
1243 		if (old_active)
1244 			bond_set_slave_inactive_flags(old_active,
1245 						      BOND_SLAVE_NOTIFY_NOW);
1246 		if (new_active)
1247 			bond_set_slave_active_flags(new_active,
1248 						    BOND_SLAVE_NOTIFY_NOW);
1249 	} else {
1250 		rcu_assign_pointer(bond->curr_active_slave, new_active);
1251 	}
1252 
1253 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1254 		if (old_active)
1255 			bond_set_slave_inactive_flags(old_active,
1256 						      BOND_SLAVE_NOTIFY_NOW);
1257 
1258 		if (new_active) {
1259 			bool should_notify_peers = false;
1260 
1261 			bond_set_slave_active_flags(new_active,
1262 						    BOND_SLAVE_NOTIFY_NOW);
1263 
1264 			if (bond->params.fail_over_mac)
1265 				bond_do_fail_over_mac(bond, new_active,
1266 						      old_active);
1267 
1268 			if (netif_running(bond->dev)) {
1269 				bond->send_peer_notif =
1270 					bond->params.num_peer_notif *
1271 					max(1, bond->params.peer_notif_delay);
1272 				should_notify_peers =
1273 					bond_should_notify_peers(bond);
1274 			}
1275 
1276 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1277 			if (should_notify_peers) {
1278 				bond->send_peer_notif--;
1279 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1280 							 bond->dev);
1281 			}
1282 		}
1283 	}
1284 
1285 #ifdef CONFIG_XFRM_OFFLOAD
1286 	bond_ipsec_add_sa_all(bond);
1287 #endif /* CONFIG_XFRM_OFFLOAD */
1288 
1289 	/* resend IGMP joins since active slave has changed or
1290 	 * all were sent on curr_active_slave.
1291 	 * resend only if bond is brought up with the affected
1292 	 * bonding modes and the retransmission is enabled
1293 	 */
1294 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1295 	    ((bond_uses_primary(bond) && new_active) ||
1296 	     BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1297 		bond->igmp_retrans = bond->params.resend_igmp;
1298 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1299 	}
1300 }
1301 
1302 /**
1303  * bond_select_active_slave - select a new active slave, if needed
1304  * @bond: our bonding struct
1305  *
1306  * This functions should be called when one of the following occurs:
1307  * - The old curr_active_slave has been released or lost its link.
1308  * - The primary_slave has got its link back.
1309  * - A slave has got its link back and there's no old curr_active_slave.
1310  *
1311  * Caller must hold RTNL.
1312  */
1313 void bond_select_active_slave(struct bonding *bond)
1314 {
1315 	struct slave *best_slave;
1316 	int rv;
1317 
1318 	ASSERT_RTNL();
1319 
1320 	best_slave = bond_find_best_slave(bond);
1321 	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1322 		bond_change_active_slave(bond, best_slave);
1323 		rv = bond_set_carrier(bond);
1324 		if (!rv)
1325 			return;
1326 
1327 		if (netif_carrier_ok(bond->dev))
1328 			netdev_info(bond->dev, "active interface up!\n");
1329 		else
1330 			netdev_info(bond->dev, "now running without any active interface!\n");
1331 	}
1332 }
1333 
1334 #ifdef CONFIG_NET_POLL_CONTROLLER
1335 static inline int slave_enable_netpoll(struct slave *slave)
1336 {
1337 	struct netpoll *np;
1338 	int err = 0;
1339 
1340 	np = kzalloc(sizeof(*np), GFP_KERNEL);
1341 	err = -ENOMEM;
1342 	if (!np)
1343 		goto out;
1344 
1345 	err = __netpoll_setup(np, slave->dev);
1346 	if (err) {
1347 		kfree(np);
1348 		goto out;
1349 	}
1350 	slave->np = np;
1351 out:
1352 	return err;
1353 }
1354 static inline void slave_disable_netpoll(struct slave *slave)
1355 {
1356 	struct netpoll *np = slave->np;
1357 
1358 	if (!np)
1359 		return;
1360 
1361 	slave->np = NULL;
1362 
1363 	__netpoll_free(np);
1364 }
1365 
1366 static void bond_poll_controller(struct net_device *bond_dev)
1367 {
1368 	struct bonding *bond = netdev_priv(bond_dev);
1369 	struct slave *slave = NULL;
1370 	struct list_head *iter;
1371 	struct ad_info ad_info;
1372 
1373 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1374 		if (bond_3ad_get_active_agg_info(bond, &ad_info))
1375 			return;
1376 
1377 	bond_for_each_slave_rcu(bond, slave, iter) {
1378 		if (!bond_slave_is_up(slave))
1379 			continue;
1380 
1381 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1382 			struct aggregator *agg =
1383 			    SLAVE_AD_INFO(slave)->port.aggregator;
1384 
1385 			if (agg &&
1386 			    agg->aggregator_identifier != ad_info.aggregator_id)
1387 				continue;
1388 		}
1389 
1390 		netpoll_poll_dev(slave->dev);
1391 	}
1392 }
1393 
1394 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1395 {
1396 	struct bonding *bond = netdev_priv(bond_dev);
1397 	struct list_head *iter;
1398 	struct slave *slave;
1399 
1400 	bond_for_each_slave(bond, slave, iter)
1401 		if (bond_slave_is_up(slave))
1402 			slave_disable_netpoll(slave);
1403 }
1404 
1405 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1406 {
1407 	struct bonding *bond = netdev_priv(dev);
1408 	struct list_head *iter;
1409 	struct slave *slave;
1410 	int err = 0;
1411 
1412 	bond_for_each_slave(bond, slave, iter) {
1413 		err = slave_enable_netpoll(slave);
1414 		if (err) {
1415 			bond_netpoll_cleanup(dev);
1416 			break;
1417 		}
1418 	}
1419 	return err;
1420 }
1421 #else
1422 static inline int slave_enable_netpoll(struct slave *slave)
1423 {
1424 	return 0;
1425 }
1426 static inline void slave_disable_netpoll(struct slave *slave)
1427 {
1428 }
1429 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1430 {
1431 }
1432 #endif
1433 
1434 /*---------------------------------- IOCTL ----------------------------------*/
1435 
1436 static netdev_features_t bond_fix_features(struct net_device *dev,
1437 					   netdev_features_t features)
1438 {
1439 	struct bonding *bond = netdev_priv(dev);
1440 	struct list_head *iter;
1441 	netdev_features_t mask;
1442 	struct slave *slave;
1443 
1444 	mask = features;
1445 
1446 	features &= ~NETIF_F_ONE_FOR_ALL;
1447 	features |= NETIF_F_ALL_FOR_ALL;
1448 
1449 	bond_for_each_slave(bond, slave, iter) {
1450 		features = netdev_increment_features(features,
1451 						     slave->dev->features,
1452 						     mask);
1453 	}
1454 	features = netdev_add_tso_features(features, mask);
1455 
1456 	return features;
1457 }
1458 
1459 #define BOND_VLAN_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1460 				 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1461 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1462 
1463 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1464 				 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1465 
1466 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1467 				 NETIF_F_GSO_SOFTWARE)
1468 
1469 
1470 static void bond_compute_features(struct bonding *bond)
1471 {
1472 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1473 					IFF_XMIT_DST_RELEASE_PERM;
1474 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1475 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
1476 #ifdef CONFIG_XFRM_OFFLOAD
1477 	netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1478 #endif /* CONFIG_XFRM_OFFLOAD */
1479 	netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1480 	struct net_device *bond_dev = bond->dev;
1481 	struct list_head *iter;
1482 	struct slave *slave;
1483 	unsigned short max_hard_header_len = ETH_HLEN;
1484 	unsigned int tso_max_size = TSO_MAX_SIZE;
1485 	u16 tso_max_segs = TSO_MAX_SEGS;
1486 
1487 	if (!bond_has_slaves(bond))
1488 		goto done;
1489 	vlan_features &= NETIF_F_ALL_FOR_ALL;
1490 	mpls_features &= NETIF_F_ALL_FOR_ALL;
1491 
1492 	bond_for_each_slave(bond, slave, iter) {
1493 		vlan_features = netdev_increment_features(vlan_features,
1494 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1495 
1496 		enc_features = netdev_increment_features(enc_features,
1497 							 slave->dev->hw_enc_features,
1498 							 BOND_ENC_FEATURES);
1499 
1500 #ifdef CONFIG_XFRM_OFFLOAD
1501 		xfrm_features = netdev_increment_features(xfrm_features,
1502 							  slave->dev->hw_enc_features,
1503 							  BOND_XFRM_FEATURES);
1504 #endif /* CONFIG_XFRM_OFFLOAD */
1505 
1506 		mpls_features = netdev_increment_features(mpls_features,
1507 							  slave->dev->mpls_features,
1508 							  BOND_MPLS_FEATURES);
1509 
1510 		dst_release_flag &= slave->dev->priv_flags;
1511 		if (slave->dev->hard_header_len > max_hard_header_len)
1512 			max_hard_header_len = slave->dev->hard_header_len;
1513 
1514 		tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1515 		tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1516 	}
1517 	bond_dev->hard_header_len = max_hard_header_len;
1518 
1519 done:
1520 	bond_dev->vlan_features = vlan_features;
1521 	bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1522 				    NETIF_F_HW_VLAN_CTAG_TX |
1523 				    NETIF_F_HW_VLAN_STAG_TX;
1524 #ifdef CONFIG_XFRM_OFFLOAD
1525 	bond_dev->hw_enc_features |= xfrm_features;
1526 #endif /* CONFIG_XFRM_OFFLOAD */
1527 	bond_dev->mpls_features = mpls_features;
1528 	netif_set_tso_max_segs(bond_dev, tso_max_segs);
1529 	netif_set_tso_max_size(bond_dev, tso_max_size);
1530 
1531 	bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1532 	if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1533 	    dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1534 		bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1535 
1536 	netdev_change_features(bond_dev);
1537 }
1538 
1539 static void bond_setup_by_slave(struct net_device *bond_dev,
1540 				struct net_device *slave_dev)
1541 {
1542 	bool was_up = !!(bond_dev->flags & IFF_UP);
1543 
1544 	dev_close(bond_dev);
1545 
1546 	bond_dev->header_ops	    = slave_dev->header_ops;
1547 
1548 	bond_dev->type		    = slave_dev->type;
1549 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1550 	bond_dev->needed_headroom   = slave_dev->needed_headroom;
1551 	bond_dev->addr_len	    = slave_dev->addr_len;
1552 
1553 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1554 		slave_dev->addr_len);
1555 
1556 	if (slave_dev->flags & IFF_POINTOPOINT) {
1557 		bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1558 		bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1559 	}
1560 	if (was_up)
1561 		dev_open(bond_dev, NULL);
1562 }
1563 
1564 /* On bonding slaves other than the currently active slave, suppress
1565  * duplicates except for alb non-mcast/bcast.
1566  */
1567 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1568 					    struct slave *slave,
1569 					    struct bonding *bond)
1570 {
1571 	if (bond_is_slave_inactive(slave)) {
1572 		if (BOND_MODE(bond) == BOND_MODE_ALB &&
1573 		    skb->pkt_type != PACKET_BROADCAST &&
1574 		    skb->pkt_type != PACKET_MULTICAST)
1575 			return false;
1576 		return true;
1577 	}
1578 	return false;
1579 }
1580 
1581 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1582 {
1583 	struct sk_buff *skb = *pskb;
1584 	struct slave *slave;
1585 	struct bonding *bond;
1586 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1587 			  struct slave *);
1588 	int ret = RX_HANDLER_ANOTHER;
1589 
1590 	skb = skb_share_check(skb, GFP_ATOMIC);
1591 	if (unlikely(!skb))
1592 		return RX_HANDLER_CONSUMED;
1593 
1594 	*pskb = skb;
1595 
1596 	slave = bond_slave_get_rcu(skb->dev);
1597 	bond = slave->bond;
1598 
1599 	recv_probe = READ_ONCE(bond->recv_probe);
1600 	if (recv_probe) {
1601 		ret = recv_probe(skb, bond, slave);
1602 		if (ret == RX_HANDLER_CONSUMED) {
1603 			consume_skb(skb);
1604 			return ret;
1605 		}
1606 	}
1607 
1608 	/*
1609 	 * For packets determined by bond_should_deliver_exact_match() call to
1610 	 * be suppressed we want to make an exception for link-local packets.
1611 	 * This is necessary for e.g. LLDP daemons to be able to monitor
1612 	 * inactive slave links without being forced to bind to them
1613 	 * explicitly.
1614 	 *
1615 	 * At the same time, packets that are passed to the bonding master
1616 	 * (including link-local ones) can have their originating interface
1617 	 * determined via PACKET_ORIGDEV socket option.
1618 	 */
1619 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1620 		if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1621 			return RX_HANDLER_PASS;
1622 		return RX_HANDLER_EXACT;
1623 	}
1624 
1625 	skb->dev = bond->dev;
1626 
1627 	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1628 	    netif_is_bridge_port(bond->dev) &&
1629 	    skb->pkt_type == PACKET_HOST) {
1630 
1631 		if (unlikely(skb_cow_head(skb,
1632 					  skb->data - skb_mac_header(skb)))) {
1633 			kfree_skb(skb);
1634 			return RX_HANDLER_CONSUMED;
1635 		}
1636 		bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1637 				  bond->dev->addr_len);
1638 	}
1639 
1640 	return ret;
1641 }
1642 
1643 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1644 {
1645 	switch (BOND_MODE(bond)) {
1646 	case BOND_MODE_ROUNDROBIN:
1647 		return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1648 	case BOND_MODE_ACTIVEBACKUP:
1649 		return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1650 	case BOND_MODE_BROADCAST:
1651 		return NETDEV_LAG_TX_TYPE_BROADCAST;
1652 	case BOND_MODE_XOR:
1653 	case BOND_MODE_8023AD:
1654 		return NETDEV_LAG_TX_TYPE_HASH;
1655 	default:
1656 		return NETDEV_LAG_TX_TYPE_UNKNOWN;
1657 	}
1658 }
1659 
1660 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1661 					       enum netdev_lag_tx_type type)
1662 {
1663 	if (type != NETDEV_LAG_TX_TYPE_HASH)
1664 		return NETDEV_LAG_HASH_NONE;
1665 
1666 	switch (bond->params.xmit_policy) {
1667 	case BOND_XMIT_POLICY_LAYER2:
1668 		return NETDEV_LAG_HASH_L2;
1669 	case BOND_XMIT_POLICY_LAYER34:
1670 		return NETDEV_LAG_HASH_L34;
1671 	case BOND_XMIT_POLICY_LAYER23:
1672 		return NETDEV_LAG_HASH_L23;
1673 	case BOND_XMIT_POLICY_ENCAP23:
1674 		return NETDEV_LAG_HASH_E23;
1675 	case BOND_XMIT_POLICY_ENCAP34:
1676 		return NETDEV_LAG_HASH_E34;
1677 	case BOND_XMIT_POLICY_VLAN_SRCMAC:
1678 		return NETDEV_LAG_HASH_VLAN_SRCMAC;
1679 	default:
1680 		return NETDEV_LAG_HASH_UNKNOWN;
1681 	}
1682 }
1683 
1684 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1685 				      struct netlink_ext_ack *extack)
1686 {
1687 	struct netdev_lag_upper_info lag_upper_info;
1688 	enum netdev_lag_tx_type type;
1689 	int err;
1690 
1691 	type = bond_lag_tx_type(bond);
1692 	lag_upper_info.tx_type = type;
1693 	lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1694 
1695 	err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1696 					   &lag_upper_info, extack);
1697 	if (err)
1698 		return err;
1699 
1700 	slave->dev->flags |= IFF_SLAVE;
1701 	return 0;
1702 }
1703 
1704 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1705 {
1706 	netdev_upper_dev_unlink(slave->dev, bond->dev);
1707 	slave->dev->flags &= ~IFF_SLAVE;
1708 }
1709 
1710 static void slave_kobj_release(struct kobject *kobj)
1711 {
1712 	struct slave *slave = to_slave(kobj);
1713 	struct bonding *bond = bond_get_bond_by_slave(slave);
1714 
1715 	cancel_delayed_work_sync(&slave->notify_work);
1716 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1717 		kfree(SLAVE_AD_INFO(slave));
1718 
1719 	kfree(slave);
1720 }
1721 
1722 static struct kobj_type slave_ktype = {
1723 	.release = slave_kobj_release,
1724 #ifdef CONFIG_SYSFS
1725 	.sysfs_ops = &slave_sysfs_ops,
1726 #endif
1727 };
1728 
1729 static int bond_kobj_init(struct slave *slave)
1730 {
1731 	int err;
1732 
1733 	err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1734 				   &(slave->dev->dev.kobj), "bonding_slave");
1735 	if (err)
1736 		kobject_put(&slave->kobj);
1737 
1738 	return err;
1739 }
1740 
1741 static struct slave *bond_alloc_slave(struct bonding *bond,
1742 				      struct net_device *slave_dev)
1743 {
1744 	struct slave *slave = NULL;
1745 
1746 	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1747 	if (!slave)
1748 		return NULL;
1749 
1750 	slave->bond = bond;
1751 	slave->dev = slave_dev;
1752 	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1753 
1754 	if (bond_kobj_init(slave))
1755 		return NULL;
1756 
1757 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1758 		SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1759 					       GFP_KERNEL);
1760 		if (!SLAVE_AD_INFO(slave)) {
1761 			kobject_put(&slave->kobj);
1762 			return NULL;
1763 		}
1764 	}
1765 
1766 	return slave;
1767 }
1768 
1769 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1770 {
1771 	info->bond_mode = BOND_MODE(bond);
1772 	info->miimon = bond->params.miimon;
1773 	info->num_slaves = bond->slave_cnt;
1774 }
1775 
1776 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1777 {
1778 	strcpy(info->slave_name, slave->dev->name);
1779 	info->link = slave->link;
1780 	info->state = bond_slave_state(slave);
1781 	info->link_failure_count = slave->link_failure_count;
1782 }
1783 
1784 static void bond_netdev_notify_work(struct work_struct *_work)
1785 {
1786 	struct slave *slave = container_of(_work, struct slave,
1787 					   notify_work.work);
1788 
1789 	if (rtnl_trylock()) {
1790 		struct netdev_bonding_info binfo;
1791 
1792 		bond_fill_ifslave(slave, &binfo.slave);
1793 		bond_fill_ifbond(slave->bond, &binfo.master);
1794 		netdev_bonding_info_change(slave->dev, &binfo);
1795 		rtnl_unlock();
1796 	} else {
1797 		queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1798 	}
1799 }
1800 
1801 void bond_queue_slave_event(struct slave *slave)
1802 {
1803 	queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1804 }
1805 
1806 void bond_lower_state_changed(struct slave *slave)
1807 {
1808 	struct netdev_lag_lower_state_info info;
1809 
1810 	info.link_up = slave->link == BOND_LINK_UP ||
1811 		       slave->link == BOND_LINK_FAIL;
1812 	info.tx_enabled = bond_is_active_slave(slave);
1813 	netdev_lower_state_changed(slave->dev, &info);
1814 }
1815 
1816 #define BOND_NL_ERR(bond_dev, extack, errmsg) do {		\
1817 	if (extack)						\
1818 		NL_SET_ERR_MSG(extack, errmsg);			\
1819 	else							\
1820 		netdev_err(bond_dev, "Error: %s\n", errmsg);	\
1821 } while (0)
1822 
1823 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {		\
1824 	if (extack)							\
1825 		NL_SET_ERR_MSG(extack, errmsg);				\
1826 	else								\
1827 		slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);	\
1828 } while (0)
1829 
1830 /* The bonding driver uses ether_setup() to convert a master bond device
1831  * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1832  * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1833  * if they were set
1834  */
1835 static void bond_ether_setup(struct net_device *bond_dev)
1836 {
1837 	unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1838 
1839 	ether_setup(bond_dev);
1840 	bond_dev->flags |= IFF_MASTER | flags;
1841 	bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1842 }
1843 
1844 void bond_xdp_set_features(struct net_device *bond_dev)
1845 {
1846 	struct bonding *bond = netdev_priv(bond_dev);
1847 	xdp_features_t val = NETDEV_XDP_ACT_MASK;
1848 	struct list_head *iter;
1849 	struct slave *slave;
1850 
1851 	ASSERT_RTNL();
1852 
1853 	if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) {
1854 		xdp_clear_features_flag(bond_dev);
1855 		return;
1856 	}
1857 
1858 	bond_for_each_slave(bond, slave, iter)
1859 		val &= slave->dev->xdp_features;
1860 
1861 	val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1862 
1863 	xdp_set_features_flag(bond_dev, val);
1864 }
1865 
1866 /* enslave device <slave> to bond device <master> */
1867 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1868 		 struct netlink_ext_ack *extack)
1869 {
1870 	struct bonding *bond = netdev_priv(bond_dev);
1871 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1872 	struct slave *new_slave = NULL, *prev_slave;
1873 	struct sockaddr_storage ss;
1874 	int link_reporting;
1875 	int res = 0, i;
1876 
1877 	if (slave_dev->flags & IFF_MASTER &&
1878 	    !netif_is_bond_master(slave_dev)) {
1879 		BOND_NL_ERR(bond_dev, extack,
1880 			    "Device type (master device) cannot be enslaved");
1881 		return -EPERM;
1882 	}
1883 
1884 	if (!bond->params.use_carrier &&
1885 	    slave_dev->ethtool_ops->get_link == NULL &&
1886 	    slave_ops->ndo_eth_ioctl == NULL) {
1887 		slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1888 	}
1889 
1890 	/* already in-use? */
1891 	if (netdev_is_rx_handler_busy(slave_dev)) {
1892 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1893 			     "Device is in use and cannot be enslaved");
1894 		return -EBUSY;
1895 	}
1896 
1897 	if (bond_dev == slave_dev) {
1898 		BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1899 		return -EPERM;
1900 	}
1901 
1902 	/* vlan challenged mutual exclusion */
1903 	/* no need to lock since we're protected by rtnl_lock */
1904 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1905 		slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1906 		if (vlan_uses_dev(bond_dev)) {
1907 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1908 				     "Can not enslave VLAN challenged device to VLAN enabled bond");
1909 			return -EPERM;
1910 		} else {
1911 			slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1912 		}
1913 	} else {
1914 		slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1915 	}
1916 
1917 	if (slave_dev->features & NETIF_F_HW_ESP)
1918 		slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1919 
1920 	/* Old ifenslave binaries are no longer supported.  These can
1921 	 * be identified with moderate accuracy by the state of the slave:
1922 	 * the current ifenslave will set the interface down prior to
1923 	 * enslaving it; the old ifenslave will not.
1924 	 */
1925 	if (slave_dev->flags & IFF_UP) {
1926 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1927 			     "Device can not be enslaved while up");
1928 		return -EPERM;
1929 	}
1930 
1931 	/* set bonding device ether type by slave - bonding netdevices are
1932 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1933 	 * there is a need to override some of the type dependent attribs/funcs.
1934 	 *
1935 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1936 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1937 	 */
1938 	if (!bond_has_slaves(bond)) {
1939 		if (bond_dev->type != slave_dev->type) {
1940 			slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1941 				  bond_dev->type, slave_dev->type);
1942 
1943 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1944 						       bond_dev);
1945 			res = notifier_to_errno(res);
1946 			if (res) {
1947 				slave_err(bond_dev, slave_dev, "refused to change device type\n");
1948 				return -EBUSY;
1949 			}
1950 
1951 			/* Flush unicast and multicast addresses */
1952 			dev_uc_flush(bond_dev);
1953 			dev_mc_flush(bond_dev);
1954 
1955 			if (slave_dev->type != ARPHRD_ETHER)
1956 				bond_setup_by_slave(bond_dev, slave_dev);
1957 			else
1958 				bond_ether_setup(bond_dev);
1959 
1960 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1961 						 bond_dev);
1962 		}
1963 	} else if (bond_dev->type != slave_dev->type) {
1964 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1965 			     "Device type is different from other slaves");
1966 		return -EINVAL;
1967 	}
1968 
1969 	if (slave_dev->type == ARPHRD_INFINIBAND &&
1970 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1971 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1972 			     "Only active-backup mode is supported for infiniband slaves");
1973 		res = -EOPNOTSUPP;
1974 		goto err_undo_flags;
1975 	}
1976 
1977 	if (!slave_ops->ndo_set_mac_address ||
1978 	    slave_dev->type == ARPHRD_INFINIBAND) {
1979 		slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1980 		if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1981 		    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1982 			if (!bond_has_slaves(bond)) {
1983 				bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1984 				slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1985 			} else {
1986 				SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1987 					     "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1988 				res = -EOPNOTSUPP;
1989 				goto err_undo_flags;
1990 			}
1991 		}
1992 	}
1993 
1994 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1995 
1996 	/* If this is the first slave, then we need to set the master's hardware
1997 	 * address to be the same as the slave's.
1998 	 */
1999 	if (!bond_has_slaves(bond) &&
2000 	    bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
2001 		res = bond_set_dev_addr(bond->dev, slave_dev);
2002 		if (res)
2003 			goto err_undo_flags;
2004 	}
2005 
2006 	new_slave = bond_alloc_slave(bond, slave_dev);
2007 	if (!new_slave) {
2008 		res = -ENOMEM;
2009 		goto err_undo_flags;
2010 	}
2011 
2012 	/* Set the new_slave's queue_id to be zero.  Queue ID mapping
2013 	 * is set via sysfs or module option if desired.
2014 	 */
2015 	new_slave->queue_id = 0;
2016 
2017 	/* Save slave's original mtu and then set it to match the bond */
2018 	new_slave->original_mtu = slave_dev->mtu;
2019 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
2020 	if (res) {
2021 		slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
2022 		goto err_free;
2023 	}
2024 
2025 	/* Save slave's original ("permanent") mac address for modes
2026 	 * that need it, and for restoring it upon release, and then
2027 	 * set it to the master's address
2028 	 */
2029 	bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
2030 			  slave_dev->addr_len);
2031 
2032 	if (!bond->params.fail_over_mac ||
2033 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2034 		/* Set slave to master's mac address.  The application already
2035 		 * set the master's mac address to that of the first slave
2036 		 */
2037 		memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
2038 		ss.ss_family = slave_dev->type;
2039 		res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
2040 					  extack);
2041 		if (res) {
2042 			slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
2043 			goto err_restore_mtu;
2044 		}
2045 	}
2046 
2047 	/* set no_addrconf flag before open to prevent IPv6 addrconf */
2048 	slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2049 
2050 	/* open the slave since the application closed it */
2051 	res = dev_open(slave_dev, extack);
2052 	if (res) {
2053 		slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2054 		goto err_restore_mac;
2055 	}
2056 
2057 	slave_dev->priv_flags |= IFF_BONDING;
2058 	/* initialize slave stats */
2059 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2060 
2061 	if (bond_is_lb(bond)) {
2062 		/* bond_alb_init_slave() must be called before all other stages since
2063 		 * it might fail and we do not want to have to undo everything
2064 		 */
2065 		res = bond_alb_init_slave(bond, new_slave);
2066 		if (res)
2067 			goto err_close;
2068 	}
2069 
2070 	res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2071 	if (res) {
2072 		slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2073 		goto err_close;
2074 	}
2075 
2076 	prev_slave = bond_last_slave(bond);
2077 
2078 	new_slave->delay = 0;
2079 	new_slave->link_failure_count = 0;
2080 
2081 	if (bond_update_speed_duplex(new_slave) &&
2082 	    bond_needs_speed_duplex(bond))
2083 		new_slave->link = BOND_LINK_DOWN;
2084 
2085 	new_slave->last_rx = jiffies -
2086 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
2087 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2088 		new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2089 
2090 	new_slave->last_tx = new_slave->last_rx;
2091 
2092 	if (bond->params.miimon && !bond->params.use_carrier) {
2093 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2094 
2095 		if ((link_reporting == -1) && !bond->params.arp_interval) {
2096 			/* miimon is set but a bonded network driver
2097 			 * does not support ETHTOOL/MII and
2098 			 * arp_interval is not set.  Note: if
2099 			 * use_carrier is enabled, we will never go
2100 			 * here (because netif_carrier is always
2101 			 * supported); thus, we don't need to change
2102 			 * the messages for netif_carrier.
2103 			 */
2104 			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");
2105 		} else if (link_reporting == -1) {
2106 			/* unable get link status using mii/ethtool */
2107 			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");
2108 		}
2109 	}
2110 
2111 	/* check for initial state */
2112 	new_slave->link = BOND_LINK_NOCHANGE;
2113 	if (bond->params.miimon) {
2114 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2115 			if (bond->params.updelay) {
2116 				bond_set_slave_link_state(new_slave,
2117 							  BOND_LINK_BACK,
2118 							  BOND_SLAVE_NOTIFY_NOW);
2119 				new_slave->delay = bond->params.updelay;
2120 			} else {
2121 				bond_set_slave_link_state(new_slave,
2122 							  BOND_LINK_UP,
2123 							  BOND_SLAVE_NOTIFY_NOW);
2124 			}
2125 		} else {
2126 			bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2127 						  BOND_SLAVE_NOTIFY_NOW);
2128 		}
2129 	} else if (bond->params.arp_interval) {
2130 		bond_set_slave_link_state(new_slave,
2131 					  (netif_carrier_ok(slave_dev) ?
2132 					  BOND_LINK_UP : BOND_LINK_DOWN),
2133 					  BOND_SLAVE_NOTIFY_NOW);
2134 	} else {
2135 		bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2136 					  BOND_SLAVE_NOTIFY_NOW);
2137 	}
2138 
2139 	if (new_slave->link != BOND_LINK_DOWN)
2140 		new_slave->last_link_up = jiffies;
2141 	slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2142 		  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2143 		  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2144 
2145 	if (bond_uses_primary(bond) && bond->params.primary[0]) {
2146 		/* if there is a primary slave, remember it */
2147 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2148 			rcu_assign_pointer(bond->primary_slave, new_slave);
2149 			bond->force_primary = true;
2150 		}
2151 	}
2152 
2153 	switch (BOND_MODE(bond)) {
2154 	case BOND_MODE_ACTIVEBACKUP:
2155 		bond_set_slave_inactive_flags(new_slave,
2156 					      BOND_SLAVE_NOTIFY_NOW);
2157 		break;
2158 	case BOND_MODE_8023AD:
2159 		/* in 802.3ad mode, the internal mechanism
2160 		 * will activate the slaves in the selected
2161 		 * aggregator
2162 		 */
2163 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2164 		/* if this is the first slave */
2165 		if (!prev_slave) {
2166 			SLAVE_AD_INFO(new_slave)->id = 1;
2167 			/* Initialize AD with the number of times that the AD timer is called in 1 second
2168 			 * can be called only after the mac address of the bond is set
2169 			 */
2170 			bond_3ad_initialize(bond);
2171 		} else {
2172 			SLAVE_AD_INFO(new_slave)->id =
2173 				SLAVE_AD_INFO(prev_slave)->id + 1;
2174 		}
2175 
2176 		bond_3ad_bind_slave(new_slave);
2177 		break;
2178 	case BOND_MODE_TLB:
2179 	case BOND_MODE_ALB:
2180 		bond_set_active_slave(new_slave);
2181 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2182 		break;
2183 	default:
2184 		slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2185 
2186 		/* always active in trunk mode */
2187 		bond_set_active_slave(new_slave);
2188 
2189 		/* In trunking mode there is little meaning to curr_active_slave
2190 		 * anyway (it holds no special properties of the bond device),
2191 		 * so we can change it without calling change_active_interface()
2192 		 */
2193 		if (!rcu_access_pointer(bond->curr_active_slave) &&
2194 		    new_slave->link == BOND_LINK_UP)
2195 			rcu_assign_pointer(bond->curr_active_slave, new_slave);
2196 
2197 		break;
2198 	} /* switch(bond_mode) */
2199 
2200 #ifdef CONFIG_NET_POLL_CONTROLLER
2201 	if (bond->dev->npinfo) {
2202 		if (slave_enable_netpoll(new_slave)) {
2203 			slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2204 			res = -EBUSY;
2205 			goto err_detach;
2206 		}
2207 	}
2208 #endif
2209 
2210 	if (!(bond_dev->features & NETIF_F_LRO))
2211 		dev_disable_lro(slave_dev);
2212 
2213 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2214 					 new_slave);
2215 	if (res) {
2216 		slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2217 		goto err_detach;
2218 	}
2219 
2220 	res = bond_master_upper_dev_link(bond, new_slave, extack);
2221 	if (res) {
2222 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2223 		goto err_unregister;
2224 	}
2225 
2226 	bond_lower_state_changed(new_slave);
2227 
2228 	res = bond_sysfs_slave_add(new_slave);
2229 	if (res) {
2230 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2231 		goto err_upper_unlink;
2232 	}
2233 
2234 	/* If the mode uses primary, then the following is handled by
2235 	 * bond_change_active_slave().
2236 	 */
2237 	if (!bond_uses_primary(bond)) {
2238 		/* set promiscuity level to new slave */
2239 		if (bond_dev->flags & IFF_PROMISC) {
2240 			res = dev_set_promiscuity(slave_dev, 1);
2241 			if (res)
2242 				goto err_sysfs_del;
2243 		}
2244 
2245 		/* set allmulti level to new slave */
2246 		if (bond_dev->flags & IFF_ALLMULTI) {
2247 			res = dev_set_allmulti(slave_dev, 1);
2248 			if (res) {
2249 				if (bond_dev->flags & IFF_PROMISC)
2250 					dev_set_promiscuity(slave_dev, -1);
2251 				goto err_sysfs_del;
2252 			}
2253 		}
2254 
2255 		if (bond_dev->flags & IFF_UP) {
2256 			netif_addr_lock_bh(bond_dev);
2257 			dev_mc_sync_multiple(slave_dev, bond_dev);
2258 			dev_uc_sync_multiple(slave_dev, bond_dev);
2259 			netif_addr_unlock_bh(bond_dev);
2260 
2261 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
2262 				dev_mc_add(slave_dev, lacpdu_mcast_addr);
2263 		}
2264 	}
2265 
2266 	bond->slave_cnt++;
2267 	bond_compute_features(bond);
2268 	bond_set_carrier(bond);
2269 
2270 	if (bond_uses_primary(bond)) {
2271 		block_netpoll_tx();
2272 		bond_select_active_slave(bond);
2273 		unblock_netpoll_tx();
2274 	}
2275 
2276 	if (bond_mode_can_use_xmit_hash(bond))
2277 		bond_update_slave_arr(bond, NULL);
2278 
2279 
2280 	if (!slave_dev->netdev_ops->ndo_bpf ||
2281 	    !slave_dev->netdev_ops->ndo_xdp_xmit) {
2282 		if (bond->xdp_prog) {
2283 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2284 				     "Slave does not support XDP");
2285 			res = -EOPNOTSUPP;
2286 			goto err_sysfs_del;
2287 		}
2288 	} else if (bond->xdp_prog) {
2289 		struct netdev_bpf xdp = {
2290 			.command = XDP_SETUP_PROG,
2291 			.flags   = 0,
2292 			.prog    = bond->xdp_prog,
2293 			.extack  = extack,
2294 		};
2295 
2296 		if (dev_xdp_prog_count(slave_dev) > 0) {
2297 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2298 				     "Slave has XDP program loaded, please unload before enslaving");
2299 			res = -EOPNOTSUPP;
2300 			goto err_sysfs_del;
2301 		}
2302 
2303 		res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2304 		if (res < 0) {
2305 			/* ndo_bpf() sets extack error message */
2306 			slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2307 			goto err_sysfs_del;
2308 		}
2309 		if (bond->xdp_prog)
2310 			bpf_prog_inc(bond->xdp_prog);
2311 	}
2312 
2313 	bond_xdp_set_features(bond_dev);
2314 
2315 	slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2316 		   bond_is_active_slave(new_slave) ? "an active" : "a backup",
2317 		   new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2318 
2319 	/* enslave is successful */
2320 	bond_queue_slave_event(new_slave);
2321 	return 0;
2322 
2323 /* Undo stages on error */
2324 err_sysfs_del:
2325 	bond_sysfs_slave_del(new_slave);
2326 
2327 err_upper_unlink:
2328 	bond_upper_dev_unlink(bond, new_slave);
2329 
2330 err_unregister:
2331 	netdev_rx_handler_unregister(slave_dev);
2332 
2333 err_detach:
2334 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2335 	if (rcu_access_pointer(bond->primary_slave) == new_slave)
2336 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2337 	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2338 		block_netpoll_tx();
2339 		bond_change_active_slave(bond, NULL);
2340 		bond_select_active_slave(bond);
2341 		unblock_netpoll_tx();
2342 	}
2343 	/* either primary_slave or curr_active_slave might've changed */
2344 	synchronize_rcu();
2345 	slave_disable_netpoll(new_slave);
2346 
2347 err_close:
2348 	if (!netif_is_bond_master(slave_dev))
2349 		slave_dev->priv_flags &= ~IFF_BONDING;
2350 	dev_close(slave_dev);
2351 
2352 err_restore_mac:
2353 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2354 	if (!bond->params.fail_over_mac ||
2355 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2356 		/* XXX TODO - fom follow mode needs to change master's
2357 		 * MAC if this slave's MAC is in use by the bond, or at
2358 		 * least print a warning.
2359 		 */
2360 		bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2361 				  new_slave->dev->addr_len);
2362 		ss.ss_family = slave_dev->type;
2363 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2364 	}
2365 
2366 err_restore_mtu:
2367 	dev_set_mtu(slave_dev, new_slave->original_mtu);
2368 
2369 err_free:
2370 	kobject_put(&new_slave->kobj);
2371 
2372 err_undo_flags:
2373 	/* Enslave of first slave has failed and we need to fix master's mac */
2374 	if (!bond_has_slaves(bond)) {
2375 		if (ether_addr_equal_64bits(bond_dev->dev_addr,
2376 					    slave_dev->dev_addr))
2377 			eth_hw_addr_random(bond_dev);
2378 		if (bond_dev->type != ARPHRD_ETHER) {
2379 			dev_close(bond_dev);
2380 			bond_ether_setup(bond_dev);
2381 		}
2382 	}
2383 
2384 	return res;
2385 }
2386 
2387 /* Try to release the slave device <slave> from the bond device <master>
2388  * It is legal to access curr_active_slave without a lock because all the function
2389  * is RTNL-locked. If "all" is true it means that the function is being called
2390  * while destroying a bond interface and all slaves are being released.
2391  *
2392  * The rules for slave state should be:
2393  *   for Active/Backup:
2394  *     Active stays on all backups go down
2395  *   for Bonded connections:
2396  *     The first up interface should be left on and all others downed.
2397  */
2398 static int __bond_release_one(struct net_device *bond_dev,
2399 			      struct net_device *slave_dev,
2400 			      bool all, bool unregister)
2401 {
2402 	struct bonding *bond = netdev_priv(bond_dev);
2403 	struct slave *slave, *oldcurrent;
2404 	struct sockaddr_storage ss;
2405 	int old_flags = bond_dev->flags;
2406 	netdev_features_t old_features = bond_dev->features;
2407 
2408 	/* slave is not a slave or master is not master of this slave */
2409 	if (!(slave_dev->flags & IFF_SLAVE) ||
2410 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
2411 		slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2412 		return -EINVAL;
2413 	}
2414 
2415 	block_netpoll_tx();
2416 
2417 	slave = bond_get_slave_by_dev(bond, slave_dev);
2418 	if (!slave) {
2419 		/* not a slave of this bond */
2420 		slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2421 		unblock_netpoll_tx();
2422 		return -EINVAL;
2423 	}
2424 
2425 	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2426 
2427 	bond_sysfs_slave_del(slave);
2428 
2429 	/* recompute stats just before removing the slave */
2430 	bond_get_stats(bond->dev, &bond->bond_stats);
2431 
2432 	if (bond->xdp_prog) {
2433 		struct netdev_bpf xdp = {
2434 			.command = XDP_SETUP_PROG,
2435 			.flags   = 0,
2436 			.prog	 = NULL,
2437 			.extack  = NULL,
2438 		};
2439 		if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2440 			slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2441 	}
2442 
2443 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2444 	 * for this slave anymore.
2445 	 */
2446 	netdev_rx_handler_unregister(slave_dev);
2447 
2448 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
2449 		bond_3ad_unbind_slave(slave);
2450 
2451 	bond_upper_dev_unlink(bond, slave);
2452 
2453 	if (bond_mode_can_use_xmit_hash(bond))
2454 		bond_update_slave_arr(bond, slave);
2455 
2456 	slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2457 		    bond_is_active_slave(slave) ? "active" : "backup");
2458 
2459 	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2460 
2461 	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2462 
2463 	if (!all && (!bond->params.fail_over_mac ||
2464 		     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2465 		if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2466 		    bond_has_slaves(bond))
2467 			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",
2468 				   slave->perm_hwaddr);
2469 	}
2470 
2471 	if (rtnl_dereference(bond->primary_slave) == slave)
2472 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2473 
2474 	if (oldcurrent == slave)
2475 		bond_change_active_slave(bond, NULL);
2476 
2477 	if (bond_is_lb(bond)) {
2478 		/* Must be called only after the slave has been
2479 		 * detached from the list and the curr_active_slave
2480 		 * has been cleared (if our_slave == old_current),
2481 		 * but before a new active slave is selected.
2482 		 */
2483 		bond_alb_deinit_slave(bond, slave);
2484 	}
2485 
2486 	if (all) {
2487 		RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2488 	} else if (oldcurrent == slave) {
2489 		/* Note that we hold RTNL over this sequence, so there
2490 		 * is no concern that another slave add/remove event
2491 		 * will interfere.
2492 		 */
2493 		bond_select_active_slave(bond);
2494 	}
2495 
2496 	bond_set_carrier(bond);
2497 	if (!bond_has_slaves(bond))
2498 		eth_hw_addr_random(bond_dev);
2499 
2500 	unblock_netpoll_tx();
2501 	synchronize_rcu();
2502 	bond->slave_cnt--;
2503 
2504 	if (!bond_has_slaves(bond)) {
2505 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2506 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2507 	}
2508 
2509 	bond_compute_features(bond);
2510 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2511 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2512 		slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2513 
2514 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2515 
2516 	/* If the mode uses primary, then this case was handled above by
2517 	 * bond_change_active_slave(..., NULL)
2518 	 */
2519 	if (!bond_uses_primary(bond)) {
2520 		/* unset promiscuity level from slave
2521 		 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2522 		 * of the IFF_PROMISC flag in the bond_dev, but we need the
2523 		 * value of that flag before that change, as that was the value
2524 		 * when this slave was attached, so we cache at the start of the
2525 		 * function and use it here. Same goes for ALLMULTI below
2526 		 */
2527 		if (old_flags & IFF_PROMISC)
2528 			dev_set_promiscuity(slave_dev, -1);
2529 
2530 		/* unset allmulti level from slave */
2531 		if (old_flags & IFF_ALLMULTI)
2532 			dev_set_allmulti(slave_dev, -1);
2533 
2534 		if (old_flags & IFF_UP)
2535 			bond_hw_addr_flush(bond_dev, slave_dev);
2536 	}
2537 
2538 	slave_disable_netpoll(slave);
2539 
2540 	/* close slave before restoring its mac address */
2541 	dev_close(slave_dev);
2542 
2543 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2544 
2545 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2546 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2547 		/* restore original ("permanent") mac address */
2548 		bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2549 				  slave->dev->addr_len);
2550 		ss.ss_family = slave_dev->type;
2551 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2552 	}
2553 
2554 	if (unregister)
2555 		__dev_set_mtu(slave_dev, slave->original_mtu);
2556 	else
2557 		dev_set_mtu(slave_dev, slave->original_mtu);
2558 
2559 	if (!netif_is_bond_master(slave_dev))
2560 		slave_dev->priv_flags &= ~IFF_BONDING;
2561 
2562 	bond_xdp_set_features(bond_dev);
2563 	kobject_put(&slave->kobj);
2564 
2565 	return 0;
2566 }
2567 
2568 /* A wrapper used because of ndo_del_link */
2569 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2570 {
2571 	return __bond_release_one(bond_dev, slave_dev, false, false);
2572 }
2573 
2574 /* First release a slave and then destroy the bond if no more slaves are left.
2575  * Must be under rtnl_lock when this function is called.
2576  */
2577 static int bond_release_and_destroy(struct net_device *bond_dev,
2578 				    struct net_device *slave_dev)
2579 {
2580 	struct bonding *bond = netdev_priv(bond_dev);
2581 	int ret;
2582 
2583 	ret = __bond_release_one(bond_dev, slave_dev, false, true);
2584 	if (ret == 0 && !bond_has_slaves(bond) &&
2585 	    bond_dev->reg_state != NETREG_UNREGISTERING) {
2586 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2587 		netdev_info(bond_dev, "Destroying bond\n");
2588 		bond_remove_proc_entry(bond);
2589 		unregister_netdevice(bond_dev);
2590 	}
2591 	return ret;
2592 }
2593 
2594 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2595 {
2596 	struct bonding *bond = netdev_priv(bond_dev);
2597 
2598 	bond_fill_ifbond(bond, info);
2599 }
2600 
2601 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2602 {
2603 	struct bonding *bond = netdev_priv(bond_dev);
2604 	struct list_head *iter;
2605 	int i = 0, res = -ENODEV;
2606 	struct slave *slave;
2607 
2608 	bond_for_each_slave(bond, slave, iter) {
2609 		if (i++ == (int)info->slave_id) {
2610 			res = 0;
2611 			bond_fill_ifslave(slave, info);
2612 			break;
2613 		}
2614 	}
2615 
2616 	return res;
2617 }
2618 
2619 /*-------------------------------- Monitoring -------------------------------*/
2620 
2621 /* called with rcu_read_lock() */
2622 static int bond_miimon_inspect(struct bonding *bond)
2623 {
2624 	bool ignore_updelay = false;
2625 	int link_state, commit = 0;
2626 	struct list_head *iter;
2627 	struct slave *slave;
2628 
2629 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2630 		ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2631 	} else {
2632 		struct bond_up_slave *usable_slaves;
2633 
2634 		usable_slaves = rcu_dereference(bond->usable_slaves);
2635 
2636 		if (usable_slaves && usable_slaves->count == 0)
2637 			ignore_updelay = true;
2638 	}
2639 
2640 	bond_for_each_slave_rcu(bond, slave, iter) {
2641 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2642 
2643 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2644 
2645 		switch (slave->link) {
2646 		case BOND_LINK_UP:
2647 			if (link_state)
2648 				continue;
2649 
2650 			bond_propose_link_state(slave, BOND_LINK_FAIL);
2651 			commit++;
2652 			slave->delay = bond->params.downdelay;
2653 			if (slave->delay && net_ratelimit()) {
2654 				slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2655 					   (BOND_MODE(bond) ==
2656 					    BOND_MODE_ACTIVEBACKUP) ?
2657 					    (bond_is_active_slave(slave) ?
2658 					     "active " : "backup ") : "",
2659 					   bond->params.downdelay * bond->params.miimon);
2660 			}
2661 			fallthrough;
2662 		case BOND_LINK_FAIL:
2663 			if (link_state) {
2664 				/* recovered before downdelay expired */
2665 				bond_propose_link_state(slave, BOND_LINK_UP);
2666 				slave->last_link_up = jiffies;
2667 				if (net_ratelimit())
2668 					slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2669 						   (bond->params.downdelay - slave->delay) *
2670 						   bond->params.miimon);
2671 				commit++;
2672 				continue;
2673 			}
2674 
2675 			if (slave->delay <= 0) {
2676 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2677 				commit++;
2678 				continue;
2679 			}
2680 
2681 			slave->delay--;
2682 			break;
2683 
2684 		case BOND_LINK_DOWN:
2685 			if (!link_state)
2686 				continue;
2687 
2688 			bond_propose_link_state(slave, BOND_LINK_BACK);
2689 			commit++;
2690 			slave->delay = bond->params.updelay;
2691 
2692 			if (slave->delay && net_ratelimit()) {
2693 				slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2694 					   ignore_updelay ? 0 :
2695 					   bond->params.updelay *
2696 					   bond->params.miimon);
2697 			}
2698 			fallthrough;
2699 		case BOND_LINK_BACK:
2700 			if (!link_state) {
2701 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2702 				if (net_ratelimit())
2703 					slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2704 						   (bond->params.updelay - slave->delay) *
2705 						   bond->params.miimon);
2706 				commit++;
2707 				continue;
2708 			}
2709 
2710 			if (ignore_updelay)
2711 				slave->delay = 0;
2712 
2713 			if (slave->delay <= 0) {
2714 				bond_propose_link_state(slave, BOND_LINK_UP);
2715 				commit++;
2716 				ignore_updelay = false;
2717 				continue;
2718 			}
2719 
2720 			slave->delay--;
2721 			break;
2722 		}
2723 	}
2724 
2725 	return commit;
2726 }
2727 
2728 static void bond_miimon_link_change(struct bonding *bond,
2729 				    struct slave *slave,
2730 				    char link)
2731 {
2732 	switch (BOND_MODE(bond)) {
2733 	case BOND_MODE_8023AD:
2734 		bond_3ad_handle_link_change(slave, link);
2735 		break;
2736 	case BOND_MODE_TLB:
2737 	case BOND_MODE_ALB:
2738 		bond_alb_handle_link_change(bond, slave, link);
2739 		break;
2740 	case BOND_MODE_XOR:
2741 		bond_update_slave_arr(bond, NULL);
2742 		break;
2743 	}
2744 }
2745 
2746 static void bond_miimon_commit(struct bonding *bond)
2747 {
2748 	struct slave *slave, *primary, *active;
2749 	bool do_failover = false;
2750 	struct list_head *iter;
2751 
2752 	ASSERT_RTNL();
2753 
2754 	bond_for_each_slave(bond, slave, iter) {
2755 		switch (slave->link_new_state) {
2756 		case BOND_LINK_NOCHANGE:
2757 			/* For 802.3ad mode, check current slave speed and
2758 			 * duplex again in case its port was disabled after
2759 			 * invalid speed/duplex reporting but recovered before
2760 			 * link monitoring could make a decision on the actual
2761 			 * link status
2762 			 */
2763 			if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2764 			    slave->link == BOND_LINK_UP)
2765 				bond_3ad_adapter_speed_duplex_changed(slave);
2766 			continue;
2767 
2768 		case BOND_LINK_UP:
2769 			if (bond_update_speed_duplex(slave) &&
2770 			    bond_needs_speed_duplex(bond)) {
2771 				slave->link = BOND_LINK_DOWN;
2772 				if (net_ratelimit())
2773 					slave_warn(bond->dev, slave->dev,
2774 						   "failed to get link speed/duplex\n");
2775 				continue;
2776 			}
2777 			bond_set_slave_link_state(slave, BOND_LINK_UP,
2778 						  BOND_SLAVE_NOTIFY_NOW);
2779 			slave->last_link_up = jiffies;
2780 
2781 			primary = rtnl_dereference(bond->primary_slave);
2782 			if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2783 				/* prevent it from being the active one */
2784 				bond_set_backup_slave(slave);
2785 			} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2786 				/* make it immediately active */
2787 				bond_set_active_slave(slave);
2788 			}
2789 
2790 			slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2791 				   slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2792 				   slave->duplex ? "full" : "half");
2793 
2794 			bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2795 
2796 			active = rtnl_dereference(bond->curr_active_slave);
2797 			if (!active || slave == primary || slave->prio > active->prio)
2798 				do_failover = true;
2799 
2800 			continue;
2801 
2802 		case BOND_LINK_DOWN:
2803 			if (slave->link_failure_count < UINT_MAX)
2804 				slave->link_failure_count++;
2805 
2806 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2807 						  BOND_SLAVE_NOTIFY_NOW);
2808 
2809 			if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2810 			    BOND_MODE(bond) == BOND_MODE_8023AD)
2811 				bond_set_slave_inactive_flags(slave,
2812 							      BOND_SLAVE_NOTIFY_NOW);
2813 
2814 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2815 
2816 			bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2817 
2818 			if (slave == rcu_access_pointer(bond->curr_active_slave))
2819 				do_failover = true;
2820 
2821 			continue;
2822 
2823 		default:
2824 			slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2825 				  slave->link_new_state);
2826 			bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2827 
2828 			continue;
2829 		}
2830 	}
2831 
2832 	if (do_failover) {
2833 		block_netpoll_tx();
2834 		bond_select_active_slave(bond);
2835 		unblock_netpoll_tx();
2836 	}
2837 
2838 	bond_set_carrier(bond);
2839 }
2840 
2841 /* bond_mii_monitor
2842  *
2843  * Really a wrapper that splits the mii monitor into two phases: an
2844  * inspection, then (if inspection indicates something needs to be done)
2845  * an acquisition of appropriate locks followed by a commit phase to
2846  * implement whatever link state changes are indicated.
2847  */
2848 static void bond_mii_monitor(struct work_struct *work)
2849 {
2850 	struct bonding *bond = container_of(work, struct bonding,
2851 					    mii_work.work);
2852 	bool should_notify_peers = false;
2853 	bool commit;
2854 	unsigned long delay;
2855 	struct slave *slave;
2856 	struct list_head *iter;
2857 
2858 	delay = msecs_to_jiffies(bond->params.miimon);
2859 
2860 	if (!bond_has_slaves(bond))
2861 		goto re_arm;
2862 
2863 	rcu_read_lock();
2864 	should_notify_peers = bond_should_notify_peers(bond);
2865 	commit = !!bond_miimon_inspect(bond);
2866 	if (bond->send_peer_notif) {
2867 		rcu_read_unlock();
2868 		if (rtnl_trylock()) {
2869 			bond->send_peer_notif--;
2870 			rtnl_unlock();
2871 		}
2872 	} else {
2873 		rcu_read_unlock();
2874 	}
2875 
2876 	if (commit) {
2877 		/* Race avoidance with bond_close cancel of workqueue */
2878 		if (!rtnl_trylock()) {
2879 			delay = 1;
2880 			should_notify_peers = false;
2881 			goto re_arm;
2882 		}
2883 
2884 		bond_for_each_slave(bond, slave, iter) {
2885 			bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2886 		}
2887 		bond_miimon_commit(bond);
2888 
2889 		rtnl_unlock();	/* might sleep, hold no other locks */
2890 	}
2891 
2892 re_arm:
2893 	if (bond->params.miimon)
2894 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2895 
2896 	if (should_notify_peers) {
2897 		if (!rtnl_trylock())
2898 			return;
2899 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2900 		rtnl_unlock();
2901 	}
2902 }
2903 
2904 static int bond_upper_dev_walk(struct net_device *upper,
2905 			       struct netdev_nested_priv *priv)
2906 {
2907 	__be32 ip = *(__be32 *)priv->data;
2908 
2909 	return ip == bond_confirm_addr(upper, 0, ip);
2910 }
2911 
2912 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2913 {
2914 	struct netdev_nested_priv priv = {
2915 		.data = (void *)&ip,
2916 	};
2917 	bool ret = false;
2918 
2919 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2920 		return true;
2921 
2922 	rcu_read_lock();
2923 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2924 		ret = true;
2925 	rcu_read_unlock();
2926 
2927 	return ret;
2928 }
2929 
2930 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
2931 
2932 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2933 			     struct sk_buff *skb)
2934 {
2935 	struct net_device *bond_dev = slave->bond->dev;
2936 	struct net_device *slave_dev = slave->dev;
2937 	struct bond_vlan_tag *outer_tag = tags;
2938 
2939 	if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
2940 		return true;
2941 
2942 	tags++;
2943 
2944 	/* Go through all the tags backwards and add them to the packet */
2945 	while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
2946 		if (!tags->vlan_id) {
2947 			tags++;
2948 			continue;
2949 		}
2950 
2951 		slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2952 			  ntohs(outer_tag->vlan_proto), tags->vlan_id);
2953 		skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2954 						tags->vlan_id);
2955 		if (!skb) {
2956 			net_err_ratelimited("failed to insert inner VLAN tag\n");
2957 			return false;
2958 		}
2959 
2960 		tags++;
2961 	}
2962 	/* Set the outer tag */
2963 	if (outer_tag->vlan_id) {
2964 		slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2965 			  ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2966 		__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2967 				       outer_tag->vlan_id);
2968 	}
2969 
2970 	return true;
2971 }
2972 
2973 /* We go to the (large) trouble of VLAN tagging ARP frames because
2974  * switches in VLAN mode (especially if ports are configured as
2975  * "native" to a VLAN) might not pass non-tagged frames.
2976  */
2977 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2978 			  __be32 src_ip, struct bond_vlan_tag *tags)
2979 {
2980 	struct net_device *bond_dev = slave->bond->dev;
2981 	struct net_device *slave_dev = slave->dev;
2982 	struct sk_buff *skb;
2983 
2984 	slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2985 		  arp_op, &dest_ip, &src_ip);
2986 
2987 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2988 			 NULL, slave_dev->dev_addr, NULL);
2989 
2990 	if (!skb) {
2991 		net_err_ratelimited("ARP packet allocation failed\n");
2992 		return;
2993 	}
2994 
2995 	if (bond_handle_vlan(slave, tags, skb)) {
2996 		slave_update_last_tx(slave);
2997 		arp_xmit(skb);
2998 	}
2999 
3000 	return;
3001 }
3002 
3003 /* Validate the device path between the @start_dev and the @end_dev.
3004  * The path is valid if the @end_dev is reachable through device
3005  * stacking.
3006  * When the path is validated, collect any vlan information in the
3007  * path.
3008  */
3009 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
3010 					      struct net_device *end_dev,
3011 					      int level)
3012 {
3013 	struct bond_vlan_tag *tags;
3014 	struct net_device *upper;
3015 	struct list_head  *iter;
3016 
3017 	if (start_dev == end_dev) {
3018 		tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
3019 		if (!tags)
3020 			return ERR_PTR(-ENOMEM);
3021 		tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
3022 		return tags;
3023 	}
3024 
3025 	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
3026 		tags = bond_verify_device_path(upper, end_dev, level + 1);
3027 		if (IS_ERR_OR_NULL(tags)) {
3028 			if (IS_ERR(tags))
3029 				return tags;
3030 			continue;
3031 		}
3032 		if (is_vlan_dev(upper)) {
3033 			tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
3034 			tags[level].vlan_id = vlan_dev_vlan_id(upper);
3035 		}
3036 
3037 		return tags;
3038 	}
3039 
3040 	return NULL;
3041 }
3042 
3043 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
3044 {
3045 	struct rtable *rt;
3046 	struct bond_vlan_tag *tags;
3047 	__be32 *targets = bond->params.arp_targets, addr;
3048 	int i;
3049 
3050 	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3051 		slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3052 			  __func__, &targets[i]);
3053 		tags = NULL;
3054 
3055 		/* Find out through which dev should the packet go */
3056 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0,
3057 				     RT_SCOPE_LINK);
3058 		if (IS_ERR(rt)) {
3059 			/* there's no route to target - try to send arp
3060 			 * probe to generate any traffic (arp_validate=0)
3061 			 */
3062 			if (bond->params.arp_validate)
3063 				pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3064 					     bond->dev->name,
3065 					     &targets[i]);
3066 			bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3067 				      0, tags);
3068 			continue;
3069 		}
3070 
3071 		/* bond device itself */
3072 		if (rt->dst.dev == bond->dev)
3073 			goto found;
3074 
3075 		rcu_read_lock();
3076 		tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3077 		rcu_read_unlock();
3078 
3079 		if (!IS_ERR_OR_NULL(tags))
3080 			goto found;
3081 
3082 		/* Not our device - skip */
3083 		slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3084 			   &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3085 
3086 		ip_rt_put(rt);
3087 		continue;
3088 
3089 found:
3090 		addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3091 		ip_rt_put(rt);
3092 		bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3093 		kfree(tags);
3094 	}
3095 }
3096 
3097 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3098 {
3099 	int i;
3100 
3101 	if (!sip || !bond_has_this_ip(bond, tip)) {
3102 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3103 			   __func__, &sip, &tip);
3104 		return;
3105 	}
3106 
3107 	i = bond_get_targets_ip(bond->params.arp_targets, sip);
3108 	if (i == -1) {
3109 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3110 			   __func__, &sip);
3111 		return;
3112 	}
3113 	slave->last_rx = jiffies;
3114 	slave->target_last_arp_rx[i] = jiffies;
3115 }
3116 
3117 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3118 			struct slave *slave)
3119 {
3120 	struct arphdr *arp = (struct arphdr *)skb->data;
3121 	struct slave *curr_active_slave, *curr_arp_slave;
3122 	unsigned char *arp_ptr;
3123 	__be32 sip, tip;
3124 	unsigned int alen;
3125 
3126 	alen = arp_hdr_len(bond->dev);
3127 
3128 	if (alen > skb_headlen(skb)) {
3129 		arp = kmalloc(alen, GFP_ATOMIC);
3130 		if (!arp)
3131 			goto out_unlock;
3132 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
3133 			goto out_unlock;
3134 	}
3135 
3136 	if (arp->ar_hln != bond->dev->addr_len ||
3137 	    skb->pkt_type == PACKET_OTHERHOST ||
3138 	    skb->pkt_type == PACKET_LOOPBACK ||
3139 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
3140 	    arp->ar_pro != htons(ETH_P_IP) ||
3141 	    arp->ar_pln != 4)
3142 		goto out_unlock;
3143 
3144 	arp_ptr = (unsigned char *)(arp + 1);
3145 	arp_ptr += bond->dev->addr_len;
3146 	memcpy(&sip, arp_ptr, 4);
3147 	arp_ptr += 4 + bond->dev->addr_len;
3148 	memcpy(&tip, arp_ptr, 4);
3149 
3150 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3151 		  __func__, slave->dev->name, bond_slave_state(slave),
3152 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3153 		  &sip, &tip);
3154 
3155 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3156 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3157 
3158 	/* We 'trust' the received ARP enough to validate it if:
3159 	 *
3160 	 * (a) the slave receiving the ARP is active (which includes the
3161 	 * current ARP slave, if any), or
3162 	 *
3163 	 * (b) the receiving slave isn't active, but there is a currently
3164 	 * active slave and it received valid arp reply(s) after it became
3165 	 * the currently active slave, or
3166 	 *
3167 	 * (c) there is an ARP slave that sent an ARP during the prior ARP
3168 	 * interval, and we receive an ARP reply on any slave.  We accept
3169 	 * these because switch FDB update delays may deliver the ARP
3170 	 * reply to a slave other than the sender of the ARP request.
3171 	 *
3172 	 * Note: for (b), backup slaves are receiving the broadcast ARP
3173 	 * request, not a reply.  This request passes from the sending
3174 	 * slave through the L2 switch(es) to the receiving slave.  Since
3175 	 * this is checking the request, sip/tip are swapped for
3176 	 * validation.
3177 	 *
3178 	 * This is done to avoid endless looping when we can't reach the
3179 	 * arp_ip_target and fool ourselves with our own arp requests.
3180 	 */
3181 	if (bond_is_active_slave(slave))
3182 		bond_validate_arp(bond, slave, sip, tip);
3183 	else if (curr_active_slave &&
3184 		 time_after(slave_last_rx(bond, curr_active_slave),
3185 			    curr_active_slave->last_link_up))
3186 		bond_validate_arp(bond, slave, tip, sip);
3187 	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3188 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3189 		bond_validate_arp(bond, slave, sip, tip);
3190 
3191 out_unlock:
3192 	if (arp != (struct arphdr *)skb->data)
3193 		kfree(arp);
3194 	return RX_HANDLER_ANOTHER;
3195 }
3196 
3197 #if IS_ENABLED(CONFIG_IPV6)
3198 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3199 			 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3200 {
3201 	struct net_device *bond_dev = slave->bond->dev;
3202 	struct net_device *slave_dev = slave->dev;
3203 	struct in6_addr mcaddr;
3204 	struct sk_buff *skb;
3205 
3206 	slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3207 		  daddr, saddr);
3208 
3209 	skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3210 	if (!skb) {
3211 		net_err_ratelimited("NS packet allocation failed\n");
3212 		return;
3213 	}
3214 
3215 	addrconf_addr_solict_mult(daddr, &mcaddr);
3216 	if (bond_handle_vlan(slave, tags, skb)) {
3217 		slave_update_last_tx(slave);
3218 		ndisc_send_skb(skb, &mcaddr, saddr);
3219 	}
3220 }
3221 
3222 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3223 {
3224 	struct in6_addr *targets = bond->params.ns_targets;
3225 	struct bond_vlan_tag *tags;
3226 	struct dst_entry *dst;
3227 	struct in6_addr saddr;
3228 	struct flowi6 fl6;
3229 	int i;
3230 
3231 	for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3232 		slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3233 			  __func__, &targets[i]);
3234 		tags = NULL;
3235 
3236 		/* Find out through which dev should the packet go */
3237 		memset(&fl6, 0, sizeof(struct flowi6));
3238 		fl6.daddr = targets[i];
3239 		fl6.flowi6_oif = bond->dev->ifindex;
3240 
3241 		dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3242 		if (dst->error) {
3243 			dst_release(dst);
3244 			/* there's no route to target - try to send arp
3245 			 * probe to generate any traffic (arp_validate=0)
3246 			 */
3247 			if (bond->params.arp_validate)
3248 				pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3249 					     bond->dev->name,
3250 					     &targets[i]);
3251 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3252 			continue;
3253 		}
3254 
3255 		/* bond device itself */
3256 		if (dst->dev == bond->dev)
3257 			goto found;
3258 
3259 		rcu_read_lock();
3260 		tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3261 		rcu_read_unlock();
3262 
3263 		if (!IS_ERR_OR_NULL(tags))
3264 			goto found;
3265 
3266 		/* Not our device - skip */
3267 		slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3268 			  &targets[i], dst->dev ? dst->dev->name : "NULL");
3269 
3270 		dst_release(dst);
3271 		continue;
3272 
3273 found:
3274 		if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3275 			bond_ns_send(slave, &targets[i], &saddr, tags);
3276 		else
3277 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3278 
3279 		dst_release(dst);
3280 		kfree(tags);
3281 	}
3282 }
3283 
3284 static int bond_confirm_addr6(struct net_device *dev,
3285 			      struct netdev_nested_priv *priv)
3286 {
3287 	struct in6_addr *addr = (struct in6_addr *)priv->data;
3288 
3289 	return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3290 }
3291 
3292 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3293 {
3294 	struct netdev_nested_priv priv = {
3295 		.data = addr,
3296 	};
3297 	int ret = false;
3298 
3299 	if (bond_confirm_addr6(bond->dev, &priv))
3300 		return true;
3301 
3302 	rcu_read_lock();
3303 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3304 		ret = true;
3305 	rcu_read_unlock();
3306 
3307 	return ret;
3308 }
3309 
3310 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3311 			     struct in6_addr *saddr, struct in6_addr *daddr)
3312 {
3313 	int i;
3314 
3315 	/* Ignore NAs that:
3316 	 * 1. Source address is unspecified address.
3317 	 * 2. Dest address is neither all-nodes multicast address nor
3318 	 *    exist on bond interface.
3319 	 */
3320 	if (ipv6_addr_any(saddr) ||
3321 	    (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3322 	     !bond_has_this_ip6(bond, daddr))) {
3323 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3324 			  __func__, saddr, daddr);
3325 		return;
3326 	}
3327 
3328 	i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3329 	if (i == -1) {
3330 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3331 			  __func__, saddr);
3332 		return;
3333 	}
3334 	slave->last_rx = jiffies;
3335 	slave->target_last_arp_rx[i] = jiffies;
3336 }
3337 
3338 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3339 		       struct slave *slave)
3340 {
3341 	struct slave *curr_active_slave, *curr_arp_slave;
3342 	struct in6_addr *saddr, *daddr;
3343 	struct {
3344 		struct ipv6hdr ip6;
3345 		struct icmp6hdr icmp6;
3346 	} *combined, _combined;
3347 
3348 	if (skb->pkt_type == PACKET_OTHERHOST ||
3349 	    skb->pkt_type == PACKET_LOOPBACK)
3350 		goto out;
3351 
3352 	combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3353 	if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3354 	    (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3355 	     combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3356 		goto out;
3357 
3358 	saddr = &combined->ip6.saddr;
3359 	daddr = &combined->ip6.daddr;
3360 
3361 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3362 		  __func__, slave->dev->name, bond_slave_state(slave),
3363 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3364 		  saddr, daddr);
3365 
3366 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3367 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3368 
3369 	/* We 'trust' the received ARP enough to validate it if:
3370 	 * see bond_arp_rcv().
3371 	 */
3372 	if (bond_is_active_slave(slave))
3373 		bond_validate_na(bond, slave, saddr, daddr);
3374 	else if (curr_active_slave &&
3375 		 time_after(slave_last_rx(bond, curr_active_slave),
3376 			    curr_active_slave->last_link_up))
3377 		bond_validate_na(bond, slave, daddr, saddr);
3378 	else if (curr_arp_slave &&
3379 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3380 		bond_validate_na(bond, slave, saddr, daddr);
3381 
3382 out:
3383 	return RX_HANDLER_ANOTHER;
3384 }
3385 #endif
3386 
3387 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3388 		      struct slave *slave)
3389 {
3390 #if IS_ENABLED(CONFIG_IPV6)
3391 	bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3392 #endif
3393 	bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3394 
3395 	slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3396 		  __func__, skb->dev->name);
3397 
3398 	/* Use arp validate logic for both ARP and NS */
3399 	if (!slave_do_arp_validate(bond, slave)) {
3400 		if ((slave_do_arp_validate_only(bond) && is_arp) ||
3401 #if IS_ENABLED(CONFIG_IPV6)
3402 		    (slave_do_arp_validate_only(bond) && is_ipv6) ||
3403 #endif
3404 		    !slave_do_arp_validate_only(bond))
3405 			slave->last_rx = jiffies;
3406 		return RX_HANDLER_ANOTHER;
3407 	} else if (is_arp) {
3408 		return bond_arp_rcv(skb, bond, slave);
3409 #if IS_ENABLED(CONFIG_IPV6)
3410 	} else if (is_ipv6) {
3411 		return bond_na_rcv(skb, bond, slave);
3412 #endif
3413 	} else {
3414 		return RX_HANDLER_ANOTHER;
3415 	}
3416 }
3417 
3418 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3419 {
3420 	bond_arp_send_all(bond, slave);
3421 #if IS_ENABLED(CONFIG_IPV6)
3422 	bond_ns_send_all(bond, slave);
3423 #endif
3424 }
3425 
3426 /* function to verify if we're in the arp_interval timeslice, returns true if
3427  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3428  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3429  */
3430 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3431 				  int mod)
3432 {
3433 	int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3434 
3435 	return time_in_range(jiffies,
3436 			     last_act - delta_in_ticks,
3437 			     last_act + mod * delta_in_ticks + delta_in_ticks/2);
3438 }
3439 
3440 /* This function is called regularly to monitor each slave's link
3441  * ensuring that traffic is being sent and received when arp monitoring
3442  * is used in load-balancing mode. if the adapter has been dormant, then an
3443  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3444  * arp monitoring in active backup mode.
3445  */
3446 static void bond_loadbalance_arp_mon(struct bonding *bond)
3447 {
3448 	struct slave *slave, *oldcurrent;
3449 	struct list_head *iter;
3450 	int do_failover = 0, slave_state_changed = 0;
3451 
3452 	if (!bond_has_slaves(bond))
3453 		goto re_arm;
3454 
3455 	rcu_read_lock();
3456 
3457 	oldcurrent = rcu_dereference(bond->curr_active_slave);
3458 	/* see if any of the previous devices are up now (i.e. they have
3459 	 * xmt and rcv traffic). the curr_active_slave does not come into
3460 	 * the picture unless it is null. also, slave->last_link_up is not
3461 	 * needed here because we send an arp on each slave and give a slave
3462 	 * as long as it needs to get the tx/rx within the delta.
3463 	 * TODO: what about up/down delay in arp mode? it wasn't here before
3464 	 *       so it can wait
3465 	 */
3466 	bond_for_each_slave_rcu(bond, slave, iter) {
3467 		unsigned long last_tx = slave_last_tx(slave);
3468 
3469 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3470 
3471 		if (slave->link != BOND_LINK_UP) {
3472 			if (bond_time_in_interval(bond, last_tx, 1) &&
3473 			    bond_time_in_interval(bond, slave->last_rx, 1)) {
3474 
3475 				bond_propose_link_state(slave, BOND_LINK_UP);
3476 				slave_state_changed = 1;
3477 
3478 				/* primary_slave has no meaning in round-robin
3479 				 * mode. the window of a slave being up and
3480 				 * curr_active_slave being null after enslaving
3481 				 * is closed.
3482 				 */
3483 				if (!oldcurrent) {
3484 					slave_info(bond->dev, slave->dev, "link status definitely up\n");
3485 					do_failover = 1;
3486 				} else {
3487 					slave_info(bond->dev, slave->dev, "interface is now up\n");
3488 				}
3489 			}
3490 		} else {
3491 			/* slave->link == BOND_LINK_UP */
3492 
3493 			/* not all switches will respond to an arp request
3494 			 * when the source ip is 0, so don't take the link down
3495 			 * if we don't know our ip yet
3496 			 */
3497 			if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3498 			    !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3499 
3500 				bond_propose_link_state(slave, BOND_LINK_DOWN);
3501 				slave_state_changed = 1;
3502 
3503 				if (slave->link_failure_count < UINT_MAX)
3504 					slave->link_failure_count++;
3505 
3506 				slave_info(bond->dev, slave->dev, "interface is now down\n");
3507 
3508 				if (slave == oldcurrent)
3509 					do_failover = 1;
3510 			}
3511 		}
3512 
3513 		/* note: if switch is in round-robin mode, all links
3514 		 * must tx arp to ensure all links rx an arp - otherwise
3515 		 * links may oscillate or not come up at all; if switch is
3516 		 * in something like xor mode, there is nothing we can
3517 		 * do - all replies will be rx'ed on same link causing slaves
3518 		 * to be unstable during low/no traffic periods
3519 		 */
3520 		if (bond_slave_is_up(slave))
3521 			bond_send_validate(bond, slave);
3522 	}
3523 
3524 	rcu_read_unlock();
3525 
3526 	if (do_failover || slave_state_changed) {
3527 		if (!rtnl_trylock())
3528 			goto re_arm;
3529 
3530 		bond_for_each_slave(bond, slave, iter) {
3531 			if (slave->link_new_state != BOND_LINK_NOCHANGE)
3532 				slave->link = slave->link_new_state;
3533 		}
3534 
3535 		if (slave_state_changed) {
3536 			bond_slave_state_change(bond);
3537 			if (BOND_MODE(bond) == BOND_MODE_XOR)
3538 				bond_update_slave_arr(bond, NULL);
3539 		}
3540 		if (do_failover) {
3541 			block_netpoll_tx();
3542 			bond_select_active_slave(bond);
3543 			unblock_netpoll_tx();
3544 		}
3545 		rtnl_unlock();
3546 	}
3547 
3548 re_arm:
3549 	if (bond->params.arp_interval)
3550 		queue_delayed_work(bond->wq, &bond->arp_work,
3551 				   msecs_to_jiffies(bond->params.arp_interval));
3552 }
3553 
3554 /* Called to inspect slaves for active-backup mode ARP monitor link state
3555  * changes.  Sets proposed link state in slaves to specify what action
3556  * should take place for the slave.  Returns 0 if no changes are found, >0
3557  * if changes to link states must be committed.
3558  *
3559  * Called with rcu_read_lock held.
3560  */
3561 static int bond_ab_arp_inspect(struct bonding *bond)
3562 {
3563 	unsigned long last_tx, last_rx;
3564 	struct list_head *iter;
3565 	struct slave *slave;
3566 	int commit = 0;
3567 
3568 	bond_for_each_slave_rcu(bond, slave, iter) {
3569 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3570 		last_rx = slave_last_rx(bond, slave);
3571 
3572 		if (slave->link != BOND_LINK_UP) {
3573 			if (bond_time_in_interval(bond, last_rx, 1)) {
3574 				bond_propose_link_state(slave, BOND_LINK_UP);
3575 				commit++;
3576 			} else if (slave->link == BOND_LINK_BACK) {
3577 				bond_propose_link_state(slave, BOND_LINK_FAIL);
3578 				commit++;
3579 			}
3580 			continue;
3581 		}
3582 
3583 		/* Give slaves 2*delta after being enslaved or made
3584 		 * active.  This avoids bouncing, as the last receive
3585 		 * times need a full ARP monitor cycle to be updated.
3586 		 */
3587 		if (bond_time_in_interval(bond, slave->last_link_up, 2))
3588 			continue;
3589 
3590 		/* Backup slave is down if:
3591 		 * - No current_arp_slave AND
3592 		 * - more than (missed_max+1)*delta since last receive AND
3593 		 * - the bond has an IP address
3594 		 *
3595 		 * Note: a non-null current_arp_slave indicates
3596 		 * the curr_active_slave went down and we are
3597 		 * searching for a new one; under this condition
3598 		 * we only take the curr_active_slave down - this
3599 		 * gives each slave a chance to tx/rx traffic
3600 		 * before being taken out
3601 		 */
3602 		if (!bond_is_active_slave(slave) &&
3603 		    !rcu_access_pointer(bond->current_arp_slave) &&
3604 		    !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3605 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3606 			commit++;
3607 		}
3608 
3609 		/* Active slave is down if:
3610 		 * - more than missed_max*delta since transmitting OR
3611 		 * - (more than missed_max*delta since receive AND
3612 		 *    the bond has an IP address)
3613 		 */
3614 		last_tx = slave_last_tx(slave);
3615 		if (bond_is_active_slave(slave) &&
3616 		    (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3617 		     !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3618 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3619 			commit++;
3620 		}
3621 	}
3622 
3623 	return commit;
3624 }
3625 
3626 /* Called to commit link state changes noted by inspection step of
3627  * active-backup mode ARP monitor.
3628  *
3629  * Called with RTNL hold.
3630  */
3631 static void bond_ab_arp_commit(struct bonding *bond)
3632 {
3633 	bool do_failover = false;
3634 	struct list_head *iter;
3635 	unsigned long last_tx;
3636 	struct slave *slave;
3637 
3638 	bond_for_each_slave(bond, slave, iter) {
3639 		switch (slave->link_new_state) {
3640 		case BOND_LINK_NOCHANGE:
3641 			continue;
3642 
3643 		case BOND_LINK_UP:
3644 			last_tx = slave_last_tx(slave);
3645 			if (rtnl_dereference(bond->curr_active_slave) != slave ||
3646 			    (!rtnl_dereference(bond->curr_active_slave) &&
3647 			     bond_time_in_interval(bond, last_tx, 1))) {
3648 				struct slave *current_arp_slave;
3649 
3650 				current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3651 				bond_set_slave_link_state(slave, BOND_LINK_UP,
3652 							  BOND_SLAVE_NOTIFY_NOW);
3653 				if (current_arp_slave) {
3654 					bond_set_slave_inactive_flags(
3655 						current_arp_slave,
3656 						BOND_SLAVE_NOTIFY_NOW);
3657 					RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3658 				}
3659 
3660 				slave_info(bond->dev, slave->dev, "link status definitely up\n");
3661 
3662 				if (!rtnl_dereference(bond->curr_active_slave) ||
3663 				    slave == rtnl_dereference(bond->primary_slave) ||
3664 				    slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3665 					do_failover = true;
3666 
3667 			}
3668 
3669 			continue;
3670 
3671 		case BOND_LINK_DOWN:
3672 			if (slave->link_failure_count < UINT_MAX)
3673 				slave->link_failure_count++;
3674 
3675 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3676 						  BOND_SLAVE_NOTIFY_NOW);
3677 			bond_set_slave_inactive_flags(slave,
3678 						      BOND_SLAVE_NOTIFY_NOW);
3679 
3680 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3681 
3682 			if (slave == rtnl_dereference(bond->curr_active_slave)) {
3683 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3684 				do_failover = true;
3685 			}
3686 
3687 			continue;
3688 
3689 		case BOND_LINK_FAIL:
3690 			bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3691 						  BOND_SLAVE_NOTIFY_NOW);
3692 			bond_set_slave_inactive_flags(slave,
3693 						      BOND_SLAVE_NOTIFY_NOW);
3694 
3695 			/* A slave has just been enslaved and has become
3696 			 * the current active slave.
3697 			 */
3698 			if (rtnl_dereference(bond->curr_active_slave))
3699 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3700 			continue;
3701 
3702 		default:
3703 			slave_err(bond->dev, slave->dev,
3704 				  "impossible: link_new_state %d on slave\n",
3705 				  slave->link_new_state);
3706 			continue;
3707 		}
3708 	}
3709 
3710 	if (do_failover) {
3711 		block_netpoll_tx();
3712 		bond_select_active_slave(bond);
3713 		unblock_netpoll_tx();
3714 	}
3715 
3716 	bond_set_carrier(bond);
3717 }
3718 
3719 /* Send ARP probes for active-backup mode ARP monitor.
3720  *
3721  * Called with rcu_read_lock held.
3722  */
3723 static bool bond_ab_arp_probe(struct bonding *bond)
3724 {
3725 	struct slave *slave, *before = NULL, *new_slave = NULL,
3726 		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3727 		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3728 	struct list_head *iter;
3729 	bool found = false;
3730 	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3731 
3732 	if (curr_arp_slave && curr_active_slave)
3733 		netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3734 			    curr_arp_slave->dev->name,
3735 			    curr_active_slave->dev->name);
3736 
3737 	if (curr_active_slave) {
3738 		bond_send_validate(bond, curr_active_slave);
3739 		return should_notify_rtnl;
3740 	}
3741 
3742 	/* if we don't have a curr_active_slave, search for the next available
3743 	 * backup slave from the current_arp_slave and make it the candidate
3744 	 * for becoming the curr_active_slave
3745 	 */
3746 
3747 	if (!curr_arp_slave) {
3748 		curr_arp_slave = bond_first_slave_rcu(bond);
3749 		if (!curr_arp_slave)
3750 			return should_notify_rtnl;
3751 	}
3752 
3753 	bond_for_each_slave_rcu(bond, slave, iter) {
3754 		if (!found && !before && bond_slave_is_up(slave))
3755 			before = slave;
3756 
3757 		if (found && !new_slave && bond_slave_is_up(slave))
3758 			new_slave = slave;
3759 		/* if the link state is up at this point, we
3760 		 * mark it down - this can happen if we have
3761 		 * simultaneous link failures and
3762 		 * reselect_active_interface doesn't make this
3763 		 * one the current slave so it is still marked
3764 		 * up when it is actually down
3765 		 */
3766 		if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3767 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3768 						  BOND_SLAVE_NOTIFY_LATER);
3769 			if (slave->link_failure_count < UINT_MAX)
3770 				slave->link_failure_count++;
3771 
3772 			bond_set_slave_inactive_flags(slave,
3773 						      BOND_SLAVE_NOTIFY_LATER);
3774 
3775 			slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3776 		}
3777 		if (slave == curr_arp_slave)
3778 			found = true;
3779 	}
3780 
3781 	if (!new_slave && before)
3782 		new_slave = before;
3783 
3784 	if (!new_slave)
3785 		goto check_state;
3786 
3787 	bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3788 				  BOND_SLAVE_NOTIFY_LATER);
3789 	bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3790 	bond_send_validate(bond, new_slave);
3791 	new_slave->last_link_up = jiffies;
3792 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
3793 
3794 check_state:
3795 	bond_for_each_slave_rcu(bond, slave, iter) {
3796 		if (slave->should_notify || slave->should_notify_link) {
3797 			should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3798 			break;
3799 		}
3800 	}
3801 	return should_notify_rtnl;
3802 }
3803 
3804 static void bond_activebackup_arp_mon(struct bonding *bond)
3805 {
3806 	bool should_notify_peers = false;
3807 	bool should_notify_rtnl = false;
3808 	int delta_in_ticks;
3809 
3810 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3811 
3812 	if (!bond_has_slaves(bond))
3813 		goto re_arm;
3814 
3815 	rcu_read_lock();
3816 
3817 	should_notify_peers = bond_should_notify_peers(bond);
3818 
3819 	if (bond_ab_arp_inspect(bond)) {
3820 		rcu_read_unlock();
3821 
3822 		/* Race avoidance with bond_close flush of workqueue */
3823 		if (!rtnl_trylock()) {
3824 			delta_in_ticks = 1;
3825 			should_notify_peers = false;
3826 			goto re_arm;
3827 		}
3828 
3829 		bond_ab_arp_commit(bond);
3830 
3831 		rtnl_unlock();
3832 		rcu_read_lock();
3833 	}
3834 
3835 	should_notify_rtnl = bond_ab_arp_probe(bond);
3836 	rcu_read_unlock();
3837 
3838 re_arm:
3839 	if (bond->params.arp_interval)
3840 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3841 
3842 	if (should_notify_peers || should_notify_rtnl) {
3843 		if (!rtnl_trylock())
3844 			return;
3845 
3846 		if (should_notify_peers) {
3847 			bond->send_peer_notif--;
3848 			call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3849 						 bond->dev);
3850 		}
3851 		if (should_notify_rtnl) {
3852 			bond_slave_state_notify(bond);
3853 			bond_slave_link_notify(bond);
3854 		}
3855 
3856 		rtnl_unlock();
3857 	}
3858 }
3859 
3860 static void bond_arp_monitor(struct work_struct *work)
3861 {
3862 	struct bonding *bond = container_of(work, struct bonding,
3863 					    arp_work.work);
3864 
3865 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3866 		bond_activebackup_arp_mon(bond);
3867 	else
3868 		bond_loadbalance_arp_mon(bond);
3869 }
3870 
3871 /*-------------------------- netdev event handling --------------------------*/
3872 
3873 /* Change device name */
3874 static int bond_event_changename(struct bonding *bond)
3875 {
3876 	bond_remove_proc_entry(bond);
3877 	bond_create_proc_entry(bond);
3878 
3879 	bond_debug_reregister(bond);
3880 
3881 	return NOTIFY_DONE;
3882 }
3883 
3884 static int bond_master_netdev_event(unsigned long event,
3885 				    struct net_device *bond_dev)
3886 {
3887 	struct bonding *event_bond = netdev_priv(bond_dev);
3888 
3889 	netdev_dbg(bond_dev, "%s called\n", __func__);
3890 
3891 	switch (event) {
3892 	case NETDEV_CHANGENAME:
3893 		return bond_event_changename(event_bond);
3894 	case NETDEV_UNREGISTER:
3895 		bond_remove_proc_entry(event_bond);
3896 #ifdef CONFIG_XFRM_OFFLOAD
3897 		xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3898 #endif /* CONFIG_XFRM_OFFLOAD */
3899 		break;
3900 	case NETDEV_REGISTER:
3901 		bond_create_proc_entry(event_bond);
3902 		break;
3903 	default:
3904 		break;
3905 	}
3906 
3907 	return NOTIFY_DONE;
3908 }
3909 
3910 static int bond_slave_netdev_event(unsigned long event,
3911 				   struct net_device *slave_dev)
3912 {
3913 	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3914 	struct bonding *bond;
3915 	struct net_device *bond_dev;
3916 
3917 	/* A netdev event can be generated while enslaving a device
3918 	 * before netdev_rx_handler_register is called in which case
3919 	 * slave will be NULL
3920 	 */
3921 	if (!slave) {
3922 		netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3923 		return NOTIFY_DONE;
3924 	}
3925 
3926 	bond_dev = slave->bond->dev;
3927 	bond = slave->bond;
3928 	primary = rtnl_dereference(bond->primary_slave);
3929 
3930 	slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3931 
3932 	switch (event) {
3933 	case NETDEV_UNREGISTER:
3934 		if (bond_dev->type != ARPHRD_ETHER)
3935 			bond_release_and_destroy(bond_dev, slave_dev);
3936 		else
3937 			__bond_release_one(bond_dev, slave_dev, false, true);
3938 		break;
3939 	case NETDEV_UP:
3940 	case NETDEV_CHANGE:
3941 		/* For 802.3ad mode only:
3942 		 * Getting invalid Speed/Duplex values here will put slave
3943 		 * in weird state. Mark it as link-fail if the link was
3944 		 * previously up or link-down if it hasn't yet come up, and
3945 		 * let link-monitoring (miimon) set it right when correct
3946 		 * speeds/duplex are available.
3947 		 */
3948 		if (bond_update_speed_duplex(slave) &&
3949 		    BOND_MODE(bond) == BOND_MODE_8023AD) {
3950 			if (slave->last_link_up)
3951 				slave->link = BOND_LINK_FAIL;
3952 			else
3953 				slave->link = BOND_LINK_DOWN;
3954 		}
3955 
3956 		if (BOND_MODE(bond) == BOND_MODE_8023AD)
3957 			bond_3ad_adapter_speed_duplex_changed(slave);
3958 		fallthrough;
3959 	case NETDEV_DOWN:
3960 		/* Refresh slave-array if applicable!
3961 		 * If the setup does not use miimon or arpmon (mode-specific!),
3962 		 * then these events will not cause the slave-array to be
3963 		 * refreshed. This will cause xmit to use a slave that is not
3964 		 * usable. Avoid such situation by refeshing the array at these
3965 		 * events. If these (miimon/arpmon) parameters are configured
3966 		 * then array gets refreshed twice and that should be fine!
3967 		 */
3968 		if (bond_mode_can_use_xmit_hash(bond))
3969 			bond_update_slave_arr(bond, NULL);
3970 		break;
3971 	case NETDEV_CHANGEMTU:
3972 		/* TODO: Should slaves be allowed to
3973 		 * independently alter their MTU?  For
3974 		 * an active-backup bond, slaves need
3975 		 * not be the same type of device, so
3976 		 * MTUs may vary.  For other modes,
3977 		 * slaves arguably should have the
3978 		 * same MTUs. To do this, we'd need to
3979 		 * take over the slave's change_mtu
3980 		 * function for the duration of their
3981 		 * servitude.
3982 		 */
3983 		break;
3984 	case NETDEV_CHANGENAME:
3985 		/* we don't care if we don't have primary set */
3986 		if (!bond_uses_primary(bond) ||
3987 		    !bond->params.primary[0])
3988 			break;
3989 
3990 		if (slave == primary) {
3991 			/* slave's name changed - he's no longer primary */
3992 			RCU_INIT_POINTER(bond->primary_slave, NULL);
3993 		} else if (!strcmp(slave_dev->name, bond->params.primary)) {
3994 			/* we have a new primary slave */
3995 			rcu_assign_pointer(bond->primary_slave, slave);
3996 		} else { /* we didn't change primary - exit */
3997 			break;
3998 		}
3999 
4000 		netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
4001 			    primary ? slave_dev->name : "none");
4002 
4003 		block_netpoll_tx();
4004 		bond_select_active_slave(bond);
4005 		unblock_netpoll_tx();
4006 		break;
4007 	case NETDEV_FEAT_CHANGE:
4008 		if (!bond->notifier_ctx) {
4009 			bond->notifier_ctx = true;
4010 			bond_compute_features(bond);
4011 			bond->notifier_ctx = false;
4012 		}
4013 		break;
4014 	case NETDEV_RESEND_IGMP:
4015 		/* Propagate to master device */
4016 		call_netdevice_notifiers(event, slave->bond->dev);
4017 		break;
4018 	case NETDEV_XDP_FEAT_CHANGE:
4019 		bond_xdp_set_features(bond_dev);
4020 		break;
4021 	default:
4022 		break;
4023 	}
4024 
4025 	return NOTIFY_DONE;
4026 }
4027 
4028 /* bond_netdev_event: handle netdev notifier chain events.
4029  *
4030  * This function receives events for the netdev chain.  The caller (an
4031  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
4032  * locks for us to safely manipulate the slave devices (RTNL lock,
4033  * dev_probe_lock).
4034  */
4035 static int bond_netdev_event(struct notifier_block *this,
4036 			     unsigned long event, void *ptr)
4037 {
4038 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
4039 
4040 	netdev_dbg(event_dev, "%s received %s\n",
4041 		   __func__, netdev_cmd_to_name(event));
4042 
4043 	if (!(event_dev->priv_flags & IFF_BONDING))
4044 		return NOTIFY_DONE;
4045 
4046 	if (event_dev->flags & IFF_MASTER) {
4047 		int ret;
4048 
4049 		ret = bond_master_netdev_event(event, event_dev);
4050 		if (ret != NOTIFY_DONE)
4051 			return ret;
4052 	}
4053 
4054 	if (event_dev->flags & IFF_SLAVE)
4055 		return bond_slave_netdev_event(event, event_dev);
4056 
4057 	return NOTIFY_DONE;
4058 }
4059 
4060 static struct notifier_block bond_netdev_notifier = {
4061 	.notifier_call = bond_netdev_event,
4062 };
4063 
4064 /*---------------------------- Hashing Policies -----------------------------*/
4065 
4066 /* Helper to access data in a packet, with or without a backing skb.
4067  * If skb is given the data is linearized if necessary via pskb_may_pull.
4068  */
4069 static inline const void *bond_pull_data(struct sk_buff *skb,
4070 					 const void *data, int hlen, int n)
4071 {
4072 	if (likely(n <= hlen))
4073 		return data;
4074 	else if (skb && likely(pskb_may_pull(skb, n)))
4075 		return skb->data;
4076 
4077 	return NULL;
4078 }
4079 
4080 /* L2 hash helper */
4081 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4082 {
4083 	struct ethhdr *ep;
4084 
4085 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4086 	if (!data)
4087 		return 0;
4088 
4089 	ep = (struct ethhdr *)(data + mhoff);
4090 	return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4091 }
4092 
4093 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4094 			 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4095 {
4096 	const struct ipv6hdr *iph6;
4097 	const struct iphdr *iph;
4098 
4099 	if (l2_proto == htons(ETH_P_IP)) {
4100 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4101 		if (!data)
4102 			return false;
4103 
4104 		iph = (const struct iphdr *)(data + *nhoff);
4105 		iph_to_flow_copy_v4addrs(fk, iph);
4106 		*nhoff += iph->ihl << 2;
4107 		if (!ip_is_fragment(iph))
4108 			*ip_proto = iph->protocol;
4109 	} else if (l2_proto == htons(ETH_P_IPV6)) {
4110 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4111 		if (!data)
4112 			return false;
4113 
4114 		iph6 = (const struct ipv6hdr *)(data + *nhoff);
4115 		iph_to_flow_copy_v6addrs(fk, iph6);
4116 		*nhoff += sizeof(*iph6);
4117 		*ip_proto = iph6->nexthdr;
4118 	} else {
4119 		return false;
4120 	}
4121 
4122 	if (l34 && *ip_proto >= 0)
4123 		fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4124 
4125 	return true;
4126 }
4127 
4128 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4129 {
4130 	u32 srcmac_vendor = 0, srcmac_dev = 0;
4131 	struct ethhdr *mac_hdr;
4132 	u16 vlan = 0;
4133 	int i;
4134 
4135 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4136 	if (!data)
4137 		return 0;
4138 	mac_hdr = (struct ethhdr *)(data + mhoff);
4139 
4140 	for (i = 0; i < 3; i++)
4141 		srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4142 
4143 	for (i = 3; i < ETH_ALEN; i++)
4144 		srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4145 
4146 	if (skb && skb_vlan_tag_present(skb))
4147 		vlan = skb_vlan_tag_get(skb);
4148 
4149 	return vlan ^ srcmac_vendor ^ srcmac_dev;
4150 }
4151 
4152 /* Extract the appropriate headers based on bond's xmit policy */
4153 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4154 			      __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4155 {
4156 	bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4157 	int ip_proto = -1;
4158 
4159 	switch (bond->params.xmit_policy) {
4160 	case BOND_XMIT_POLICY_ENCAP23:
4161 	case BOND_XMIT_POLICY_ENCAP34:
4162 		memset(fk, 0, sizeof(*fk));
4163 		return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4164 					  fk, data, l2_proto, nhoff, hlen, 0);
4165 	default:
4166 		break;
4167 	}
4168 
4169 	fk->ports.ports = 0;
4170 	memset(&fk->icmp, 0, sizeof(fk->icmp));
4171 	if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4172 		return false;
4173 
4174 	/* ICMP error packets contains at least 8 bytes of the header
4175 	 * of the packet which generated the error. Use this information
4176 	 * to correlate ICMP error packets within the same flow which
4177 	 * generated the error.
4178 	 */
4179 	if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4180 		skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4181 		if (ip_proto == IPPROTO_ICMP) {
4182 			if (!icmp_is_err(fk->icmp.type))
4183 				return true;
4184 
4185 			nhoff += sizeof(struct icmphdr);
4186 		} else if (ip_proto == IPPROTO_ICMPV6) {
4187 			if (!icmpv6_is_err(fk->icmp.type))
4188 				return true;
4189 
4190 			nhoff += sizeof(struct icmp6hdr);
4191 		}
4192 		return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4193 	}
4194 
4195 	return true;
4196 }
4197 
4198 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4199 {
4200 	hash ^= (__force u32)flow_get_u32_dst(flow) ^
4201 		(__force u32)flow_get_u32_src(flow);
4202 	hash ^= (hash >> 16);
4203 	hash ^= (hash >> 8);
4204 
4205 	/* discard lowest hash bit to deal with the common even ports pattern */
4206 	if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4207 		xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4208 		return hash >> 1;
4209 
4210 	return hash;
4211 }
4212 
4213 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4214  * the data as required, but this function can be used without it if the data is
4215  * known to be linear (e.g. with xdp_buff).
4216  */
4217 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4218 			    __be16 l2_proto, int mhoff, int nhoff, int hlen)
4219 {
4220 	struct flow_keys flow;
4221 	u32 hash;
4222 
4223 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4224 		return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4225 
4226 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4227 	    !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4228 		return bond_eth_hash(skb, data, mhoff, hlen);
4229 
4230 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4231 	    bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4232 		hash = bond_eth_hash(skb, data, mhoff, hlen);
4233 	} else {
4234 		if (flow.icmp.id)
4235 			memcpy(&hash, &flow.icmp, sizeof(hash));
4236 		else
4237 			memcpy(&hash, &flow.ports.ports, sizeof(hash));
4238 	}
4239 
4240 	return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4241 }
4242 
4243 /**
4244  * bond_xmit_hash - generate a hash value based on the xmit policy
4245  * @bond: bonding device
4246  * @skb: buffer to use for headers
4247  *
4248  * This function will extract the necessary headers from the skb buffer and use
4249  * them to generate a hash based on the xmit_policy set in the bonding device
4250  */
4251 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4252 {
4253 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4254 	    skb->l4_hash)
4255 		return skb->hash;
4256 
4257 	return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4258 				0, skb_network_offset(skb),
4259 				skb_headlen(skb));
4260 }
4261 
4262 /**
4263  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4264  * @bond: bonding device
4265  * @xdp: buffer to use for headers
4266  *
4267  * The XDP variant of bond_xmit_hash.
4268  */
4269 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4270 {
4271 	struct ethhdr *eth;
4272 
4273 	if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4274 		return 0;
4275 
4276 	eth = (struct ethhdr *)xdp->data;
4277 
4278 	return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4279 				sizeof(struct ethhdr), xdp->data_end - xdp->data);
4280 }
4281 
4282 /*-------------------------- Device entry points ----------------------------*/
4283 
4284 void bond_work_init_all(struct bonding *bond)
4285 {
4286 	INIT_DELAYED_WORK(&bond->mcast_work,
4287 			  bond_resend_igmp_join_requests_delayed);
4288 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4289 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4290 	INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4291 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4292 	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4293 }
4294 
4295 static void bond_work_cancel_all(struct bonding *bond)
4296 {
4297 	cancel_delayed_work_sync(&bond->mii_work);
4298 	cancel_delayed_work_sync(&bond->arp_work);
4299 	cancel_delayed_work_sync(&bond->alb_work);
4300 	cancel_delayed_work_sync(&bond->ad_work);
4301 	cancel_delayed_work_sync(&bond->mcast_work);
4302 	cancel_delayed_work_sync(&bond->slave_arr_work);
4303 }
4304 
4305 static int bond_open(struct net_device *bond_dev)
4306 {
4307 	struct bonding *bond = netdev_priv(bond_dev);
4308 	struct list_head *iter;
4309 	struct slave *slave;
4310 
4311 	if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4312 		bond->rr_tx_counter = alloc_percpu(u32);
4313 		if (!bond->rr_tx_counter)
4314 			return -ENOMEM;
4315 	}
4316 
4317 	/* reset slave->backup and slave->inactive */
4318 	if (bond_has_slaves(bond)) {
4319 		bond_for_each_slave(bond, slave, iter) {
4320 			if (bond_uses_primary(bond) &&
4321 			    slave != rcu_access_pointer(bond->curr_active_slave)) {
4322 				bond_set_slave_inactive_flags(slave,
4323 							      BOND_SLAVE_NOTIFY_NOW);
4324 			} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4325 				bond_set_slave_active_flags(slave,
4326 							    BOND_SLAVE_NOTIFY_NOW);
4327 			}
4328 		}
4329 	}
4330 
4331 	if (bond_is_lb(bond)) {
4332 		/* bond_alb_initialize must be called before the timer
4333 		 * is started.
4334 		 */
4335 		if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4336 			return -ENOMEM;
4337 		if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4338 			queue_delayed_work(bond->wq, &bond->alb_work, 0);
4339 	}
4340 
4341 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
4342 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
4343 
4344 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4345 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
4346 		bond->recv_probe = bond_rcv_validate;
4347 	}
4348 
4349 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4350 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
4351 		/* register to receive LACPDUs */
4352 		bond->recv_probe = bond_3ad_lacpdu_recv;
4353 		bond_3ad_initiate_agg_selection(bond, 1);
4354 
4355 		bond_for_each_slave(bond, slave, iter)
4356 			dev_mc_add(slave->dev, lacpdu_mcast_addr);
4357 	}
4358 
4359 	if (bond_mode_can_use_xmit_hash(bond))
4360 		bond_update_slave_arr(bond, NULL);
4361 
4362 	return 0;
4363 }
4364 
4365 static int bond_close(struct net_device *bond_dev)
4366 {
4367 	struct bonding *bond = netdev_priv(bond_dev);
4368 	struct slave *slave;
4369 
4370 	bond_work_cancel_all(bond);
4371 	bond->send_peer_notif = 0;
4372 	if (bond_is_lb(bond))
4373 		bond_alb_deinitialize(bond);
4374 	bond->recv_probe = NULL;
4375 
4376 	if (bond_uses_primary(bond)) {
4377 		rcu_read_lock();
4378 		slave = rcu_dereference(bond->curr_active_slave);
4379 		if (slave)
4380 			bond_hw_addr_flush(bond_dev, slave->dev);
4381 		rcu_read_unlock();
4382 	} else {
4383 		struct list_head *iter;
4384 
4385 		bond_for_each_slave(bond, slave, iter)
4386 			bond_hw_addr_flush(bond_dev, slave->dev);
4387 	}
4388 
4389 	return 0;
4390 }
4391 
4392 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4393  * that some drivers can provide 32bit values only.
4394  */
4395 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4396 			    const struct rtnl_link_stats64 *_new,
4397 			    const struct rtnl_link_stats64 *_old)
4398 {
4399 	const u64 *new = (const u64 *)_new;
4400 	const u64 *old = (const u64 *)_old;
4401 	u64 *res = (u64 *)_res;
4402 	int i;
4403 
4404 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4405 		u64 nv = new[i];
4406 		u64 ov = old[i];
4407 		s64 delta = nv - ov;
4408 
4409 		/* detects if this particular field is 32bit only */
4410 		if (((nv | ov) >> 32) == 0)
4411 			delta = (s64)(s32)((u32)nv - (u32)ov);
4412 
4413 		/* filter anomalies, some drivers reset their stats
4414 		 * at down/up events.
4415 		 */
4416 		if (delta > 0)
4417 			res[i] += delta;
4418 	}
4419 }
4420 
4421 #ifdef CONFIG_LOCKDEP
4422 static int bond_get_lowest_level_rcu(struct net_device *dev)
4423 {
4424 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4425 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4426 	int cur = 0, max = 0;
4427 
4428 	now = dev;
4429 	iter = &dev->adj_list.lower;
4430 
4431 	while (1) {
4432 		next = NULL;
4433 		while (1) {
4434 			ldev = netdev_next_lower_dev_rcu(now, &iter);
4435 			if (!ldev)
4436 				break;
4437 
4438 			next = ldev;
4439 			niter = &ldev->adj_list.lower;
4440 			dev_stack[cur] = now;
4441 			iter_stack[cur++] = iter;
4442 			if (max <= cur)
4443 				max = cur;
4444 			break;
4445 		}
4446 
4447 		if (!next) {
4448 			if (!cur)
4449 				return max;
4450 			next = dev_stack[--cur];
4451 			niter = iter_stack[cur];
4452 		}
4453 
4454 		now = next;
4455 		iter = niter;
4456 	}
4457 
4458 	return max;
4459 }
4460 #endif
4461 
4462 static void bond_get_stats(struct net_device *bond_dev,
4463 			   struct rtnl_link_stats64 *stats)
4464 {
4465 	struct bonding *bond = netdev_priv(bond_dev);
4466 	struct rtnl_link_stats64 temp;
4467 	struct list_head *iter;
4468 	struct slave *slave;
4469 	int nest_level = 0;
4470 
4471 
4472 	rcu_read_lock();
4473 #ifdef CONFIG_LOCKDEP
4474 	nest_level = bond_get_lowest_level_rcu(bond_dev);
4475 #endif
4476 
4477 	spin_lock_nested(&bond->stats_lock, nest_level);
4478 	memcpy(stats, &bond->bond_stats, sizeof(*stats));
4479 
4480 	bond_for_each_slave_rcu(bond, slave, iter) {
4481 		const struct rtnl_link_stats64 *new =
4482 			dev_get_stats(slave->dev, &temp);
4483 
4484 		bond_fold_stats(stats, new, &slave->slave_stats);
4485 
4486 		/* save off the slave stats for the next run */
4487 		memcpy(&slave->slave_stats, new, sizeof(*new));
4488 	}
4489 
4490 	memcpy(&bond->bond_stats, stats, sizeof(*stats));
4491 	spin_unlock(&bond->stats_lock);
4492 	rcu_read_unlock();
4493 }
4494 
4495 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4496 {
4497 	struct bonding *bond = netdev_priv(bond_dev);
4498 	struct mii_ioctl_data *mii = NULL;
4499 
4500 	netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4501 
4502 	switch (cmd) {
4503 	case SIOCGMIIPHY:
4504 		mii = if_mii(ifr);
4505 		if (!mii)
4506 			return -EINVAL;
4507 
4508 		mii->phy_id = 0;
4509 		fallthrough;
4510 	case SIOCGMIIREG:
4511 		/* We do this again just in case we were called by SIOCGMIIREG
4512 		 * instead of SIOCGMIIPHY.
4513 		 */
4514 		mii = if_mii(ifr);
4515 		if (!mii)
4516 			return -EINVAL;
4517 
4518 		if (mii->reg_num == 1) {
4519 			mii->val_out = 0;
4520 			if (netif_carrier_ok(bond->dev))
4521 				mii->val_out = BMSR_LSTATUS;
4522 		}
4523 
4524 		break;
4525 	default:
4526 		return -EOPNOTSUPP;
4527 	}
4528 
4529 	return 0;
4530 }
4531 
4532 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4533 {
4534 	struct bonding *bond = netdev_priv(bond_dev);
4535 	struct net_device *slave_dev = NULL;
4536 	struct ifbond k_binfo;
4537 	struct ifbond __user *u_binfo = NULL;
4538 	struct ifslave k_sinfo;
4539 	struct ifslave __user *u_sinfo = NULL;
4540 	struct bond_opt_value newval;
4541 	struct net *net;
4542 	int res = 0;
4543 
4544 	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4545 
4546 	switch (cmd) {
4547 	case SIOCBONDINFOQUERY:
4548 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
4549 
4550 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4551 			return -EFAULT;
4552 
4553 		bond_info_query(bond_dev, &k_binfo);
4554 		if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4555 			return -EFAULT;
4556 
4557 		return 0;
4558 	case SIOCBONDSLAVEINFOQUERY:
4559 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4560 
4561 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4562 			return -EFAULT;
4563 
4564 		res = bond_slave_info_query(bond_dev, &k_sinfo);
4565 		if (res == 0 &&
4566 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4567 			return -EFAULT;
4568 
4569 		return res;
4570 	default:
4571 		break;
4572 	}
4573 
4574 	net = dev_net(bond_dev);
4575 
4576 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4577 		return -EPERM;
4578 
4579 	slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4580 
4581 	slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4582 
4583 	if (!slave_dev)
4584 		return -ENODEV;
4585 
4586 	switch (cmd) {
4587 	case SIOCBONDENSLAVE:
4588 		res = bond_enslave(bond_dev, slave_dev, NULL);
4589 		break;
4590 	case SIOCBONDRELEASE:
4591 		res = bond_release(bond_dev, slave_dev);
4592 		break;
4593 	case SIOCBONDSETHWADDR:
4594 		res = bond_set_dev_addr(bond_dev, slave_dev);
4595 		break;
4596 	case SIOCBONDCHANGEACTIVE:
4597 		bond_opt_initstr(&newval, slave_dev->name);
4598 		res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4599 					    &newval);
4600 		break;
4601 	default:
4602 		res = -EOPNOTSUPP;
4603 	}
4604 
4605 	return res;
4606 }
4607 
4608 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4609 			       void __user *data, int cmd)
4610 {
4611 	struct ifreq ifrdata = { .ifr_data = data };
4612 
4613 	switch (cmd) {
4614 	case BOND_INFO_QUERY_OLD:
4615 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4616 	case BOND_SLAVE_INFO_QUERY_OLD:
4617 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4618 	case BOND_ENSLAVE_OLD:
4619 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4620 	case BOND_RELEASE_OLD:
4621 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4622 	case BOND_SETHWADDR_OLD:
4623 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4624 	case BOND_CHANGE_ACTIVE_OLD:
4625 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4626 	}
4627 
4628 	return -EOPNOTSUPP;
4629 }
4630 
4631 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4632 {
4633 	struct bonding *bond = netdev_priv(bond_dev);
4634 
4635 	if (change & IFF_PROMISC)
4636 		bond_set_promiscuity(bond,
4637 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
4638 
4639 	if (change & IFF_ALLMULTI)
4640 		bond_set_allmulti(bond,
4641 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4642 }
4643 
4644 static void bond_set_rx_mode(struct net_device *bond_dev)
4645 {
4646 	struct bonding *bond = netdev_priv(bond_dev);
4647 	struct list_head *iter;
4648 	struct slave *slave;
4649 
4650 	rcu_read_lock();
4651 	if (bond_uses_primary(bond)) {
4652 		slave = rcu_dereference(bond->curr_active_slave);
4653 		if (slave) {
4654 			dev_uc_sync(slave->dev, bond_dev);
4655 			dev_mc_sync(slave->dev, bond_dev);
4656 		}
4657 	} else {
4658 		bond_for_each_slave_rcu(bond, slave, iter) {
4659 			dev_uc_sync_multiple(slave->dev, bond_dev);
4660 			dev_mc_sync_multiple(slave->dev, bond_dev);
4661 		}
4662 	}
4663 	rcu_read_unlock();
4664 }
4665 
4666 static int bond_neigh_init(struct neighbour *n)
4667 {
4668 	struct bonding *bond = netdev_priv(n->dev);
4669 	const struct net_device_ops *slave_ops;
4670 	struct neigh_parms parms;
4671 	struct slave *slave;
4672 	int ret = 0;
4673 
4674 	rcu_read_lock();
4675 	slave = bond_first_slave_rcu(bond);
4676 	if (!slave)
4677 		goto out;
4678 	slave_ops = slave->dev->netdev_ops;
4679 	if (!slave_ops->ndo_neigh_setup)
4680 		goto out;
4681 
4682 	/* TODO: find another way [1] to implement this.
4683 	 * Passing a zeroed structure is fragile,
4684 	 * but at least we do not pass garbage.
4685 	 *
4686 	 * [1] One way would be that ndo_neigh_setup() never touch
4687 	 *     struct neigh_parms, but propagate the new neigh_setup()
4688 	 *     back to ___neigh_create() / neigh_parms_alloc()
4689 	 */
4690 	memset(&parms, 0, sizeof(parms));
4691 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4692 
4693 	if (ret)
4694 		goto out;
4695 
4696 	if (parms.neigh_setup)
4697 		ret = parms.neigh_setup(n);
4698 out:
4699 	rcu_read_unlock();
4700 	return ret;
4701 }
4702 
4703 /* The bonding ndo_neigh_setup is called at init time beofre any
4704  * slave exists. So we must declare proxy setup function which will
4705  * be used at run time to resolve the actual slave neigh param setup.
4706  *
4707  * It's also called by master devices (such as vlans) to setup their
4708  * underlying devices. In that case - do nothing, we're already set up from
4709  * our init.
4710  */
4711 static int bond_neigh_setup(struct net_device *dev,
4712 			    struct neigh_parms *parms)
4713 {
4714 	/* modify only our neigh_parms */
4715 	if (parms->dev == dev)
4716 		parms->neigh_setup = bond_neigh_init;
4717 
4718 	return 0;
4719 }
4720 
4721 /* Change the MTU of all of a master's slaves to match the master */
4722 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4723 {
4724 	struct bonding *bond = netdev_priv(bond_dev);
4725 	struct slave *slave, *rollback_slave;
4726 	struct list_head *iter;
4727 	int res = 0;
4728 
4729 	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4730 
4731 	bond_for_each_slave(bond, slave, iter) {
4732 		slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4733 			   slave, slave->dev->netdev_ops->ndo_change_mtu);
4734 
4735 		res = dev_set_mtu(slave->dev, new_mtu);
4736 
4737 		if (res) {
4738 			/* If we failed to set the slave's mtu to the new value
4739 			 * we must abort the operation even in ACTIVE_BACKUP
4740 			 * mode, because if we allow the backup slaves to have
4741 			 * different mtu values than the active slave we'll
4742 			 * need to change their mtu when doing a failover. That
4743 			 * means changing their mtu from timer context, which
4744 			 * is probably not a good idea.
4745 			 */
4746 			slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4747 				  res, new_mtu);
4748 			goto unwind;
4749 		}
4750 	}
4751 
4752 	WRITE_ONCE(bond_dev->mtu, new_mtu);
4753 
4754 	return 0;
4755 
4756 unwind:
4757 	/* unwind from head to the slave that failed */
4758 	bond_for_each_slave(bond, rollback_slave, iter) {
4759 		int tmp_res;
4760 
4761 		if (rollback_slave == slave)
4762 			break;
4763 
4764 		tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4765 		if (tmp_res)
4766 			slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4767 				  tmp_res);
4768 	}
4769 
4770 	return res;
4771 }
4772 
4773 /* Change HW address
4774  *
4775  * Note that many devices must be down to change the HW address, and
4776  * downing the master releases all slaves.  We can make bonds full of
4777  * bonding devices to test this, however.
4778  */
4779 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4780 {
4781 	struct bonding *bond = netdev_priv(bond_dev);
4782 	struct slave *slave, *rollback_slave;
4783 	struct sockaddr_storage *ss = addr, tmp_ss;
4784 	struct list_head *iter;
4785 	int res = 0;
4786 
4787 	if (BOND_MODE(bond) == BOND_MODE_ALB)
4788 		return bond_alb_set_mac_address(bond_dev, addr);
4789 
4790 
4791 	netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4792 
4793 	/* If fail_over_mac is enabled, do nothing and return success.
4794 	 * Returning an error causes ifenslave to fail.
4795 	 */
4796 	if (bond->params.fail_over_mac &&
4797 	    BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4798 		return 0;
4799 
4800 	if (!is_valid_ether_addr(ss->__data))
4801 		return -EADDRNOTAVAIL;
4802 
4803 	bond_for_each_slave(bond, slave, iter) {
4804 		slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4805 			  __func__, slave);
4806 		res = dev_set_mac_address(slave->dev, addr, NULL);
4807 		if (res) {
4808 			/* TODO: consider downing the slave
4809 			 * and retry ?
4810 			 * User should expect communications
4811 			 * breakage anyway until ARP finish
4812 			 * updating, so...
4813 			 */
4814 			slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4815 				  __func__, res);
4816 			goto unwind;
4817 		}
4818 	}
4819 
4820 	/* success */
4821 	dev_addr_set(bond_dev, ss->__data);
4822 	return 0;
4823 
4824 unwind:
4825 	memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4826 	tmp_ss.ss_family = bond_dev->type;
4827 
4828 	/* unwind from head to the slave that failed */
4829 	bond_for_each_slave(bond, rollback_slave, iter) {
4830 		int tmp_res;
4831 
4832 		if (rollback_slave == slave)
4833 			break;
4834 
4835 		tmp_res = dev_set_mac_address(rollback_slave->dev,
4836 					      (struct sockaddr *)&tmp_ss, NULL);
4837 		if (tmp_res) {
4838 			slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4839 				   __func__, tmp_res);
4840 		}
4841 	}
4842 
4843 	return res;
4844 }
4845 
4846 /**
4847  * bond_get_slave_by_id - get xmit slave with slave_id
4848  * @bond: bonding device that is transmitting
4849  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4850  *
4851  * This function tries to get slave with slave_id but in case
4852  * it fails, it tries to find the first available slave for transmission.
4853  */
4854 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4855 					  int slave_id)
4856 {
4857 	struct list_head *iter;
4858 	struct slave *slave;
4859 	int i = slave_id;
4860 
4861 	/* Here we start from the slave with slave_id */
4862 	bond_for_each_slave_rcu(bond, slave, iter) {
4863 		if (--i < 0) {
4864 			if (bond_slave_can_tx(slave))
4865 				return slave;
4866 		}
4867 	}
4868 
4869 	/* Here we start from the first slave up to slave_id */
4870 	i = slave_id;
4871 	bond_for_each_slave_rcu(bond, slave, iter) {
4872 		if (--i < 0)
4873 			break;
4874 		if (bond_slave_can_tx(slave))
4875 			return slave;
4876 	}
4877 	/* no slave that can tx has been found */
4878 	return NULL;
4879 }
4880 
4881 /**
4882  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4883  * @bond: bonding device to use
4884  *
4885  * Based on the value of the bonding device's packets_per_slave parameter
4886  * this function generates a slave id, which is usually used as the next
4887  * slave to transmit through.
4888  */
4889 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4890 {
4891 	u32 slave_id;
4892 	struct reciprocal_value reciprocal_packets_per_slave;
4893 	int packets_per_slave = bond->params.packets_per_slave;
4894 
4895 	switch (packets_per_slave) {
4896 	case 0:
4897 		slave_id = get_random_u32();
4898 		break;
4899 	case 1:
4900 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4901 		break;
4902 	default:
4903 		reciprocal_packets_per_slave =
4904 			bond->params.reciprocal_packets_per_slave;
4905 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4906 		slave_id = reciprocal_divide(slave_id,
4907 					     reciprocal_packets_per_slave);
4908 		break;
4909 	}
4910 
4911 	return slave_id;
4912 }
4913 
4914 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4915 						    struct sk_buff *skb)
4916 {
4917 	struct slave *slave;
4918 	int slave_cnt;
4919 	u32 slave_id;
4920 
4921 	/* Start with the curr_active_slave that joined the bond as the
4922 	 * default for sending IGMP traffic.  For failover purposes one
4923 	 * needs to maintain some consistency for the interface that will
4924 	 * send the join/membership reports.  The curr_active_slave found
4925 	 * will send all of this type of traffic.
4926 	 */
4927 	if (skb->protocol == htons(ETH_P_IP)) {
4928 		int noff = skb_network_offset(skb);
4929 		struct iphdr *iph;
4930 
4931 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4932 			goto non_igmp;
4933 
4934 		iph = ip_hdr(skb);
4935 		if (iph->protocol == IPPROTO_IGMP) {
4936 			slave = rcu_dereference(bond->curr_active_slave);
4937 			if (slave)
4938 				return slave;
4939 			return bond_get_slave_by_id(bond, 0);
4940 		}
4941 	}
4942 
4943 non_igmp:
4944 	slave_cnt = READ_ONCE(bond->slave_cnt);
4945 	if (likely(slave_cnt)) {
4946 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4947 		return bond_get_slave_by_id(bond, slave_id);
4948 	}
4949 	return NULL;
4950 }
4951 
4952 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4953 							struct xdp_buff *xdp)
4954 {
4955 	struct slave *slave;
4956 	int slave_cnt;
4957 	u32 slave_id;
4958 	const struct ethhdr *eth;
4959 	void *data = xdp->data;
4960 
4961 	if (data + sizeof(struct ethhdr) > xdp->data_end)
4962 		goto non_igmp;
4963 
4964 	eth = (struct ethhdr *)data;
4965 	data += sizeof(struct ethhdr);
4966 
4967 	/* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4968 	if (eth->h_proto == htons(ETH_P_IP)) {
4969 		const struct iphdr *iph;
4970 
4971 		if (data + sizeof(struct iphdr) > xdp->data_end)
4972 			goto non_igmp;
4973 
4974 		iph = (struct iphdr *)data;
4975 
4976 		if (iph->protocol == IPPROTO_IGMP) {
4977 			slave = rcu_dereference(bond->curr_active_slave);
4978 			if (slave)
4979 				return slave;
4980 			return bond_get_slave_by_id(bond, 0);
4981 		}
4982 	}
4983 
4984 non_igmp:
4985 	slave_cnt = READ_ONCE(bond->slave_cnt);
4986 	if (likely(slave_cnt)) {
4987 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4988 		return bond_get_slave_by_id(bond, slave_id);
4989 	}
4990 	return NULL;
4991 }
4992 
4993 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4994 					struct net_device *bond_dev)
4995 {
4996 	struct bonding *bond = netdev_priv(bond_dev);
4997 	struct slave *slave;
4998 
4999 	slave = bond_xmit_roundrobin_slave_get(bond, skb);
5000 	if (likely(slave))
5001 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5002 
5003 	return bond_tx_drop(bond_dev, skb);
5004 }
5005 
5006 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
5007 {
5008 	return rcu_dereference(bond->curr_active_slave);
5009 }
5010 
5011 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
5012  * the bond has a usable interface.
5013  */
5014 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
5015 					  struct net_device *bond_dev)
5016 {
5017 	struct bonding *bond = netdev_priv(bond_dev);
5018 	struct slave *slave;
5019 
5020 	slave = bond_xmit_activebackup_slave_get(bond);
5021 	if (slave)
5022 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5023 
5024 	return bond_tx_drop(bond_dev, skb);
5025 }
5026 
5027 /* Use this to update slave_array when (a) it's not appropriate to update
5028  * slave_array right away (note that update_slave_array() may sleep)
5029  * and / or (b) RTNL is not held.
5030  */
5031 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
5032 {
5033 	queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
5034 }
5035 
5036 /* Slave array work handler. Holds only RTNL */
5037 static void bond_slave_arr_handler(struct work_struct *work)
5038 {
5039 	struct bonding *bond = container_of(work, struct bonding,
5040 					    slave_arr_work.work);
5041 	int ret;
5042 
5043 	if (!rtnl_trylock())
5044 		goto err;
5045 
5046 	ret = bond_update_slave_arr(bond, NULL);
5047 	rtnl_unlock();
5048 	if (ret) {
5049 		pr_warn_ratelimited("Failed to update slave array from WT\n");
5050 		goto err;
5051 	}
5052 	return;
5053 
5054 err:
5055 	bond_slave_arr_work_rearm(bond, 1);
5056 }
5057 
5058 static void bond_skip_slave(struct bond_up_slave *slaves,
5059 			    struct slave *skipslave)
5060 {
5061 	int idx;
5062 
5063 	/* Rare situation where caller has asked to skip a specific
5064 	 * slave but allocation failed (most likely!). BTW this is
5065 	 * only possible when the call is initiated from
5066 	 * __bond_release_one(). In this situation; overwrite the
5067 	 * skipslave entry in the array with the last entry from the
5068 	 * array to avoid a situation where the xmit path may choose
5069 	 * this to-be-skipped slave to send a packet out.
5070 	 */
5071 	for (idx = 0; slaves && idx < slaves->count; idx++) {
5072 		if (skipslave == slaves->arr[idx]) {
5073 			slaves->arr[idx] =
5074 				slaves->arr[slaves->count - 1];
5075 			slaves->count--;
5076 			break;
5077 		}
5078 	}
5079 }
5080 
5081 static void bond_set_slave_arr(struct bonding *bond,
5082 			       struct bond_up_slave *usable_slaves,
5083 			       struct bond_up_slave *all_slaves)
5084 {
5085 	struct bond_up_slave *usable, *all;
5086 
5087 	usable = rtnl_dereference(bond->usable_slaves);
5088 	rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5089 	kfree_rcu(usable, rcu);
5090 
5091 	all = rtnl_dereference(bond->all_slaves);
5092 	rcu_assign_pointer(bond->all_slaves, all_slaves);
5093 	kfree_rcu(all, rcu);
5094 }
5095 
5096 static void bond_reset_slave_arr(struct bonding *bond)
5097 {
5098 	bond_set_slave_arr(bond, NULL, NULL);
5099 }
5100 
5101 /* Build the usable slaves array in control path for modes that use xmit-hash
5102  * to determine the slave interface -
5103  * (a) BOND_MODE_8023AD
5104  * (b) BOND_MODE_XOR
5105  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5106  *
5107  * The caller is expected to hold RTNL only and NO other lock!
5108  */
5109 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5110 {
5111 	struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5112 	struct slave *slave;
5113 	struct list_head *iter;
5114 	int agg_id = 0;
5115 	int ret = 0;
5116 
5117 	might_sleep();
5118 
5119 	usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5120 					    bond->slave_cnt), GFP_KERNEL);
5121 	all_slaves = kzalloc(struct_size(all_slaves, arr,
5122 					 bond->slave_cnt), GFP_KERNEL);
5123 	if (!usable_slaves || !all_slaves) {
5124 		ret = -ENOMEM;
5125 		goto out;
5126 	}
5127 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5128 		struct ad_info ad_info;
5129 
5130 		spin_lock_bh(&bond->mode_lock);
5131 		if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5132 			spin_unlock_bh(&bond->mode_lock);
5133 			pr_debug("bond_3ad_get_active_agg_info failed\n");
5134 			/* No active aggragator means it's not safe to use
5135 			 * the previous array.
5136 			 */
5137 			bond_reset_slave_arr(bond);
5138 			goto out;
5139 		}
5140 		spin_unlock_bh(&bond->mode_lock);
5141 		agg_id = ad_info.aggregator_id;
5142 	}
5143 	bond_for_each_slave(bond, slave, iter) {
5144 		if (skipslave == slave)
5145 			continue;
5146 
5147 		all_slaves->arr[all_slaves->count++] = slave;
5148 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5149 			struct aggregator *agg;
5150 
5151 			agg = SLAVE_AD_INFO(slave)->port.aggregator;
5152 			if (!agg || agg->aggregator_identifier != agg_id)
5153 				continue;
5154 		}
5155 		if (!bond_slave_can_tx(slave))
5156 			continue;
5157 
5158 		slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5159 			  usable_slaves->count);
5160 
5161 		usable_slaves->arr[usable_slaves->count++] = slave;
5162 	}
5163 
5164 	bond_set_slave_arr(bond, usable_slaves, all_slaves);
5165 	return ret;
5166 out:
5167 	if (ret != 0 && skipslave) {
5168 		bond_skip_slave(rtnl_dereference(bond->all_slaves),
5169 				skipslave);
5170 		bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5171 				skipslave);
5172 	}
5173 	kfree_rcu(all_slaves, rcu);
5174 	kfree_rcu(usable_slaves, rcu);
5175 
5176 	return ret;
5177 }
5178 
5179 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5180 						 struct sk_buff *skb,
5181 						 struct bond_up_slave *slaves)
5182 {
5183 	struct slave *slave;
5184 	unsigned int count;
5185 	u32 hash;
5186 
5187 	hash = bond_xmit_hash(bond, skb);
5188 	count = slaves ? READ_ONCE(slaves->count) : 0;
5189 	if (unlikely(!count))
5190 		return NULL;
5191 
5192 	slave = slaves->arr[hash % count];
5193 	return slave;
5194 }
5195 
5196 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5197 						     struct xdp_buff *xdp)
5198 {
5199 	struct bond_up_slave *slaves;
5200 	unsigned int count;
5201 	u32 hash;
5202 
5203 	hash = bond_xmit_hash_xdp(bond, xdp);
5204 	slaves = rcu_dereference(bond->usable_slaves);
5205 	count = slaves ? READ_ONCE(slaves->count) : 0;
5206 	if (unlikely(!count))
5207 		return NULL;
5208 
5209 	return slaves->arr[hash % count];
5210 }
5211 
5212 /* Use this Xmit function for 3AD as well as XOR modes. The current
5213  * usable slave array is formed in the control path. The xmit function
5214  * just calculates hash and sends the packet out.
5215  */
5216 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5217 				     struct net_device *dev)
5218 {
5219 	struct bonding *bond = netdev_priv(dev);
5220 	struct bond_up_slave *slaves;
5221 	struct slave *slave;
5222 
5223 	slaves = rcu_dereference(bond->usable_slaves);
5224 	slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5225 	if (likely(slave))
5226 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5227 
5228 	return bond_tx_drop(dev, skb);
5229 }
5230 
5231 /* in broadcast mode, we send everything to all usable interfaces. */
5232 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5233 				       struct net_device *bond_dev)
5234 {
5235 	struct bonding *bond = netdev_priv(bond_dev);
5236 	struct slave *slave = NULL;
5237 	struct list_head *iter;
5238 	bool xmit_suc = false;
5239 	bool skb_used = false;
5240 
5241 	bond_for_each_slave_rcu(bond, slave, iter) {
5242 		struct sk_buff *skb2;
5243 
5244 		if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5245 			continue;
5246 
5247 		if (bond_is_last_slave(bond, slave)) {
5248 			skb2 = skb;
5249 			skb_used = true;
5250 		} else {
5251 			skb2 = skb_clone(skb, GFP_ATOMIC);
5252 			if (!skb2) {
5253 				net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5254 						    bond_dev->name, __func__);
5255 				continue;
5256 			}
5257 		}
5258 
5259 		if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5260 			xmit_suc = true;
5261 	}
5262 
5263 	if (!skb_used)
5264 		dev_kfree_skb_any(skb);
5265 
5266 	if (xmit_suc)
5267 		return NETDEV_TX_OK;
5268 
5269 	dev_core_stats_tx_dropped_inc(bond_dev);
5270 	return NET_XMIT_DROP;
5271 }
5272 
5273 /*------------------------- Device initialization ---------------------------*/
5274 
5275 /* Lookup the slave that corresponds to a qid */
5276 static inline int bond_slave_override(struct bonding *bond,
5277 				      struct sk_buff *skb)
5278 {
5279 	struct slave *slave = NULL;
5280 	struct list_head *iter;
5281 
5282 	if (!skb_rx_queue_recorded(skb))
5283 		return 1;
5284 
5285 	/* Find out if any slaves have the same mapping as this skb. */
5286 	bond_for_each_slave_rcu(bond, slave, iter) {
5287 		if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) {
5288 			if (bond_slave_is_up(slave) &&
5289 			    slave->link == BOND_LINK_UP) {
5290 				bond_dev_queue_xmit(bond, skb, slave->dev);
5291 				return 0;
5292 			}
5293 			/* If the slave isn't UP, use default transmit policy. */
5294 			break;
5295 		}
5296 	}
5297 
5298 	return 1;
5299 }
5300 
5301 
5302 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5303 			     struct net_device *sb_dev)
5304 {
5305 	/* This helper function exists to help dev_pick_tx get the correct
5306 	 * destination queue.  Using a helper function skips a call to
5307 	 * skb_tx_hash and will put the skbs in the queue we expect on their
5308 	 * way down to the bonding driver.
5309 	 */
5310 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5311 
5312 	/* Save the original txq to restore before passing to the driver */
5313 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5314 
5315 	if (unlikely(txq >= dev->real_num_tx_queues)) {
5316 		do {
5317 			txq -= dev->real_num_tx_queues;
5318 		} while (txq >= dev->real_num_tx_queues);
5319 	}
5320 	return txq;
5321 }
5322 
5323 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5324 					      struct sk_buff *skb,
5325 					      bool all_slaves)
5326 {
5327 	struct bonding *bond = netdev_priv(master_dev);
5328 	struct bond_up_slave *slaves;
5329 	struct slave *slave = NULL;
5330 
5331 	switch (BOND_MODE(bond)) {
5332 	case BOND_MODE_ROUNDROBIN:
5333 		slave = bond_xmit_roundrobin_slave_get(bond, skb);
5334 		break;
5335 	case BOND_MODE_ACTIVEBACKUP:
5336 		slave = bond_xmit_activebackup_slave_get(bond);
5337 		break;
5338 	case BOND_MODE_8023AD:
5339 	case BOND_MODE_XOR:
5340 		if (all_slaves)
5341 			slaves = rcu_dereference(bond->all_slaves);
5342 		else
5343 			slaves = rcu_dereference(bond->usable_slaves);
5344 		slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5345 		break;
5346 	case BOND_MODE_BROADCAST:
5347 		break;
5348 	case BOND_MODE_ALB:
5349 		slave = bond_xmit_alb_slave_get(bond, skb);
5350 		break;
5351 	case BOND_MODE_TLB:
5352 		slave = bond_xmit_tlb_slave_get(bond, skb);
5353 		break;
5354 	default:
5355 		/* Should never happen, mode already checked */
5356 		WARN_ONCE(true, "Unknown bonding mode");
5357 		break;
5358 	}
5359 
5360 	if (slave)
5361 		return slave->dev;
5362 	return NULL;
5363 }
5364 
5365 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5366 {
5367 	switch (sk->sk_family) {
5368 #if IS_ENABLED(CONFIG_IPV6)
5369 	case AF_INET6:
5370 		if (ipv6_only_sock(sk) ||
5371 		    ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5372 			flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5373 			flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5374 			flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5375 			break;
5376 		}
5377 		fallthrough;
5378 #endif
5379 	default: /* AF_INET */
5380 		flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5381 		flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5382 		flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5383 		break;
5384 	}
5385 
5386 	flow->ports.src = inet_sk(sk)->inet_sport;
5387 	flow->ports.dst = inet_sk(sk)->inet_dport;
5388 }
5389 
5390 /**
5391  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5392  * @sk: socket to use for headers
5393  *
5394  * This function will extract the necessary field from the socket and use
5395  * them to generate a hash based on the LAYER34 xmit_policy.
5396  * Assumes that sk is a TCP or UDP socket.
5397  */
5398 static u32 bond_sk_hash_l34(struct sock *sk)
5399 {
5400 	struct flow_keys flow;
5401 	u32 hash;
5402 
5403 	bond_sk_to_flow(sk, &flow);
5404 
5405 	/* L4 */
5406 	memcpy(&hash, &flow.ports.ports, sizeof(hash));
5407 	/* L3 */
5408 	return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5409 }
5410 
5411 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5412 						  struct sock *sk)
5413 {
5414 	struct bond_up_slave *slaves;
5415 	struct slave *slave;
5416 	unsigned int count;
5417 	u32 hash;
5418 
5419 	slaves = rcu_dereference(bond->usable_slaves);
5420 	count = slaves ? READ_ONCE(slaves->count) : 0;
5421 	if (unlikely(!count))
5422 		return NULL;
5423 
5424 	hash = bond_sk_hash_l34(sk);
5425 	slave = slaves->arr[hash % count];
5426 
5427 	return slave->dev;
5428 }
5429 
5430 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5431 						struct sock *sk)
5432 {
5433 	struct bonding *bond = netdev_priv(dev);
5434 	struct net_device *lower = NULL;
5435 
5436 	rcu_read_lock();
5437 	if (bond_sk_check(bond))
5438 		lower = __bond_sk_get_lower_dev(bond, sk);
5439 	rcu_read_unlock();
5440 
5441 	return lower;
5442 }
5443 
5444 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5445 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5446 					struct net_device *dev)
5447 {
5448 	struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5449 
5450 	/* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5451 	 * was true, if tls_device_down is running in parallel, but it's OK,
5452 	 * because bond_get_slave_by_dev has a NULL check.
5453 	 */
5454 	if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5455 		return bond_dev_queue_xmit(bond, skb, tls_netdev);
5456 	return bond_tx_drop(dev, skb);
5457 }
5458 #endif
5459 
5460 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5461 {
5462 	struct bonding *bond = netdev_priv(dev);
5463 
5464 	if (bond_should_override_tx_queue(bond) &&
5465 	    !bond_slave_override(bond, skb))
5466 		return NETDEV_TX_OK;
5467 
5468 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5469 	if (tls_is_skb_tx_device_offloaded(skb))
5470 		return bond_tls_device_xmit(bond, skb, dev);
5471 #endif
5472 
5473 	switch (BOND_MODE(bond)) {
5474 	case BOND_MODE_ROUNDROBIN:
5475 		return bond_xmit_roundrobin(skb, dev);
5476 	case BOND_MODE_ACTIVEBACKUP:
5477 		return bond_xmit_activebackup(skb, dev);
5478 	case BOND_MODE_8023AD:
5479 	case BOND_MODE_XOR:
5480 		return bond_3ad_xor_xmit(skb, dev);
5481 	case BOND_MODE_BROADCAST:
5482 		return bond_xmit_broadcast(skb, dev);
5483 	case BOND_MODE_ALB:
5484 		return bond_alb_xmit(skb, dev);
5485 	case BOND_MODE_TLB:
5486 		return bond_tlb_xmit(skb, dev);
5487 	default:
5488 		/* Should never happen, mode already checked */
5489 		netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5490 		WARN_ON_ONCE(1);
5491 		return bond_tx_drop(dev, skb);
5492 	}
5493 }
5494 
5495 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5496 {
5497 	struct bonding *bond = netdev_priv(dev);
5498 	netdev_tx_t ret = NETDEV_TX_OK;
5499 
5500 	/* If we risk deadlock from transmitting this in the
5501 	 * netpoll path, tell netpoll to queue the frame for later tx
5502 	 */
5503 	if (unlikely(is_netpoll_tx_blocked(dev)))
5504 		return NETDEV_TX_BUSY;
5505 
5506 	rcu_read_lock();
5507 	if (bond_has_slaves(bond))
5508 		ret = __bond_start_xmit(skb, dev);
5509 	else
5510 		ret = bond_tx_drop(dev, skb);
5511 	rcu_read_unlock();
5512 
5513 	return ret;
5514 }
5515 
5516 static struct net_device *
5517 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5518 {
5519 	struct bonding *bond = netdev_priv(bond_dev);
5520 	struct slave *slave;
5521 
5522 	/* Caller needs to hold rcu_read_lock() */
5523 
5524 	switch (BOND_MODE(bond)) {
5525 	case BOND_MODE_ROUNDROBIN:
5526 		slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5527 		break;
5528 
5529 	case BOND_MODE_ACTIVEBACKUP:
5530 		slave = bond_xmit_activebackup_slave_get(bond);
5531 		break;
5532 
5533 	case BOND_MODE_8023AD:
5534 	case BOND_MODE_XOR:
5535 		slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5536 		break;
5537 
5538 	default:
5539 		/* Should never happen. Mode guarded by bond_xdp_check() */
5540 		netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5541 		WARN_ON_ONCE(1);
5542 		return NULL;
5543 	}
5544 
5545 	if (slave)
5546 		return slave->dev;
5547 
5548 	return NULL;
5549 }
5550 
5551 static int bond_xdp_xmit(struct net_device *bond_dev,
5552 			 int n, struct xdp_frame **frames, u32 flags)
5553 {
5554 	int nxmit, err = -ENXIO;
5555 
5556 	rcu_read_lock();
5557 
5558 	for (nxmit = 0; nxmit < n; nxmit++) {
5559 		struct xdp_frame *frame = frames[nxmit];
5560 		struct xdp_frame *frames1[] = {frame};
5561 		struct net_device *slave_dev;
5562 		struct xdp_buff xdp;
5563 
5564 		xdp_convert_frame_to_buff(frame, &xdp);
5565 
5566 		slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5567 		if (!slave_dev) {
5568 			err = -ENXIO;
5569 			break;
5570 		}
5571 
5572 		err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5573 		if (err < 1)
5574 			break;
5575 	}
5576 
5577 	rcu_read_unlock();
5578 
5579 	/* If error happened on the first frame then we can pass the error up, otherwise
5580 	 * report the number of frames that were xmitted.
5581 	 */
5582 	if (err < 0)
5583 		return (nxmit == 0 ? err : nxmit);
5584 
5585 	return nxmit;
5586 }
5587 
5588 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5589 			struct netlink_ext_ack *extack)
5590 {
5591 	struct bonding *bond = netdev_priv(dev);
5592 	struct list_head *iter;
5593 	struct slave *slave, *rollback_slave;
5594 	struct bpf_prog *old_prog;
5595 	struct netdev_bpf xdp = {
5596 		.command = XDP_SETUP_PROG,
5597 		.flags   = 0,
5598 		.prog    = prog,
5599 		.extack  = extack,
5600 	};
5601 	int err;
5602 
5603 	ASSERT_RTNL();
5604 
5605 	if (!bond_xdp_check(bond))
5606 		return -EOPNOTSUPP;
5607 
5608 	old_prog = bond->xdp_prog;
5609 	bond->xdp_prog = prog;
5610 
5611 	bond_for_each_slave(bond, slave, iter) {
5612 		struct net_device *slave_dev = slave->dev;
5613 
5614 		if (!slave_dev->netdev_ops->ndo_bpf ||
5615 		    !slave_dev->netdev_ops->ndo_xdp_xmit) {
5616 			SLAVE_NL_ERR(dev, slave_dev, extack,
5617 				     "Slave device does not support XDP");
5618 			err = -EOPNOTSUPP;
5619 			goto err;
5620 		}
5621 
5622 		if (dev_xdp_prog_count(slave_dev) > 0) {
5623 			SLAVE_NL_ERR(dev, slave_dev, extack,
5624 				     "Slave has XDP program loaded, please unload before enslaving");
5625 			err = -EOPNOTSUPP;
5626 			goto err;
5627 		}
5628 
5629 		err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5630 		if (err < 0) {
5631 			/* ndo_bpf() sets extack error message */
5632 			slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5633 			goto err;
5634 		}
5635 		if (prog)
5636 			bpf_prog_inc(prog);
5637 	}
5638 
5639 	if (prog) {
5640 		static_branch_inc(&bpf_master_redirect_enabled_key);
5641 	} else if (old_prog) {
5642 		bpf_prog_put(old_prog);
5643 		static_branch_dec(&bpf_master_redirect_enabled_key);
5644 	}
5645 
5646 	return 0;
5647 
5648 err:
5649 	/* unwind the program changes */
5650 	bond->xdp_prog = old_prog;
5651 	xdp.prog = old_prog;
5652 	xdp.extack = NULL; /* do not overwrite original error */
5653 
5654 	bond_for_each_slave(bond, rollback_slave, iter) {
5655 		struct net_device *slave_dev = rollback_slave->dev;
5656 		int err_unwind;
5657 
5658 		if (slave == rollback_slave)
5659 			break;
5660 
5661 		err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5662 		if (err_unwind < 0)
5663 			slave_err(dev, slave_dev,
5664 				  "Error %d when unwinding XDP program change\n", err_unwind);
5665 		else if (xdp.prog)
5666 			bpf_prog_inc(xdp.prog);
5667 	}
5668 	return err;
5669 }
5670 
5671 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5672 {
5673 	switch (xdp->command) {
5674 	case XDP_SETUP_PROG:
5675 		return bond_xdp_set(dev, xdp->prog, xdp->extack);
5676 	default:
5677 		return -EINVAL;
5678 	}
5679 }
5680 
5681 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5682 {
5683 	if (speed == 0 || speed == SPEED_UNKNOWN)
5684 		speed = slave->speed;
5685 	else
5686 		speed = min(speed, slave->speed);
5687 
5688 	return speed;
5689 }
5690 
5691 /* Set the BOND_PHC_INDEX flag to notify user space */
5692 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5693 {
5694 	struct ifreq *ifr = kernel_cfg->ifr;
5695 	struct hwtstamp_config cfg;
5696 
5697 	if (kernel_cfg->copied_to_user) {
5698 		/* Lower device has a legacy implementation */
5699 		if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
5700 			return -EFAULT;
5701 
5702 		cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5703 		if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)))
5704 			return -EFAULT;
5705 	} else {
5706 		kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5707 	}
5708 
5709 	return 0;
5710 }
5711 
5712 static int bond_hwtstamp_get(struct net_device *dev,
5713 			     struct kernel_hwtstamp_config *cfg)
5714 {
5715 	struct bonding *bond = netdev_priv(dev);
5716 	struct net_device *real_dev;
5717 	int err;
5718 
5719 	real_dev = bond_option_active_slave_get_rcu(bond);
5720 	if (!real_dev)
5721 		return -EOPNOTSUPP;
5722 
5723 	err = generic_hwtstamp_get_lower(real_dev, cfg);
5724 	if (err)
5725 		return err;
5726 
5727 	return bond_set_phc_index_flag(cfg);
5728 }
5729 
5730 static int bond_hwtstamp_set(struct net_device *dev,
5731 			     struct kernel_hwtstamp_config *cfg,
5732 			     struct netlink_ext_ack *extack)
5733 {
5734 	struct bonding *bond = netdev_priv(dev);
5735 	struct net_device *real_dev;
5736 	int err;
5737 
5738 	if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5739 		return -EOPNOTSUPP;
5740 
5741 	real_dev = bond_option_active_slave_get_rcu(bond);
5742 	if (!real_dev)
5743 		return -EOPNOTSUPP;
5744 
5745 	err = generic_hwtstamp_set_lower(real_dev, cfg, extack);
5746 	if (err)
5747 		return err;
5748 
5749 	return bond_set_phc_index_flag(cfg);
5750 }
5751 
5752 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5753 					   struct ethtool_link_ksettings *cmd)
5754 {
5755 	struct bonding *bond = netdev_priv(bond_dev);
5756 	struct list_head *iter;
5757 	struct slave *slave;
5758 	u32 speed = 0;
5759 
5760 	cmd->base.duplex = DUPLEX_UNKNOWN;
5761 	cmd->base.port = PORT_OTHER;
5762 
5763 	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5764 	 * do not need to check mode.  Though link speed might not represent
5765 	 * the true receive or transmit bandwidth (not all modes are symmetric)
5766 	 * this is an accurate maximum.
5767 	 */
5768 	bond_for_each_slave(bond, slave, iter) {
5769 		if (bond_slave_can_tx(slave)) {
5770 			bond_update_speed_duplex(slave);
5771 			if (slave->speed != SPEED_UNKNOWN) {
5772 				if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5773 					speed = bond_mode_bcast_speed(slave,
5774 								      speed);
5775 				else
5776 					speed += slave->speed;
5777 			}
5778 			if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5779 			    slave->duplex != DUPLEX_UNKNOWN)
5780 				cmd->base.duplex = slave->duplex;
5781 		}
5782 	}
5783 	cmd->base.speed = speed ? : SPEED_UNKNOWN;
5784 
5785 	return 0;
5786 }
5787 
5788 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5789 				     struct ethtool_drvinfo *drvinfo)
5790 {
5791 	strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5792 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5793 		 BOND_ABI_VERSION);
5794 }
5795 
5796 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5797 				    struct kernel_ethtool_ts_info *info)
5798 {
5799 	struct bonding *bond = netdev_priv(bond_dev);
5800 	struct kernel_ethtool_ts_info ts_info;
5801 	struct net_device *real_dev;
5802 	bool sw_tx_support = false;
5803 	struct list_head *iter;
5804 	struct slave *slave;
5805 	int ret = 0;
5806 
5807 	rcu_read_lock();
5808 	real_dev = bond_option_active_slave_get_rcu(bond);
5809 	dev_hold(real_dev);
5810 	rcu_read_unlock();
5811 
5812 	if (real_dev) {
5813 		ret = ethtool_get_ts_info_by_layer(real_dev, info);
5814 	} else {
5815 		info->phc_index = -1;
5816 		info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5817 					SOF_TIMESTAMPING_SOFTWARE;
5818 		/* Check if all slaves support software tx timestamping */
5819 		rcu_read_lock();
5820 		bond_for_each_slave_rcu(bond, slave, iter) {
5821 			ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info);
5822 			if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5823 				sw_tx_support = true;
5824 				continue;
5825 			}
5826 
5827 			sw_tx_support = false;
5828 			break;
5829 		}
5830 		rcu_read_unlock();
5831 	}
5832 
5833 	if (sw_tx_support)
5834 		info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5835 
5836 	dev_put(real_dev);
5837 	return ret;
5838 }
5839 
5840 static const struct ethtool_ops bond_ethtool_ops = {
5841 	.get_drvinfo		= bond_ethtool_get_drvinfo,
5842 	.get_link		= ethtool_op_get_link,
5843 	.get_link_ksettings	= bond_ethtool_get_link_ksettings,
5844 	.get_ts_info		= bond_ethtool_get_ts_info,
5845 };
5846 
5847 static const struct net_device_ops bond_netdev_ops = {
5848 	.ndo_init		= bond_init,
5849 	.ndo_uninit		= bond_uninit,
5850 	.ndo_open		= bond_open,
5851 	.ndo_stop		= bond_close,
5852 	.ndo_start_xmit		= bond_start_xmit,
5853 	.ndo_select_queue	= bond_select_queue,
5854 	.ndo_get_stats64	= bond_get_stats,
5855 	.ndo_eth_ioctl		= bond_eth_ioctl,
5856 	.ndo_siocbond		= bond_do_ioctl,
5857 	.ndo_siocdevprivate	= bond_siocdevprivate,
5858 	.ndo_change_rx_flags	= bond_change_rx_flags,
5859 	.ndo_set_rx_mode	= bond_set_rx_mode,
5860 	.ndo_change_mtu		= bond_change_mtu,
5861 	.ndo_set_mac_address	= bond_set_mac_address,
5862 	.ndo_neigh_setup	= bond_neigh_setup,
5863 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
5864 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
5865 #ifdef CONFIG_NET_POLL_CONTROLLER
5866 	.ndo_netpoll_setup	= bond_netpoll_setup,
5867 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
5868 	.ndo_poll_controller	= bond_poll_controller,
5869 #endif
5870 	.ndo_add_slave		= bond_enslave,
5871 	.ndo_del_slave		= bond_release,
5872 	.ndo_fix_features	= bond_fix_features,
5873 	.ndo_features_check	= passthru_features_check,
5874 	.ndo_get_xmit_slave	= bond_xmit_get_slave,
5875 	.ndo_sk_get_lower_dev	= bond_sk_get_lower_dev,
5876 	.ndo_bpf		= bond_xdp,
5877 	.ndo_xdp_xmit           = bond_xdp_xmit,
5878 	.ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5879 	.ndo_hwtstamp_get	= bond_hwtstamp_get,
5880 	.ndo_hwtstamp_set	= bond_hwtstamp_set,
5881 };
5882 
5883 static const struct device_type bond_type = {
5884 	.name = "bond",
5885 };
5886 
5887 static void bond_destructor(struct net_device *bond_dev)
5888 {
5889 	struct bonding *bond = netdev_priv(bond_dev);
5890 
5891 	if (bond->wq)
5892 		destroy_workqueue(bond->wq);
5893 
5894 	free_percpu(bond->rr_tx_counter);
5895 }
5896 
5897 void bond_setup(struct net_device *bond_dev)
5898 {
5899 	struct bonding *bond = netdev_priv(bond_dev);
5900 
5901 	spin_lock_init(&bond->mode_lock);
5902 	bond->params = bonding_defaults;
5903 
5904 	/* Initialize pointers */
5905 	bond->dev = bond_dev;
5906 
5907 	/* Initialize the device entry points */
5908 	ether_setup(bond_dev);
5909 	bond_dev->max_mtu = ETH_MAX_MTU;
5910 	bond_dev->netdev_ops = &bond_netdev_ops;
5911 	bond_dev->ethtool_ops = &bond_ethtool_ops;
5912 
5913 	bond_dev->needs_free_netdev = true;
5914 	bond_dev->priv_destructor = bond_destructor;
5915 
5916 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5917 
5918 	/* Initialize the device options */
5919 	bond_dev->flags |= IFF_MASTER;
5920 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5921 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5922 
5923 #ifdef CONFIG_XFRM_OFFLOAD
5924 	/* set up xfrm device ops (only supported in active-backup right now) */
5925 	bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5926 	INIT_LIST_HEAD(&bond->ipsec_list);
5927 	mutex_init(&bond->ipsec_lock);
5928 #endif /* CONFIG_XFRM_OFFLOAD */
5929 
5930 	/* don't acquire bond device's netif_tx_lock when transmitting */
5931 	bond_dev->features |= NETIF_F_LLTX;
5932 
5933 	/* By default, we declare the bond to be fully
5934 	 * VLAN hardware accelerated capable. Special
5935 	 * care is taken in the various xmit functions
5936 	 * when there are slaves that are not hw accel
5937 	 * capable
5938 	 */
5939 
5940 	/* Don't allow bond devices to change network namespaces. */
5941 	bond_dev->features |= NETIF_F_NETNS_LOCAL;
5942 
5943 	bond_dev->hw_features = BOND_VLAN_FEATURES |
5944 				NETIF_F_HW_VLAN_CTAG_RX |
5945 				NETIF_F_HW_VLAN_CTAG_FILTER |
5946 				NETIF_F_HW_VLAN_STAG_RX |
5947 				NETIF_F_HW_VLAN_STAG_FILTER;
5948 
5949 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5950 	bond_dev->features |= bond_dev->hw_features;
5951 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5952 #ifdef CONFIG_XFRM_OFFLOAD
5953 	bond_dev->hw_features |= BOND_XFRM_FEATURES;
5954 	/* Only enable XFRM features if this is an active-backup config */
5955 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5956 		bond_dev->features |= BOND_XFRM_FEATURES;
5957 #endif /* CONFIG_XFRM_OFFLOAD */
5958 }
5959 
5960 /* Destroy a bonding device.
5961  * Must be under rtnl_lock when this function is called.
5962  */
5963 static void bond_uninit(struct net_device *bond_dev)
5964 {
5965 	struct bonding *bond = netdev_priv(bond_dev);
5966 	struct list_head *iter;
5967 	struct slave *slave;
5968 
5969 	bond_netpoll_cleanup(bond_dev);
5970 
5971 	/* Release the bonded slaves */
5972 	bond_for_each_slave(bond, slave, iter)
5973 		__bond_release_one(bond_dev, slave->dev, true, true);
5974 	netdev_info(bond_dev, "Released all slaves\n");
5975 
5976 #ifdef CONFIG_XFRM_OFFLOAD
5977 	mutex_destroy(&bond->ipsec_lock);
5978 #endif /* CONFIG_XFRM_OFFLOAD */
5979 
5980 	bond_set_slave_arr(bond, NULL, NULL);
5981 
5982 	list_del_rcu(&bond->bond_list);
5983 
5984 	bond_debug_unregister(bond);
5985 }
5986 
5987 /*------------------------- Module initialization ---------------------------*/
5988 
5989 static int __init bond_check_params(struct bond_params *params)
5990 {
5991 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5992 	struct bond_opt_value newval;
5993 	const struct bond_opt_value *valptr;
5994 	int arp_all_targets_value = 0;
5995 	u16 ad_actor_sys_prio = 0;
5996 	u16 ad_user_port_key = 0;
5997 	__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5998 	int arp_ip_count;
5999 	int bond_mode	= BOND_MODE_ROUNDROBIN;
6000 	int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
6001 	int lacp_fast = 0;
6002 	int tlb_dynamic_lb;
6003 
6004 	/* Convert string parameters. */
6005 	if (mode) {
6006 		bond_opt_initstr(&newval, mode);
6007 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
6008 		if (!valptr) {
6009 			pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
6010 			return -EINVAL;
6011 		}
6012 		bond_mode = valptr->value;
6013 	}
6014 
6015 	if (xmit_hash_policy) {
6016 		if (bond_mode == BOND_MODE_ROUNDROBIN ||
6017 		    bond_mode == BOND_MODE_ACTIVEBACKUP ||
6018 		    bond_mode == BOND_MODE_BROADCAST) {
6019 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
6020 				bond_mode_name(bond_mode));
6021 		} else {
6022 			bond_opt_initstr(&newval, xmit_hash_policy);
6023 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
6024 						&newval);
6025 			if (!valptr) {
6026 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
6027 				       xmit_hash_policy);
6028 				return -EINVAL;
6029 			}
6030 			xmit_hashtype = valptr->value;
6031 		}
6032 	}
6033 
6034 	if (lacp_rate) {
6035 		if (bond_mode != BOND_MODE_8023AD) {
6036 			pr_info("lacp_rate param is irrelevant in mode %s\n",
6037 				bond_mode_name(bond_mode));
6038 		} else {
6039 			bond_opt_initstr(&newval, lacp_rate);
6040 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
6041 						&newval);
6042 			if (!valptr) {
6043 				pr_err("Error: Invalid lacp rate \"%s\"\n",
6044 				       lacp_rate);
6045 				return -EINVAL;
6046 			}
6047 			lacp_fast = valptr->value;
6048 		}
6049 	}
6050 
6051 	if (ad_select) {
6052 		bond_opt_initstr(&newval, ad_select);
6053 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6054 					&newval);
6055 		if (!valptr) {
6056 			pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6057 			return -EINVAL;
6058 		}
6059 		params->ad_select = valptr->value;
6060 		if (bond_mode != BOND_MODE_8023AD)
6061 			pr_warn("ad_select param only affects 802.3ad mode\n");
6062 	} else {
6063 		params->ad_select = BOND_AD_STABLE;
6064 	}
6065 
6066 	if (max_bonds < 0) {
6067 		pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6068 			max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6069 		max_bonds = BOND_DEFAULT_MAX_BONDS;
6070 	}
6071 
6072 	if (miimon < 0) {
6073 		pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6074 			miimon, INT_MAX);
6075 		miimon = 0;
6076 	}
6077 
6078 	if (updelay < 0) {
6079 		pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6080 			updelay, INT_MAX);
6081 		updelay = 0;
6082 	}
6083 
6084 	if (downdelay < 0) {
6085 		pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6086 			downdelay, INT_MAX);
6087 		downdelay = 0;
6088 	}
6089 
6090 	if ((use_carrier != 0) && (use_carrier != 1)) {
6091 		pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6092 			use_carrier);
6093 		use_carrier = 1;
6094 	}
6095 
6096 	if (num_peer_notif < 0 || num_peer_notif > 255) {
6097 		pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6098 			num_peer_notif);
6099 		num_peer_notif = 1;
6100 	}
6101 
6102 	/* reset values for 802.3ad/TLB/ALB */
6103 	if (!bond_mode_uses_arp(bond_mode)) {
6104 		if (!miimon) {
6105 			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");
6106 			pr_warn("Forcing miimon to 100msec\n");
6107 			miimon = BOND_DEFAULT_MIIMON;
6108 		}
6109 	}
6110 
6111 	if (tx_queues < 1 || tx_queues > 255) {
6112 		pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6113 			tx_queues, BOND_DEFAULT_TX_QUEUES);
6114 		tx_queues = BOND_DEFAULT_TX_QUEUES;
6115 	}
6116 
6117 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6118 		pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6119 			all_slaves_active);
6120 		all_slaves_active = 0;
6121 	}
6122 
6123 	if (resend_igmp < 0 || resend_igmp > 255) {
6124 		pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6125 			resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6126 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6127 	}
6128 
6129 	bond_opt_initval(&newval, packets_per_slave);
6130 	if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6131 		pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6132 			packets_per_slave, USHRT_MAX);
6133 		packets_per_slave = 1;
6134 	}
6135 
6136 	if (bond_mode == BOND_MODE_ALB) {
6137 		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",
6138 			  updelay);
6139 	}
6140 
6141 	if (!miimon) {
6142 		if (updelay || downdelay) {
6143 			/* just warn the user the up/down delay will have
6144 			 * no effect since miimon is zero...
6145 			 */
6146 			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",
6147 				updelay, downdelay);
6148 		}
6149 	} else {
6150 		/* don't allow arp monitoring */
6151 		if (arp_interval) {
6152 			pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6153 				miimon, arp_interval);
6154 			arp_interval = 0;
6155 		}
6156 
6157 		if ((updelay % miimon) != 0) {
6158 			pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6159 				updelay, miimon, (updelay / miimon) * miimon);
6160 		}
6161 
6162 		updelay /= miimon;
6163 
6164 		if ((downdelay % miimon) != 0) {
6165 			pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6166 				downdelay, miimon,
6167 				(downdelay / miimon) * miimon);
6168 		}
6169 
6170 		downdelay /= miimon;
6171 	}
6172 
6173 	if (arp_interval < 0) {
6174 		pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6175 			arp_interval, INT_MAX);
6176 		arp_interval = 0;
6177 	}
6178 
6179 	for (arp_ip_count = 0, i = 0;
6180 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6181 		__be32 ip;
6182 
6183 		/* not a complete check, but good enough to catch mistakes */
6184 		if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6185 		    !bond_is_ip_target_ok(ip)) {
6186 			pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6187 				arp_ip_target[i]);
6188 			arp_interval = 0;
6189 		} else {
6190 			if (bond_get_targets_ip(arp_target, ip) == -1)
6191 				arp_target[arp_ip_count++] = ip;
6192 			else
6193 				pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6194 					&ip);
6195 		}
6196 	}
6197 
6198 	if (arp_interval && !arp_ip_count) {
6199 		/* don't allow arping if no arp_ip_target given... */
6200 		pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6201 			arp_interval);
6202 		arp_interval = 0;
6203 	}
6204 
6205 	if (arp_validate) {
6206 		if (!arp_interval) {
6207 			pr_err("arp_validate requires arp_interval\n");
6208 			return -EINVAL;
6209 		}
6210 
6211 		bond_opt_initstr(&newval, arp_validate);
6212 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6213 					&newval);
6214 		if (!valptr) {
6215 			pr_err("Error: invalid arp_validate \"%s\"\n",
6216 			       arp_validate);
6217 			return -EINVAL;
6218 		}
6219 		arp_validate_value = valptr->value;
6220 	} else {
6221 		arp_validate_value = 0;
6222 	}
6223 
6224 	if (arp_all_targets) {
6225 		bond_opt_initstr(&newval, arp_all_targets);
6226 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6227 					&newval);
6228 		if (!valptr) {
6229 			pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6230 			       arp_all_targets);
6231 			arp_all_targets_value = 0;
6232 		} else {
6233 			arp_all_targets_value = valptr->value;
6234 		}
6235 	}
6236 
6237 	if (miimon) {
6238 		pr_info("MII link monitoring set to %d ms\n", miimon);
6239 	} else if (arp_interval) {
6240 		valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6241 					  arp_validate_value);
6242 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6243 			arp_interval, valptr->string, arp_ip_count);
6244 
6245 		for (i = 0; i < arp_ip_count; i++)
6246 			pr_cont(" %s", arp_ip_target[i]);
6247 
6248 		pr_cont("\n");
6249 
6250 	} else if (max_bonds) {
6251 		/* miimon and arp_interval not set, we need one so things
6252 		 * work as expected, see bonding.txt for details
6253 		 */
6254 		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");
6255 	}
6256 
6257 	if (primary && !bond_mode_uses_primary(bond_mode)) {
6258 		/* currently, using a primary only makes sense
6259 		 * in active backup, TLB or ALB modes
6260 		 */
6261 		pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6262 			primary, bond_mode_name(bond_mode));
6263 		primary = NULL;
6264 	}
6265 
6266 	if (primary && primary_reselect) {
6267 		bond_opt_initstr(&newval, primary_reselect);
6268 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6269 					&newval);
6270 		if (!valptr) {
6271 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
6272 			       primary_reselect);
6273 			return -EINVAL;
6274 		}
6275 		primary_reselect_value = valptr->value;
6276 	} else {
6277 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6278 	}
6279 
6280 	if (fail_over_mac) {
6281 		bond_opt_initstr(&newval, fail_over_mac);
6282 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6283 					&newval);
6284 		if (!valptr) {
6285 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
6286 			       fail_over_mac);
6287 			return -EINVAL;
6288 		}
6289 		fail_over_mac_value = valptr->value;
6290 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6291 			pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6292 	} else {
6293 		fail_over_mac_value = BOND_FOM_NONE;
6294 	}
6295 
6296 	bond_opt_initstr(&newval, "default");
6297 	valptr = bond_opt_parse(
6298 			bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6299 				     &newval);
6300 	if (!valptr) {
6301 		pr_err("Error: No ad_actor_sys_prio default value");
6302 		return -EINVAL;
6303 	}
6304 	ad_actor_sys_prio = valptr->value;
6305 
6306 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6307 				&newval);
6308 	if (!valptr) {
6309 		pr_err("Error: No ad_user_port_key default value");
6310 		return -EINVAL;
6311 	}
6312 	ad_user_port_key = valptr->value;
6313 
6314 	bond_opt_initstr(&newval, "default");
6315 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6316 	if (!valptr) {
6317 		pr_err("Error: No tlb_dynamic_lb default value");
6318 		return -EINVAL;
6319 	}
6320 	tlb_dynamic_lb = valptr->value;
6321 
6322 	if (lp_interval == 0) {
6323 		pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6324 			INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6325 		lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6326 	}
6327 
6328 	/* fill params struct with the proper values */
6329 	params->mode = bond_mode;
6330 	params->xmit_policy = xmit_hashtype;
6331 	params->miimon = miimon;
6332 	params->num_peer_notif = num_peer_notif;
6333 	params->arp_interval = arp_interval;
6334 	params->arp_validate = arp_validate_value;
6335 	params->arp_all_targets = arp_all_targets_value;
6336 	params->missed_max = 2;
6337 	params->updelay = updelay;
6338 	params->downdelay = downdelay;
6339 	params->peer_notif_delay = 0;
6340 	params->use_carrier = use_carrier;
6341 	params->lacp_active = 1;
6342 	params->lacp_fast = lacp_fast;
6343 	params->primary[0] = 0;
6344 	params->primary_reselect = primary_reselect_value;
6345 	params->fail_over_mac = fail_over_mac_value;
6346 	params->tx_queues = tx_queues;
6347 	params->all_slaves_active = all_slaves_active;
6348 	params->resend_igmp = resend_igmp;
6349 	params->min_links = min_links;
6350 	params->lp_interval = lp_interval;
6351 	params->packets_per_slave = packets_per_slave;
6352 	params->tlb_dynamic_lb = tlb_dynamic_lb;
6353 	params->ad_actor_sys_prio = ad_actor_sys_prio;
6354 	eth_zero_addr(params->ad_actor_system);
6355 	params->ad_user_port_key = ad_user_port_key;
6356 	params->coupled_control = 1;
6357 	if (packets_per_slave > 0) {
6358 		params->reciprocal_packets_per_slave =
6359 			reciprocal_value(packets_per_slave);
6360 	} else {
6361 		/* reciprocal_packets_per_slave is unused if
6362 		 * packets_per_slave is 0 or 1, just initialize it
6363 		 */
6364 		params->reciprocal_packets_per_slave =
6365 			(struct reciprocal_value) { 0 };
6366 	}
6367 
6368 	if (primary)
6369 		strscpy_pad(params->primary, primary, sizeof(params->primary));
6370 
6371 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6372 #if IS_ENABLED(CONFIG_IPV6)
6373 	memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6374 #endif
6375 
6376 	return 0;
6377 }
6378 
6379 /* Called from registration process */
6380 static int bond_init(struct net_device *bond_dev)
6381 {
6382 	struct bonding *bond = netdev_priv(bond_dev);
6383 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6384 
6385 	netdev_dbg(bond_dev, "Begin bond_init\n");
6386 
6387 	bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6388 	if (!bond->wq)
6389 		return -ENOMEM;
6390 
6391 	bond->notifier_ctx = false;
6392 
6393 	spin_lock_init(&bond->stats_lock);
6394 	netdev_lockdep_set_classes(bond_dev);
6395 
6396 	list_add_tail_rcu(&bond->bond_list, &bn->dev_list);
6397 
6398 	bond_prepare_sysfs_group(bond);
6399 
6400 	bond_debug_register(bond);
6401 
6402 	/* Ensure valid dev_addr */
6403 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
6404 	    bond_dev->addr_assign_type == NET_ADDR_PERM)
6405 		eth_hw_addr_random(bond_dev);
6406 
6407 	return 0;
6408 }
6409 
6410 unsigned int bond_get_num_tx_queues(void)
6411 {
6412 	return tx_queues;
6413 }
6414 
6415 /* Create a new bond based on the specified name and bonding parameters.
6416  * If name is NULL, obtain a suitable "bond%d" name for us.
6417  * Caller must NOT hold rtnl_lock; we need to release it here before we
6418  * set up our sysfs entries.
6419  */
6420 int bond_create(struct net *net, const char *name)
6421 {
6422 	struct net_device *bond_dev;
6423 	struct bonding *bond;
6424 	int res = -ENOMEM;
6425 
6426 	rtnl_lock();
6427 
6428 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6429 				   name ? name : "bond%d", NET_NAME_UNKNOWN,
6430 				   bond_setup, tx_queues);
6431 	if (!bond_dev)
6432 		goto out;
6433 
6434 	bond = netdev_priv(bond_dev);
6435 	dev_net_set(bond_dev, net);
6436 	bond_dev->rtnl_link_ops = &bond_link_ops;
6437 
6438 	res = register_netdevice(bond_dev);
6439 	if (res < 0) {
6440 		free_netdev(bond_dev);
6441 		goto out;
6442 	}
6443 
6444 	netif_carrier_off(bond_dev);
6445 
6446 	bond_work_init_all(bond);
6447 
6448 out:
6449 	rtnl_unlock();
6450 	return res;
6451 }
6452 
6453 static int __net_init bond_net_init(struct net *net)
6454 {
6455 	struct bond_net *bn = net_generic(net, bond_net_id);
6456 
6457 	bn->net = net;
6458 	INIT_LIST_HEAD(&bn->dev_list);
6459 
6460 	bond_create_proc_dir(bn);
6461 	bond_create_sysfs(bn);
6462 
6463 	return 0;
6464 }
6465 
6466 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters
6467  * race condition in bond unloading") we need to remove sysfs files
6468  * before we remove our devices (done later in bond_net_exit_batch_rtnl())
6469  */
6470 static void __net_exit bond_net_pre_exit(struct net *net)
6471 {
6472 	struct bond_net *bn = net_generic(net, bond_net_id);
6473 
6474 	bond_destroy_sysfs(bn);
6475 }
6476 
6477 static void __net_exit bond_net_exit_batch_rtnl(struct list_head *net_list,
6478 						struct list_head *dev_kill_list)
6479 {
6480 	struct bond_net *bn;
6481 	struct net *net;
6482 
6483 	/* Kill off any bonds created after unregistering bond rtnl ops */
6484 	list_for_each_entry(net, net_list, exit_list) {
6485 		struct bonding *bond, *tmp_bond;
6486 
6487 		bn = net_generic(net, bond_net_id);
6488 		list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6489 			unregister_netdevice_queue(bond->dev, dev_kill_list);
6490 	}
6491 }
6492 
6493 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory
6494  * only after all bonds are gone") bond_destroy_proc_dir() is called
6495  * after bond_net_exit_batch_rtnl() has completed.
6496  */
6497 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6498 {
6499 	struct bond_net *bn;
6500 	struct net *net;
6501 
6502 	list_for_each_entry(net, net_list, exit_list) {
6503 		bn = net_generic(net, bond_net_id);
6504 		bond_destroy_proc_dir(bn);
6505 	}
6506 }
6507 
6508 static struct pernet_operations bond_net_ops = {
6509 	.init = bond_net_init,
6510 	.pre_exit = bond_net_pre_exit,
6511 	.exit_batch_rtnl = bond_net_exit_batch_rtnl,
6512 	.exit_batch = bond_net_exit_batch,
6513 	.id   = &bond_net_id,
6514 	.size = sizeof(struct bond_net),
6515 };
6516 
6517 static int __init bonding_init(void)
6518 {
6519 	int i;
6520 	int res;
6521 
6522 	res = bond_check_params(&bonding_defaults);
6523 	if (res)
6524 		goto out;
6525 
6526 	bond_create_debugfs();
6527 
6528 	res = register_pernet_subsys(&bond_net_ops);
6529 	if (res)
6530 		goto err_net_ops;
6531 
6532 	res = bond_netlink_init();
6533 	if (res)
6534 		goto err_link;
6535 
6536 	for (i = 0; i < max_bonds; i++) {
6537 		res = bond_create(&init_net, NULL);
6538 		if (res)
6539 			goto err;
6540 	}
6541 
6542 	skb_flow_dissector_init(&flow_keys_bonding,
6543 				flow_keys_bonding_keys,
6544 				ARRAY_SIZE(flow_keys_bonding_keys));
6545 
6546 	register_netdevice_notifier(&bond_netdev_notifier);
6547 out:
6548 	return res;
6549 err:
6550 	bond_netlink_fini();
6551 err_link:
6552 	unregister_pernet_subsys(&bond_net_ops);
6553 err_net_ops:
6554 	bond_destroy_debugfs();
6555 	goto out;
6556 
6557 }
6558 
6559 static void __exit bonding_exit(void)
6560 {
6561 	unregister_netdevice_notifier(&bond_netdev_notifier);
6562 
6563 	bond_netlink_fini();
6564 	unregister_pernet_subsys(&bond_net_ops);
6565 
6566 	bond_destroy_debugfs();
6567 
6568 #ifdef CONFIG_NET_POLL_CONTROLLER
6569 	/* Make sure we don't have an imbalance on our netpoll blocking */
6570 	WARN_ON(atomic_read(&netpoll_block_tx));
6571 #endif
6572 }
6573 
6574 module_init(bonding_init);
6575 module_exit(bonding_exit);
6576 MODULE_LICENSE("GPL");
6577 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6578 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
6579