xref: /linux/drivers/net/bonding/bond_main.c (revision b803c4a4f78834b31ebfbbcea350473333760559)
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/netdev_lock.h>
94 #include <net/xdp.h>
95 
96 #include "bonding_priv.h"
97 
98 /*---------------------------- Module parameters ----------------------------*/
99 
100 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
101 
102 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
103 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
104 static int num_peer_notif = 1;
105 static int miimon;
106 static int updelay;
107 static int downdelay;
108 static int use_carrier	= 1;
109 static char *mode;
110 static char *primary;
111 static char *primary_reselect;
112 static char *lacp_rate;
113 static int min_links;
114 static char *ad_select;
115 static char *xmit_hash_policy;
116 static int arp_interval;
117 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
118 static char *arp_validate;
119 static char *arp_all_targets;
120 static char *fail_over_mac;
121 static int all_slaves_active;
122 static struct bond_params bonding_defaults;
123 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
124 static int packets_per_slave = 1;
125 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
126 
127 module_param(max_bonds, int, 0);
128 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
129 module_param(tx_queues, int, 0);
130 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
131 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
132 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
133 			       "failover event (alias of num_unsol_na)");
134 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
135 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
136 			       "failover event (alias of num_grat_arp)");
137 module_param(miimon, int, 0);
138 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
139 module_param(updelay, int, 0);
140 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
141 module_param(downdelay, int, 0);
142 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
143 			    "in milliseconds");
144 module_param(use_carrier, int, 0);
145 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
146 			      "0 for off, 1 for on (default)");
147 module_param(mode, charp, 0);
148 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
149 		       "1 for active-backup, 2 for balance-xor, "
150 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
151 		       "6 for balance-alb");
152 module_param(primary, charp, 0);
153 MODULE_PARM_DESC(primary, "Primary network device to use");
154 module_param(primary_reselect, charp, 0);
155 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
156 				   "once it comes up; "
157 				   "0 for always (default), "
158 				   "1 for only if speed of primary is "
159 				   "better, "
160 				   "2 for only on active slave "
161 				   "failure");
162 module_param(lacp_rate, charp, 0);
163 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
164 			    "0 for slow, 1 for fast");
165 module_param(ad_select, charp, 0);
166 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
167 			    "0 for stable (default), 1 for bandwidth, "
168 			    "2 for count");
169 module_param(min_links, int, 0);
170 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
171 
172 module_param(xmit_hash_policy, charp, 0);
173 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
174 				   "0 for layer 2 (default), 1 for layer 3+4, "
175 				   "2 for layer 2+3, 3 for encap layer 2+3, "
176 				   "4 for encap layer 3+4, 5 for vlan+srcmac");
177 module_param(arp_interval, int, 0);
178 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
179 module_param_array(arp_ip_target, charp, NULL, 0);
180 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
181 module_param(arp_validate, charp, 0);
182 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
183 			       "0 for none (default), 1 for active, "
184 			       "2 for backup, 3 for all");
185 module_param(arp_all_targets, charp, 0);
186 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
187 module_param(fail_over_mac, charp, 0);
188 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
189 				"the same MAC; 0 for none (default), "
190 				"1 for active, 2 for follow");
191 module_param(all_slaves_active, int, 0);
192 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
193 				     "by setting active flag for all slaves; "
194 				     "0 for never (default), 1 for always.");
195 module_param(resend_igmp, int, 0);
196 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
197 			      "link failure");
198 module_param(packets_per_slave, int, 0);
199 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
200 				    "mode; 0 for a random slave, 1 packet per "
201 				    "slave (default), >1 packets per slave.");
202 module_param(lp_interval, uint, 0);
203 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
204 			      "the bonding driver sends learning packets to "
205 			      "each slaves peer switch. The default is 1.");
206 
207 /*----------------------------- Global variables ----------------------------*/
208 
209 #ifdef CONFIG_NET_POLL_CONTROLLER
210 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
211 #endif
212 
213 unsigned int bond_net_id __read_mostly;
214 
215 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
216 	{
217 		.key_id = FLOW_DISSECTOR_KEY_CONTROL,
218 		.offset = offsetof(struct flow_keys, control),
219 	},
220 	{
221 		.key_id = FLOW_DISSECTOR_KEY_BASIC,
222 		.offset = offsetof(struct flow_keys, basic),
223 	},
224 	{
225 		.key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
226 		.offset = offsetof(struct flow_keys, addrs.v4addrs),
227 	},
228 	{
229 		.key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
230 		.offset = offsetof(struct flow_keys, addrs.v6addrs),
231 	},
232 	{
233 		.key_id = FLOW_DISSECTOR_KEY_TIPC,
234 		.offset = offsetof(struct flow_keys, addrs.tipckey),
235 	},
236 	{
237 		.key_id = FLOW_DISSECTOR_KEY_PORTS,
238 		.offset = offsetof(struct flow_keys, ports),
239 	},
240 	{
241 		.key_id = FLOW_DISSECTOR_KEY_ICMP,
242 		.offset = offsetof(struct flow_keys, icmp),
243 	},
244 	{
245 		.key_id = FLOW_DISSECTOR_KEY_VLAN,
246 		.offset = offsetof(struct flow_keys, vlan),
247 	},
248 	{
249 		.key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
250 		.offset = offsetof(struct flow_keys, tags),
251 	},
252 	{
253 		.key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
254 		.offset = offsetof(struct flow_keys, keyid),
255 	},
256 };
257 
258 static struct flow_dissector flow_keys_bonding __read_mostly;
259 
260 /*-------------------------- Forward declarations ---------------------------*/
261 
262 static int bond_init(struct net_device *bond_dev);
263 static void bond_uninit(struct net_device *bond_dev);
264 static void bond_get_stats(struct net_device *bond_dev,
265 			   struct rtnl_link_stats64 *stats);
266 static void bond_slave_arr_handler(struct work_struct *work);
267 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
268 				  int mod);
269 static void bond_netdev_notify_work(struct work_struct *work);
270 
271 /*---------------------------- General routines -----------------------------*/
272 
273 const char *bond_mode_name(int mode)
274 {
275 	static const char *names[] = {
276 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
277 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
278 		[BOND_MODE_XOR] = "load balancing (xor)",
279 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
280 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
281 		[BOND_MODE_TLB] = "transmit load balancing",
282 		[BOND_MODE_ALB] = "adaptive load balancing",
283 	};
284 
285 	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
286 		return "unknown";
287 
288 	return names[mode];
289 }
290 
291 /**
292  * bond_dev_queue_xmit - Prepare skb for xmit.
293  *
294  * @bond: bond device that got this skb for tx.
295  * @skb: hw accel VLAN tagged skb to transmit
296  * @slave_dev: slave that is supposed to xmit this skbuff
297  */
298 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
299 			struct net_device *slave_dev)
300 {
301 	skb->dev = slave_dev;
302 
303 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
304 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
305 	skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
306 
307 	if (unlikely(netpoll_tx_running(bond->dev)))
308 		return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
309 
310 	return dev_queue_xmit(skb);
311 }
312 
313 static bool bond_sk_check(struct bonding *bond)
314 {
315 	switch (BOND_MODE(bond)) {
316 	case BOND_MODE_8023AD:
317 	case BOND_MODE_XOR:
318 		if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
319 			return true;
320 		fallthrough;
321 	default:
322 		return false;
323 	}
324 }
325 
326 bool bond_xdp_check(struct bonding *bond, int mode)
327 {
328 	switch (mode) {
329 	case BOND_MODE_ROUNDROBIN:
330 	case BOND_MODE_ACTIVEBACKUP:
331 		return true;
332 	case BOND_MODE_8023AD:
333 	case BOND_MODE_XOR:
334 		/* vlan+srcmac is not supported with XDP as in most cases the 802.1q
335 		 * payload is not in the packet due to hardware offload.
336 		 */
337 		if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
338 			return true;
339 		fallthrough;
340 	default:
341 		return false;
342 	}
343 }
344 
345 /*---------------------------------- VLAN -----------------------------------*/
346 
347 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
348  * We don't protect the slave list iteration with a lock because:
349  * a. This operation is performed in IOCTL context,
350  * b. The operation is protected by the RTNL semaphore in the 8021q code,
351  * c. Holding a lock with BH disabled while directly calling a base driver
352  *    entry point is generally a BAD idea.
353  *
354  * The design of synchronization/protection for this operation in the 8021q
355  * module is good for one or more VLAN devices over a single physical device
356  * and cannot be extended for a teaming solution like bonding, so there is a
357  * potential race condition here where a net device from the vlan group might
358  * be referenced (either by a base driver or the 8021q code) while it is being
359  * removed from the system. However, it turns out we're not making matters
360  * worse, and if it works for regular VLAN usage it will work here too.
361 */
362 
363 /**
364  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
365  * @bond_dev: bonding net device that got called
366  * @proto: network protocol ID
367  * @vid: vlan id being added
368  */
369 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
370 				__be16 proto, u16 vid)
371 {
372 	struct bonding *bond = netdev_priv(bond_dev);
373 	struct slave *slave, *rollback_slave;
374 	struct list_head *iter;
375 	int res;
376 
377 	bond_for_each_slave(bond, slave, iter) {
378 		res = vlan_vid_add(slave->dev, proto, vid);
379 		if (res)
380 			goto unwind;
381 	}
382 
383 	return 0;
384 
385 unwind:
386 	/* unwind to the slave that failed */
387 	bond_for_each_slave(bond, rollback_slave, iter) {
388 		if (rollback_slave == slave)
389 			break;
390 
391 		vlan_vid_del(rollback_slave->dev, proto, vid);
392 	}
393 
394 	return res;
395 }
396 
397 /**
398  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
399  * @bond_dev: bonding net device that got called
400  * @proto: network protocol ID
401  * @vid: vlan id being removed
402  */
403 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
404 				 __be16 proto, u16 vid)
405 {
406 	struct bonding *bond = netdev_priv(bond_dev);
407 	struct list_head *iter;
408 	struct slave *slave;
409 
410 	bond_for_each_slave(bond, slave, iter)
411 		vlan_vid_del(slave->dev, proto, vid);
412 
413 	if (bond_is_lb(bond))
414 		bond_alb_clear_vlan(bond, vid);
415 
416 	return 0;
417 }
418 
419 /*---------------------------------- XFRM -----------------------------------*/
420 
421 #ifdef CONFIG_XFRM_OFFLOAD
422 /**
423  * bond_ipsec_dev - Get active device for IPsec offload
424  * @xs: pointer to transformer state struct
425  *
426  * Context: caller must hold rcu_read_lock.
427  *
428  * Return: the device for ipsec offload, or NULL if not exist.
429  **/
430 static struct net_device *bond_ipsec_dev(struct xfrm_state *xs)
431 {
432 	struct net_device *bond_dev = xs->xso.dev;
433 	struct bonding *bond;
434 	struct slave *slave;
435 
436 	bond = netdev_priv(bond_dev);
437 	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
438 		return NULL;
439 
440 	slave = rcu_dereference(bond->curr_active_slave);
441 	if (!slave)
442 		return NULL;
443 
444 	if (!xs->xso.real_dev)
445 		return NULL;
446 
447 	if (xs->xso.real_dev != slave->dev)
448 		pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n",
449 				    bond_dev->name, slave->dev->name, xs->xso.real_dev->name);
450 
451 	return slave->dev;
452 }
453 
454 /**
455  * bond_ipsec_add_sa - program device with a security association
456  * @xs: pointer to transformer state struct
457  * @extack: extack point to fill failure reason
458  **/
459 static int bond_ipsec_add_sa(struct xfrm_state *xs,
460 			     struct netlink_ext_ack *extack)
461 {
462 	struct net_device *bond_dev = xs->xso.dev;
463 	struct net_device *real_dev;
464 	netdevice_tracker tracker;
465 	struct bond_ipsec *ipsec;
466 	struct bonding *bond;
467 	struct slave *slave;
468 	int err;
469 
470 	if (!bond_dev)
471 		return -EINVAL;
472 
473 	rcu_read_lock();
474 	bond = netdev_priv(bond_dev);
475 	slave = rcu_dereference(bond->curr_active_slave);
476 	real_dev = slave ? slave->dev : NULL;
477 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
478 	rcu_read_unlock();
479 	if (!real_dev) {
480 		err = -ENODEV;
481 		goto out;
482 	}
483 
484 	if (!real_dev->xfrmdev_ops ||
485 	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
486 	    netif_is_bond_master(real_dev)) {
487 		NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
488 		err = -EINVAL;
489 		goto out;
490 	}
491 
492 	ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL);
493 	if (!ipsec) {
494 		err = -ENOMEM;
495 		goto out;
496 	}
497 
498 	xs->xso.real_dev = real_dev;
499 	err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
500 	if (!err) {
501 		ipsec->xs = xs;
502 		INIT_LIST_HEAD(&ipsec->list);
503 		mutex_lock(&bond->ipsec_lock);
504 		list_add(&ipsec->list, &bond->ipsec_list);
505 		mutex_unlock(&bond->ipsec_lock);
506 	} else {
507 		kfree(ipsec);
508 	}
509 out:
510 	netdev_put(real_dev, &tracker);
511 	return err;
512 }
513 
514 static void bond_ipsec_add_sa_all(struct bonding *bond)
515 {
516 	struct net_device *bond_dev = bond->dev;
517 	struct net_device *real_dev;
518 	struct bond_ipsec *ipsec;
519 	struct slave *slave;
520 
521 	slave = rtnl_dereference(bond->curr_active_slave);
522 	real_dev = slave ? slave->dev : NULL;
523 	if (!real_dev)
524 		return;
525 
526 	mutex_lock(&bond->ipsec_lock);
527 	if (!real_dev->xfrmdev_ops ||
528 	    !real_dev->xfrmdev_ops->xdo_dev_state_add ||
529 	    netif_is_bond_master(real_dev)) {
530 		if (!list_empty(&bond->ipsec_list))
531 			slave_warn(bond_dev, real_dev,
532 				   "%s: no slave xdo_dev_state_add\n",
533 				   __func__);
534 		goto out;
535 	}
536 
537 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
538 		/* If new state is added before ipsec_lock acquired */
539 		if (ipsec->xs->xso.real_dev == real_dev)
540 			continue;
541 
542 		ipsec->xs->xso.real_dev = real_dev;
543 		if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
544 			slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__);
545 			ipsec->xs->xso.real_dev = NULL;
546 		}
547 	}
548 out:
549 	mutex_unlock(&bond->ipsec_lock);
550 }
551 
552 /**
553  * bond_ipsec_del_sa - clear out this specific SA
554  * @xs: pointer to transformer state struct
555  **/
556 static void bond_ipsec_del_sa(struct xfrm_state *xs)
557 {
558 	struct net_device *bond_dev = xs->xso.dev;
559 	struct net_device *real_dev;
560 	netdevice_tracker tracker;
561 	struct bond_ipsec *ipsec;
562 	struct bonding *bond;
563 	struct slave *slave;
564 
565 	if (!bond_dev)
566 		return;
567 
568 	rcu_read_lock();
569 	bond = netdev_priv(bond_dev);
570 	slave = rcu_dereference(bond->curr_active_slave);
571 	real_dev = slave ? slave->dev : NULL;
572 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
573 	rcu_read_unlock();
574 
575 	if (!slave)
576 		goto out;
577 
578 	if (!xs->xso.real_dev)
579 		goto out;
580 
581 	WARN_ON(xs->xso.real_dev != real_dev);
582 
583 	if (!real_dev->xfrmdev_ops ||
584 	    !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
585 	    netif_is_bond_master(real_dev)) {
586 		slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__);
587 		goto out;
588 	}
589 
590 	real_dev->xfrmdev_ops->xdo_dev_state_delete(xs);
591 out:
592 	netdev_put(real_dev, &tracker);
593 	mutex_lock(&bond->ipsec_lock);
594 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
595 		if (ipsec->xs == xs) {
596 			list_del(&ipsec->list);
597 			kfree(ipsec);
598 			break;
599 		}
600 	}
601 	mutex_unlock(&bond->ipsec_lock);
602 }
603 
604 static void bond_ipsec_del_sa_all(struct bonding *bond)
605 {
606 	struct net_device *bond_dev = bond->dev;
607 	struct net_device *real_dev;
608 	struct bond_ipsec *ipsec;
609 	struct slave *slave;
610 
611 	slave = rtnl_dereference(bond->curr_active_slave);
612 	real_dev = slave ? slave->dev : NULL;
613 	if (!real_dev)
614 		return;
615 
616 	mutex_lock(&bond->ipsec_lock);
617 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
618 		if (!ipsec->xs->xso.real_dev)
619 			continue;
620 
621 		if (!real_dev->xfrmdev_ops ||
622 		    !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
623 		    netif_is_bond_master(real_dev)) {
624 			slave_warn(bond_dev, real_dev,
625 				   "%s: no slave xdo_dev_state_delete\n",
626 				   __func__);
627 		} else {
628 			real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
629 			if (real_dev->xfrmdev_ops->xdo_dev_state_free)
630 				real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs);
631 		}
632 	}
633 	mutex_unlock(&bond->ipsec_lock);
634 }
635 
636 static void bond_ipsec_free_sa(struct xfrm_state *xs)
637 {
638 	struct net_device *bond_dev = xs->xso.dev;
639 	struct net_device *real_dev;
640 	netdevice_tracker tracker;
641 	struct bonding *bond;
642 	struct slave *slave;
643 
644 	if (!bond_dev)
645 		return;
646 
647 	rcu_read_lock();
648 	bond = netdev_priv(bond_dev);
649 	slave = rcu_dereference(bond->curr_active_slave);
650 	real_dev = slave ? slave->dev : NULL;
651 	netdev_hold(real_dev, &tracker, GFP_ATOMIC);
652 	rcu_read_unlock();
653 
654 	if (!slave)
655 		goto out;
656 
657 	if (!xs->xso.real_dev)
658 		goto out;
659 
660 	WARN_ON(xs->xso.real_dev != real_dev);
661 
662 	if (real_dev && real_dev->xfrmdev_ops &&
663 	    real_dev->xfrmdev_ops->xdo_dev_state_free)
664 		real_dev->xfrmdev_ops->xdo_dev_state_free(xs);
665 out:
666 	netdev_put(real_dev, &tracker);
667 }
668 
669 /**
670  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
671  * @skb: current data packet
672  * @xs: pointer to transformer state struct
673  **/
674 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
675 {
676 	struct net_device *real_dev;
677 
678 	rcu_read_lock();
679 	real_dev = bond_ipsec_dev(xs);
680 	if (!real_dev || netif_is_bond_master(real_dev)) {
681 		rcu_read_unlock();
682 		return false;
683 	}
684 
685 	rcu_read_unlock();
686 	return true;
687 }
688 
689 /**
690  * bond_advance_esn_state - ESN support for IPSec HW offload
691  * @xs: pointer to transformer state struct
692  **/
693 static void bond_advance_esn_state(struct xfrm_state *xs)
694 {
695 	struct net_device *real_dev;
696 
697 	rcu_read_lock();
698 	real_dev = bond_ipsec_dev(xs);
699 	if (!real_dev)
700 		goto out;
701 
702 	if (!real_dev->xfrmdev_ops ||
703 	    !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
704 		pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name);
705 		goto out;
706 	}
707 
708 	real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs);
709 out:
710 	rcu_read_unlock();
711 }
712 
713 /**
714  * bond_xfrm_update_stats - Update xfrm state
715  * @xs: pointer to transformer state struct
716  **/
717 static void bond_xfrm_update_stats(struct xfrm_state *xs)
718 {
719 	struct net_device *real_dev;
720 
721 	rcu_read_lock();
722 	real_dev = bond_ipsec_dev(xs);
723 	if (!real_dev)
724 		goto out;
725 
726 	if (!real_dev->xfrmdev_ops ||
727 	    !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) {
728 		pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name);
729 		goto out;
730 	}
731 
732 	real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs);
733 out:
734 	rcu_read_unlock();
735 }
736 
737 static const struct xfrmdev_ops bond_xfrmdev_ops = {
738 	.xdo_dev_state_add = bond_ipsec_add_sa,
739 	.xdo_dev_state_delete = bond_ipsec_del_sa,
740 	.xdo_dev_state_free = bond_ipsec_free_sa,
741 	.xdo_dev_offload_ok = bond_ipsec_offload_ok,
742 	.xdo_dev_state_advance_esn = bond_advance_esn_state,
743 	.xdo_dev_state_update_stats = bond_xfrm_update_stats,
744 };
745 #endif /* CONFIG_XFRM_OFFLOAD */
746 
747 /*------------------------------- Link status -------------------------------*/
748 
749 /* Set the carrier state for the master according to the state of its
750  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
751  * do special 802.3ad magic.
752  *
753  * Returns zero if carrier state does not change, nonzero if it does.
754  */
755 int bond_set_carrier(struct bonding *bond)
756 {
757 	struct list_head *iter;
758 	struct slave *slave;
759 
760 	if (!bond_has_slaves(bond))
761 		goto down;
762 
763 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
764 		return bond_3ad_set_carrier(bond);
765 
766 	bond_for_each_slave(bond, slave, iter) {
767 		if (slave->link == BOND_LINK_UP) {
768 			if (!netif_carrier_ok(bond->dev)) {
769 				netif_carrier_on(bond->dev);
770 				return 1;
771 			}
772 			return 0;
773 		}
774 	}
775 
776 down:
777 	if (netif_carrier_ok(bond->dev)) {
778 		netif_carrier_off(bond->dev);
779 		return 1;
780 	}
781 	return 0;
782 }
783 
784 /* Get link speed and duplex from the slave's base driver
785  * using ethtool. If for some reason the call fails or the
786  * values are invalid, set speed and duplex to -1,
787  * and return. Return 1 if speed or duplex settings are
788  * UNKNOWN; 0 otherwise.
789  */
790 static int bond_update_speed_duplex(struct slave *slave)
791 {
792 	struct net_device *slave_dev = slave->dev;
793 	struct ethtool_link_ksettings ecmd;
794 	int res;
795 
796 	slave->speed = SPEED_UNKNOWN;
797 	slave->duplex = DUPLEX_UNKNOWN;
798 
799 	res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
800 	if (res < 0)
801 		return 1;
802 	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
803 		return 1;
804 	switch (ecmd.base.duplex) {
805 	case DUPLEX_FULL:
806 	case DUPLEX_HALF:
807 		break;
808 	default:
809 		return 1;
810 	}
811 
812 	slave->speed = ecmd.base.speed;
813 	slave->duplex = ecmd.base.duplex;
814 
815 	return 0;
816 }
817 
818 const char *bond_slave_link_status(s8 link)
819 {
820 	switch (link) {
821 	case BOND_LINK_UP:
822 		return "up";
823 	case BOND_LINK_FAIL:
824 		return "going down";
825 	case BOND_LINK_DOWN:
826 		return "down";
827 	case BOND_LINK_BACK:
828 		return "going back";
829 	default:
830 		return "unknown";
831 	}
832 }
833 
834 /* if <dev> supports MII link status reporting, check its link status.
835  *
836  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
837  * depending upon the setting of the use_carrier parameter.
838  *
839  * Return either BMSR_LSTATUS, meaning that the link is up (or we
840  * can't tell and just pretend it is), or 0, meaning that the link is
841  * down.
842  *
843  * If reporting is non-zero, instead of faking link up, return -1 if
844  * both ETHTOOL and MII ioctls fail (meaning the device does not
845  * support them).  If use_carrier is set, return whatever it says.
846  * It'd be nice if there was a good way to tell if a driver supports
847  * netif_carrier, but there really isn't.
848  */
849 static int bond_check_dev_link(struct bonding *bond,
850 			       struct net_device *slave_dev, int reporting)
851 {
852 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
853 	struct mii_ioctl_data *mii;
854 	struct ifreq ifr;
855 	int ret;
856 
857 	if (!reporting && !netif_running(slave_dev))
858 		return 0;
859 
860 	if (bond->params.use_carrier)
861 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
862 
863 	/* Try to get link status using Ethtool first. */
864 	if (slave_dev->ethtool_ops->get_link) {
865 		netdev_lock_ops(slave_dev);
866 		ret = slave_dev->ethtool_ops->get_link(slave_dev);
867 		netdev_unlock_ops(slave_dev);
868 
869 		return ret ? BMSR_LSTATUS : 0;
870 	}
871 
872 	/* Ethtool can't be used, fallback to MII ioctls. */
873 	if (slave_ops->ndo_eth_ioctl) {
874 		/* TODO: set pointer to correct ioctl on a per team member
875 		 *       bases to make this more efficient. that is, once
876 		 *       we determine the correct ioctl, we will always
877 		 *       call it and not the others for that team
878 		 *       member.
879 		 */
880 
881 		/* We cannot assume that SIOCGMIIPHY will also read a
882 		 * register; not all network drivers (e.g., e100)
883 		 * support that.
884 		 */
885 
886 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
887 		strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
888 		mii = if_mii(&ifr);
889 
890 		if (dev_eth_ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
891 			mii->reg_num = MII_BMSR;
892 			if (dev_eth_ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
893 				return mii->val_out & BMSR_LSTATUS;
894 		}
895 	}
896 
897 	/* If reporting, report that either there's no ndo_eth_ioctl,
898 	 * or both SIOCGMIIREG and get_link failed (meaning that we
899 	 * cannot report link status).  If not reporting, pretend
900 	 * we're ok.
901 	 */
902 	return reporting ? -1 : BMSR_LSTATUS;
903 }
904 
905 /*----------------------------- Multicast list ------------------------------*/
906 
907 /* Push the promiscuity flag down to appropriate slaves */
908 static int bond_set_promiscuity(struct bonding *bond, int inc)
909 {
910 	struct list_head *iter;
911 	int err = 0;
912 
913 	if (bond_uses_primary(bond)) {
914 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
915 
916 		if (curr_active)
917 			err = dev_set_promiscuity(curr_active->dev, inc);
918 	} else {
919 		struct slave *slave;
920 
921 		bond_for_each_slave(bond, slave, iter) {
922 			err = dev_set_promiscuity(slave->dev, inc);
923 			if (err)
924 				return err;
925 		}
926 	}
927 	return err;
928 }
929 
930 /* Push the allmulti flag down to all slaves */
931 static int bond_set_allmulti(struct bonding *bond, int inc)
932 {
933 	struct list_head *iter;
934 	int err = 0;
935 
936 	if (bond_uses_primary(bond)) {
937 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
938 
939 		if (curr_active)
940 			err = dev_set_allmulti(curr_active->dev, inc);
941 	} else {
942 		struct slave *slave;
943 
944 		bond_for_each_slave(bond, slave, iter) {
945 			err = dev_set_allmulti(slave->dev, inc);
946 			if (err)
947 				return err;
948 		}
949 	}
950 	return err;
951 }
952 
953 /* Retrieve the list of registered multicast addresses for the bonding
954  * device and retransmit an IGMP JOIN request to the current active
955  * slave.
956  */
957 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
958 {
959 	struct bonding *bond = container_of(work, struct bonding,
960 					    mcast_work.work);
961 
962 	if (!rtnl_trylock()) {
963 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
964 		return;
965 	}
966 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
967 
968 	if (bond->igmp_retrans > 1) {
969 		bond->igmp_retrans--;
970 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
971 	}
972 	rtnl_unlock();
973 }
974 
975 /* Flush bond's hardware addresses from slave */
976 static void bond_hw_addr_flush(struct net_device *bond_dev,
977 			       struct net_device *slave_dev)
978 {
979 	struct bonding *bond = netdev_priv(bond_dev);
980 
981 	dev_uc_unsync(slave_dev, bond_dev);
982 	dev_mc_unsync(slave_dev, bond_dev);
983 
984 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
985 		dev_mc_del(slave_dev, lacpdu_mcast_addr);
986 }
987 
988 /*--------------------------- Active slave change ---------------------------*/
989 
990 /* Update the hardware address list and promisc/allmulti for the new and
991  * old active slaves (if any).  Modes that are not using primary keep all
992  * slaves up date at all times; only the modes that use primary need to call
993  * this function to swap these settings during a failover.
994  */
995 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
996 			      struct slave *old_active)
997 {
998 	if (old_active) {
999 		if (bond->dev->flags & IFF_PROMISC)
1000 			dev_set_promiscuity(old_active->dev, -1);
1001 
1002 		if (bond->dev->flags & IFF_ALLMULTI)
1003 			dev_set_allmulti(old_active->dev, -1);
1004 
1005 		if (bond->dev->flags & IFF_UP)
1006 			bond_hw_addr_flush(bond->dev, old_active->dev);
1007 
1008 		bond_slave_ns_maddrs_add(bond, old_active);
1009 	}
1010 
1011 	if (new_active) {
1012 		/* FIXME: Signal errors upstream. */
1013 		if (bond->dev->flags & IFF_PROMISC)
1014 			dev_set_promiscuity(new_active->dev, 1);
1015 
1016 		if (bond->dev->flags & IFF_ALLMULTI)
1017 			dev_set_allmulti(new_active->dev, 1);
1018 
1019 		if (bond->dev->flags & IFF_UP) {
1020 			netif_addr_lock_bh(bond->dev);
1021 			dev_uc_sync(new_active->dev, bond->dev);
1022 			dev_mc_sync(new_active->dev, bond->dev);
1023 			netif_addr_unlock_bh(bond->dev);
1024 		}
1025 
1026 		bond_slave_ns_maddrs_del(bond, new_active);
1027 	}
1028 }
1029 
1030 /**
1031  * bond_set_dev_addr - clone slave's address to bond
1032  * @bond_dev: bond net device
1033  * @slave_dev: slave net device
1034  *
1035  * Should be called with RTNL held.
1036  */
1037 static int bond_set_dev_addr(struct net_device *bond_dev,
1038 			     struct net_device *slave_dev)
1039 {
1040 	int err;
1041 
1042 	slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
1043 		  bond_dev, slave_dev, slave_dev->addr_len);
1044 	err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
1045 	if (err)
1046 		return err;
1047 
1048 	__dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
1049 	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
1050 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1051 	return 0;
1052 }
1053 
1054 static struct slave *bond_get_old_active(struct bonding *bond,
1055 					 struct slave *new_active)
1056 {
1057 	struct slave *slave;
1058 	struct list_head *iter;
1059 
1060 	bond_for_each_slave(bond, slave, iter) {
1061 		if (slave == new_active)
1062 			continue;
1063 
1064 		if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
1065 			return slave;
1066 	}
1067 
1068 	return NULL;
1069 }
1070 
1071 /* bond_do_fail_over_mac
1072  *
1073  * Perform special MAC address swapping for fail_over_mac settings
1074  *
1075  * Called with RTNL
1076  */
1077 static void bond_do_fail_over_mac(struct bonding *bond,
1078 				  struct slave *new_active,
1079 				  struct slave *old_active)
1080 {
1081 	u8 tmp_mac[MAX_ADDR_LEN];
1082 	struct sockaddr_storage ss;
1083 	int rv;
1084 
1085 	switch (bond->params.fail_over_mac) {
1086 	case BOND_FOM_ACTIVE:
1087 		if (new_active) {
1088 			rv = bond_set_dev_addr(bond->dev, new_active->dev);
1089 			if (rv)
1090 				slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
1091 					  -rv);
1092 		}
1093 		break;
1094 	case BOND_FOM_FOLLOW:
1095 		/* if new_active && old_active, swap them
1096 		 * if just old_active, do nothing (going to no active slave)
1097 		 * if just new_active, set new_active to bond's MAC
1098 		 */
1099 		if (!new_active)
1100 			return;
1101 
1102 		if (!old_active)
1103 			old_active = bond_get_old_active(bond, new_active);
1104 
1105 		if (old_active) {
1106 			bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
1107 					  new_active->dev->addr_len);
1108 			bond_hw_addr_copy(ss.__data,
1109 					  old_active->dev->dev_addr,
1110 					  old_active->dev->addr_len);
1111 			ss.ss_family = new_active->dev->type;
1112 		} else {
1113 			bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1114 					  bond->dev->addr_len);
1115 			ss.ss_family = bond->dev->type;
1116 		}
1117 
1118 		rv = dev_set_mac_address(new_active->dev,
1119 					 (struct sockaddr *)&ss, NULL);
1120 		if (rv) {
1121 			slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1122 				  -rv);
1123 			goto out;
1124 		}
1125 
1126 		if (!old_active)
1127 			goto out;
1128 
1129 		bond_hw_addr_copy(ss.__data, tmp_mac,
1130 				  new_active->dev->addr_len);
1131 		ss.ss_family = old_active->dev->type;
1132 
1133 		rv = dev_set_mac_address(old_active->dev,
1134 					 (struct sockaddr *)&ss, NULL);
1135 		if (rv)
1136 			slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1137 				  -rv);
1138 out:
1139 		break;
1140 	default:
1141 		netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1142 			   bond->params.fail_over_mac);
1143 		break;
1144 	}
1145 
1146 }
1147 
1148 /**
1149  * bond_choose_primary_or_current - select the primary or high priority slave
1150  * @bond: our bonding struct
1151  *
1152  * - Check if there is a primary link. If the primary link was set and is up,
1153  *   go on and do link reselection.
1154  *
1155  * - If primary link is not set or down, find the highest priority link.
1156  *   If the highest priority link is not current slave, set it as primary
1157  *   link and do link reselection.
1158  */
1159 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1160 {
1161 	struct slave *prim = rtnl_dereference(bond->primary_slave);
1162 	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1163 	struct slave *slave, *hprio = NULL;
1164 	struct list_head *iter;
1165 
1166 	if (!prim || prim->link != BOND_LINK_UP) {
1167 		bond_for_each_slave(bond, slave, iter) {
1168 			if (slave->link == BOND_LINK_UP) {
1169 				hprio = hprio ?: slave;
1170 				if (slave->prio > hprio->prio)
1171 					hprio = slave;
1172 			}
1173 		}
1174 
1175 		if (hprio && hprio != curr) {
1176 			prim = hprio;
1177 			goto link_reselect;
1178 		}
1179 
1180 		if (!curr || curr->link != BOND_LINK_UP)
1181 			return NULL;
1182 		return curr;
1183 	}
1184 
1185 	if (bond->force_primary) {
1186 		bond->force_primary = false;
1187 		return prim;
1188 	}
1189 
1190 link_reselect:
1191 	if (!curr || curr->link != BOND_LINK_UP)
1192 		return prim;
1193 
1194 	/* At this point, prim and curr are both up */
1195 	switch (bond->params.primary_reselect) {
1196 	case BOND_PRI_RESELECT_ALWAYS:
1197 		return prim;
1198 	case BOND_PRI_RESELECT_BETTER:
1199 		if (prim->speed < curr->speed)
1200 			return curr;
1201 		if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1202 			return curr;
1203 		return prim;
1204 	case BOND_PRI_RESELECT_FAILURE:
1205 		return curr;
1206 	default:
1207 		netdev_err(bond->dev, "impossible primary_reselect %d\n",
1208 			   bond->params.primary_reselect);
1209 		return curr;
1210 	}
1211 }
1212 
1213 /**
1214  * bond_find_best_slave - select the best available slave to be the active one
1215  * @bond: our bonding struct
1216  */
1217 static struct slave *bond_find_best_slave(struct bonding *bond)
1218 {
1219 	struct slave *slave, *bestslave = NULL;
1220 	struct list_head *iter;
1221 	int mintime = bond->params.updelay;
1222 
1223 	slave = bond_choose_primary_or_current(bond);
1224 	if (slave)
1225 		return slave;
1226 
1227 	bond_for_each_slave(bond, slave, iter) {
1228 		if (slave->link == BOND_LINK_UP)
1229 			return slave;
1230 		if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1231 		    slave->delay < mintime) {
1232 			mintime = slave->delay;
1233 			bestslave = slave;
1234 		}
1235 	}
1236 
1237 	return bestslave;
1238 }
1239 
1240 /* must be called in RCU critical section or with RTNL held */
1241 static bool bond_should_notify_peers(struct bonding *bond)
1242 {
1243 	struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
1244 
1245 	if (!slave || !bond->send_peer_notif ||
1246 	    bond->send_peer_notif %
1247 	    max(1, bond->params.peer_notif_delay) != 0 ||
1248 	    !netif_carrier_ok(bond->dev) ||
1249 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1250 		return false;
1251 
1252 	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1253 		   slave ? slave->dev->name : "NULL");
1254 
1255 	return true;
1256 }
1257 
1258 /**
1259  * bond_change_active_slave - change the active slave into the specified one
1260  * @bond: our bonding struct
1261  * @new_active: the new slave to make the active one
1262  *
1263  * Set the new slave to the bond's settings and unset them on the old
1264  * curr_active_slave.
1265  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1266  *
1267  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1268  * because it is apparently the best available slave we have, even though its
1269  * updelay hasn't timed out yet.
1270  *
1271  * Caller must hold RTNL.
1272  */
1273 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1274 {
1275 	struct slave *old_active;
1276 
1277 	ASSERT_RTNL();
1278 
1279 	old_active = rtnl_dereference(bond->curr_active_slave);
1280 
1281 	if (old_active == new_active)
1282 		return;
1283 
1284 #ifdef CONFIG_XFRM_OFFLOAD
1285 	bond_ipsec_del_sa_all(bond);
1286 #endif /* CONFIG_XFRM_OFFLOAD */
1287 
1288 	if (new_active) {
1289 		new_active->last_link_up = jiffies;
1290 
1291 		if (new_active->link == BOND_LINK_BACK) {
1292 			if (bond_uses_primary(bond)) {
1293 				slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1294 					   (bond->params.updelay - new_active->delay) * bond->params.miimon);
1295 			}
1296 
1297 			new_active->delay = 0;
1298 			bond_set_slave_link_state(new_active, BOND_LINK_UP,
1299 						  BOND_SLAVE_NOTIFY_NOW);
1300 
1301 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
1302 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1303 
1304 			if (bond_is_lb(bond))
1305 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1306 		} else {
1307 			if (bond_uses_primary(bond))
1308 				slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1309 		}
1310 	}
1311 
1312 	if (bond_uses_primary(bond))
1313 		bond_hw_addr_swap(bond, new_active, old_active);
1314 
1315 	if (bond_is_lb(bond)) {
1316 		bond_alb_handle_active_change(bond, new_active);
1317 		if (old_active)
1318 			bond_set_slave_inactive_flags(old_active,
1319 						      BOND_SLAVE_NOTIFY_NOW);
1320 		if (new_active)
1321 			bond_set_slave_active_flags(new_active,
1322 						    BOND_SLAVE_NOTIFY_NOW);
1323 	} else {
1324 		rcu_assign_pointer(bond->curr_active_slave, new_active);
1325 	}
1326 
1327 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1328 		if (old_active)
1329 			bond_set_slave_inactive_flags(old_active,
1330 						      BOND_SLAVE_NOTIFY_NOW);
1331 
1332 		if (new_active) {
1333 			bool should_notify_peers = false;
1334 
1335 			bond_set_slave_active_flags(new_active,
1336 						    BOND_SLAVE_NOTIFY_NOW);
1337 
1338 			if (bond->params.fail_over_mac)
1339 				bond_do_fail_over_mac(bond, new_active,
1340 						      old_active);
1341 
1342 			if (netif_running(bond->dev)) {
1343 				bond->send_peer_notif =
1344 					bond->params.num_peer_notif *
1345 					max(1, bond->params.peer_notif_delay);
1346 				should_notify_peers =
1347 					bond_should_notify_peers(bond);
1348 			}
1349 
1350 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1351 			if (should_notify_peers) {
1352 				bond->send_peer_notif--;
1353 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1354 							 bond->dev);
1355 			}
1356 		}
1357 	}
1358 
1359 #ifdef CONFIG_XFRM_OFFLOAD
1360 	bond_ipsec_add_sa_all(bond);
1361 #endif /* CONFIG_XFRM_OFFLOAD */
1362 
1363 	/* resend IGMP joins since active slave has changed or
1364 	 * all were sent on curr_active_slave.
1365 	 * resend only if bond is brought up with the affected
1366 	 * bonding modes and the retransmission is enabled
1367 	 */
1368 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1369 	    ((bond_uses_primary(bond) && new_active) ||
1370 	     BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1371 		bond->igmp_retrans = bond->params.resend_igmp;
1372 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1373 	}
1374 }
1375 
1376 /**
1377  * bond_select_active_slave - select a new active slave, if needed
1378  * @bond: our bonding struct
1379  *
1380  * This functions should be called when one of the following occurs:
1381  * - The old curr_active_slave has been released or lost its link.
1382  * - The primary_slave has got its link back.
1383  * - A slave has got its link back and there's no old curr_active_slave.
1384  *
1385  * Caller must hold RTNL.
1386  */
1387 void bond_select_active_slave(struct bonding *bond)
1388 {
1389 	struct slave *best_slave;
1390 	int rv;
1391 
1392 	ASSERT_RTNL();
1393 
1394 	best_slave = bond_find_best_slave(bond);
1395 	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1396 		bond_change_active_slave(bond, best_slave);
1397 		rv = bond_set_carrier(bond);
1398 		if (!rv)
1399 			return;
1400 
1401 		if (netif_carrier_ok(bond->dev))
1402 			netdev_info(bond->dev, "active interface up!\n");
1403 		else
1404 			netdev_info(bond->dev, "now running without any active interface!\n");
1405 	}
1406 }
1407 
1408 #ifdef CONFIG_NET_POLL_CONTROLLER
1409 static inline int slave_enable_netpoll(struct slave *slave)
1410 {
1411 	struct netpoll *np;
1412 	int err = 0;
1413 
1414 	np = kzalloc(sizeof(*np), GFP_KERNEL);
1415 	err = -ENOMEM;
1416 	if (!np)
1417 		goto out;
1418 
1419 	err = __netpoll_setup(np, slave->dev);
1420 	if (err) {
1421 		kfree(np);
1422 		goto out;
1423 	}
1424 	slave->np = np;
1425 out:
1426 	return err;
1427 }
1428 static inline void slave_disable_netpoll(struct slave *slave)
1429 {
1430 	struct netpoll *np = slave->np;
1431 
1432 	if (!np)
1433 		return;
1434 
1435 	slave->np = NULL;
1436 
1437 	__netpoll_free(np);
1438 }
1439 
1440 static void bond_poll_controller(struct net_device *bond_dev)
1441 {
1442 	struct bonding *bond = netdev_priv(bond_dev);
1443 	struct slave *slave = NULL;
1444 	struct list_head *iter;
1445 	struct ad_info ad_info;
1446 
1447 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1448 		if (bond_3ad_get_active_agg_info(bond, &ad_info))
1449 			return;
1450 
1451 	bond_for_each_slave_rcu(bond, slave, iter) {
1452 		if (!bond_slave_is_up(slave))
1453 			continue;
1454 
1455 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1456 			struct aggregator *agg =
1457 			    SLAVE_AD_INFO(slave)->port.aggregator;
1458 
1459 			if (agg &&
1460 			    agg->aggregator_identifier != ad_info.aggregator_id)
1461 				continue;
1462 		}
1463 
1464 		netpoll_poll_dev(slave->dev);
1465 	}
1466 }
1467 
1468 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1469 {
1470 	struct bonding *bond = netdev_priv(bond_dev);
1471 	struct list_head *iter;
1472 	struct slave *slave;
1473 
1474 	bond_for_each_slave(bond, slave, iter)
1475 		if (bond_slave_is_up(slave))
1476 			slave_disable_netpoll(slave);
1477 }
1478 
1479 static int bond_netpoll_setup(struct net_device *dev)
1480 {
1481 	struct bonding *bond = netdev_priv(dev);
1482 	struct list_head *iter;
1483 	struct slave *slave;
1484 	int err = 0;
1485 
1486 	bond_for_each_slave(bond, slave, iter) {
1487 		err = slave_enable_netpoll(slave);
1488 		if (err) {
1489 			bond_netpoll_cleanup(dev);
1490 			break;
1491 		}
1492 	}
1493 	return err;
1494 }
1495 #else
1496 static inline int slave_enable_netpoll(struct slave *slave)
1497 {
1498 	return 0;
1499 }
1500 static inline void slave_disable_netpoll(struct slave *slave)
1501 {
1502 }
1503 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1504 {
1505 }
1506 #endif
1507 
1508 /*---------------------------------- IOCTL ----------------------------------*/
1509 
1510 static netdev_features_t bond_fix_features(struct net_device *dev,
1511 					   netdev_features_t features)
1512 {
1513 	struct bonding *bond = netdev_priv(dev);
1514 	struct list_head *iter;
1515 	netdev_features_t mask;
1516 	struct slave *slave;
1517 
1518 	mask = features;
1519 	features = netdev_base_features(features);
1520 
1521 	bond_for_each_slave(bond, slave, iter) {
1522 		features = netdev_increment_features(features,
1523 						     slave->dev->features,
1524 						     mask);
1525 	}
1526 	features = netdev_add_tso_features(features, mask);
1527 
1528 	return features;
1529 }
1530 
1531 #define BOND_VLAN_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1532 				 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1533 				 NETIF_F_GSO_ENCAP_ALL | \
1534 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1535 
1536 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1537 				 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE | \
1538 				 NETIF_F_GSO_PARTIAL)
1539 
1540 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1541 				 NETIF_F_GSO_SOFTWARE)
1542 
1543 #define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP)
1544 
1545 
1546 static void bond_compute_features(struct bonding *bond)
1547 {
1548 	netdev_features_t gso_partial_features = BOND_GSO_PARTIAL_FEATURES;
1549 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1550 					IFF_XMIT_DST_RELEASE_PERM;
1551 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1552 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
1553 #ifdef CONFIG_XFRM_OFFLOAD
1554 	netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1555 #endif /* CONFIG_XFRM_OFFLOAD */
1556 	netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1557 	struct net_device *bond_dev = bond->dev;
1558 	struct list_head *iter;
1559 	struct slave *slave;
1560 	unsigned short max_hard_header_len = ETH_HLEN;
1561 	unsigned int tso_max_size = TSO_MAX_SIZE;
1562 	u16 tso_max_segs = TSO_MAX_SEGS;
1563 
1564 	if (!bond_has_slaves(bond))
1565 		goto done;
1566 
1567 	vlan_features = netdev_base_features(vlan_features);
1568 	mpls_features = netdev_base_features(mpls_features);
1569 
1570 	bond_for_each_slave(bond, slave, iter) {
1571 		vlan_features = netdev_increment_features(vlan_features,
1572 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1573 
1574 		enc_features = netdev_increment_features(enc_features,
1575 							 slave->dev->hw_enc_features,
1576 							 BOND_ENC_FEATURES);
1577 
1578 #ifdef CONFIG_XFRM_OFFLOAD
1579 		xfrm_features = netdev_increment_features(xfrm_features,
1580 							  slave->dev->hw_enc_features,
1581 							  BOND_XFRM_FEATURES);
1582 #endif /* CONFIG_XFRM_OFFLOAD */
1583 
1584 		gso_partial_features = netdev_increment_features(gso_partial_features,
1585 								 slave->dev->gso_partial_features,
1586 								 BOND_GSO_PARTIAL_FEATURES);
1587 
1588 		mpls_features = netdev_increment_features(mpls_features,
1589 							  slave->dev->mpls_features,
1590 							  BOND_MPLS_FEATURES);
1591 
1592 		dst_release_flag &= slave->dev->priv_flags;
1593 		if (slave->dev->hard_header_len > max_hard_header_len)
1594 			max_hard_header_len = slave->dev->hard_header_len;
1595 
1596 		tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1597 		tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1598 	}
1599 	bond_dev->hard_header_len = max_hard_header_len;
1600 
1601 done:
1602 	bond_dev->gso_partial_features = gso_partial_features;
1603 	bond_dev->vlan_features = vlan_features;
1604 	bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1605 				    NETIF_F_HW_VLAN_CTAG_TX |
1606 				    NETIF_F_HW_VLAN_STAG_TX;
1607 #ifdef CONFIG_XFRM_OFFLOAD
1608 	bond_dev->hw_enc_features |= xfrm_features;
1609 #endif /* CONFIG_XFRM_OFFLOAD */
1610 	bond_dev->mpls_features = mpls_features;
1611 	netif_set_tso_max_segs(bond_dev, tso_max_segs);
1612 	netif_set_tso_max_size(bond_dev, tso_max_size);
1613 
1614 	bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1615 	if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1616 	    dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1617 		bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1618 
1619 	netdev_change_features(bond_dev);
1620 }
1621 
1622 static void bond_setup_by_slave(struct net_device *bond_dev,
1623 				struct net_device *slave_dev)
1624 {
1625 	bool was_up = !!(bond_dev->flags & IFF_UP);
1626 
1627 	dev_close(bond_dev);
1628 
1629 	bond_dev->header_ops	    = slave_dev->header_ops;
1630 
1631 	bond_dev->type		    = slave_dev->type;
1632 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1633 	bond_dev->needed_headroom   = slave_dev->needed_headroom;
1634 	bond_dev->addr_len	    = slave_dev->addr_len;
1635 
1636 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1637 		slave_dev->addr_len);
1638 
1639 	if (slave_dev->flags & IFF_POINTOPOINT) {
1640 		bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1641 		bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1642 	}
1643 	if (was_up)
1644 		dev_open(bond_dev, NULL);
1645 }
1646 
1647 /* On bonding slaves other than the currently active slave, suppress
1648  * duplicates except for alb non-mcast/bcast.
1649  */
1650 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1651 					    struct slave *slave,
1652 					    struct bonding *bond)
1653 {
1654 	if (bond_is_slave_inactive(slave)) {
1655 		if (BOND_MODE(bond) == BOND_MODE_ALB &&
1656 		    skb->pkt_type != PACKET_BROADCAST &&
1657 		    skb->pkt_type != PACKET_MULTICAST)
1658 			return false;
1659 		return true;
1660 	}
1661 	return false;
1662 }
1663 
1664 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1665 {
1666 	struct sk_buff *skb = *pskb;
1667 	struct slave *slave;
1668 	struct bonding *bond;
1669 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1670 			  struct slave *);
1671 	int ret = RX_HANDLER_ANOTHER;
1672 
1673 	skb = skb_share_check(skb, GFP_ATOMIC);
1674 	if (unlikely(!skb))
1675 		return RX_HANDLER_CONSUMED;
1676 
1677 	*pskb = skb;
1678 
1679 	slave = bond_slave_get_rcu(skb->dev);
1680 	bond = slave->bond;
1681 
1682 	recv_probe = READ_ONCE(bond->recv_probe);
1683 	if (recv_probe) {
1684 		ret = recv_probe(skb, bond, slave);
1685 		if (ret == RX_HANDLER_CONSUMED) {
1686 			consume_skb(skb);
1687 			return ret;
1688 		}
1689 	}
1690 
1691 	/*
1692 	 * For packets determined by bond_should_deliver_exact_match() call to
1693 	 * be suppressed we want to make an exception for link-local packets.
1694 	 * This is necessary for e.g. LLDP daemons to be able to monitor
1695 	 * inactive slave links without being forced to bind to them
1696 	 * explicitly.
1697 	 *
1698 	 * At the same time, packets that are passed to the bonding master
1699 	 * (including link-local ones) can have their originating interface
1700 	 * determined via PACKET_ORIGDEV socket option.
1701 	 */
1702 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1703 		if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1704 			return RX_HANDLER_PASS;
1705 		return RX_HANDLER_EXACT;
1706 	}
1707 
1708 	skb->dev = bond->dev;
1709 
1710 	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1711 	    netif_is_bridge_port(bond->dev) &&
1712 	    skb->pkt_type == PACKET_HOST) {
1713 
1714 		if (unlikely(skb_cow_head(skb,
1715 					  skb->data - skb_mac_header(skb)))) {
1716 			kfree_skb(skb);
1717 			return RX_HANDLER_CONSUMED;
1718 		}
1719 		bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1720 				  bond->dev->addr_len);
1721 	}
1722 
1723 	return ret;
1724 }
1725 
1726 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1727 {
1728 	switch (BOND_MODE(bond)) {
1729 	case BOND_MODE_ROUNDROBIN:
1730 		return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1731 	case BOND_MODE_ACTIVEBACKUP:
1732 		return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1733 	case BOND_MODE_BROADCAST:
1734 		return NETDEV_LAG_TX_TYPE_BROADCAST;
1735 	case BOND_MODE_XOR:
1736 	case BOND_MODE_8023AD:
1737 		return NETDEV_LAG_TX_TYPE_HASH;
1738 	default:
1739 		return NETDEV_LAG_TX_TYPE_UNKNOWN;
1740 	}
1741 }
1742 
1743 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1744 					       enum netdev_lag_tx_type type)
1745 {
1746 	if (type != NETDEV_LAG_TX_TYPE_HASH)
1747 		return NETDEV_LAG_HASH_NONE;
1748 
1749 	switch (bond->params.xmit_policy) {
1750 	case BOND_XMIT_POLICY_LAYER2:
1751 		return NETDEV_LAG_HASH_L2;
1752 	case BOND_XMIT_POLICY_LAYER34:
1753 		return NETDEV_LAG_HASH_L34;
1754 	case BOND_XMIT_POLICY_LAYER23:
1755 		return NETDEV_LAG_HASH_L23;
1756 	case BOND_XMIT_POLICY_ENCAP23:
1757 		return NETDEV_LAG_HASH_E23;
1758 	case BOND_XMIT_POLICY_ENCAP34:
1759 		return NETDEV_LAG_HASH_E34;
1760 	case BOND_XMIT_POLICY_VLAN_SRCMAC:
1761 		return NETDEV_LAG_HASH_VLAN_SRCMAC;
1762 	default:
1763 		return NETDEV_LAG_HASH_UNKNOWN;
1764 	}
1765 }
1766 
1767 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1768 				      struct netlink_ext_ack *extack)
1769 {
1770 	struct netdev_lag_upper_info lag_upper_info;
1771 	enum netdev_lag_tx_type type;
1772 	int err;
1773 
1774 	type = bond_lag_tx_type(bond);
1775 	lag_upper_info.tx_type = type;
1776 	lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1777 
1778 	err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1779 					   &lag_upper_info, extack);
1780 	if (err)
1781 		return err;
1782 
1783 	slave->dev->flags |= IFF_SLAVE;
1784 	return 0;
1785 }
1786 
1787 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1788 {
1789 	netdev_upper_dev_unlink(slave->dev, bond->dev);
1790 	slave->dev->flags &= ~IFF_SLAVE;
1791 }
1792 
1793 static void slave_kobj_release(struct kobject *kobj)
1794 {
1795 	struct slave *slave = to_slave(kobj);
1796 	struct bonding *bond = bond_get_bond_by_slave(slave);
1797 
1798 	cancel_delayed_work_sync(&slave->notify_work);
1799 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1800 		kfree(SLAVE_AD_INFO(slave));
1801 
1802 	kfree(slave);
1803 }
1804 
1805 static struct kobj_type slave_ktype = {
1806 	.release = slave_kobj_release,
1807 #ifdef CONFIG_SYSFS
1808 	.sysfs_ops = &slave_sysfs_ops,
1809 #endif
1810 };
1811 
1812 static int bond_kobj_init(struct slave *slave)
1813 {
1814 	int err;
1815 
1816 	err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1817 				   &(slave->dev->dev.kobj), "bonding_slave");
1818 	if (err)
1819 		kobject_put(&slave->kobj);
1820 
1821 	return err;
1822 }
1823 
1824 static struct slave *bond_alloc_slave(struct bonding *bond,
1825 				      struct net_device *slave_dev)
1826 {
1827 	struct slave *slave = NULL;
1828 
1829 	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1830 	if (!slave)
1831 		return NULL;
1832 
1833 	slave->bond = bond;
1834 	slave->dev = slave_dev;
1835 	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1836 
1837 	if (bond_kobj_init(slave))
1838 		return NULL;
1839 
1840 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1841 		SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1842 					       GFP_KERNEL);
1843 		if (!SLAVE_AD_INFO(slave)) {
1844 			kobject_put(&slave->kobj);
1845 			return NULL;
1846 		}
1847 	}
1848 
1849 	return slave;
1850 }
1851 
1852 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1853 {
1854 	info->bond_mode = BOND_MODE(bond);
1855 	info->miimon = bond->params.miimon;
1856 	info->num_slaves = bond->slave_cnt;
1857 }
1858 
1859 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1860 {
1861 	strcpy(info->slave_name, slave->dev->name);
1862 	info->link = slave->link;
1863 	info->state = bond_slave_state(slave);
1864 	info->link_failure_count = slave->link_failure_count;
1865 }
1866 
1867 static void bond_netdev_notify_work(struct work_struct *_work)
1868 {
1869 	struct slave *slave = container_of(_work, struct slave,
1870 					   notify_work.work);
1871 
1872 	if (rtnl_trylock()) {
1873 		struct netdev_bonding_info binfo;
1874 
1875 		bond_fill_ifslave(slave, &binfo.slave);
1876 		bond_fill_ifbond(slave->bond, &binfo.master);
1877 		netdev_bonding_info_change(slave->dev, &binfo);
1878 		rtnl_unlock();
1879 	} else {
1880 		queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1881 	}
1882 }
1883 
1884 void bond_queue_slave_event(struct slave *slave)
1885 {
1886 	queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1887 }
1888 
1889 void bond_lower_state_changed(struct slave *slave)
1890 {
1891 	struct netdev_lag_lower_state_info info;
1892 
1893 	info.link_up = slave->link == BOND_LINK_UP ||
1894 		       slave->link == BOND_LINK_FAIL;
1895 	info.tx_enabled = bond_is_active_slave(slave);
1896 	netdev_lower_state_changed(slave->dev, &info);
1897 }
1898 
1899 #define BOND_NL_ERR(bond_dev, extack, errmsg) do {		\
1900 	if (extack)						\
1901 		NL_SET_ERR_MSG(extack, errmsg);			\
1902 	else							\
1903 		netdev_err(bond_dev, "Error: %s\n", errmsg);	\
1904 } while (0)
1905 
1906 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {		\
1907 	if (extack)							\
1908 		NL_SET_ERR_MSG(extack, errmsg);				\
1909 	else								\
1910 		slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);	\
1911 } while (0)
1912 
1913 /* The bonding driver uses ether_setup() to convert a master bond device
1914  * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1915  * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1916  * if they were set
1917  */
1918 static void bond_ether_setup(struct net_device *bond_dev)
1919 {
1920 	unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1921 
1922 	ether_setup(bond_dev);
1923 	bond_dev->flags |= IFF_MASTER | flags;
1924 	bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1925 }
1926 
1927 void bond_xdp_set_features(struct net_device *bond_dev)
1928 {
1929 	struct bonding *bond = netdev_priv(bond_dev);
1930 	xdp_features_t val = NETDEV_XDP_ACT_MASK;
1931 	struct list_head *iter;
1932 	struct slave *slave;
1933 
1934 	ASSERT_RTNL();
1935 
1936 	if (!bond_xdp_check(bond, BOND_MODE(bond)) || !bond_has_slaves(bond)) {
1937 		xdp_clear_features_flag(bond_dev);
1938 		return;
1939 	}
1940 
1941 	bond_for_each_slave(bond, slave, iter)
1942 		val &= slave->dev->xdp_features;
1943 
1944 	val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1945 
1946 	xdp_set_features_flag(bond_dev, val);
1947 }
1948 
1949 /* enslave device <slave> to bond device <master> */
1950 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1951 		 struct netlink_ext_ack *extack)
1952 {
1953 	struct bonding *bond = netdev_priv(bond_dev);
1954 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1955 	struct slave *new_slave = NULL, *prev_slave;
1956 	struct sockaddr_storage ss;
1957 	int link_reporting;
1958 	int res = 0, i;
1959 
1960 	if (slave_dev->flags & IFF_MASTER &&
1961 	    !netif_is_bond_master(slave_dev)) {
1962 		BOND_NL_ERR(bond_dev, extack,
1963 			    "Device type (master device) cannot be enslaved");
1964 		return -EPERM;
1965 	}
1966 
1967 	if (!bond->params.use_carrier &&
1968 	    slave_dev->ethtool_ops->get_link == NULL &&
1969 	    slave_ops->ndo_eth_ioctl == NULL) {
1970 		slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1971 	}
1972 
1973 	/* already in-use? */
1974 	if (netdev_is_rx_handler_busy(slave_dev)) {
1975 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1976 			     "Device is in use and cannot be enslaved");
1977 		return -EBUSY;
1978 	}
1979 
1980 	if (bond_dev == slave_dev) {
1981 		BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1982 		return -EPERM;
1983 	}
1984 
1985 	/* vlan challenged mutual exclusion */
1986 	/* no need to lock since we're protected by rtnl_lock */
1987 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1988 		slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1989 		if (vlan_uses_dev(bond_dev)) {
1990 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1991 				     "Can not enslave VLAN challenged device to VLAN enabled bond");
1992 			return -EPERM;
1993 		} else {
1994 			slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1995 		}
1996 	} else {
1997 		slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1998 	}
1999 
2000 	if (slave_dev->features & NETIF_F_HW_ESP)
2001 		slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
2002 
2003 	/* Old ifenslave binaries are no longer supported.  These can
2004 	 * be identified with moderate accuracy by the state of the slave:
2005 	 * the current ifenslave will set the interface down prior to
2006 	 * enslaving it; the old ifenslave will not.
2007 	 */
2008 	if (slave_dev->flags & IFF_UP) {
2009 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2010 			     "Device can not be enslaved while up");
2011 		return -EPERM;
2012 	}
2013 
2014 	/* set bonding device ether type by slave - bonding netdevices are
2015 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
2016 	 * there is a need to override some of the type dependent attribs/funcs.
2017 	 *
2018 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
2019 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
2020 	 */
2021 	if (!bond_has_slaves(bond)) {
2022 		if (bond_dev->type != slave_dev->type) {
2023 			slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
2024 				  bond_dev->type, slave_dev->type);
2025 
2026 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
2027 						       bond_dev);
2028 			res = notifier_to_errno(res);
2029 			if (res) {
2030 				slave_err(bond_dev, slave_dev, "refused to change device type\n");
2031 				return -EBUSY;
2032 			}
2033 
2034 			/* Flush unicast and multicast addresses */
2035 			dev_uc_flush(bond_dev);
2036 			dev_mc_flush(bond_dev);
2037 
2038 			if (slave_dev->type != ARPHRD_ETHER)
2039 				bond_setup_by_slave(bond_dev, slave_dev);
2040 			else
2041 				bond_ether_setup(bond_dev);
2042 
2043 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
2044 						 bond_dev);
2045 		}
2046 	} else if (bond_dev->type != slave_dev->type) {
2047 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2048 			     "Device type is different from other slaves");
2049 		return -EINVAL;
2050 	}
2051 
2052 	if (slave_dev->type == ARPHRD_INFINIBAND &&
2053 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2054 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2055 			     "Only active-backup mode is supported for infiniband slaves");
2056 		res = -EOPNOTSUPP;
2057 		goto err_undo_flags;
2058 	}
2059 
2060 	if (!slave_ops->ndo_set_mac_address ||
2061 	    slave_dev->type == ARPHRD_INFINIBAND) {
2062 		slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
2063 		if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
2064 		    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2065 			if (!bond_has_slaves(bond)) {
2066 				bond->params.fail_over_mac = BOND_FOM_ACTIVE;
2067 				slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
2068 			} else {
2069 				SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2070 					     "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
2071 				res = -EOPNOTSUPP;
2072 				goto err_undo_flags;
2073 			}
2074 		}
2075 	}
2076 
2077 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
2078 
2079 	/* If this is the first slave, then we need to set the master's hardware
2080 	 * address to be the same as the slave's.
2081 	 */
2082 	if (!bond_has_slaves(bond) &&
2083 	    bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
2084 		res = bond_set_dev_addr(bond->dev, slave_dev);
2085 		if (res)
2086 			goto err_undo_flags;
2087 	}
2088 
2089 	new_slave = bond_alloc_slave(bond, slave_dev);
2090 	if (!new_slave) {
2091 		res = -ENOMEM;
2092 		goto err_undo_flags;
2093 	}
2094 
2095 	/* Set the new_slave's queue_id to be zero.  Queue ID mapping
2096 	 * is set via sysfs or module option if desired.
2097 	 */
2098 	new_slave->queue_id = 0;
2099 
2100 	/* Save slave's original mtu and then set it to match the bond */
2101 	new_slave->original_mtu = slave_dev->mtu;
2102 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
2103 	if (res) {
2104 		slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
2105 		goto err_free;
2106 	}
2107 
2108 	/* Save slave's original ("permanent") mac address for modes
2109 	 * that need it, and for restoring it upon release, and then
2110 	 * set it to the master's address
2111 	 */
2112 	bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
2113 			  slave_dev->addr_len);
2114 
2115 	if (!bond->params.fail_over_mac ||
2116 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2117 		/* Set slave to master's mac address.  The application already
2118 		 * set the master's mac address to that of the first slave
2119 		 */
2120 		memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
2121 	} else if (bond->params.fail_over_mac == BOND_FOM_FOLLOW &&
2122 		   BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
2123 		   memcmp(slave_dev->dev_addr, bond_dev->dev_addr, bond_dev->addr_len) == 0) {
2124 		/* Set slave to random address to avoid duplicate mac
2125 		 * address in later fail over.
2126 		 */
2127 		eth_random_addr(ss.__data);
2128 	} else {
2129 		goto skip_mac_set;
2130 	}
2131 
2132 	ss.ss_family = slave_dev->type;
2133 	res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, extack);
2134 	if (res) {
2135 		slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
2136 		goto err_restore_mtu;
2137 	}
2138 
2139 skip_mac_set:
2140 
2141 	/* set no_addrconf flag before open to prevent IPv6 addrconf */
2142 	slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2143 
2144 	/* open the slave since the application closed it */
2145 	res = dev_open(slave_dev, extack);
2146 	if (res) {
2147 		slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2148 		goto err_restore_mac;
2149 	}
2150 
2151 	slave_dev->priv_flags |= IFF_BONDING;
2152 	/* initialize slave stats */
2153 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2154 
2155 	if (bond_is_lb(bond)) {
2156 		/* bond_alb_init_slave() must be called before all other stages since
2157 		 * it might fail and we do not want to have to undo everything
2158 		 */
2159 		res = bond_alb_init_slave(bond, new_slave);
2160 		if (res)
2161 			goto err_close;
2162 	}
2163 
2164 	res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2165 	if (res) {
2166 		slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2167 		goto err_close;
2168 	}
2169 
2170 	prev_slave = bond_last_slave(bond);
2171 
2172 	new_slave->delay = 0;
2173 	new_slave->link_failure_count = 0;
2174 
2175 	if (bond_update_speed_duplex(new_slave) &&
2176 	    bond_needs_speed_duplex(bond))
2177 		new_slave->link = BOND_LINK_DOWN;
2178 
2179 	new_slave->last_rx = jiffies -
2180 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
2181 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2182 		new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2183 
2184 	new_slave->last_tx = new_slave->last_rx;
2185 
2186 	if (bond->params.miimon && !bond->params.use_carrier) {
2187 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2188 
2189 		if ((link_reporting == -1) && !bond->params.arp_interval) {
2190 			/* miimon is set but a bonded network driver
2191 			 * does not support ETHTOOL/MII and
2192 			 * arp_interval is not set.  Note: if
2193 			 * use_carrier is enabled, we will never go
2194 			 * here (because netif_carrier is always
2195 			 * supported); thus, we don't need to change
2196 			 * the messages for netif_carrier.
2197 			 */
2198 			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");
2199 		} else if (link_reporting == -1) {
2200 			/* unable get link status using mii/ethtool */
2201 			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");
2202 		}
2203 	}
2204 
2205 	/* check for initial state */
2206 	new_slave->link = BOND_LINK_NOCHANGE;
2207 	if (bond->params.miimon) {
2208 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2209 			if (bond->params.updelay) {
2210 				bond_set_slave_link_state(new_slave,
2211 							  BOND_LINK_BACK,
2212 							  BOND_SLAVE_NOTIFY_NOW);
2213 				new_slave->delay = bond->params.updelay;
2214 			} else {
2215 				bond_set_slave_link_state(new_slave,
2216 							  BOND_LINK_UP,
2217 							  BOND_SLAVE_NOTIFY_NOW);
2218 			}
2219 		} else {
2220 			bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2221 						  BOND_SLAVE_NOTIFY_NOW);
2222 		}
2223 	} else if (bond->params.arp_interval) {
2224 		bond_set_slave_link_state(new_slave,
2225 					  (netif_carrier_ok(slave_dev) ?
2226 					  BOND_LINK_UP : BOND_LINK_DOWN),
2227 					  BOND_SLAVE_NOTIFY_NOW);
2228 	} else {
2229 		bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2230 					  BOND_SLAVE_NOTIFY_NOW);
2231 	}
2232 
2233 	if (new_slave->link != BOND_LINK_DOWN)
2234 		new_slave->last_link_up = jiffies;
2235 	slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2236 		  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2237 		  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2238 
2239 	if (bond_uses_primary(bond) && bond->params.primary[0]) {
2240 		/* if there is a primary slave, remember it */
2241 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2242 			rcu_assign_pointer(bond->primary_slave, new_slave);
2243 			bond->force_primary = true;
2244 		}
2245 	}
2246 
2247 	switch (BOND_MODE(bond)) {
2248 	case BOND_MODE_ACTIVEBACKUP:
2249 		bond_set_slave_inactive_flags(new_slave,
2250 					      BOND_SLAVE_NOTIFY_NOW);
2251 		break;
2252 	case BOND_MODE_8023AD:
2253 		/* in 802.3ad mode, the internal mechanism
2254 		 * will activate the slaves in the selected
2255 		 * aggregator
2256 		 */
2257 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2258 		/* if this is the first slave */
2259 		if (!prev_slave) {
2260 			SLAVE_AD_INFO(new_slave)->id = 1;
2261 			/* Initialize AD with the number of times that the AD timer is called in 1 second
2262 			 * can be called only after the mac address of the bond is set
2263 			 */
2264 			bond_3ad_initialize(bond);
2265 		} else {
2266 			SLAVE_AD_INFO(new_slave)->id =
2267 				SLAVE_AD_INFO(prev_slave)->id + 1;
2268 		}
2269 
2270 		bond_3ad_bind_slave(new_slave);
2271 		break;
2272 	case BOND_MODE_TLB:
2273 	case BOND_MODE_ALB:
2274 		bond_set_active_slave(new_slave);
2275 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2276 		break;
2277 	default:
2278 		slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2279 
2280 		/* always active in trunk mode */
2281 		bond_set_active_slave(new_slave);
2282 
2283 		/* In trunking mode there is little meaning to curr_active_slave
2284 		 * anyway (it holds no special properties of the bond device),
2285 		 * so we can change it without calling change_active_interface()
2286 		 */
2287 		if (!rcu_access_pointer(bond->curr_active_slave) &&
2288 		    new_slave->link == BOND_LINK_UP)
2289 			rcu_assign_pointer(bond->curr_active_slave, new_slave);
2290 
2291 		break;
2292 	} /* switch(bond_mode) */
2293 
2294 #ifdef CONFIG_NET_POLL_CONTROLLER
2295 	if (bond->dev->npinfo) {
2296 		if (slave_enable_netpoll(new_slave)) {
2297 			slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2298 			res = -EBUSY;
2299 			goto err_detach;
2300 		}
2301 	}
2302 #endif
2303 
2304 	if (!(bond_dev->features & NETIF_F_LRO))
2305 		dev_disable_lro(slave_dev);
2306 
2307 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2308 					 new_slave);
2309 	if (res) {
2310 		slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2311 		goto err_detach;
2312 	}
2313 
2314 	res = bond_master_upper_dev_link(bond, new_slave, extack);
2315 	if (res) {
2316 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2317 		goto err_unregister;
2318 	}
2319 
2320 	bond_lower_state_changed(new_slave);
2321 
2322 	res = bond_sysfs_slave_add(new_slave);
2323 	if (res) {
2324 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2325 		goto err_upper_unlink;
2326 	}
2327 
2328 	/* If the mode uses primary, then the following is handled by
2329 	 * bond_change_active_slave().
2330 	 */
2331 	if (!bond_uses_primary(bond)) {
2332 		/* set promiscuity level to new slave */
2333 		if (bond_dev->flags & IFF_PROMISC) {
2334 			res = dev_set_promiscuity(slave_dev, 1);
2335 			if (res)
2336 				goto err_sysfs_del;
2337 		}
2338 
2339 		/* set allmulti level to new slave */
2340 		if (bond_dev->flags & IFF_ALLMULTI) {
2341 			res = dev_set_allmulti(slave_dev, 1);
2342 			if (res) {
2343 				if (bond_dev->flags & IFF_PROMISC)
2344 					dev_set_promiscuity(slave_dev, -1);
2345 				goto err_sysfs_del;
2346 			}
2347 		}
2348 
2349 		if (bond_dev->flags & IFF_UP) {
2350 			netif_addr_lock_bh(bond_dev);
2351 			dev_mc_sync_multiple(slave_dev, bond_dev);
2352 			dev_uc_sync_multiple(slave_dev, bond_dev);
2353 			netif_addr_unlock_bh(bond_dev);
2354 
2355 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
2356 				dev_mc_add(slave_dev, lacpdu_mcast_addr);
2357 		}
2358 	}
2359 
2360 	bond->slave_cnt++;
2361 	bond_compute_features(bond);
2362 	bond_set_carrier(bond);
2363 
2364 	/* Needs to be called before bond_select_active_slave(), which will
2365 	 * remove the maddrs if the slave is selected as active slave.
2366 	 */
2367 	bond_slave_ns_maddrs_add(bond, new_slave);
2368 
2369 	if (bond_uses_primary(bond)) {
2370 		block_netpoll_tx();
2371 		bond_select_active_slave(bond);
2372 		unblock_netpoll_tx();
2373 	}
2374 
2375 	if (bond_mode_can_use_xmit_hash(bond))
2376 		bond_update_slave_arr(bond, NULL);
2377 
2378 	if (!slave_dev->netdev_ops->ndo_bpf ||
2379 	    !slave_dev->netdev_ops->ndo_xdp_xmit) {
2380 		if (bond->xdp_prog) {
2381 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2382 				     "Slave does not support XDP");
2383 			res = -EOPNOTSUPP;
2384 			goto err_sysfs_del;
2385 		}
2386 	} else if (bond->xdp_prog) {
2387 		struct netdev_bpf xdp = {
2388 			.command = XDP_SETUP_PROG,
2389 			.flags   = 0,
2390 			.prog    = bond->xdp_prog,
2391 			.extack  = extack,
2392 		};
2393 
2394 		if (dev_xdp_prog_count(slave_dev) > 0) {
2395 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2396 				     "Slave has XDP program loaded, please unload before enslaving");
2397 			res = -EOPNOTSUPP;
2398 			goto err_sysfs_del;
2399 		}
2400 
2401 		res = dev_xdp_propagate(slave_dev, &xdp);
2402 		if (res < 0) {
2403 			/* ndo_bpf() sets extack error message */
2404 			slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2405 			goto err_sysfs_del;
2406 		}
2407 		if (bond->xdp_prog)
2408 			bpf_prog_inc(bond->xdp_prog);
2409 	}
2410 
2411 	bond_xdp_set_features(bond_dev);
2412 
2413 	slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2414 		   bond_is_active_slave(new_slave) ? "an active" : "a backup",
2415 		   new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2416 
2417 	/* enslave is successful */
2418 	bond_queue_slave_event(new_slave);
2419 	return 0;
2420 
2421 /* Undo stages on error */
2422 err_sysfs_del:
2423 	bond_sysfs_slave_del(new_slave);
2424 
2425 err_upper_unlink:
2426 	bond_upper_dev_unlink(bond, new_slave);
2427 
2428 err_unregister:
2429 	netdev_rx_handler_unregister(slave_dev);
2430 
2431 err_detach:
2432 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2433 	if (rcu_access_pointer(bond->primary_slave) == new_slave)
2434 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2435 	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2436 		block_netpoll_tx();
2437 		bond_change_active_slave(bond, NULL);
2438 		bond_select_active_slave(bond);
2439 		unblock_netpoll_tx();
2440 	}
2441 	/* either primary_slave or curr_active_slave might've changed */
2442 	synchronize_rcu();
2443 	slave_disable_netpoll(new_slave);
2444 
2445 err_close:
2446 	if (!netif_is_bond_master(slave_dev))
2447 		slave_dev->priv_flags &= ~IFF_BONDING;
2448 	dev_close(slave_dev);
2449 
2450 err_restore_mac:
2451 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2452 	if (!bond->params.fail_over_mac ||
2453 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2454 		/* XXX TODO - fom follow mode needs to change master's
2455 		 * MAC if this slave's MAC is in use by the bond, or at
2456 		 * least print a warning.
2457 		 */
2458 		bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2459 				  new_slave->dev->addr_len);
2460 		ss.ss_family = slave_dev->type;
2461 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2462 	}
2463 
2464 err_restore_mtu:
2465 	dev_set_mtu(slave_dev, new_slave->original_mtu);
2466 
2467 err_free:
2468 	kobject_put(&new_slave->kobj);
2469 
2470 err_undo_flags:
2471 	/* Enslave of first slave has failed and we need to fix master's mac */
2472 	if (!bond_has_slaves(bond)) {
2473 		if (ether_addr_equal_64bits(bond_dev->dev_addr,
2474 					    slave_dev->dev_addr))
2475 			eth_hw_addr_random(bond_dev);
2476 		if (bond_dev->type != ARPHRD_ETHER) {
2477 			dev_close(bond_dev);
2478 			bond_ether_setup(bond_dev);
2479 		}
2480 	}
2481 
2482 	return res;
2483 }
2484 
2485 /* Try to release the slave device <slave> from the bond device <master>
2486  * It is legal to access curr_active_slave without a lock because all the function
2487  * is RTNL-locked. If "all" is true it means that the function is being called
2488  * while destroying a bond interface and all slaves are being released.
2489  *
2490  * The rules for slave state should be:
2491  *   for Active/Backup:
2492  *     Active stays on all backups go down
2493  *   for Bonded connections:
2494  *     The first up interface should be left on and all others downed.
2495  */
2496 static int __bond_release_one(struct net_device *bond_dev,
2497 			      struct net_device *slave_dev,
2498 			      bool all, bool unregister)
2499 {
2500 	struct bonding *bond = netdev_priv(bond_dev);
2501 	struct slave *slave, *oldcurrent;
2502 	struct sockaddr_storage ss;
2503 	int old_flags = bond_dev->flags;
2504 	netdev_features_t old_features = bond_dev->features;
2505 
2506 	/* slave is not a slave or master is not master of this slave */
2507 	if (!(slave_dev->flags & IFF_SLAVE) ||
2508 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
2509 		slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2510 		return -EINVAL;
2511 	}
2512 
2513 	block_netpoll_tx();
2514 
2515 	slave = bond_get_slave_by_dev(bond, slave_dev);
2516 	if (!slave) {
2517 		/* not a slave of this bond */
2518 		slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2519 		unblock_netpoll_tx();
2520 		return -EINVAL;
2521 	}
2522 
2523 	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2524 
2525 	bond_sysfs_slave_del(slave);
2526 
2527 	/* recompute stats just before removing the slave */
2528 	bond_get_stats(bond->dev, &bond->bond_stats);
2529 
2530 	if (bond->xdp_prog) {
2531 		struct netdev_bpf xdp = {
2532 			.command = XDP_SETUP_PROG,
2533 			.flags   = 0,
2534 			.prog	 = NULL,
2535 			.extack  = NULL,
2536 		};
2537 		if (dev_xdp_propagate(slave_dev, &xdp))
2538 			slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2539 	}
2540 
2541 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2542 	 * for this slave anymore.
2543 	 */
2544 	netdev_rx_handler_unregister(slave_dev);
2545 
2546 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
2547 		bond_3ad_unbind_slave(slave);
2548 
2549 	bond_upper_dev_unlink(bond, slave);
2550 
2551 	if (bond_mode_can_use_xmit_hash(bond))
2552 		bond_update_slave_arr(bond, slave);
2553 
2554 	slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2555 		    bond_is_active_slave(slave) ? "active" : "backup");
2556 
2557 	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2558 
2559 	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2560 
2561 	if (!all && (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2562 		     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2563 		if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2564 		    bond_has_slaves(bond))
2565 			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",
2566 				   slave->perm_hwaddr);
2567 	}
2568 
2569 	if (rtnl_dereference(bond->primary_slave) == slave)
2570 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2571 
2572 	if (oldcurrent == slave)
2573 		bond_change_active_slave(bond, NULL);
2574 
2575 	/* Must be called after bond_change_active_slave () as the slave
2576 	 * might change from an active slave to a backup slave. Then it is
2577 	 * necessary to clear the maddrs on the backup slave.
2578 	 */
2579 	bond_slave_ns_maddrs_del(bond, slave);
2580 
2581 	if (bond_is_lb(bond)) {
2582 		/* Must be called only after the slave has been
2583 		 * detached from the list and the curr_active_slave
2584 		 * has been cleared (if our_slave == old_current),
2585 		 * but before a new active slave is selected.
2586 		 */
2587 		bond_alb_deinit_slave(bond, slave);
2588 	}
2589 
2590 	if (all) {
2591 		RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2592 	} else if (oldcurrent == slave) {
2593 		/* Note that we hold RTNL over this sequence, so there
2594 		 * is no concern that another slave add/remove event
2595 		 * will interfere.
2596 		 */
2597 		bond_select_active_slave(bond);
2598 	}
2599 
2600 	bond_set_carrier(bond);
2601 	if (!bond_has_slaves(bond))
2602 		eth_hw_addr_random(bond_dev);
2603 
2604 	unblock_netpoll_tx();
2605 	synchronize_rcu();
2606 	bond->slave_cnt--;
2607 
2608 	if (!bond_has_slaves(bond)) {
2609 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2610 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2611 	}
2612 
2613 	bond_compute_features(bond);
2614 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2615 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2616 		slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2617 
2618 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2619 
2620 	/* If the mode uses primary, then this case was handled above by
2621 	 * bond_change_active_slave(..., NULL)
2622 	 */
2623 	if (!bond_uses_primary(bond)) {
2624 		/* unset promiscuity level from slave
2625 		 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2626 		 * of the IFF_PROMISC flag in the bond_dev, but we need the
2627 		 * value of that flag before that change, as that was the value
2628 		 * when this slave was attached, so we cache at the start of the
2629 		 * function and use it here. Same goes for ALLMULTI below
2630 		 */
2631 		if (old_flags & IFF_PROMISC)
2632 			dev_set_promiscuity(slave_dev, -1);
2633 
2634 		/* unset allmulti level from slave */
2635 		if (old_flags & IFF_ALLMULTI)
2636 			dev_set_allmulti(slave_dev, -1);
2637 
2638 		if (old_flags & IFF_UP)
2639 			bond_hw_addr_flush(bond_dev, slave_dev);
2640 	}
2641 
2642 	slave_disable_netpoll(slave);
2643 
2644 	/* close slave before restoring its mac address */
2645 	dev_close(slave_dev);
2646 
2647 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2648 
2649 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2650 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2651 		/* restore original ("permanent") mac address */
2652 		bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2653 				  slave->dev->addr_len);
2654 		ss.ss_family = slave_dev->type;
2655 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2656 	}
2657 
2658 	if (unregister) {
2659 		netdev_lock_ops(slave_dev);
2660 		__dev_set_mtu(slave_dev, slave->original_mtu);
2661 		netdev_unlock_ops(slave_dev);
2662 	} else {
2663 		dev_set_mtu(slave_dev, slave->original_mtu);
2664 	}
2665 
2666 	if (!netif_is_bond_master(slave_dev))
2667 		slave_dev->priv_flags &= ~IFF_BONDING;
2668 
2669 	bond_xdp_set_features(bond_dev);
2670 	kobject_put(&slave->kobj);
2671 
2672 	return 0;
2673 }
2674 
2675 /* A wrapper used because of ndo_del_link */
2676 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2677 {
2678 	return __bond_release_one(bond_dev, slave_dev, false, false);
2679 }
2680 
2681 /* First release a slave and then destroy the bond if no more slaves are left.
2682  * Must be under rtnl_lock when this function is called.
2683  */
2684 static int bond_release_and_destroy(struct net_device *bond_dev,
2685 				    struct net_device *slave_dev)
2686 {
2687 	struct bonding *bond = netdev_priv(bond_dev);
2688 	int ret;
2689 
2690 	ret = __bond_release_one(bond_dev, slave_dev, false, true);
2691 	if (ret == 0 && !bond_has_slaves(bond) &&
2692 	    bond_dev->reg_state != NETREG_UNREGISTERING) {
2693 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2694 		netdev_info(bond_dev, "Destroying bond\n");
2695 		bond_remove_proc_entry(bond);
2696 		unregister_netdevice(bond_dev);
2697 	}
2698 	return ret;
2699 }
2700 
2701 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2702 {
2703 	struct bonding *bond = netdev_priv(bond_dev);
2704 
2705 	bond_fill_ifbond(bond, info);
2706 }
2707 
2708 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2709 {
2710 	struct bonding *bond = netdev_priv(bond_dev);
2711 	struct list_head *iter;
2712 	int i = 0, res = -ENODEV;
2713 	struct slave *slave;
2714 
2715 	bond_for_each_slave(bond, slave, iter) {
2716 		if (i++ == (int)info->slave_id) {
2717 			res = 0;
2718 			bond_fill_ifslave(slave, info);
2719 			break;
2720 		}
2721 	}
2722 
2723 	return res;
2724 }
2725 
2726 /*-------------------------------- Monitoring -------------------------------*/
2727 
2728 /* called with rcu_read_lock() */
2729 static int bond_miimon_inspect(struct bonding *bond)
2730 {
2731 	bool ignore_updelay = false;
2732 	int link_state, commit = 0;
2733 	struct list_head *iter;
2734 	struct slave *slave;
2735 
2736 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2737 		ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2738 	} else {
2739 		struct bond_up_slave *usable_slaves;
2740 
2741 		usable_slaves = rcu_dereference(bond->usable_slaves);
2742 
2743 		if (usable_slaves && usable_slaves->count == 0)
2744 			ignore_updelay = true;
2745 	}
2746 
2747 	bond_for_each_slave_rcu(bond, slave, iter) {
2748 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2749 
2750 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2751 
2752 		switch (slave->link) {
2753 		case BOND_LINK_UP:
2754 			if (link_state)
2755 				continue;
2756 
2757 			bond_propose_link_state(slave, BOND_LINK_FAIL);
2758 			commit++;
2759 			slave->delay = bond->params.downdelay;
2760 			if (slave->delay && net_ratelimit()) {
2761 				slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2762 					   (BOND_MODE(bond) ==
2763 					    BOND_MODE_ACTIVEBACKUP) ?
2764 					    (bond_is_active_slave(slave) ?
2765 					     "active " : "backup ") : "",
2766 					   bond->params.downdelay * bond->params.miimon);
2767 			}
2768 			fallthrough;
2769 		case BOND_LINK_FAIL:
2770 			if (link_state) {
2771 				/* recovered before downdelay expired */
2772 				bond_propose_link_state(slave, BOND_LINK_UP);
2773 				slave->last_link_up = jiffies;
2774 				if (net_ratelimit())
2775 					slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2776 						   (bond->params.downdelay - slave->delay) *
2777 						   bond->params.miimon);
2778 				commit++;
2779 				continue;
2780 			}
2781 
2782 			if (slave->delay <= 0) {
2783 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2784 				commit++;
2785 				continue;
2786 			}
2787 
2788 			slave->delay--;
2789 			break;
2790 
2791 		case BOND_LINK_DOWN:
2792 			if (!link_state)
2793 				continue;
2794 
2795 			bond_propose_link_state(slave, BOND_LINK_BACK);
2796 			commit++;
2797 			slave->delay = bond->params.updelay;
2798 
2799 			if (slave->delay && net_ratelimit()) {
2800 				slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2801 					   ignore_updelay ? 0 :
2802 					   bond->params.updelay *
2803 					   bond->params.miimon);
2804 			}
2805 			fallthrough;
2806 		case BOND_LINK_BACK:
2807 			if (!link_state) {
2808 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2809 				if (net_ratelimit())
2810 					slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2811 						   (bond->params.updelay - slave->delay) *
2812 						   bond->params.miimon);
2813 				commit++;
2814 				continue;
2815 			}
2816 
2817 			if (ignore_updelay)
2818 				slave->delay = 0;
2819 
2820 			if (slave->delay <= 0) {
2821 				bond_propose_link_state(slave, BOND_LINK_UP);
2822 				commit++;
2823 				ignore_updelay = false;
2824 				continue;
2825 			}
2826 
2827 			slave->delay--;
2828 			break;
2829 		}
2830 	}
2831 
2832 	return commit;
2833 }
2834 
2835 static void bond_miimon_link_change(struct bonding *bond,
2836 				    struct slave *slave,
2837 				    char link)
2838 {
2839 	switch (BOND_MODE(bond)) {
2840 	case BOND_MODE_8023AD:
2841 		bond_3ad_handle_link_change(slave, link);
2842 		break;
2843 	case BOND_MODE_TLB:
2844 	case BOND_MODE_ALB:
2845 		bond_alb_handle_link_change(bond, slave, link);
2846 		break;
2847 	case BOND_MODE_XOR:
2848 		bond_update_slave_arr(bond, NULL);
2849 		break;
2850 	}
2851 }
2852 
2853 static void bond_miimon_commit(struct bonding *bond)
2854 {
2855 	struct slave *slave, *primary, *active;
2856 	bool do_failover = false;
2857 	struct list_head *iter;
2858 
2859 	ASSERT_RTNL();
2860 
2861 	bond_for_each_slave(bond, slave, iter) {
2862 		switch (slave->link_new_state) {
2863 		case BOND_LINK_NOCHANGE:
2864 			/* For 802.3ad mode, check current slave speed and
2865 			 * duplex again in case its port was disabled after
2866 			 * invalid speed/duplex reporting but recovered before
2867 			 * link monitoring could make a decision on the actual
2868 			 * link status
2869 			 */
2870 			if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2871 			    slave->link == BOND_LINK_UP)
2872 				bond_3ad_adapter_speed_duplex_changed(slave);
2873 			continue;
2874 
2875 		case BOND_LINK_UP:
2876 			if (bond_update_speed_duplex(slave) &&
2877 			    bond_needs_speed_duplex(bond)) {
2878 				slave->link = BOND_LINK_DOWN;
2879 				if (net_ratelimit())
2880 					slave_warn(bond->dev, slave->dev,
2881 						   "failed to get link speed/duplex\n");
2882 				continue;
2883 			}
2884 			bond_set_slave_link_state(slave, BOND_LINK_UP,
2885 						  BOND_SLAVE_NOTIFY_NOW);
2886 			slave->last_link_up = jiffies;
2887 
2888 			primary = rtnl_dereference(bond->primary_slave);
2889 			if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2890 				/* prevent it from being the active one */
2891 				bond_set_backup_slave(slave);
2892 			} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2893 				/* make it immediately active */
2894 				bond_set_active_slave(slave);
2895 			}
2896 
2897 			slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2898 				   slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2899 				   slave->duplex ? "full" : "half");
2900 
2901 			bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2902 
2903 			active = rtnl_dereference(bond->curr_active_slave);
2904 			if (!active || slave == primary || slave->prio > active->prio)
2905 				do_failover = true;
2906 
2907 			continue;
2908 
2909 		case BOND_LINK_DOWN:
2910 			if (slave->link_failure_count < UINT_MAX)
2911 				slave->link_failure_count++;
2912 
2913 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2914 						  BOND_SLAVE_NOTIFY_NOW);
2915 
2916 			if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2917 			    BOND_MODE(bond) == BOND_MODE_8023AD)
2918 				bond_set_slave_inactive_flags(slave,
2919 							      BOND_SLAVE_NOTIFY_NOW);
2920 
2921 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2922 
2923 			bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2924 
2925 			if (slave == rcu_access_pointer(bond->curr_active_slave))
2926 				do_failover = true;
2927 
2928 			continue;
2929 
2930 		default:
2931 			slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2932 				  slave->link_new_state);
2933 			bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2934 
2935 			continue;
2936 		}
2937 	}
2938 
2939 	if (do_failover) {
2940 		block_netpoll_tx();
2941 		bond_select_active_slave(bond);
2942 		unblock_netpoll_tx();
2943 	}
2944 
2945 	bond_set_carrier(bond);
2946 }
2947 
2948 /* bond_mii_monitor
2949  *
2950  * Really a wrapper that splits the mii monitor into two phases: an
2951  * inspection, then (if inspection indicates something needs to be done)
2952  * an acquisition of appropriate locks followed by a commit phase to
2953  * implement whatever link state changes are indicated.
2954  */
2955 static void bond_mii_monitor(struct work_struct *work)
2956 {
2957 	struct bonding *bond = container_of(work, struct bonding,
2958 					    mii_work.work);
2959 	bool should_notify_peers = false;
2960 	bool commit;
2961 	unsigned long delay;
2962 	struct slave *slave;
2963 	struct list_head *iter;
2964 
2965 	delay = msecs_to_jiffies(bond->params.miimon);
2966 
2967 	if (!bond_has_slaves(bond))
2968 		goto re_arm;
2969 
2970 	rcu_read_lock();
2971 	should_notify_peers = bond_should_notify_peers(bond);
2972 	commit = !!bond_miimon_inspect(bond);
2973 	if (bond->send_peer_notif) {
2974 		rcu_read_unlock();
2975 		if (rtnl_trylock()) {
2976 			bond->send_peer_notif--;
2977 			rtnl_unlock();
2978 		}
2979 	} else {
2980 		rcu_read_unlock();
2981 	}
2982 
2983 	if (commit) {
2984 		/* Race avoidance with bond_close cancel of workqueue */
2985 		if (!rtnl_trylock()) {
2986 			delay = 1;
2987 			should_notify_peers = false;
2988 			goto re_arm;
2989 		}
2990 
2991 		bond_for_each_slave(bond, slave, iter) {
2992 			bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2993 		}
2994 		bond_miimon_commit(bond);
2995 
2996 		rtnl_unlock();	/* might sleep, hold no other locks */
2997 	}
2998 
2999 re_arm:
3000 	if (bond->params.miimon)
3001 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
3002 
3003 	if (should_notify_peers) {
3004 		if (!rtnl_trylock())
3005 			return;
3006 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3007 		rtnl_unlock();
3008 	}
3009 }
3010 
3011 static int bond_upper_dev_walk(struct net_device *upper,
3012 			       struct netdev_nested_priv *priv)
3013 {
3014 	__be32 ip = *(__be32 *)priv->data;
3015 
3016 	return ip == bond_confirm_addr(upper, 0, ip);
3017 }
3018 
3019 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
3020 {
3021 	struct netdev_nested_priv priv = {
3022 		.data = (void *)&ip,
3023 	};
3024 	bool ret = false;
3025 
3026 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
3027 		return true;
3028 
3029 	rcu_read_lock();
3030 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
3031 		ret = true;
3032 	rcu_read_unlock();
3033 
3034 	return ret;
3035 }
3036 
3037 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
3038 
3039 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
3040 			     struct sk_buff *skb)
3041 {
3042 	struct net_device *bond_dev = slave->bond->dev;
3043 	struct net_device *slave_dev = slave->dev;
3044 	struct bond_vlan_tag *outer_tag = tags;
3045 
3046 	if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
3047 		return true;
3048 
3049 	tags++;
3050 
3051 	/* Go through all the tags backwards and add them to the packet */
3052 	while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
3053 		if (!tags->vlan_id) {
3054 			tags++;
3055 			continue;
3056 		}
3057 
3058 		slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
3059 			  ntohs(outer_tag->vlan_proto), tags->vlan_id);
3060 		skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
3061 						tags->vlan_id);
3062 		if (!skb) {
3063 			net_err_ratelimited("failed to insert inner VLAN tag\n");
3064 			return false;
3065 		}
3066 
3067 		tags++;
3068 	}
3069 	/* Set the outer tag */
3070 	if (outer_tag->vlan_id) {
3071 		slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
3072 			  ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
3073 		__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
3074 				       outer_tag->vlan_id);
3075 	}
3076 
3077 	return true;
3078 }
3079 
3080 /* We go to the (large) trouble of VLAN tagging ARP frames because
3081  * switches in VLAN mode (especially if ports are configured as
3082  * "native" to a VLAN) might not pass non-tagged frames.
3083  */
3084 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
3085 			  __be32 src_ip, struct bond_vlan_tag *tags)
3086 {
3087 	struct net_device *bond_dev = slave->bond->dev;
3088 	struct net_device *slave_dev = slave->dev;
3089 	struct sk_buff *skb;
3090 
3091 	slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
3092 		  arp_op, &dest_ip, &src_ip);
3093 
3094 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
3095 			 NULL, slave_dev->dev_addr, NULL);
3096 
3097 	if (!skb) {
3098 		net_err_ratelimited("ARP packet allocation failed\n");
3099 		return;
3100 	}
3101 
3102 	if (bond_handle_vlan(slave, tags, skb)) {
3103 		slave_update_last_tx(slave);
3104 		arp_xmit(skb);
3105 	}
3106 
3107 	return;
3108 }
3109 
3110 /* Validate the device path between the @start_dev and the @end_dev.
3111  * The path is valid if the @end_dev is reachable through device
3112  * stacking.
3113  * When the path is validated, collect any vlan information in the
3114  * path.
3115  */
3116 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
3117 					      struct net_device *end_dev,
3118 					      int level)
3119 {
3120 	struct bond_vlan_tag *tags;
3121 	struct net_device *upper;
3122 	struct list_head  *iter;
3123 
3124 	if (start_dev == end_dev) {
3125 		tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
3126 		if (!tags)
3127 			return ERR_PTR(-ENOMEM);
3128 		tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
3129 		return tags;
3130 	}
3131 
3132 	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
3133 		tags = bond_verify_device_path(upper, end_dev, level + 1);
3134 		if (IS_ERR_OR_NULL(tags)) {
3135 			if (IS_ERR(tags))
3136 				return tags;
3137 			continue;
3138 		}
3139 		if (is_vlan_dev(upper)) {
3140 			tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
3141 			tags[level].vlan_id = vlan_dev_vlan_id(upper);
3142 		}
3143 
3144 		return tags;
3145 	}
3146 
3147 	return NULL;
3148 }
3149 
3150 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
3151 {
3152 	struct rtable *rt;
3153 	struct bond_vlan_tag *tags;
3154 	__be32 *targets = bond->params.arp_targets, addr;
3155 	int i;
3156 
3157 	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3158 		slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3159 			  __func__, &targets[i]);
3160 		tags = NULL;
3161 
3162 		/* Find out through which dev should the packet go */
3163 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0,
3164 				     RT_SCOPE_LINK);
3165 		if (IS_ERR(rt)) {
3166 			/* there's no route to target - try to send arp
3167 			 * probe to generate any traffic (arp_validate=0)
3168 			 */
3169 			if (bond->params.arp_validate)
3170 				pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3171 					     bond->dev->name,
3172 					     &targets[i]);
3173 			bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3174 				      0, tags);
3175 			continue;
3176 		}
3177 
3178 		/* bond device itself */
3179 		if (rt->dst.dev == bond->dev)
3180 			goto found;
3181 
3182 		rcu_read_lock();
3183 		tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3184 		rcu_read_unlock();
3185 
3186 		if (!IS_ERR_OR_NULL(tags))
3187 			goto found;
3188 
3189 		/* Not our device - skip */
3190 		slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3191 			   &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3192 
3193 		ip_rt_put(rt);
3194 		continue;
3195 
3196 found:
3197 		addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3198 		ip_rt_put(rt);
3199 		bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3200 		kfree(tags);
3201 	}
3202 }
3203 
3204 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3205 {
3206 	int i;
3207 
3208 	if (!sip || !bond_has_this_ip(bond, tip)) {
3209 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3210 			   __func__, &sip, &tip);
3211 		return;
3212 	}
3213 
3214 	i = bond_get_targets_ip(bond->params.arp_targets, sip);
3215 	if (i == -1) {
3216 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3217 			   __func__, &sip);
3218 		return;
3219 	}
3220 	slave->last_rx = jiffies;
3221 	slave->target_last_arp_rx[i] = jiffies;
3222 }
3223 
3224 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3225 			struct slave *slave)
3226 {
3227 	struct arphdr *arp = (struct arphdr *)skb->data;
3228 	struct slave *curr_active_slave, *curr_arp_slave;
3229 	unsigned char *arp_ptr;
3230 	__be32 sip, tip;
3231 	unsigned int alen;
3232 
3233 	alen = arp_hdr_len(bond->dev);
3234 
3235 	if (alen > skb_headlen(skb)) {
3236 		arp = kmalloc(alen, GFP_ATOMIC);
3237 		if (!arp)
3238 			goto out_unlock;
3239 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
3240 			goto out_unlock;
3241 	}
3242 
3243 	if (arp->ar_hln != bond->dev->addr_len ||
3244 	    skb->pkt_type == PACKET_OTHERHOST ||
3245 	    skb->pkt_type == PACKET_LOOPBACK ||
3246 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
3247 	    arp->ar_pro != htons(ETH_P_IP) ||
3248 	    arp->ar_pln != 4)
3249 		goto out_unlock;
3250 
3251 	arp_ptr = (unsigned char *)(arp + 1);
3252 	arp_ptr += bond->dev->addr_len;
3253 	memcpy(&sip, arp_ptr, 4);
3254 	arp_ptr += 4 + bond->dev->addr_len;
3255 	memcpy(&tip, arp_ptr, 4);
3256 
3257 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3258 		  __func__, slave->dev->name, bond_slave_state(slave),
3259 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3260 		  &sip, &tip);
3261 
3262 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3263 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3264 
3265 	/* We 'trust' the received ARP enough to validate it if:
3266 	 *
3267 	 * (a) the slave receiving the ARP is active (which includes the
3268 	 * current ARP slave, if any), or
3269 	 *
3270 	 * (b) the receiving slave isn't active, but there is a currently
3271 	 * active slave and it received valid arp reply(s) after it became
3272 	 * the currently active slave, or
3273 	 *
3274 	 * (c) there is an ARP slave that sent an ARP during the prior ARP
3275 	 * interval, and we receive an ARP reply on any slave.  We accept
3276 	 * these because switch FDB update delays may deliver the ARP
3277 	 * reply to a slave other than the sender of the ARP request.
3278 	 *
3279 	 * Note: for (b), backup slaves are receiving the broadcast ARP
3280 	 * request, not a reply.  This request passes from the sending
3281 	 * slave through the L2 switch(es) to the receiving slave.  Since
3282 	 * this is checking the request, sip/tip are swapped for
3283 	 * validation.
3284 	 *
3285 	 * This is done to avoid endless looping when we can't reach the
3286 	 * arp_ip_target and fool ourselves with our own arp requests.
3287 	 */
3288 	if (bond_is_active_slave(slave))
3289 		bond_validate_arp(bond, slave, sip, tip);
3290 	else if (curr_active_slave &&
3291 		 time_after(slave_last_rx(bond, curr_active_slave),
3292 			    curr_active_slave->last_link_up))
3293 		bond_validate_arp(bond, slave, tip, sip);
3294 	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3295 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3296 		bond_validate_arp(bond, slave, sip, tip);
3297 
3298 out_unlock:
3299 	if (arp != (struct arphdr *)skb->data)
3300 		kfree(arp);
3301 	return RX_HANDLER_ANOTHER;
3302 }
3303 
3304 #if IS_ENABLED(CONFIG_IPV6)
3305 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3306 			 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3307 {
3308 	struct net_device *bond_dev = slave->bond->dev;
3309 	struct net_device *slave_dev = slave->dev;
3310 	struct in6_addr mcaddr;
3311 	struct sk_buff *skb;
3312 
3313 	slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3314 		  daddr, saddr);
3315 
3316 	skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3317 	if (!skb) {
3318 		net_err_ratelimited("NS packet allocation failed\n");
3319 		return;
3320 	}
3321 
3322 	addrconf_addr_solict_mult(daddr, &mcaddr);
3323 	if (bond_handle_vlan(slave, tags, skb)) {
3324 		slave_update_last_tx(slave);
3325 		ndisc_send_skb(skb, &mcaddr, saddr);
3326 	}
3327 }
3328 
3329 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3330 {
3331 	struct in6_addr *targets = bond->params.ns_targets;
3332 	struct bond_vlan_tag *tags;
3333 	struct dst_entry *dst;
3334 	struct in6_addr saddr;
3335 	struct flowi6 fl6;
3336 	int i;
3337 
3338 	for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3339 		slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3340 			  __func__, &targets[i]);
3341 		tags = NULL;
3342 
3343 		/* Find out through which dev should the packet go */
3344 		memset(&fl6, 0, sizeof(struct flowi6));
3345 		fl6.daddr = targets[i];
3346 		fl6.flowi6_oif = bond->dev->ifindex;
3347 
3348 		dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3349 		if (dst->error) {
3350 			dst_release(dst);
3351 			/* there's no route to target - try to send arp
3352 			 * probe to generate any traffic (arp_validate=0)
3353 			 */
3354 			if (bond->params.arp_validate)
3355 				pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3356 					     bond->dev->name,
3357 					     &targets[i]);
3358 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3359 			continue;
3360 		}
3361 
3362 		/* bond device itself */
3363 		if (dst->dev == bond->dev)
3364 			goto found;
3365 
3366 		rcu_read_lock();
3367 		tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3368 		rcu_read_unlock();
3369 
3370 		if (!IS_ERR_OR_NULL(tags))
3371 			goto found;
3372 
3373 		/* Not our device - skip */
3374 		slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3375 			  &targets[i], dst->dev ? dst->dev->name : "NULL");
3376 
3377 		dst_release(dst);
3378 		continue;
3379 
3380 found:
3381 		if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3382 			bond_ns_send(slave, &targets[i], &saddr, tags);
3383 		else
3384 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3385 
3386 		dst_release(dst);
3387 		kfree(tags);
3388 	}
3389 }
3390 
3391 static int bond_confirm_addr6(struct net_device *dev,
3392 			      struct netdev_nested_priv *priv)
3393 {
3394 	struct in6_addr *addr = (struct in6_addr *)priv->data;
3395 
3396 	return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3397 }
3398 
3399 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3400 {
3401 	struct netdev_nested_priv priv = {
3402 		.data = addr,
3403 	};
3404 	int ret = false;
3405 
3406 	if (bond_confirm_addr6(bond->dev, &priv))
3407 		return true;
3408 
3409 	rcu_read_lock();
3410 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3411 		ret = true;
3412 	rcu_read_unlock();
3413 
3414 	return ret;
3415 }
3416 
3417 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3418 			     struct in6_addr *saddr, struct in6_addr *daddr)
3419 {
3420 	int i;
3421 
3422 	/* Ignore NAs that:
3423 	 * 1. Source address is unspecified address.
3424 	 * 2. Dest address is neither all-nodes multicast address nor
3425 	 *    exist on bond interface.
3426 	 */
3427 	if (ipv6_addr_any(saddr) ||
3428 	    (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3429 	     !bond_has_this_ip6(bond, daddr))) {
3430 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3431 			  __func__, saddr, daddr);
3432 		return;
3433 	}
3434 
3435 	i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3436 	if (i == -1) {
3437 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3438 			  __func__, saddr);
3439 		return;
3440 	}
3441 	slave->last_rx = jiffies;
3442 	slave->target_last_arp_rx[i] = jiffies;
3443 }
3444 
3445 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3446 		       struct slave *slave)
3447 {
3448 	struct slave *curr_active_slave, *curr_arp_slave;
3449 	struct in6_addr *saddr, *daddr;
3450 	struct {
3451 		struct ipv6hdr ip6;
3452 		struct icmp6hdr icmp6;
3453 	} *combined, _combined;
3454 
3455 	if (skb->pkt_type == PACKET_OTHERHOST ||
3456 	    skb->pkt_type == PACKET_LOOPBACK)
3457 		goto out;
3458 
3459 	combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3460 	if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3461 	    (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3462 	     combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3463 		goto out;
3464 
3465 	saddr = &combined->ip6.saddr;
3466 	daddr = &combined->ip6.daddr;
3467 
3468 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3469 		  __func__, slave->dev->name, bond_slave_state(slave),
3470 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3471 		  saddr, daddr);
3472 
3473 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3474 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3475 
3476 	/* We 'trust' the received ARP enough to validate it if:
3477 	 * see bond_arp_rcv().
3478 	 */
3479 	if (bond_is_active_slave(slave))
3480 		bond_validate_na(bond, slave, saddr, daddr);
3481 	else if (curr_active_slave &&
3482 		 time_after(slave_last_rx(bond, curr_active_slave),
3483 			    curr_active_slave->last_link_up))
3484 		bond_validate_na(bond, slave, daddr, saddr);
3485 	else if (curr_arp_slave &&
3486 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3487 		bond_validate_na(bond, slave, saddr, daddr);
3488 
3489 out:
3490 	return RX_HANDLER_ANOTHER;
3491 }
3492 #endif
3493 
3494 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3495 		      struct slave *slave)
3496 {
3497 #if IS_ENABLED(CONFIG_IPV6)
3498 	bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3499 #endif
3500 	bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3501 
3502 	slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3503 		  __func__, skb->dev->name);
3504 
3505 	/* Use arp validate logic for both ARP and NS */
3506 	if (!slave_do_arp_validate(bond, slave)) {
3507 		if ((slave_do_arp_validate_only(bond) && is_arp) ||
3508 #if IS_ENABLED(CONFIG_IPV6)
3509 		    (slave_do_arp_validate_only(bond) && is_ipv6) ||
3510 #endif
3511 		    !slave_do_arp_validate_only(bond))
3512 			slave->last_rx = jiffies;
3513 		return RX_HANDLER_ANOTHER;
3514 	} else if (is_arp) {
3515 		return bond_arp_rcv(skb, bond, slave);
3516 #if IS_ENABLED(CONFIG_IPV6)
3517 	} else if (is_ipv6) {
3518 		return bond_na_rcv(skb, bond, slave);
3519 #endif
3520 	} else {
3521 		return RX_HANDLER_ANOTHER;
3522 	}
3523 }
3524 
3525 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3526 {
3527 	bond_arp_send_all(bond, slave);
3528 #if IS_ENABLED(CONFIG_IPV6)
3529 	bond_ns_send_all(bond, slave);
3530 #endif
3531 }
3532 
3533 /* function to verify if we're in the arp_interval timeslice, returns true if
3534  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3535  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3536  */
3537 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3538 				  int mod)
3539 {
3540 	int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3541 
3542 	return time_in_range(jiffies,
3543 			     last_act - delta_in_ticks,
3544 			     last_act + mod * delta_in_ticks + delta_in_ticks/2);
3545 }
3546 
3547 /* This function is called regularly to monitor each slave's link
3548  * ensuring that traffic is being sent and received when arp monitoring
3549  * is used in load-balancing mode. if the adapter has been dormant, then an
3550  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3551  * arp monitoring in active backup mode.
3552  */
3553 static void bond_loadbalance_arp_mon(struct bonding *bond)
3554 {
3555 	struct slave *slave, *oldcurrent;
3556 	struct list_head *iter;
3557 	int do_failover = 0, slave_state_changed = 0;
3558 
3559 	if (!bond_has_slaves(bond))
3560 		goto re_arm;
3561 
3562 	rcu_read_lock();
3563 
3564 	oldcurrent = rcu_dereference(bond->curr_active_slave);
3565 	/* see if any of the previous devices are up now (i.e. they have
3566 	 * xmt and rcv traffic). the curr_active_slave does not come into
3567 	 * the picture unless it is null. also, slave->last_link_up is not
3568 	 * needed here because we send an arp on each slave and give a slave
3569 	 * as long as it needs to get the tx/rx within the delta.
3570 	 * TODO: what about up/down delay in arp mode? it wasn't here before
3571 	 *       so it can wait
3572 	 */
3573 	bond_for_each_slave_rcu(bond, slave, iter) {
3574 		unsigned long last_tx = slave_last_tx(slave);
3575 
3576 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3577 
3578 		if (slave->link != BOND_LINK_UP) {
3579 			if (bond_time_in_interval(bond, last_tx, 1) &&
3580 			    bond_time_in_interval(bond, slave->last_rx, 1)) {
3581 
3582 				bond_propose_link_state(slave, BOND_LINK_UP);
3583 				slave_state_changed = 1;
3584 
3585 				/* primary_slave has no meaning in round-robin
3586 				 * mode. the window of a slave being up and
3587 				 * curr_active_slave being null after enslaving
3588 				 * is closed.
3589 				 */
3590 				if (!oldcurrent) {
3591 					slave_info(bond->dev, slave->dev, "link status definitely up\n");
3592 					do_failover = 1;
3593 				} else {
3594 					slave_info(bond->dev, slave->dev, "interface is now up\n");
3595 				}
3596 			}
3597 		} else {
3598 			/* slave->link == BOND_LINK_UP */
3599 
3600 			/* not all switches will respond to an arp request
3601 			 * when the source ip is 0, so don't take the link down
3602 			 * if we don't know our ip yet
3603 			 */
3604 			if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3605 			    !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3606 
3607 				bond_propose_link_state(slave, BOND_LINK_DOWN);
3608 				slave_state_changed = 1;
3609 
3610 				if (slave->link_failure_count < UINT_MAX)
3611 					slave->link_failure_count++;
3612 
3613 				slave_info(bond->dev, slave->dev, "interface is now down\n");
3614 
3615 				if (slave == oldcurrent)
3616 					do_failover = 1;
3617 			}
3618 		}
3619 
3620 		/* note: if switch is in round-robin mode, all links
3621 		 * must tx arp to ensure all links rx an arp - otherwise
3622 		 * links may oscillate or not come up at all; if switch is
3623 		 * in something like xor mode, there is nothing we can
3624 		 * do - all replies will be rx'ed on same link causing slaves
3625 		 * to be unstable during low/no traffic periods
3626 		 */
3627 		if (bond_slave_is_up(slave))
3628 			bond_send_validate(bond, slave);
3629 	}
3630 
3631 	rcu_read_unlock();
3632 
3633 	if (do_failover || slave_state_changed) {
3634 		if (!rtnl_trylock())
3635 			goto re_arm;
3636 
3637 		bond_for_each_slave(bond, slave, iter) {
3638 			if (slave->link_new_state != BOND_LINK_NOCHANGE)
3639 				slave->link = slave->link_new_state;
3640 		}
3641 
3642 		if (slave_state_changed) {
3643 			bond_slave_state_change(bond);
3644 			if (BOND_MODE(bond) == BOND_MODE_XOR)
3645 				bond_update_slave_arr(bond, NULL);
3646 		}
3647 		if (do_failover) {
3648 			block_netpoll_tx();
3649 			bond_select_active_slave(bond);
3650 			unblock_netpoll_tx();
3651 		}
3652 		rtnl_unlock();
3653 	}
3654 
3655 re_arm:
3656 	if (bond->params.arp_interval)
3657 		queue_delayed_work(bond->wq, &bond->arp_work,
3658 				   msecs_to_jiffies(bond->params.arp_interval));
3659 }
3660 
3661 /* Called to inspect slaves for active-backup mode ARP monitor link state
3662  * changes.  Sets proposed link state in slaves to specify what action
3663  * should take place for the slave.  Returns 0 if no changes are found, >0
3664  * if changes to link states must be committed.
3665  *
3666  * Called with rcu_read_lock held.
3667  */
3668 static int bond_ab_arp_inspect(struct bonding *bond)
3669 {
3670 	unsigned long last_tx, last_rx;
3671 	struct list_head *iter;
3672 	struct slave *slave;
3673 	int commit = 0;
3674 
3675 	bond_for_each_slave_rcu(bond, slave, iter) {
3676 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3677 		last_rx = slave_last_rx(bond, slave);
3678 
3679 		if (slave->link != BOND_LINK_UP) {
3680 			if (bond_time_in_interval(bond, last_rx, 1)) {
3681 				bond_propose_link_state(slave, BOND_LINK_UP);
3682 				commit++;
3683 			} else if (slave->link == BOND_LINK_BACK) {
3684 				bond_propose_link_state(slave, BOND_LINK_FAIL);
3685 				commit++;
3686 			}
3687 			continue;
3688 		}
3689 
3690 		/* Give slaves 2*delta after being enslaved or made
3691 		 * active.  This avoids bouncing, as the last receive
3692 		 * times need a full ARP monitor cycle to be updated.
3693 		 */
3694 		if (bond_time_in_interval(bond, slave->last_link_up, 2))
3695 			continue;
3696 
3697 		/* Backup slave is down if:
3698 		 * - No current_arp_slave AND
3699 		 * - more than (missed_max+1)*delta since last receive AND
3700 		 * - the bond has an IP address
3701 		 *
3702 		 * Note: a non-null current_arp_slave indicates
3703 		 * the curr_active_slave went down and we are
3704 		 * searching for a new one; under this condition
3705 		 * we only take the curr_active_slave down - this
3706 		 * gives each slave a chance to tx/rx traffic
3707 		 * before being taken out
3708 		 */
3709 		if (!bond_is_active_slave(slave) &&
3710 		    !rcu_access_pointer(bond->current_arp_slave) &&
3711 		    !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3712 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3713 			commit++;
3714 		}
3715 
3716 		/* Active slave is down if:
3717 		 * - more than missed_max*delta since transmitting OR
3718 		 * - (more than missed_max*delta since receive AND
3719 		 *    the bond has an IP address)
3720 		 */
3721 		last_tx = slave_last_tx(slave);
3722 		if (bond_is_active_slave(slave) &&
3723 		    (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3724 		     !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3725 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3726 			commit++;
3727 		}
3728 	}
3729 
3730 	return commit;
3731 }
3732 
3733 /* Called to commit link state changes noted by inspection step of
3734  * active-backup mode ARP monitor.
3735  *
3736  * Called with RTNL hold.
3737  */
3738 static void bond_ab_arp_commit(struct bonding *bond)
3739 {
3740 	bool do_failover = false;
3741 	struct list_head *iter;
3742 	unsigned long last_tx;
3743 	struct slave *slave;
3744 
3745 	bond_for_each_slave(bond, slave, iter) {
3746 		switch (slave->link_new_state) {
3747 		case BOND_LINK_NOCHANGE:
3748 			continue;
3749 
3750 		case BOND_LINK_UP:
3751 			last_tx = slave_last_tx(slave);
3752 			if (rtnl_dereference(bond->curr_active_slave) != slave ||
3753 			    (!rtnl_dereference(bond->curr_active_slave) &&
3754 			     bond_time_in_interval(bond, last_tx, 1))) {
3755 				struct slave *current_arp_slave;
3756 
3757 				current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3758 				bond_set_slave_link_state(slave, BOND_LINK_UP,
3759 							  BOND_SLAVE_NOTIFY_NOW);
3760 				if (current_arp_slave) {
3761 					bond_set_slave_inactive_flags(
3762 						current_arp_slave,
3763 						BOND_SLAVE_NOTIFY_NOW);
3764 					RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3765 				}
3766 
3767 				slave_info(bond->dev, slave->dev, "link status definitely up\n");
3768 
3769 				if (!rtnl_dereference(bond->curr_active_slave) ||
3770 				    slave == rtnl_dereference(bond->primary_slave) ||
3771 				    slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3772 					do_failover = true;
3773 
3774 			}
3775 
3776 			continue;
3777 
3778 		case BOND_LINK_DOWN:
3779 			if (slave->link_failure_count < UINT_MAX)
3780 				slave->link_failure_count++;
3781 
3782 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3783 						  BOND_SLAVE_NOTIFY_NOW);
3784 			bond_set_slave_inactive_flags(slave,
3785 						      BOND_SLAVE_NOTIFY_NOW);
3786 
3787 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3788 
3789 			if (slave == rtnl_dereference(bond->curr_active_slave)) {
3790 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3791 				do_failover = true;
3792 			}
3793 
3794 			continue;
3795 
3796 		case BOND_LINK_FAIL:
3797 			bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3798 						  BOND_SLAVE_NOTIFY_NOW);
3799 			bond_set_slave_inactive_flags(slave,
3800 						      BOND_SLAVE_NOTIFY_NOW);
3801 
3802 			/* A slave has just been enslaved and has become
3803 			 * the current active slave.
3804 			 */
3805 			if (rtnl_dereference(bond->curr_active_slave))
3806 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3807 			continue;
3808 
3809 		default:
3810 			slave_err(bond->dev, slave->dev,
3811 				  "impossible: link_new_state %d on slave\n",
3812 				  slave->link_new_state);
3813 			continue;
3814 		}
3815 	}
3816 
3817 	if (do_failover) {
3818 		block_netpoll_tx();
3819 		bond_select_active_slave(bond);
3820 		unblock_netpoll_tx();
3821 	}
3822 
3823 	bond_set_carrier(bond);
3824 }
3825 
3826 /* Send ARP probes for active-backup mode ARP monitor.
3827  *
3828  * Called with rcu_read_lock held.
3829  */
3830 static bool bond_ab_arp_probe(struct bonding *bond)
3831 {
3832 	struct slave *slave, *before = NULL, *new_slave = NULL,
3833 		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3834 		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3835 	struct list_head *iter;
3836 	bool found = false;
3837 	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3838 
3839 	if (curr_arp_slave && curr_active_slave)
3840 		netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3841 			    curr_arp_slave->dev->name,
3842 			    curr_active_slave->dev->name);
3843 
3844 	if (curr_active_slave) {
3845 		bond_send_validate(bond, curr_active_slave);
3846 		return should_notify_rtnl;
3847 	}
3848 
3849 	/* if we don't have a curr_active_slave, search for the next available
3850 	 * backup slave from the current_arp_slave and make it the candidate
3851 	 * for becoming the curr_active_slave
3852 	 */
3853 
3854 	if (!curr_arp_slave) {
3855 		curr_arp_slave = bond_first_slave_rcu(bond);
3856 		if (!curr_arp_slave)
3857 			return should_notify_rtnl;
3858 	}
3859 
3860 	bond_for_each_slave_rcu(bond, slave, iter) {
3861 		if (!found && !before && bond_slave_is_up(slave))
3862 			before = slave;
3863 
3864 		if (found && !new_slave && bond_slave_is_up(slave))
3865 			new_slave = slave;
3866 		/* if the link state is up at this point, we
3867 		 * mark it down - this can happen if we have
3868 		 * simultaneous link failures and
3869 		 * reselect_active_interface doesn't make this
3870 		 * one the current slave so it is still marked
3871 		 * up when it is actually down
3872 		 */
3873 		if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3874 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3875 						  BOND_SLAVE_NOTIFY_LATER);
3876 			if (slave->link_failure_count < UINT_MAX)
3877 				slave->link_failure_count++;
3878 
3879 			bond_set_slave_inactive_flags(slave,
3880 						      BOND_SLAVE_NOTIFY_LATER);
3881 
3882 			slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3883 		}
3884 		if (slave == curr_arp_slave)
3885 			found = true;
3886 	}
3887 
3888 	if (!new_slave && before)
3889 		new_slave = before;
3890 
3891 	if (!new_slave)
3892 		goto check_state;
3893 
3894 	bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3895 				  BOND_SLAVE_NOTIFY_LATER);
3896 	bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3897 	bond_send_validate(bond, new_slave);
3898 	new_slave->last_link_up = jiffies;
3899 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
3900 
3901 check_state:
3902 	bond_for_each_slave_rcu(bond, slave, iter) {
3903 		if (slave->should_notify || slave->should_notify_link) {
3904 			should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3905 			break;
3906 		}
3907 	}
3908 	return should_notify_rtnl;
3909 }
3910 
3911 static void bond_activebackup_arp_mon(struct bonding *bond)
3912 {
3913 	bool should_notify_peers = false;
3914 	bool should_notify_rtnl = false;
3915 	int delta_in_ticks;
3916 
3917 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3918 
3919 	if (!bond_has_slaves(bond))
3920 		goto re_arm;
3921 
3922 	rcu_read_lock();
3923 
3924 	should_notify_peers = bond_should_notify_peers(bond);
3925 
3926 	if (bond_ab_arp_inspect(bond)) {
3927 		rcu_read_unlock();
3928 
3929 		/* Race avoidance with bond_close flush of workqueue */
3930 		if (!rtnl_trylock()) {
3931 			delta_in_ticks = 1;
3932 			should_notify_peers = false;
3933 			goto re_arm;
3934 		}
3935 
3936 		bond_ab_arp_commit(bond);
3937 
3938 		rtnl_unlock();
3939 		rcu_read_lock();
3940 	}
3941 
3942 	should_notify_rtnl = bond_ab_arp_probe(bond);
3943 	rcu_read_unlock();
3944 
3945 re_arm:
3946 	if (bond->params.arp_interval)
3947 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3948 
3949 	if (should_notify_peers || should_notify_rtnl) {
3950 		if (!rtnl_trylock())
3951 			return;
3952 
3953 		if (should_notify_peers) {
3954 			bond->send_peer_notif--;
3955 			call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3956 						 bond->dev);
3957 		}
3958 		if (should_notify_rtnl) {
3959 			bond_slave_state_notify(bond);
3960 			bond_slave_link_notify(bond);
3961 		}
3962 
3963 		rtnl_unlock();
3964 	}
3965 }
3966 
3967 static void bond_arp_monitor(struct work_struct *work)
3968 {
3969 	struct bonding *bond = container_of(work, struct bonding,
3970 					    arp_work.work);
3971 
3972 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3973 		bond_activebackup_arp_mon(bond);
3974 	else
3975 		bond_loadbalance_arp_mon(bond);
3976 }
3977 
3978 /*-------------------------- netdev event handling --------------------------*/
3979 
3980 /* Change device name */
3981 static int bond_event_changename(struct bonding *bond)
3982 {
3983 	bond_remove_proc_entry(bond);
3984 	bond_create_proc_entry(bond);
3985 
3986 	bond_debug_reregister(bond);
3987 
3988 	return NOTIFY_DONE;
3989 }
3990 
3991 static int bond_master_netdev_event(unsigned long event,
3992 				    struct net_device *bond_dev)
3993 {
3994 	struct bonding *event_bond = netdev_priv(bond_dev);
3995 
3996 	netdev_dbg(bond_dev, "%s called\n", __func__);
3997 
3998 	switch (event) {
3999 	case NETDEV_CHANGENAME:
4000 		return bond_event_changename(event_bond);
4001 	case NETDEV_UNREGISTER:
4002 		bond_remove_proc_entry(event_bond);
4003 #ifdef CONFIG_XFRM_OFFLOAD
4004 		xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
4005 #endif /* CONFIG_XFRM_OFFLOAD */
4006 		break;
4007 	case NETDEV_REGISTER:
4008 		bond_create_proc_entry(event_bond);
4009 		break;
4010 	default:
4011 		break;
4012 	}
4013 
4014 	return NOTIFY_DONE;
4015 }
4016 
4017 static int bond_slave_netdev_event(unsigned long event,
4018 				   struct net_device *slave_dev)
4019 {
4020 	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
4021 	struct bonding *bond;
4022 	struct net_device *bond_dev;
4023 
4024 	/* A netdev event can be generated while enslaving a device
4025 	 * before netdev_rx_handler_register is called in which case
4026 	 * slave will be NULL
4027 	 */
4028 	if (!slave) {
4029 		netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
4030 		return NOTIFY_DONE;
4031 	}
4032 
4033 	bond_dev = slave->bond->dev;
4034 	bond = slave->bond;
4035 	primary = rtnl_dereference(bond->primary_slave);
4036 
4037 	slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
4038 
4039 	switch (event) {
4040 	case NETDEV_UNREGISTER:
4041 		if (bond_dev->type != ARPHRD_ETHER)
4042 			bond_release_and_destroy(bond_dev, slave_dev);
4043 		else
4044 			__bond_release_one(bond_dev, slave_dev, false, true);
4045 		break;
4046 	case NETDEV_UP:
4047 	case NETDEV_CHANGE:
4048 		/* For 802.3ad mode only:
4049 		 * Getting invalid Speed/Duplex values here will put slave
4050 		 * in weird state. Mark it as link-fail if the link was
4051 		 * previously up or link-down if it hasn't yet come up, and
4052 		 * let link-monitoring (miimon) set it right when correct
4053 		 * speeds/duplex are available.
4054 		 */
4055 		if (bond_update_speed_duplex(slave) &&
4056 		    BOND_MODE(bond) == BOND_MODE_8023AD) {
4057 			if (slave->last_link_up)
4058 				slave->link = BOND_LINK_FAIL;
4059 			else
4060 				slave->link = BOND_LINK_DOWN;
4061 		}
4062 
4063 		if (BOND_MODE(bond) == BOND_MODE_8023AD)
4064 			bond_3ad_adapter_speed_duplex_changed(slave);
4065 		fallthrough;
4066 	case NETDEV_DOWN:
4067 		/* Refresh slave-array if applicable!
4068 		 * If the setup does not use miimon or arpmon (mode-specific!),
4069 		 * then these events will not cause the slave-array to be
4070 		 * refreshed. This will cause xmit to use a slave that is not
4071 		 * usable. Avoid such situation by refeshing the array at these
4072 		 * events. If these (miimon/arpmon) parameters are configured
4073 		 * then array gets refreshed twice and that should be fine!
4074 		 */
4075 		if (bond_mode_can_use_xmit_hash(bond))
4076 			bond_update_slave_arr(bond, NULL);
4077 		break;
4078 	case NETDEV_CHANGEMTU:
4079 		/* TODO: Should slaves be allowed to
4080 		 * independently alter their MTU?  For
4081 		 * an active-backup bond, slaves need
4082 		 * not be the same type of device, so
4083 		 * MTUs may vary.  For other modes,
4084 		 * slaves arguably should have the
4085 		 * same MTUs. To do this, we'd need to
4086 		 * take over the slave's change_mtu
4087 		 * function for the duration of their
4088 		 * servitude.
4089 		 */
4090 		break;
4091 	case NETDEV_CHANGENAME:
4092 		/* we don't care if we don't have primary set */
4093 		if (!bond_uses_primary(bond) ||
4094 		    !bond->params.primary[0])
4095 			break;
4096 
4097 		if (slave == primary) {
4098 			/* slave's name changed - he's no longer primary */
4099 			RCU_INIT_POINTER(bond->primary_slave, NULL);
4100 		} else if (!strcmp(slave_dev->name, bond->params.primary)) {
4101 			/* we have a new primary slave */
4102 			rcu_assign_pointer(bond->primary_slave, slave);
4103 		} else { /* we didn't change primary - exit */
4104 			break;
4105 		}
4106 
4107 		netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
4108 			    primary ? slave_dev->name : "none");
4109 
4110 		block_netpoll_tx();
4111 		bond_select_active_slave(bond);
4112 		unblock_netpoll_tx();
4113 		break;
4114 	case NETDEV_FEAT_CHANGE:
4115 		if (!bond->notifier_ctx) {
4116 			bond->notifier_ctx = true;
4117 			bond_compute_features(bond);
4118 			bond->notifier_ctx = false;
4119 		}
4120 		break;
4121 	case NETDEV_RESEND_IGMP:
4122 		/* Propagate to master device */
4123 		call_netdevice_notifiers(event, slave->bond->dev);
4124 		break;
4125 	case NETDEV_XDP_FEAT_CHANGE:
4126 		bond_xdp_set_features(bond_dev);
4127 		break;
4128 	default:
4129 		break;
4130 	}
4131 
4132 	return NOTIFY_DONE;
4133 }
4134 
4135 /* bond_netdev_event: handle netdev notifier chain events.
4136  *
4137  * This function receives events for the netdev chain.  The caller (an
4138  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
4139  * locks for us to safely manipulate the slave devices (RTNL lock,
4140  * dev_probe_lock).
4141  */
4142 static int bond_netdev_event(struct notifier_block *this,
4143 			     unsigned long event, void *ptr)
4144 {
4145 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
4146 
4147 	netdev_dbg(event_dev, "%s received %s\n",
4148 		   __func__, netdev_cmd_to_name(event));
4149 
4150 	if (!(event_dev->priv_flags & IFF_BONDING))
4151 		return NOTIFY_DONE;
4152 
4153 	if (event_dev->flags & IFF_MASTER) {
4154 		int ret;
4155 
4156 		ret = bond_master_netdev_event(event, event_dev);
4157 		if (ret != NOTIFY_DONE)
4158 			return ret;
4159 	}
4160 
4161 	if (event_dev->flags & IFF_SLAVE)
4162 		return bond_slave_netdev_event(event, event_dev);
4163 
4164 	return NOTIFY_DONE;
4165 }
4166 
4167 static struct notifier_block bond_netdev_notifier = {
4168 	.notifier_call = bond_netdev_event,
4169 };
4170 
4171 /*---------------------------- Hashing Policies -----------------------------*/
4172 
4173 /* Helper to access data in a packet, with or without a backing skb.
4174  * If skb is given the data is linearized if necessary via pskb_may_pull.
4175  */
4176 static inline const void *bond_pull_data(struct sk_buff *skb,
4177 					 const void *data, int hlen, int n)
4178 {
4179 	if (likely(n <= hlen))
4180 		return data;
4181 	else if (skb && likely(pskb_may_pull(skb, n)))
4182 		return skb->data;
4183 
4184 	return NULL;
4185 }
4186 
4187 /* L2 hash helper */
4188 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4189 {
4190 	struct ethhdr *ep;
4191 
4192 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4193 	if (!data)
4194 		return 0;
4195 
4196 	ep = (struct ethhdr *)(data + mhoff);
4197 	return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4198 }
4199 
4200 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4201 			 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4202 {
4203 	const struct ipv6hdr *iph6;
4204 	const struct iphdr *iph;
4205 
4206 	if (l2_proto == htons(ETH_P_IP)) {
4207 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4208 		if (!data)
4209 			return false;
4210 
4211 		iph = (const struct iphdr *)(data + *nhoff);
4212 		iph_to_flow_copy_v4addrs(fk, iph);
4213 		*nhoff += iph->ihl << 2;
4214 		if (!ip_is_fragment(iph))
4215 			*ip_proto = iph->protocol;
4216 	} else if (l2_proto == htons(ETH_P_IPV6)) {
4217 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4218 		if (!data)
4219 			return false;
4220 
4221 		iph6 = (const struct ipv6hdr *)(data + *nhoff);
4222 		iph_to_flow_copy_v6addrs(fk, iph6);
4223 		*nhoff += sizeof(*iph6);
4224 		*ip_proto = iph6->nexthdr;
4225 	} else {
4226 		return false;
4227 	}
4228 
4229 	if (l34 && *ip_proto >= 0)
4230 		fk->ports.ports = skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4231 
4232 	return true;
4233 }
4234 
4235 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4236 {
4237 	u32 srcmac_vendor = 0, srcmac_dev = 0;
4238 	struct ethhdr *mac_hdr;
4239 	u16 vlan = 0;
4240 	int i;
4241 
4242 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4243 	if (!data)
4244 		return 0;
4245 	mac_hdr = (struct ethhdr *)(data + mhoff);
4246 
4247 	for (i = 0; i < 3; i++)
4248 		srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4249 
4250 	for (i = 3; i < ETH_ALEN; i++)
4251 		srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4252 
4253 	if (skb && skb_vlan_tag_present(skb))
4254 		vlan = skb_vlan_tag_get(skb);
4255 
4256 	return vlan ^ srcmac_vendor ^ srcmac_dev;
4257 }
4258 
4259 /* Extract the appropriate headers based on bond's xmit policy */
4260 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4261 			      __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4262 {
4263 	bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4264 	int ip_proto = -1;
4265 
4266 	switch (bond->params.xmit_policy) {
4267 	case BOND_XMIT_POLICY_ENCAP23:
4268 	case BOND_XMIT_POLICY_ENCAP34:
4269 		memset(fk, 0, sizeof(*fk));
4270 		return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4271 					  fk, data, l2_proto, nhoff, hlen, 0);
4272 	default:
4273 		break;
4274 	}
4275 
4276 	fk->ports.ports = 0;
4277 	memset(&fk->icmp, 0, sizeof(fk->icmp));
4278 	if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4279 		return false;
4280 
4281 	/* ICMP error packets contains at least 8 bytes of the header
4282 	 * of the packet which generated the error. Use this information
4283 	 * to correlate ICMP error packets within the same flow which
4284 	 * generated the error.
4285 	 */
4286 	if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4287 		skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4288 		if (ip_proto == IPPROTO_ICMP) {
4289 			if (!icmp_is_err(fk->icmp.type))
4290 				return true;
4291 
4292 			nhoff += sizeof(struct icmphdr);
4293 		} else if (ip_proto == IPPROTO_ICMPV6) {
4294 			if (!icmpv6_is_err(fk->icmp.type))
4295 				return true;
4296 
4297 			nhoff += sizeof(struct icmp6hdr);
4298 		}
4299 		return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4300 	}
4301 
4302 	return true;
4303 }
4304 
4305 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4306 {
4307 	hash ^= (__force u32)flow_get_u32_dst(flow) ^
4308 		(__force u32)flow_get_u32_src(flow);
4309 	hash ^= (hash >> 16);
4310 	hash ^= (hash >> 8);
4311 
4312 	/* discard lowest hash bit to deal with the common even ports pattern */
4313 	if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4314 		xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4315 		return hash >> 1;
4316 
4317 	return hash;
4318 }
4319 
4320 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4321  * the data as required, but this function can be used without it if the data is
4322  * known to be linear (e.g. with xdp_buff).
4323  */
4324 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4325 			    __be16 l2_proto, int mhoff, int nhoff, int hlen)
4326 {
4327 	struct flow_keys flow;
4328 	u32 hash;
4329 
4330 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4331 		return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4332 
4333 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4334 	    !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4335 		return bond_eth_hash(skb, data, mhoff, hlen);
4336 
4337 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4338 	    bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4339 		hash = bond_eth_hash(skb, data, mhoff, hlen);
4340 	} else {
4341 		if (flow.icmp.id)
4342 			memcpy(&hash, &flow.icmp, sizeof(hash));
4343 		else
4344 			memcpy(&hash, &flow.ports.ports, sizeof(hash));
4345 	}
4346 
4347 	return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4348 }
4349 
4350 /**
4351  * bond_xmit_hash - generate a hash value based on the xmit policy
4352  * @bond: bonding device
4353  * @skb: buffer to use for headers
4354  *
4355  * This function will extract the necessary headers from the skb buffer and use
4356  * them to generate a hash based on the xmit_policy set in the bonding device
4357  */
4358 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4359 {
4360 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4361 	    skb->l4_hash)
4362 		return skb->hash;
4363 
4364 	return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4365 				0, skb_network_offset(skb),
4366 				skb_headlen(skb));
4367 }
4368 
4369 /**
4370  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4371  * @bond: bonding device
4372  * @xdp: buffer to use for headers
4373  *
4374  * The XDP variant of bond_xmit_hash.
4375  */
4376 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4377 {
4378 	struct ethhdr *eth;
4379 
4380 	if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4381 		return 0;
4382 
4383 	eth = (struct ethhdr *)xdp->data;
4384 
4385 	return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4386 				sizeof(struct ethhdr), xdp->data_end - xdp->data);
4387 }
4388 
4389 /*-------------------------- Device entry points ----------------------------*/
4390 
4391 void bond_work_init_all(struct bonding *bond)
4392 {
4393 	INIT_DELAYED_WORK(&bond->mcast_work,
4394 			  bond_resend_igmp_join_requests_delayed);
4395 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4396 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4397 	INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4398 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4399 	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4400 }
4401 
4402 static void bond_work_cancel_all(struct bonding *bond)
4403 {
4404 	cancel_delayed_work_sync(&bond->mii_work);
4405 	cancel_delayed_work_sync(&bond->arp_work);
4406 	cancel_delayed_work_sync(&bond->alb_work);
4407 	cancel_delayed_work_sync(&bond->ad_work);
4408 	cancel_delayed_work_sync(&bond->mcast_work);
4409 	cancel_delayed_work_sync(&bond->slave_arr_work);
4410 }
4411 
4412 static int bond_open(struct net_device *bond_dev)
4413 {
4414 	struct bonding *bond = netdev_priv(bond_dev);
4415 	struct list_head *iter;
4416 	struct slave *slave;
4417 
4418 	if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4419 		bond->rr_tx_counter = alloc_percpu(u32);
4420 		if (!bond->rr_tx_counter)
4421 			return -ENOMEM;
4422 	}
4423 
4424 	/* reset slave->backup and slave->inactive */
4425 	if (bond_has_slaves(bond)) {
4426 		bond_for_each_slave(bond, slave, iter) {
4427 			if (bond_uses_primary(bond) &&
4428 			    slave != rcu_access_pointer(bond->curr_active_slave)) {
4429 				bond_set_slave_inactive_flags(slave,
4430 							      BOND_SLAVE_NOTIFY_NOW);
4431 			} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4432 				bond_set_slave_active_flags(slave,
4433 							    BOND_SLAVE_NOTIFY_NOW);
4434 			}
4435 		}
4436 	}
4437 
4438 	if (bond_is_lb(bond)) {
4439 		/* bond_alb_initialize must be called before the timer
4440 		 * is started.
4441 		 */
4442 		if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4443 			return -ENOMEM;
4444 		if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4445 			queue_delayed_work(bond->wq, &bond->alb_work, 0);
4446 	}
4447 
4448 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
4449 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
4450 
4451 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4452 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
4453 		bond->recv_probe = bond_rcv_validate;
4454 	}
4455 
4456 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4457 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
4458 		/* register to receive LACPDUs */
4459 		bond->recv_probe = bond_3ad_lacpdu_recv;
4460 		bond_3ad_initiate_agg_selection(bond, 1);
4461 
4462 		bond_for_each_slave(bond, slave, iter)
4463 			dev_mc_add(slave->dev, lacpdu_mcast_addr);
4464 	}
4465 
4466 	if (bond_mode_can_use_xmit_hash(bond))
4467 		bond_update_slave_arr(bond, NULL);
4468 
4469 	return 0;
4470 }
4471 
4472 static int bond_close(struct net_device *bond_dev)
4473 {
4474 	struct bonding *bond = netdev_priv(bond_dev);
4475 	struct slave *slave;
4476 
4477 	bond_work_cancel_all(bond);
4478 	bond->send_peer_notif = 0;
4479 	if (bond_is_lb(bond))
4480 		bond_alb_deinitialize(bond);
4481 	bond->recv_probe = NULL;
4482 
4483 	if (bond_uses_primary(bond)) {
4484 		rcu_read_lock();
4485 		slave = rcu_dereference(bond->curr_active_slave);
4486 		if (slave)
4487 			bond_hw_addr_flush(bond_dev, slave->dev);
4488 		rcu_read_unlock();
4489 	} else {
4490 		struct list_head *iter;
4491 
4492 		bond_for_each_slave(bond, slave, iter)
4493 			bond_hw_addr_flush(bond_dev, slave->dev);
4494 	}
4495 
4496 	return 0;
4497 }
4498 
4499 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4500  * that some drivers can provide 32bit values only.
4501  */
4502 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4503 			    const struct rtnl_link_stats64 *_new,
4504 			    const struct rtnl_link_stats64 *_old)
4505 {
4506 	const u64 *new = (const u64 *)_new;
4507 	const u64 *old = (const u64 *)_old;
4508 	u64 *res = (u64 *)_res;
4509 	int i;
4510 
4511 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4512 		u64 nv = new[i];
4513 		u64 ov = old[i];
4514 		s64 delta = nv - ov;
4515 
4516 		/* detects if this particular field is 32bit only */
4517 		if (((nv | ov) >> 32) == 0)
4518 			delta = (s64)(s32)((u32)nv - (u32)ov);
4519 
4520 		/* filter anomalies, some drivers reset their stats
4521 		 * at down/up events.
4522 		 */
4523 		if (delta > 0)
4524 			res[i] += delta;
4525 	}
4526 }
4527 
4528 #ifdef CONFIG_LOCKDEP
4529 static int bond_get_lowest_level_rcu(struct net_device *dev)
4530 {
4531 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4532 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4533 	int cur = 0, max = 0;
4534 
4535 	now = dev;
4536 	iter = &dev->adj_list.lower;
4537 
4538 	while (1) {
4539 		next = NULL;
4540 		while (1) {
4541 			ldev = netdev_next_lower_dev_rcu(now, &iter);
4542 			if (!ldev)
4543 				break;
4544 
4545 			next = ldev;
4546 			niter = &ldev->adj_list.lower;
4547 			dev_stack[cur] = now;
4548 			iter_stack[cur++] = iter;
4549 			if (max <= cur)
4550 				max = cur;
4551 			break;
4552 		}
4553 
4554 		if (!next) {
4555 			if (!cur)
4556 				return max;
4557 			next = dev_stack[--cur];
4558 			niter = iter_stack[cur];
4559 		}
4560 
4561 		now = next;
4562 		iter = niter;
4563 	}
4564 
4565 	return max;
4566 }
4567 #endif
4568 
4569 static void bond_get_stats(struct net_device *bond_dev,
4570 			   struct rtnl_link_stats64 *stats)
4571 {
4572 	struct bonding *bond = netdev_priv(bond_dev);
4573 	struct rtnl_link_stats64 temp;
4574 	struct list_head *iter;
4575 	struct slave *slave;
4576 	int nest_level = 0;
4577 
4578 
4579 	rcu_read_lock();
4580 #ifdef CONFIG_LOCKDEP
4581 	nest_level = bond_get_lowest_level_rcu(bond_dev);
4582 #endif
4583 
4584 	spin_lock_nested(&bond->stats_lock, nest_level);
4585 	memcpy(stats, &bond->bond_stats, sizeof(*stats));
4586 
4587 	bond_for_each_slave_rcu(bond, slave, iter) {
4588 		const struct rtnl_link_stats64 *new =
4589 			dev_get_stats(slave->dev, &temp);
4590 
4591 		bond_fold_stats(stats, new, &slave->slave_stats);
4592 
4593 		/* save off the slave stats for the next run */
4594 		memcpy(&slave->slave_stats, new, sizeof(*new));
4595 	}
4596 
4597 	memcpy(&bond->bond_stats, stats, sizeof(*stats));
4598 	spin_unlock(&bond->stats_lock);
4599 	rcu_read_unlock();
4600 }
4601 
4602 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4603 {
4604 	struct bonding *bond = netdev_priv(bond_dev);
4605 	struct mii_ioctl_data *mii = NULL;
4606 
4607 	netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4608 
4609 	switch (cmd) {
4610 	case SIOCGMIIPHY:
4611 		mii = if_mii(ifr);
4612 		if (!mii)
4613 			return -EINVAL;
4614 
4615 		mii->phy_id = 0;
4616 		fallthrough;
4617 	case SIOCGMIIREG:
4618 		/* We do this again just in case we were called by SIOCGMIIREG
4619 		 * instead of SIOCGMIIPHY.
4620 		 */
4621 		mii = if_mii(ifr);
4622 		if (!mii)
4623 			return -EINVAL;
4624 
4625 		if (mii->reg_num == 1) {
4626 			mii->val_out = 0;
4627 			if (netif_carrier_ok(bond->dev))
4628 				mii->val_out = BMSR_LSTATUS;
4629 		}
4630 
4631 		break;
4632 	default:
4633 		return -EOPNOTSUPP;
4634 	}
4635 
4636 	return 0;
4637 }
4638 
4639 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4640 {
4641 	struct bonding *bond = netdev_priv(bond_dev);
4642 	struct net_device *slave_dev = NULL;
4643 	struct ifbond k_binfo;
4644 	struct ifbond __user *u_binfo = NULL;
4645 	struct ifslave k_sinfo;
4646 	struct ifslave __user *u_sinfo = NULL;
4647 	struct bond_opt_value newval;
4648 	struct net *net;
4649 	int res = 0;
4650 
4651 	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4652 
4653 	switch (cmd) {
4654 	case SIOCBONDINFOQUERY:
4655 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
4656 
4657 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4658 			return -EFAULT;
4659 
4660 		bond_info_query(bond_dev, &k_binfo);
4661 		if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4662 			return -EFAULT;
4663 
4664 		return 0;
4665 	case SIOCBONDSLAVEINFOQUERY:
4666 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4667 
4668 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4669 			return -EFAULT;
4670 
4671 		res = bond_slave_info_query(bond_dev, &k_sinfo);
4672 		if (res == 0 &&
4673 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4674 			return -EFAULT;
4675 
4676 		return res;
4677 	default:
4678 		break;
4679 	}
4680 
4681 	net = dev_net(bond_dev);
4682 
4683 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4684 		return -EPERM;
4685 
4686 	slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4687 
4688 	slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4689 
4690 	if (!slave_dev)
4691 		return -ENODEV;
4692 
4693 	switch (cmd) {
4694 	case SIOCBONDENSLAVE:
4695 		res = bond_enslave(bond_dev, slave_dev, NULL);
4696 		break;
4697 	case SIOCBONDRELEASE:
4698 		res = bond_release(bond_dev, slave_dev);
4699 		break;
4700 	case SIOCBONDSETHWADDR:
4701 		res = bond_set_dev_addr(bond_dev, slave_dev);
4702 		break;
4703 	case SIOCBONDCHANGEACTIVE:
4704 		bond_opt_initstr(&newval, slave_dev->name);
4705 		res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4706 					    &newval);
4707 		break;
4708 	default:
4709 		res = -EOPNOTSUPP;
4710 	}
4711 
4712 	return res;
4713 }
4714 
4715 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4716 			       void __user *data, int cmd)
4717 {
4718 	struct ifreq ifrdata = { .ifr_data = data };
4719 
4720 	switch (cmd) {
4721 	case BOND_INFO_QUERY_OLD:
4722 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4723 	case BOND_SLAVE_INFO_QUERY_OLD:
4724 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4725 	case BOND_ENSLAVE_OLD:
4726 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4727 	case BOND_RELEASE_OLD:
4728 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4729 	case BOND_SETHWADDR_OLD:
4730 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4731 	case BOND_CHANGE_ACTIVE_OLD:
4732 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4733 	}
4734 
4735 	return -EOPNOTSUPP;
4736 }
4737 
4738 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4739 {
4740 	struct bonding *bond = netdev_priv(bond_dev);
4741 
4742 	if (change & IFF_PROMISC)
4743 		bond_set_promiscuity(bond,
4744 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
4745 
4746 	if (change & IFF_ALLMULTI)
4747 		bond_set_allmulti(bond,
4748 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4749 }
4750 
4751 static void bond_set_rx_mode(struct net_device *bond_dev)
4752 {
4753 	struct bonding *bond = netdev_priv(bond_dev);
4754 	struct list_head *iter;
4755 	struct slave *slave;
4756 
4757 	rcu_read_lock();
4758 	if (bond_uses_primary(bond)) {
4759 		slave = rcu_dereference(bond->curr_active_slave);
4760 		if (slave) {
4761 			dev_uc_sync(slave->dev, bond_dev);
4762 			dev_mc_sync(slave->dev, bond_dev);
4763 		}
4764 	} else {
4765 		bond_for_each_slave_rcu(bond, slave, iter) {
4766 			dev_uc_sync_multiple(slave->dev, bond_dev);
4767 			dev_mc_sync_multiple(slave->dev, bond_dev);
4768 		}
4769 	}
4770 	rcu_read_unlock();
4771 }
4772 
4773 static int bond_neigh_init(struct neighbour *n)
4774 {
4775 	struct bonding *bond = netdev_priv(n->dev);
4776 	const struct net_device_ops *slave_ops;
4777 	struct neigh_parms parms;
4778 	struct slave *slave;
4779 	int ret = 0;
4780 
4781 	rcu_read_lock();
4782 	slave = bond_first_slave_rcu(bond);
4783 	if (!slave)
4784 		goto out;
4785 	slave_ops = slave->dev->netdev_ops;
4786 	if (!slave_ops->ndo_neigh_setup)
4787 		goto out;
4788 
4789 	/* TODO: find another way [1] to implement this.
4790 	 * Passing a zeroed structure is fragile,
4791 	 * but at least we do not pass garbage.
4792 	 *
4793 	 * [1] One way would be that ndo_neigh_setup() never touch
4794 	 *     struct neigh_parms, but propagate the new neigh_setup()
4795 	 *     back to ___neigh_create() / neigh_parms_alloc()
4796 	 */
4797 	memset(&parms, 0, sizeof(parms));
4798 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4799 
4800 	if (ret)
4801 		goto out;
4802 
4803 	if (parms.neigh_setup)
4804 		ret = parms.neigh_setup(n);
4805 out:
4806 	rcu_read_unlock();
4807 	return ret;
4808 }
4809 
4810 /* The bonding ndo_neigh_setup is called at init time beofre any
4811  * slave exists. So we must declare proxy setup function which will
4812  * be used at run time to resolve the actual slave neigh param setup.
4813  *
4814  * It's also called by master devices (such as vlans) to setup their
4815  * underlying devices. In that case - do nothing, we're already set up from
4816  * our init.
4817  */
4818 static int bond_neigh_setup(struct net_device *dev,
4819 			    struct neigh_parms *parms)
4820 {
4821 	/* modify only our neigh_parms */
4822 	if (parms->dev == dev)
4823 		parms->neigh_setup = bond_neigh_init;
4824 
4825 	return 0;
4826 }
4827 
4828 /* Change the MTU of all of a master's slaves to match the master */
4829 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4830 {
4831 	struct bonding *bond = netdev_priv(bond_dev);
4832 	struct slave *slave, *rollback_slave;
4833 	struct list_head *iter;
4834 	int res = 0;
4835 
4836 	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4837 
4838 	bond_for_each_slave(bond, slave, iter) {
4839 		slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4840 			   slave, slave->dev->netdev_ops->ndo_change_mtu);
4841 
4842 		res = dev_set_mtu(slave->dev, new_mtu);
4843 
4844 		if (res) {
4845 			/* If we failed to set the slave's mtu to the new value
4846 			 * we must abort the operation even in ACTIVE_BACKUP
4847 			 * mode, because if we allow the backup slaves to have
4848 			 * different mtu values than the active slave we'll
4849 			 * need to change their mtu when doing a failover. That
4850 			 * means changing their mtu from timer context, which
4851 			 * is probably not a good idea.
4852 			 */
4853 			slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4854 				  res, new_mtu);
4855 			goto unwind;
4856 		}
4857 	}
4858 
4859 	WRITE_ONCE(bond_dev->mtu, new_mtu);
4860 
4861 	return 0;
4862 
4863 unwind:
4864 	/* unwind from head to the slave that failed */
4865 	bond_for_each_slave(bond, rollback_slave, iter) {
4866 		int tmp_res;
4867 
4868 		if (rollback_slave == slave)
4869 			break;
4870 
4871 		tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4872 		if (tmp_res)
4873 			slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4874 				  tmp_res);
4875 	}
4876 
4877 	return res;
4878 }
4879 
4880 /* Change HW address
4881  *
4882  * Note that many devices must be down to change the HW address, and
4883  * downing the master releases all slaves.  We can make bonds full of
4884  * bonding devices to test this, however.
4885  */
4886 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4887 {
4888 	struct bonding *bond = netdev_priv(bond_dev);
4889 	struct slave *slave, *rollback_slave;
4890 	struct sockaddr_storage *ss = addr, tmp_ss;
4891 	struct list_head *iter;
4892 	int res = 0;
4893 
4894 	if (BOND_MODE(bond) == BOND_MODE_ALB)
4895 		return bond_alb_set_mac_address(bond_dev, addr);
4896 
4897 
4898 	netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4899 
4900 	/* If fail_over_mac is enabled, do nothing and return success.
4901 	 * Returning an error causes ifenslave to fail.
4902 	 */
4903 	if (bond->params.fail_over_mac &&
4904 	    BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4905 		return 0;
4906 
4907 	if (!is_valid_ether_addr(ss->__data))
4908 		return -EADDRNOTAVAIL;
4909 
4910 	bond_for_each_slave(bond, slave, iter) {
4911 		slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4912 			  __func__, slave);
4913 		res = dev_set_mac_address(slave->dev, addr, NULL);
4914 		if (res) {
4915 			/* TODO: consider downing the slave
4916 			 * and retry ?
4917 			 * User should expect communications
4918 			 * breakage anyway until ARP finish
4919 			 * updating, so...
4920 			 */
4921 			slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4922 				  __func__, res);
4923 			goto unwind;
4924 		}
4925 	}
4926 
4927 	/* success */
4928 	dev_addr_set(bond_dev, ss->__data);
4929 	return 0;
4930 
4931 unwind:
4932 	memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4933 	tmp_ss.ss_family = bond_dev->type;
4934 
4935 	/* unwind from head to the slave that failed */
4936 	bond_for_each_slave(bond, rollback_slave, iter) {
4937 		int tmp_res;
4938 
4939 		if (rollback_slave == slave)
4940 			break;
4941 
4942 		tmp_res = dev_set_mac_address(rollback_slave->dev,
4943 					      (struct sockaddr *)&tmp_ss, NULL);
4944 		if (tmp_res) {
4945 			slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4946 				   __func__, tmp_res);
4947 		}
4948 	}
4949 
4950 	return res;
4951 }
4952 
4953 /**
4954  * bond_get_slave_by_id - get xmit slave with slave_id
4955  * @bond: bonding device that is transmitting
4956  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4957  *
4958  * This function tries to get slave with slave_id but in case
4959  * it fails, it tries to find the first available slave for transmission.
4960  */
4961 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4962 					  int slave_id)
4963 {
4964 	struct list_head *iter;
4965 	struct slave *slave;
4966 	int i = slave_id;
4967 
4968 	/* Here we start from the slave with slave_id */
4969 	bond_for_each_slave_rcu(bond, slave, iter) {
4970 		if (--i < 0) {
4971 			if (bond_slave_can_tx(slave))
4972 				return slave;
4973 		}
4974 	}
4975 
4976 	/* Here we start from the first slave up to slave_id */
4977 	i = slave_id;
4978 	bond_for_each_slave_rcu(bond, slave, iter) {
4979 		if (--i < 0)
4980 			break;
4981 		if (bond_slave_can_tx(slave))
4982 			return slave;
4983 	}
4984 	/* no slave that can tx has been found */
4985 	return NULL;
4986 }
4987 
4988 /**
4989  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4990  * @bond: bonding device to use
4991  *
4992  * Based on the value of the bonding device's packets_per_slave parameter
4993  * this function generates a slave id, which is usually used as the next
4994  * slave to transmit through.
4995  */
4996 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4997 {
4998 	u32 slave_id;
4999 	struct reciprocal_value reciprocal_packets_per_slave;
5000 	int packets_per_slave = bond->params.packets_per_slave;
5001 
5002 	switch (packets_per_slave) {
5003 	case 0:
5004 		slave_id = get_random_u32();
5005 		break;
5006 	case 1:
5007 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
5008 		break;
5009 	default:
5010 		reciprocal_packets_per_slave =
5011 			bond->params.reciprocal_packets_per_slave;
5012 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
5013 		slave_id = reciprocal_divide(slave_id,
5014 					     reciprocal_packets_per_slave);
5015 		break;
5016 	}
5017 
5018 	return slave_id;
5019 }
5020 
5021 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
5022 						    struct sk_buff *skb)
5023 {
5024 	struct slave *slave;
5025 	int slave_cnt;
5026 	u32 slave_id;
5027 
5028 	/* Start with the curr_active_slave that joined the bond as the
5029 	 * default for sending IGMP traffic.  For failover purposes one
5030 	 * needs to maintain some consistency for the interface that will
5031 	 * send the join/membership reports.  The curr_active_slave found
5032 	 * will send all of this type of traffic.
5033 	 */
5034 	if (skb->protocol == htons(ETH_P_IP)) {
5035 		int noff = skb_network_offset(skb);
5036 		struct iphdr *iph;
5037 
5038 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
5039 			goto non_igmp;
5040 
5041 		iph = ip_hdr(skb);
5042 		if (iph->protocol == IPPROTO_IGMP) {
5043 			slave = rcu_dereference(bond->curr_active_slave);
5044 			if (slave)
5045 				return slave;
5046 			return bond_get_slave_by_id(bond, 0);
5047 		}
5048 	}
5049 
5050 non_igmp:
5051 	slave_cnt = READ_ONCE(bond->slave_cnt);
5052 	if (likely(slave_cnt)) {
5053 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
5054 		return bond_get_slave_by_id(bond, slave_id);
5055 	}
5056 	return NULL;
5057 }
5058 
5059 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
5060 							struct xdp_buff *xdp)
5061 {
5062 	struct slave *slave;
5063 	int slave_cnt;
5064 	u32 slave_id;
5065 	const struct ethhdr *eth;
5066 	void *data = xdp->data;
5067 
5068 	if (data + sizeof(struct ethhdr) > xdp->data_end)
5069 		goto non_igmp;
5070 
5071 	eth = (struct ethhdr *)data;
5072 	data += sizeof(struct ethhdr);
5073 
5074 	/* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
5075 	if (eth->h_proto == htons(ETH_P_IP)) {
5076 		const struct iphdr *iph;
5077 
5078 		if (data + sizeof(struct iphdr) > xdp->data_end)
5079 			goto non_igmp;
5080 
5081 		iph = (struct iphdr *)data;
5082 
5083 		if (iph->protocol == IPPROTO_IGMP) {
5084 			slave = rcu_dereference(bond->curr_active_slave);
5085 			if (slave)
5086 				return slave;
5087 			return bond_get_slave_by_id(bond, 0);
5088 		}
5089 	}
5090 
5091 non_igmp:
5092 	slave_cnt = READ_ONCE(bond->slave_cnt);
5093 	if (likely(slave_cnt)) {
5094 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
5095 		return bond_get_slave_by_id(bond, slave_id);
5096 	}
5097 	return NULL;
5098 }
5099 
5100 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
5101 					struct net_device *bond_dev)
5102 {
5103 	struct bonding *bond = netdev_priv(bond_dev);
5104 	struct slave *slave;
5105 
5106 	slave = bond_xmit_roundrobin_slave_get(bond, skb);
5107 	if (likely(slave))
5108 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5109 
5110 	return bond_tx_drop(bond_dev, skb);
5111 }
5112 
5113 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
5114 {
5115 	return rcu_dereference(bond->curr_active_slave);
5116 }
5117 
5118 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
5119  * the bond has a usable interface.
5120  */
5121 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
5122 					  struct net_device *bond_dev)
5123 {
5124 	struct bonding *bond = netdev_priv(bond_dev);
5125 	struct slave *slave;
5126 
5127 	slave = bond_xmit_activebackup_slave_get(bond);
5128 	if (slave)
5129 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5130 
5131 	return bond_tx_drop(bond_dev, skb);
5132 }
5133 
5134 /* Use this to update slave_array when (a) it's not appropriate to update
5135  * slave_array right away (note that update_slave_array() may sleep)
5136  * and / or (b) RTNL is not held.
5137  */
5138 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
5139 {
5140 	queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
5141 }
5142 
5143 /* Slave array work handler. Holds only RTNL */
5144 static void bond_slave_arr_handler(struct work_struct *work)
5145 {
5146 	struct bonding *bond = container_of(work, struct bonding,
5147 					    slave_arr_work.work);
5148 	int ret;
5149 
5150 	if (!rtnl_trylock())
5151 		goto err;
5152 
5153 	ret = bond_update_slave_arr(bond, NULL);
5154 	rtnl_unlock();
5155 	if (ret) {
5156 		pr_warn_ratelimited("Failed to update slave array from WT\n");
5157 		goto err;
5158 	}
5159 	return;
5160 
5161 err:
5162 	bond_slave_arr_work_rearm(bond, 1);
5163 }
5164 
5165 static void bond_skip_slave(struct bond_up_slave *slaves,
5166 			    struct slave *skipslave)
5167 {
5168 	int idx;
5169 
5170 	/* Rare situation where caller has asked to skip a specific
5171 	 * slave but allocation failed (most likely!). BTW this is
5172 	 * only possible when the call is initiated from
5173 	 * __bond_release_one(). In this situation; overwrite the
5174 	 * skipslave entry in the array with the last entry from the
5175 	 * array to avoid a situation where the xmit path may choose
5176 	 * this to-be-skipped slave to send a packet out.
5177 	 */
5178 	for (idx = 0; slaves && idx < slaves->count; idx++) {
5179 		if (skipslave == slaves->arr[idx]) {
5180 			slaves->arr[idx] =
5181 				slaves->arr[slaves->count - 1];
5182 			slaves->count--;
5183 			break;
5184 		}
5185 	}
5186 }
5187 
5188 static void bond_set_slave_arr(struct bonding *bond,
5189 			       struct bond_up_slave *usable_slaves,
5190 			       struct bond_up_slave *all_slaves)
5191 {
5192 	struct bond_up_slave *usable, *all;
5193 
5194 	usable = rtnl_dereference(bond->usable_slaves);
5195 	rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5196 	kfree_rcu(usable, rcu);
5197 
5198 	all = rtnl_dereference(bond->all_slaves);
5199 	rcu_assign_pointer(bond->all_slaves, all_slaves);
5200 	kfree_rcu(all, rcu);
5201 }
5202 
5203 static void bond_reset_slave_arr(struct bonding *bond)
5204 {
5205 	bond_set_slave_arr(bond, NULL, NULL);
5206 }
5207 
5208 /* Build the usable slaves array in control path for modes that use xmit-hash
5209  * to determine the slave interface -
5210  * (a) BOND_MODE_8023AD
5211  * (b) BOND_MODE_XOR
5212  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5213  *
5214  * The caller is expected to hold RTNL only and NO other lock!
5215  */
5216 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5217 {
5218 	struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5219 	struct slave *slave;
5220 	struct list_head *iter;
5221 	int agg_id = 0;
5222 	int ret = 0;
5223 
5224 	might_sleep();
5225 
5226 	usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5227 					    bond->slave_cnt), GFP_KERNEL);
5228 	all_slaves = kzalloc(struct_size(all_slaves, arr,
5229 					 bond->slave_cnt), GFP_KERNEL);
5230 	if (!usable_slaves || !all_slaves) {
5231 		ret = -ENOMEM;
5232 		goto out;
5233 	}
5234 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5235 		struct ad_info ad_info;
5236 
5237 		spin_lock_bh(&bond->mode_lock);
5238 		if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5239 			spin_unlock_bh(&bond->mode_lock);
5240 			pr_debug("bond_3ad_get_active_agg_info failed\n");
5241 			/* No active aggragator means it's not safe to use
5242 			 * the previous array.
5243 			 */
5244 			bond_reset_slave_arr(bond);
5245 			goto out;
5246 		}
5247 		spin_unlock_bh(&bond->mode_lock);
5248 		agg_id = ad_info.aggregator_id;
5249 	}
5250 	bond_for_each_slave(bond, slave, iter) {
5251 		if (skipslave == slave)
5252 			continue;
5253 
5254 		all_slaves->arr[all_slaves->count++] = slave;
5255 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5256 			struct aggregator *agg;
5257 
5258 			agg = SLAVE_AD_INFO(slave)->port.aggregator;
5259 			if (!agg || agg->aggregator_identifier != agg_id)
5260 				continue;
5261 		}
5262 		if (!bond_slave_can_tx(slave))
5263 			continue;
5264 
5265 		slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5266 			  usable_slaves->count);
5267 
5268 		usable_slaves->arr[usable_slaves->count++] = slave;
5269 	}
5270 
5271 	bond_set_slave_arr(bond, usable_slaves, all_slaves);
5272 	return ret;
5273 out:
5274 	if (ret != 0 && skipslave) {
5275 		bond_skip_slave(rtnl_dereference(bond->all_slaves),
5276 				skipslave);
5277 		bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5278 				skipslave);
5279 	}
5280 	kfree_rcu(all_slaves, rcu);
5281 	kfree_rcu(usable_slaves, rcu);
5282 
5283 	return ret;
5284 }
5285 
5286 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5287 						 struct sk_buff *skb,
5288 						 struct bond_up_slave *slaves)
5289 {
5290 	struct slave *slave;
5291 	unsigned int count;
5292 	u32 hash;
5293 
5294 	hash = bond_xmit_hash(bond, skb);
5295 	count = slaves ? READ_ONCE(slaves->count) : 0;
5296 	if (unlikely(!count))
5297 		return NULL;
5298 
5299 	slave = slaves->arr[hash % count];
5300 	return slave;
5301 }
5302 
5303 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5304 						     struct xdp_buff *xdp)
5305 {
5306 	struct bond_up_slave *slaves;
5307 	unsigned int count;
5308 	u32 hash;
5309 
5310 	hash = bond_xmit_hash_xdp(bond, xdp);
5311 	slaves = rcu_dereference(bond->usable_slaves);
5312 	count = slaves ? READ_ONCE(slaves->count) : 0;
5313 	if (unlikely(!count))
5314 		return NULL;
5315 
5316 	return slaves->arr[hash % count];
5317 }
5318 
5319 /* Use this Xmit function for 3AD as well as XOR modes. The current
5320  * usable slave array is formed in the control path. The xmit function
5321  * just calculates hash and sends the packet out.
5322  */
5323 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5324 				     struct net_device *dev)
5325 {
5326 	struct bonding *bond = netdev_priv(dev);
5327 	struct bond_up_slave *slaves;
5328 	struct slave *slave;
5329 
5330 	slaves = rcu_dereference(bond->usable_slaves);
5331 	slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5332 	if (likely(slave))
5333 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5334 
5335 	return bond_tx_drop(dev, skb);
5336 }
5337 
5338 /* in broadcast mode, we send everything to all usable interfaces. */
5339 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5340 				       struct net_device *bond_dev)
5341 {
5342 	struct bonding *bond = netdev_priv(bond_dev);
5343 	struct slave *slave = NULL;
5344 	struct list_head *iter;
5345 	bool xmit_suc = false;
5346 	bool skb_used = false;
5347 
5348 	bond_for_each_slave_rcu(bond, slave, iter) {
5349 		struct sk_buff *skb2;
5350 
5351 		if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5352 			continue;
5353 
5354 		if (bond_is_last_slave(bond, slave)) {
5355 			skb2 = skb;
5356 			skb_used = true;
5357 		} else {
5358 			skb2 = skb_clone(skb, GFP_ATOMIC);
5359 			if (!skb2) {
5360 				net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5361 						    bond_dev->name, __func__);
5362 				continue;
5363 			}
5364 		}
5365 
5366 		if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5367 			xmit_suc = true;
5368 	}
5369 
5370 	if (!skb_used)
5371 		dev_kfree_skb_any(skb);
5372 
5373 	if (xmit_suc)
5374 		return NETDEV_TX_OK;
5375 
5376 	dev_core_stats_tx_dropped_inc(bond_dev);
5377 	return NET_XMIT_DROP;
5378 }
5379 
5380 /*------------------------- Device initialization ---------------------------*/
5381 
5382 /* Lookup the slave that corresponds to a qid */
5383 static inline int bond_slave_override(struct bonding *bond,
5384 				      struct sk_buff *skb)
5385 {
5386 	struct slave *slave = NULL;
5387 	struct list_head *iter;
5388 
5389 	if (!skb_rx_queue_recorded(skb))
5390 		return 1;
5391 
5392 	/* Find out if any slaves have the same mapping as this skb. */
5393 	bond_for_each_slave_rcu(bond, slave, iter) {
5394 		if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) {
5395 			if (bond_slave_is_up(slave) &&
5396 			    slave->link == BOND_LINK_UP) {
5397 				bond_dev_queue_xmit(bond, skb, slave->dev);
5398 				return 0;
5399 			}
5400 			/* If the slave isn't UP, use default transmit policy. */
5401 			break;
5402 		}
5403 	}
5404 
5405 	return 1;
5406 }
5407 
5408 
5409 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5410 			     struct net_device *sb_dev)
5411 {
5412 	/* This helper function exists to help dev_pick_tx get the correct
5413 	 * destination queue.  Using a helper function skips a call to
5414 	 * skb_tx_hash and will put the skbs in the queue we expect on their
5415 	 * way down to the bonding driver.
5416 	 */
5417 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5418 
5419 	/* Save the original txq to restore before passing to the driver */
5420 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5421 
5422 	if (unlikely(txq >= dev->real_num_tx_queues)) {
5423 		do {
5424 			txq -= dev->real_num_tx_queues;
5425 		} while (txq >= dev->real_num_tx_queues);
5426 	}
5427 	return txq;
5428 }
5429 
5430 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5431 					      struct sk_buff *skb,
5432 					      bool all_slaves)
5433 {
5434 	struct bonding *bond = netdev_priv(master_dev);
5435 	struct bond_up_slave *slaves;
5436 	struct slave *slave = NULL;
5437 
5438 	switch (BOND_MODE(bond)) {
5439 	case BOND_MODE_ROUNDROBIN:
5440 		slave = bond_xmit_roundrobin_slave_get(bond, skb);
5441 		break;
5442 	case BOND_MODE_ACTIVEBACKUP:
5443 		slave = bond_xmit_activebackup_slave_get(bond);
5444 		break;
5445 	case BOND_MODE_8023AD:
5446 	case BOND_MODE_XOR:
5447 		if (all_slaves)
5448 			slaves = rcu_dereference(bond->all_slaves);
5449 		else
5450 			slaves = rcu_dereference(bond->usable_slaves);
5451 		slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5452 		break;
5453 	case BOND_MODE_BROADCAST:
5454 		break;
5455 	case BOND_MODE_ALB:
5456 		slave = bond_xmit_alb_slave_get(bond, skb);
5457 		break;
5458 	case BOND_MODE_TLB:
5459 		slave = bond_xmit_tlb_slave_get(bond, skb);
5460 		break;
5461 	default:
5462 		/* Should never happen, mode already checked */
5463 		WARN_ONCE(true, "Unknown bonding mode");
5464 		break;
5465 	}
5466 
5467 	if (slave)
5468 		return slave->dev;
5469 	return NULL;
5470 }
5471 
5472 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5473 {
5474 	switch (sk->sk_family) {
5475 #if IS_ENABLED(CONFIG_IPV6)
5476 	case AF_INET6:
5477 		if (ipv6_only_sock(sk) ||
5478 		    ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5479 			flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5480 			flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5481 			flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5482 			break;
5483 		}
5484 		fallthrough;
5485 #endif
5486 	default: /* AF_INET */
5487 		flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5488 		flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5489 		flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5490 		break;
5491 	}
5492 
5493 	flow->ports.src = inet_sk(sk)->inet_sport;
5494 	flow->ports.dst = inet_sk(sk)->inet_dport;
5495 }
5496 
5497 /**
5498  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5499  * @sk: socket to use for headers
5500  *
5501  * This function will extract the necessary field from the socket and use
5502  * them to generate a hash based on the LAYER34 xmit_policy.
5503  * Assumes that sk is a TCP or UDP socket.
5504  */
5505 static u32 bond_sk_hash_l34(struct sock *sk)
5506 {
5507 	struct flow_keys flow;
5508 	u32 hash;
5509 
5510 	bond_sk_to_flow(sk, &flow);
5511 
5512 	/* L4 */
5513 	memcpy(&hash, &flow.ports.ports, sizeof(hash));
5514 	/* L3 */
5515 	return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5516 }
5517 
5518 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5519 						  struct sock *sk)
5520 {
5521 	struct bond_up_slave *slaves;
5522 	struct slave *slave;
5523 	unsigned int count;
5524 	u32 hash;
5525 
5526 	slaves = rcu_dereference(bond->usable_slaves);
5527 	count = slaves ? READ_ONCE(slaves->count) : 0;
5528 	if (unlikely(!count))
5529 		return NULL;
5530 
5531 	hash = bond_sk_hash_l34(sk);
5532 	slave = slaves->arr[hash % count];
5533 
5534 	return slave->dev;
5535 }
5536 
5537 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5538 						struct sock *sk)
5539 {
5540 	struct bonding *bond = netdev_priv(dev);
5541 	struct net_device *lower = NULL;
5542 
5543 	rcu_read_lock();
5544 	if (bond_sk_check(bond))
5545 		lower = __bond_sk_get_lower_dev(bond, sk);
5546 	rcu_read_unlock();
5547 
5548 	return lower;
5549 }
5550 
5551 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5552 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5553 					struct net_device *dev)
5554 {
5555 	struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5556 
5557 	/* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5558 	 * was true, if tls_device_down is running in parallel, but it's OK,
5559 	 * because bond_get_slave_by_dev has a NULL check.
5560 	 */
5561 	if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5562 		return bond_dev_queue_xmit(bond, skb, tls_netdev);
5563 	return bond_tx_drop(dev, skb);
5564 }
5565 #endif
5566 
5567 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5568 {
5569 	struct bonding *bond = netdev_priv(dev);
5570 
5571 	if (bond_should_override_tx_queue(bond) &&
5572 	    !bond_slave_override(bond, skb))
5573 		return NETDEV_TX_OK;
5574 
5575 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5576 	if (tls_is_skb_tx_device_offloaded(skb))
5577 		return bond_tls_device_xmit(bond, skb, dev);
5578 #endif
5579 
5580 	switch (BOND_MODE(bond)) {
5581 	case BOND_MODE_ROUNDROBIN:
5582 		return bond_xmit_roundrobin(skb, dev);
5583 	case BOND_MODE_ACTIVEBACKUP:
5584 		return bond_xmit_activebackup(skb, dev);
5585 	case BOND_MODE_8023AD:
5586 	case BOND_MODE_XOR:
5587 		return bond_3ad_xor_xmit(skb, dev);
5588 	case BOND_MODE_BROADCAST:
5589 		return bond_xmit_broadcast(skb, dev);
5590 	case BOND_MODE_ALB:
5591 		return bond_alb_xmit(skb, dev);
5592 	case BOND_MODE_TLB:
5593 		return bond_tlb_xmit(skb, dev);
5594 	default:
5595 		/* Should never happen, mode already checked */
5596 		netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5597 		WARN_ON_ONCE(1);
5598 		return bond_tx_drop(dev, skb);
5599 	}
5600 }
5601 
5602 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5603 {
5604 	struct bonding *bond = netdev_priv(dev);
5605 	netdev_tx_t ret = NETDEV_TX_OK;
5606 
5607 	/* If we risk deadlock from transmitting this in the
5608 	 * netpoll path, tell netpoll to queue the frame for later tx
5609 	 */
5610 	if (unlikely(is_netpoll_tx_blocked(dev)))
5611 		return NETDEV_TX_BUSY;
5612 
5613 	rcu_read_lock();
5614 	if (bond_has_slaves(bond))
5615 		ret = __bond_start_xmit(skb, dev);
5616 	else
5617 		ret = bond_tx_drop(dev, skb);
5618 	rcu_read_unlock();
5619 
5620 	return ret;
5621 }
5622 
5623 static struct net_device *
5624 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5625 {
5626 	struct bonding *bond = netdev_priv(bond_dev);
5627 	struct slave *slave;
5628 
5629 	/* Caller needs to hold rcu_read_lock() */
5630 
5631 	switch (BOND_MODE(bond)) {
5632 	case BOND_MODE_ROUNDROBIN:
5633 		slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5634 		break;
5635 
5636 	case BOND_MODE_ACTIVEBACKUP:
5637 		slave = bond_xmit_activebackup_slave_get(bond);
5638 		break;
5639 
5640 	case BOND_MODE_8023AD:
5641 	case BOND_MODE_XOR:
5642 		slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5643 		break;
5644 
5645 	default:
5646 		if (net_ratelimit())
5647 			netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n",
5648 				   BOND_MODE(bond));
5649 		return NULL;
5650 	}
5651 
5652 	if (slave)
5653 		return slave->dev;
5654 
5655 	return NULL;
5656 }
5657 
5658 static int bond_xdp_xmit(struct net_device *bond_dev,
5659 			 int n, struct xdp_frame **frames, u32 flags)
5660 {
5661 	int nxmit, err = -ENXIO;
5662 
5663 	rcu_read_lock();
5664 
5665 	for (nxmit = 0; nxmit < n; nxmit++) {
5666 		struct xdp_frame *frame = frames[nxmit];
5667 		struct xdp_frame *frames1[] = {frame};
5668 		struct net_device *slave_dev;
5669 		struct xdp_buff xdp;
5670 
5671 		xdp_convert_frame_to_buff(frame, &xdp);
5672 
5673 		slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5674 		if (!slave_dev) {
5675 			err = -ENXIO;
5676 			break;
5677 		}
5678 
5679 		err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5680 		if (err < 1)
5681 			break;
5682 	}
5683 
5684 	rcu_read_unlock();
5685 
5686 	/* If error happened on the first frame then we can pass the error up, otherwise
5687 	 * report the number of frames that were xmitted.
5688 	 */
5689 	if (err < 0)
5690 		return (nxmit == 0 ? err : nxmit);
5691 
5692 	return nxmit;
5693 }
5694 
5695 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5696 			struct netlink_ext_ack *extack)
5697 {
5698 	struct bonding *bond = netdev_priv(dev);
5699 	struct list_head *iter;
5700 	struct slave *slave, *rollback_slave;
5701 	struct bpf_prog *old_prog;
5702 	struct netdev_bpf xdp = {
5703 		.command = XDP_SETUP_PROG,
5704 		.flags   = 0,
5705 		.prog    = prog,
5706 		.extack  = extack,
5707 	};
5708 	int err;
5709 
5710 	ASSERT_RTNL();
5711 
5712 	if (!bond_xdp_check(bond, BOND_MODE(bond))) {
5713 		BOND_NL_ERR(dev, extack,
5714 			    "No native XDP support for the current bonding mode");
5715 		return -EOPNOTSUPP;
5716 	}
5717 
5718 	old_prog = bond->xdp_prog;
5719 	bond->xdp_prog = prog;
5720 
5721 	bond_for_each_slave(bond, slave, iter) {
5722 		struct net_device *slave_dev = slave->dev;
5723 
5724 		if (!slave_dev->netdev_ops->ndo_bpf ||
5725 		    !slave_dev->netdev_ops->ndo_xdp_xmit) {
5726 			SLAVE_NL_ERR(dev, slave_dev, extack,
5727 				     "Slave device does not support XDP");
5728 			err = -EOPNOTSUPP;
5729 			goto err;
5730 		}
5731 
5732 		if (dev_xdp_prog_count(slave_dev) > 0) {
5733 			SLAVE_NL_ERR(dev, slave_dev, extack,
5734 				     "Slave has XDP program loaded, please unload before enslaving");
5735 			err = -EOPNOTSUPP;
5736 			goto err;
5737 		}
5738 
5739 		err = dev_xdp_propagate(slave_dev, &xdp);
5740 		if (err < 0) {
5741 			/* ndo_bpf() sets extack error message */
5742 			slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5743 			goto err;
5744 		}
5745 		if (prog)
5746 			bpf_prog_inc(prog);
5747 	}
5748 
5749 	if (prog) {
5750 		static_branch_inc(&bpf_master_redirect_enabled_key);
5751 	} else if (old_prog) {
5752 		bpf_prog_put(old_prog);
5753 		static_branch_dec(&bpf_master_redirect_enabled_key);
5754 	}
5755 
5756 	return 0;
5757 
5758 err:
5759 	/* unwind the program changes */
5760 	bond->xdp_prog = old_prog;
5761 	xdp.prog = old_prog;
5762 	xdp.extack = NULL; /* do not overwrite original error */
5763 
5764 	bond_for_each_slave(bond, rollback_slave, iter) {
5765 		struct net_device *slave_dev = rollback_slave->dev;
5766 		int err_unwind;
5767 
5768 		if (slave == rollback_slave)
5769 			break;
5770 
5771 		err_unwind = dev_xdp_propagate(slave_dev, &xdp);
5772 		if (err_unwind < 0)
5773 			slave_err(dev, slave_dev,
5774 				  "Error %d when unwinding XDP program change\n", err_unwind);
5775 		else if (xdp.prog)
5776 			bpf_prog_inc(xdp.prog);
5777 	}
5778 	return err;
5779 }
5780 
5781 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5782 {
5783 	switch (xdp->command) {
5784 	case XDP_SETUP_PROG:
5785 		return bond_xdp_set(dev, xdp->prog, xdp->extack);
5786 	default:
5787 		return -EINVAL;
5788 	}
5789 }
5790 
5791 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5792 {
5793 	if (speed == 0 || speed == SPEED_UNKNOWN)
5794 		speed = slave->speed;
5795 	else
5796 		speed = min(speed, slave->speed);
5797 
5798 	return speed;
5799 }
5800 
5801 /* Set the BOND_PHC_INDEX flag to notify user space */
5802 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5803 {
5804 	struct ifreq *ifr = kernel_cfg->ifr;
5805 	struct hwtstamp_config cfg;
5806 
5807 	if (kernel_cfg->copied_to_user) {
5808 		/* Lower device has a legacy implementation */
5809 		if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
5810 			return -EFAULT;
5811 
5812 		cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5813 		if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)))
5814 			return -EFAULT;
5815 	} else {
5816 		kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5817 	}
5818 
5819 	return 0;
5820 }
5821 
5822 static int bond_hwtstamp_get(struct net_device *dev,
5823 			     struct kernel_hwtstamp_config *cfg)
5824 {
5825 	struct bonding *bond = netdev_priv(dev);
5826 	struct net_device *real_dev;
5827 	int err;
5828 
5829 	real_dev = bond_option_active_slave_get_rcu(bond);
5830 	if (!real_dev)
5831 		return -EOPNOTSUPP;
5832 
5833 	err = generic_hwtstamp_get_lower(real_dev, cfg);
5834 	if (err)
5835 		return err;
5836 
5837 	return bond_set_phc_index_flag(cfg);
5838 }
5839 
5840 static int bond_hwtstamp_set(struct net_device *dev,
5841 			     struct kernel_hwtstamp_config *cfg,
5842 			     struct netlink_ext_ack *extack)
5843 {
5844 	struct bonding *bond = netdev_priv(dev);
5845 	struct net_device *real_dev;
5846 	int err;
5847 
5848 	if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5849 		return -EOPNOTSUPP;
5850 
5851 	real_dev = bond_option_active_slave_get_rcu(bond);
5852 	if (!real_dev)
5853 		return -EOPNOTSUPP;
5854 
5855 	err = generic_hwtstamp_set_lower(real_dev, cfg, extack);
5856 	if (err)
5857 		return err;
5858 
5859 	return bond_set_phc_index_flag(cfg);
5860 }
5861 
5862 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5863 					   struct ethtool_link_ksettings *cmd)
5864 {
5865 	struct bonding *bond = netdev_priv(bond_dev);
5866 	struct list_head *iter;
5867 	struct slave *slave;
5868 	u32 speed = 0;
5869 
5870 	cmd->base.duplex = DUPLEX_UNKNOWN;
5871 	cmd->base.port = PORT_OTHER;
5872 
5873 	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5874 	 * do not need to check mode.  Though link speed might not represent
5875 	 * the true receive or transmit bandwidth (not all modes are symmetric)
5876 	 * this is an accurate maximum.
5877 	 */
5878 	bond_for_each_slave(bond, slave, iter) {
5879 		if (bond_slave_can_tx(slave)) {
5880 			bond_update_speed_duplex(slave);
5881 			if (slave->speed != SPEED_UNKNOWN) {
5882 				if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5883 					speed = bond_mode_bcast_speed(slave,
5884 								      speed);
5885 				else
5886 					speed += slave->speed;
5887 			}
5888 			if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5889 			    slave->duplex != DUPLEX_UNKNOWN)
5890 				cmd->base.duplex = slave->duplex;
5891 		}
5892 	}
5893 	cmd->base.speed = speed ? : SPEED_UNKNOWN;
5894 
5895 	return 0;
5896 }
5897 
5898 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5899 				     struct ethtool_drvinfo *drvinfo)
5900 {
5901 	strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5902 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5903 		 BOND_ABI_VERSION);
5904 }
5905 
5906 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5907 				    struct kernel_ethtool_ts_info *info)
5908 {
5909 	struct bonding *bond = netdev_priv(bond_dev);
5910 	struct kernel_ethtool_ts_info ts_info;
5911 	struct net_device *real_dev;
5912 	bool sw_tx_support = false;
5913 	struct list_head *iter;
5914 	struct slave *slave;
5915 	int ret = 0;
5916 
5917 	rcu_read_lock();
5918 	real_dev = bond_option_active_slave_get_rcu(bond);
5919 	dev_hold(real_dev);
5920 	rcu_read_unlock();
5921 
5922 	if (real_dev) {
5923 		ret = ethtool_get_ts_info_by_layer(real_dev, info);
5924 	} else {
5925 		/* Check if all slaves support software tx timestamping */
5926 		rcu_read_lock();
5927 		bond_for_each_slave_rcu(bond, slave, iter) {
5928 			ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info);
5929 			if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5930 				sw_tx_support = true;
5931 				continue;
5932 			}
5933 
5934 			sw_tx_support = false;
5935 			break;
5936 		}
5937 		rcu_read_unlock();
5938 	}
5939 
5940 	if (sw_tx_support)
5941 		info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5942 
5943 	dev_put(real_dev);
5944 	return ret;
5945 }
5946 
5947 static const struct ethtool_ops bond_ethtool_ops = {
5948 	.get_drvinfo		= bond_ethtool_get_drvinfo,
5949 	.get_link		= ethtool_op_get_link,
5950 	.get_link_ksettings	= bond_ethtool_get_link_ksettings,
5951 	.get_ts_info		= bond_ethtool_get_ts_info,
5952 };
5953 
5954 static const struct net_device_ops bond_netdev_ops = {
5955 	.ndo_init		= bond_init,
5956 	.ndo_uninit		= bond_uninit,
5957 	.ndo_open		= bond_open,
5958 	.ndo_stop		= bond_close,
5959 	.ndo_start_xmit		= bond_start_xmit,
5960 	.ndo_select_queue	= bond_select_queue,
5961 	.ndo_get_stats64	= bond_get_stats,
5962 	.ndo_eth_ioctl		= bond_eth_ioctl,
5963 	.ndo_siocbond		= bond_do_ioctl,
5964 	.ndo_siocdevprivate	= bond_siocdevprivate,
5965 	.ndo_change_rx_flags	= bond_change_rx_flags,
5966 	.ndo_set_rx_mode	= bond_set_rx_mode,
5967 	.ndo_change_mtu		= bond_change_mtu,
5968 	.ndo_set_mac_address	= bond_set_mac_address,
5969 	.ndo_neigh_setup	= bond_neigh_setup,
5970 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
5971 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
5972 #ifdef CONFIG_NET_POLL_CONTROLLER
5973 	.ndo_netpoll_setup	= bond_netpoll_setup,
5974 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
5975 	.ndo_poll_controller	= bond_poll_controller,
5976 #endif
5977 	.ndo_add_slave		= bond_enslave,
5978 	.ndo_del_slave		= bond_release,
5979 	.ndo_fix_features	= bond_fix_features,
5980 	.ndo_features_check	= passthru_features_check,
5981 	.ndo_get_xmit_slave	= bond_xmit_get_slave,
5982 	.ndo_sk_get_lower_dev	= bond_sk_get_lower_dev,
5983 	.ndo_bpf		= bond_xdp,
5984 	.ndo_xdp_xmit           = bond_xdp_xmit,
5985 	.ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5986 	.ndo_hwtstamp_get	= bond_hwtstamp_get,
5987 	.ndo_hwtstamp_set	= bond_hwtstamp_set,
5988 };
5989 
5990 static const struct device_type bond_type = {
5991 	.name = "bond",
5992 };
5993 
5994 static void bond_destructor(struct net_device *bond_dev)
5995 {
5996 	struct bonding *bond = netdev_priv(bond_dev);
5997 
5998 	if (bond->wq)
5999 		destroy_workqueue(bond->wq);
6000 
6001 	free_percpu(bond->rr_tx_counter);
6002 }
6003 
6004 void bond_setup(struct net_device *bond_dev)
6005 {
6006 	struct bonding *bond = netdev_priv(bond_dev);
6007 
6008 	spin_lock_init(&bond->mode_lock);
6009 	bond->params = bonding_defaults;
6010 
6011 	/* Initialize pointers */
6012 	bond->dev = bond_dev;
6013 
6014 	/* Initialize the device entry points */
6015 	ether_setup(bond_dev);
6016 	bond_dev->max_mtu = ETH_MAX_MTU;
6017 	bond_dev->netdev_ops = &bond_netdev_ops;
6018 	bond_dev->ethtool_ops = &bond_ethtool_ops;
6019 
6020 	bond_dev->needs_free_netdev = true;
6021 	bond_dev->priv_destructor = bond_destructor;
6022 
6023 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
6024 
6025 	/* Initialize the device options */
6026 	bond_dev->flags |= IFF_MASTER;
6027 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
6028 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
6029 
6030 #ifdef CONFIG_XFRM_OFFLOAD
6031 	/* set up xfrm device ops (only supported in active-backup right now) */
6032 	bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
6033 	INIT_LIST_HEAD(&bond->ipsec_list);
6034 	mutex_init(&bond->ipsec_lock);
6035 #endif /* CONFIG_XFRM_OFFLOAD */
6036 
6037 	/* don't acquire bond device's netif_tx_lock when transmitting */
6038 	bond_dev->lltx = true;
6039 
6040 	/* Don't allow bond devices to change network namespaces. */
6041 	bond_dev->netns_immutable = true;
6042 
6043 	/* By default, we declare the bond to be fully
6044 	 * VLAN hardware accelerated capable. Special
6045 	 * care is taken in the various xmit functions
6046 	 * when there are slaves that are not hw accel
6047 	 * capable
6048 	 */
6049 
6050 	bond_dev->hw_features = BOND_VLAN_FEATURES |
6051 				NETIF_F_HW_VLAN_CTAG_RX |
6052 				NETIF_F_HW_VLAN_CTAG_FILTER |
6053 				NETIF_F_HW_VLAN_STAG_RX |
6054 				NETIF_F_HW_VLAN_STAG_FILTER;
6055 
6056 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
6057 	bond_dev->features |= bond_dev->hw_features;
6058 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
6059 	bond_dev->features |= NETIF_F_GSO_PARTIAL;
6060 #ifdef CONFIG_XFRM_OFFLOAD
6061 	bond_dev->hw_features |= BOND_XFRM_FEATURES;
6062 	/* Only enable XFRM features if this is an active-backup config */
6063 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
6064 		bond_dev->features |= BOND_XFRM_FEATURES;
6065 #endif /* CONFIG_XFRM_OFFLOAD */
6066 }
6067 
6068 /* Destroy a bonding device.
6069  * Must be under rtnl_lock when this function is called.
6070  */
6071 static void bond_uninit(struct net_device *bond_dev)
6072 {
6073 	struct bonding *bond = netdev_priv(bond_dev);
6074 	struct list_head *iter;
6075 	struct slave *slave;
6076 
6077 	bond_netpoll_cleanup(bond_dev);
6078 
6079 	/* Release the bonded slaves */
6080 	bond_for_each_slave(bond, slave, iter)
6081 		__bond_release_one(bond_dev, slave->dev, true, true);
6082 	netdev_info(bond_dev, "Released all slaves\n");
6083 
6084 #ifdef CONFIG_XFRM_OFFLOAD
6085 	mutex_destroy(&bond->ipsec_lock);
6086 #endif /* CONFIG_XFRM_OFFLOAD */
6087 
6088 	bond_set_slave_arr(bond, NULL, NULL);
6089 
6090 	list_del_rcu(&bond->bond_list);
6091 
6092 	bond_debug_unregister(bond);
6093 }
6094 
6095 /*------------------------- Module initialization ---------------------------*/
6096 
6097 static int __init bond_check_params(struct bond_params *params)
6098 {
6099 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
6100 	struct bond_opt_value newval;
6101 	const struct bond_opt_value *valptr;
6102 	int arp_all_targets_value = 0;
6103 	u16 ad_actor_sys_prio = 0;
6104 	u16 ad_user_port_key = 0;
6105 	__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
6106 	int arp_ip_count;
6107 	int bond_mode	= BOND_MODE_ROUNDROBIN;
6108 	int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
6109 	int lacp_fast = 0;
6110 	int tlb_dynamic_lb;
6111 
6112 	/* Convert string parameters. */
6113 	if (mode) {
6114 		bond_opt_initstr(&newval, mode);
6115 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
6116 		if (!valptr) {
6117 			pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
6118 			return -EINVAL;
6119 		}
6120 		bond_mode = valptr->value;
6121 	}
6122 
6123 	if (xmit_hash_policy) {
6124 		if (bond_mode == BOND_MODE_ROUNDROBIN ||
6125 		    bond_mode == BOND_MODE_ACTIVEBACKUP ||
6126 		    bond_mode == BOND_MODE_BROADCAST) {
6127 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
6128 				bond_mode_name(bond_mode));
6129 		} else {
6130 			bond_opt_initstr(&newval, xmit_hash_policy);
6131 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
6132 						&newval);
6133 			if (!valptr) {
6134 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
6135 				       xmit_hash_policy);
6136 				return -EINVAL;
6137 			}
6138 			xmit_hashtype = valptr->value;
6139 		}
6140 	}
6141 
6142 	if (lacp_rate) {
6143 		if (bond_mode != BOND_MODE_8023AD) {
6144 			pr_info("lacp_rate param is irrelevant in mode %s\n",
6145 				bond_mode_name(bond_mode));
6146 		} else {
6147 			bond_opt_initstr(&newval, lacp_rate);
6148 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
6149 						&newval);
6150 			if (!valptr) {
6151 				pr_err("Error: Invalid lacp rate \"%s\"\n",
6152 				       lacp_rate);
6153 				return -EINVAL;
6154 			}
6155 			lacp_fast = valptr->value;
6156 		}
6157 	}
6158 
6159 	if (ad_select) {
6160 		bond_opt_initstr(&newval, ad_select);
6161 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6162 					&newval);
6163 		if (!valptr) {
6164 			pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6165 			return -EINVAL;
6166 		}
6167 		params->ad_select = valptr->value;
6168 		if (bond_mode != BOND_MODE_8023AD)
6169 			pr_warn("ad_select param only affects 802.3ad mode\n");
6170 	} else {
6171 		params->ad_select = BOND_AD_STABLE;
6172 	}
6173 
6174 	if (max_bonds < 0) {
6175 		pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6176 			max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6177 		max_bonds = BOND_DEFAULT_MAX_BONDS;
6178 	}
6179 
6180 	if (miimon < 0) {
6181 		pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6182 			miimon, INT_MAX);
6183 		miimon = 0;
6184 	}
6185 
6186 	if (updelay < 0) {
6187 		pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6188 			updelay, INT_MAX);
6189 		updelay = 0;
6190 	}
6191 
6192 	if (downdelay < 0) {
6193 		pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6194 			downdelay, INT_MAX);
6195 		downdelay = 0;
6196 	}
6197 
6198 	if ((use_carrier != 0) && (use_carrier != 1)) {
6199 		pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6200 			use_carrier);
6201 		use_carrier = 1;
6202 	}
6203 
6204 	if (num_peer_notif < 0 || num_peer_notif > 255) {
6205 		pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6206 			num_peer_notif);
6207 		num_peer_notif = 1;
6208 	}
6209 
6210 	/* reset values for 802.3ad/TLB/ALB */
6211 	if (!bond_mode_uses_arp(bond_mode)) {
6212 		if (!miimon) {
6213 			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");
6214 			pr_warn("Forcing miimon to 100msec\n");
6215 			miimon = BOND_DEFAULT_MIIMON;
6216 		}
6217 	}
6218 
6219 	if (tx_queues < 1 || tx_queues > 255) {
6220 		pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6221 			tx_queues, BOND_DEFAULT_TX_QUEUES);
6222 		tx_queues = BOND_DEFAULT_TX_QUEUES;
6223 	}
6224 
6225 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6226 		pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6227 			all_slaves_active);
6228 		all_slaves_active = 0;
6229 	}
6230 
6231 	if (resend_igmp < 0 || resend_igmp > 255) {
6232 		pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6233 			resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6234 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6235 	}
6236 
6237 	bond_opt_initval(&newval, packets_per_slave);
6238 	if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6239 		pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6240 			packets_per_slave, USHRT_MAX);
6241 		packets_per_slave = 1;
6242 	}
6243 
6244 	if (bond_mode == BOND_MODE_ALB) {
6245 		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",
6246 			  updelay);
6247 	}
6248 
6249 	if (!miimon) {
6250 		if (updelay || downdelay) {
6251 			/* just warn the user the up/down delay will have
6252 			 * no effect since miimon is zero...
6253 			 */
6254 			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",
6255 				updelay, downdelay);
6256 		}
6257 	} else {
6258 		/* don't allow arp monitoring */
6259 		if (arp_interval) {
6260 			pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6261 				miimon, arp_interval);
6262 			arp_interval = 0;
6263 		}
6264 
6265 		if ((updelay % miimon) != 0) {
6266 			pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6267 				updelay, miimon, (updelay / miimon) * miimon);
6268 		}
6269 
6270 		updelay /= miimon;
6271 
6272 		if ((downdelay % miimon) != 0) {
6273 			pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6274 				downdelay, miimon,
6275 				(downdelay / miimon) * miimon);
6276 		}
6277 
6278 		downdelay /= miimon;
6279 	}
6280 
6281 	if (arp_interval < 0) {
6282 		pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6283 			arp_interval, INT_MAX);
6284 		arp_interval = 0;
6285 	}
6286 
6287 	for (arp_ip_count = 0, i = 0;
6288 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6289 		__be32 ip;
6290 
6291 		/* not a complete check, but good enough to catch mistakes */
6292 		if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6293 		    !bond_is_ip_target_ok(ip)) {
6294 			pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6295 				arp_ip_target[i]);
6296 			arp_interval = 0;
6297 		} else {
6298 			if (bond_get_targets_ip(arp_target, ip) == -1)
6299 				arp_target[arp_ip_count++] = ip;
6300 			else
6301 				pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6302 					&ip);
6303 		}
6304 	}
6305 
6306 	if (arp_interval && !arp_ip_count) {
6307 		/* don't allow arping if no arp_ip_target given... */
6308 		pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6309 			arp_interval);
6310 		arp_interval = 0;
6311 	}
6312 
6313 	if (arp_validate) {
6314 		if (!arp_interval) {
6315 			pr_err("arp_validate requires arp_interval\n");
6316 			return -EINVAL;
6317 		}
6318 
6319 		bond_opt_initstr(&newval, arp_validate);
6320 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6321 					&newval);
6322 		if (!valptr) {
6323 			pr_err("Error: invalid arp_validate \"%s\"\n",
6324 			       arp_validate);
6325 			return -EINVAL;
6326 		}
6327 		arp_validate_value = valptr->value;
6328 	} else {
6329 		arp_validate_value = 0;
6330 	}
6331 
6332 	if (arp_all_targets) {
6333 		bond_opt_initstr(&newval, arp_all_targets);
6334 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6335 					&newval);
6336 		if (!valptr) {
6337 			pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6338 			       arp_all_targets);
6339 			arp_all_targets_value = 0;
6340 		} else {
6341 			arp_all_targets_value = valptr->value;
6342 		}
6343 	}
6344 
6345 	if (miimon) {
6346 		pr_info("MII link monitoring set to %d ms\n", miimon);
6347 	} else if (arp_interval) {
6348 		valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6349 					  arp_validate_value);
6350 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6351 			arp_interval, valptr->string, arp_ip_count);
6352 
6353 		for (i = 0; i < arp_ip_count; i++)
6354 			pr_cont(" %s", arp_ip_target[i]);
6355 
6356 		pr_cont("\n");
6357 
6358 	} else if (max_bonds) {
6359 		/* miimon and arp_interval not set, we need one so things
6360 		 * work as expected, see bonding.txt for details
6361 		 */
6362 		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");
6363 	}
6364 
6365 	if (primary && !bond_mode_uses_primary(bond_mode)) {
6366 		/* currently, using a primary only makes sense
6367 		 * in active backup, TLB or ALB modes
6368 		 */
6369 		pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6370 			primary, bond_mode_name(bond_mode));
6371 		primary = NULL;
6372 	}
6373 
6374 	if (primary && primary_reselect) {
6375 		bond_opt_initstr(&newval, primary_reselect);
6376 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6377 					&newval);
6378 		if (!valptr) {
6379 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
6380 			       primary_reselect);
6381 			return -EINVAL;
6382 		}
6383 		primary_reselect_value = valptr->value;
6384 	} else {
6385 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6386 	}
6387 
6388 	if (fail_over_mac) {
6389 		bond_opt_initstr(&newval, fail_over_mac);
6390 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6391 					&newval);
6392 		if (!valptr) {
6393 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
6394 			       fail_over_mac);
6395 			return -EINVAL;
6396 		}
6397 		fail_over_mac_value = valptr->value;
6398 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6399 			pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6400 	} else {
6401 		fail_over_mac_value = BOND_FOM_NONE;
6402 	}
6403 
6404 	bond_opt_initstr(&newval, "default");
6405 	valptr = bond_opt_parse(
6406 			bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6407 				     &newval);
6408 	if (!valptr) {
6409 		pr_err("Error: No ad_actor_sys_prio default value");
6410 		return -EINVAL;
6411 	}
6412 	ad_actor_sys_prio = valptr->value;
6413 
6414 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6415 				&newval);
6416 	if (!valptr) {
6417 		pr_err("Error: No ad_user_port_key default value");
6418 		return -EINVAL;
6419 	}
6420 	ad_user_port_key = valptr->value;
6421 
6422 	bond_opt_initstr(&newval, "default");
6423 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6424 	if (!valptr) {
6425 		pr_err("Error: No tlb_dynamic_lb default value");
6426 		return -EINVAL;
6427 	}
6428 	tlb_dynamic_lb = valptr->value;
6429 
6430 	if (lp_interval == 0) {
6431 		pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6432 			INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6433 		lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6434 	}
6435 
6436 	/* fill params struct with the proper values */
6437 	params->mode = bond_mode;
6438 	params->xmit_policy = xmit_hashtype;
6439 	params->miimon = miimon;
6440 	params->num_peer_notif = num_peer_notif;
6441 	params->arp_interval = arp_interval;
6442 	params->arp_validate = arp_validate_value;
6443 	params->arp_all_targets = arp_all_targets_value;
6444 	params->missed_max = 2;
6445 	params->updelay = updelay;
6446 	params->downdelay = downdelay;
6447 	params->peer_notif_delay = 0;
6448 	params->use_carrier = use_carrier;
6449 	params->lacp_active = 1;
6450 	params->lacp_fast = lacp_fast;
6451 	params->primary[0] = 0;
6452 	params->primary_reselect = primary_reselect_value;
6453 	params->fail_over_mac = fail_over_mac_value;
6454 	params->tx_queues = tx_queues;
6455 	params->all_slaves_active = all_slaves_active;
6456 	params->resend_igmp = resend_igmp;
6457 	params->min_links = min_links;
6458 	params->lp_interval = lp_interval;
6459 	params->packets_per_slave = packets_per_slave;
6460 	params->tlb_dynamic_lb = tlb_dynamic_lb;
6461 	params->ad_actor_sys_prio = ad_actor_sys_prio;
6462 	eth_zero_addr(params->ad_actor_system);
6463 	params->ad_user_port_key = ad_user_port_key;
6464 	params->coupled_control = 1;
6465 	if (packets_per_slave > 0) {
6466 		params->reciprocal_packets_per_slave =
6467 			reciprocal_value(packets_per_slave);
6468 	} else {
6469 		/* reciprocal_packets_per_slave is unused if
6470 		 * packets_per_slave is 0 or 1, just initialize it
6471 		 */
6472 		params->reciprocal_packets_per_slave =
6473 			(struct reciprocal_value) { 0 };
6474 	}
6475 
6476 	if (primary)
6477 		strscpy_pad(params->primary, primary, sizeof(params->primary));
6478 
6479 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6480 #if IS_ENABLED(CONFIG_IPV6)
6481 	memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6482 #endif
6483 
6484 	return 0;
6485 }
6486 
6487 /* Called from registration process */
6488 static int bond_init(struct net_device *bond_dev)
6489 {
6490 	struct bonding *bond = netdev_priv(bond_dev);
6491 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6492 
6493 	netdev_dbg(bond_dev, "Begin bond_init\n");
6494 
6495 	bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM,
6496 					   bond_dev->name);
6497 	if (!bond->wq)
6498 		return -ENOMEM;
6499 
6500 	bond->notifier_ctx = false;
6501 
6502 	spin_lock_init(&bond->stats_lock);
6503 	netdev_lockdep_set_classes(bond_dev);
6504 
6505 	list_add_tail_rcu(&bond->bond_list, &bn->dev_list);
6506 
6507 	bond_prepare_sysfs_group(bond);
6508 
6509 	bond_debug_register(bond);
6510 
6511 	/* Ensure valid dev_addr */
6512 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
6513 	    bond_dev->addr_assign_type == NET_ADDR_PERM)
6514 		eth_hw_addr_random(bond_dev);
6515 
6516 	return 0;
6517 }
6518 
6519 unsigned int bond_get_num_tx_queues(void)
6520 {
6521 	return tx_queues;
6522 }
6523 
6524 /* Create a new bond based on the specified name and bonding parameters.
6525  * If name is NULL, obtain a suitable "bond%d" name for us.
6526  * Caller must NOT hold rtnl_lock; we need to release it here before we
6527  * set up our sysfs entries.
6528  */
6529 int bond_create(struct net *net, const char *name)
6530 {
6531 	struct net_device *bond_dev;
6532 	struct bonding *bond;
6533 	int res = -ENOMEM;
6534 
6535 	rtnl_lock();
6536 
6537 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6538 				   name ? name : "bond%d", NET_NAME_UNKNOWN,
6539 				   bond_setup, tx_queues);
6540 	if (!bond_dev)
6541 		goto out;
6542 
6543 	bond = netdev_priv(bond_dev);
6544 	dev_net_set(bond_dev, net);
6545 	bond_dev->rtnl_link_ops = &bond_link_ops;
6546 
6547 	res = register_netdevice(bond_dev);
6548 	if (res < 0) {
6549 		free_netdev(bond_dev);
6550 		goto out;
6551 	}
6552 
6553 	netif_carrier_off(bond_dev);
6554 
6555 	bond_work_init_all(bond);
6556 
6557 out:
6558 	rtnl_unlock();
6559 	return res;
6560 }
6561 
6562 static int __net_init bond_net_init(struct net *net)
6563 {
6564 	struct bond_net *bn = net_generic(net, bond_net_id);
6565 
6566 	bn->net = net;
6567 	INIT_LIST_HEAD(&bn->dev_list);
6568 
6569 	bond_create_proc_dir(bn);
6570 	bond_create_sysfs(bn);
6571 
6572 	return 0;
6573 }
6574 
6575 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters
6576  * race condition in bond unloading") we need to remove sysfs files
6577  * before we remove our devices (done later in bond_net_exit_rtnl())
6578  */
6579 static void __net_exit bond_net_pre_exit(struct net *net)
6580 {
6581 	struct bond_net *bn = net_generic(net, bond_net_id);
6582 
6583 	bond_destroy_sysfs(bn);
6584 }
6585 
6586 static void __net_exit bond_net_exit_rtnl(struct net *net,
6587 					  struct list_head *dev_kill_list)
6588 {
6589 	struct bond_net *bn = net_generic(net, bond_net_id);
6590 	struct bonding *bond, *tmp_bond;
6591 
6592 	/* Kill off any bonds created after unregistering bond rtnl ops */
6593 	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6594 		unregister_netdevice_queue(bond->dev, dev_kill_list);
6595 }
6596 
6597 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory
6598  * only after all bonds are gone") bond_destroy_proc_dir() is called
6599  * after bond_net_exit_rtnl() has completed.
6600  */
6601 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6602 {
6603 	struct bond_net *bn;
6604 	struct net *net;
6605 
6606 	list_for_each_entry(net, net_list, exit_list) {
6607 		bn = net_generic(net, bond_net_id);
6608 		bond_destroy_proc_dir(bn);
6609 	}
6610 }
6611 
6612 static struct pernet_operations bond_net_ops = {
6613 	.init = bond_net_init,
6614 	.pre_exit = bond_net_pre_exit,
6615 	.exit_rtnl = bond_net_exit_rtnl,
6616 	.exit_batch = bond_net_exit_batch,
6617 	.id   = &bond_net_id,
6618 	.size = sizeof(struct bond_net),
6619 };
6620 
6621 static int __init bonding_init(void)
6622 {
6623 	int i;
6624 	int res;
6625 
6626 	res = bond_check_params(&bonding_defaults);
6627 	if (res)
6628 		goto out;
6629 
6630 	bond_create_debugfs();
6631 
6632 	res = register_pernet_subsys(&bond_net_ops);
6633 	if (res)
6634 		goto err_net_ops;
6635 
6636 	res = bond_netlink_init();
6637 	if (res)
6638 		goto err_link;
6639 
6640 	for (i = 0; i < max_bonds; i++) {
6641 		res = bond_create(&init_net, NULL);
6642 		if (res)
6643 			goto err;
6644 	}
6645 
6646 	skb_flow_dissector_init(&flow_keys_bonding,
6647 				flow_keys_bonding_keys,
6648 				ARRAY_SIZE(flow_keys_bonding_keys));
6649 
6650 	register_netdevice_notifier(&bond_netdev_notifier);
6651 out:
6652 	return res;
6653 err:
6654 	bond_netlink_fini();
6655 err_link:
6656 	unregister_pernet_subsys(&bond_net_ops);
6657 err_net_ops:
6658 	bond_destroy_debugfs();
6659 	goto out;
6660 
6661 }
6662 
6663 static void __exit bonding_exit(void)
6664 {
6665 	unregister_netdevice_notifier(&bond_netdev_notifier);
6666 
6667 	bond_netlink_fini();
6668 	unregister_pernet_subsys(&bond_net_ops);
6669 
6670 	bond_destroy_debugfs();
6671 
6672 #ifdef CONFIG_NET_POLL_CONTROLLER
6673 	/* Make sure we don't have an imbalance on our netpoll blocking */
6674 	WARN_ON(atomic_read(&netpoll_block_tx));
6675 #endif
6676 }
6677 
6678 module_init(bonding_init);
6679 module_exit(bonding_exit);
6680 MODULE_LICENSE("GPL");
6681 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6682 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
6683