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, "option obsolete, use_carrier cannot be disabled");
146	module_param(mode, charp, 0);
147	MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
148	"1 for active-backup, 2 for balance-xor, "
149	"3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
150	"6 for balance-alb");
151	module_param(primary, charp, 0);
152	MODULE_PARM_DESC(primary, "Primary network device to use");
153	module_param(primary_reselect, charp, 0);
154	MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
155	"once it comes up; "
156	"0 for always (default), "
157	"1 for only if speed of primary is "
158	"better, "
159	"2 for only on active slave "
160	"failure");
161	module_param(lacp_rate, charp, 0);
162	MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163	"0 for slow, 1 for fast");
164	module_param(ad_select, charp, 0);
165	MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166	"0 for stable (default), 1 for bandwidth, "
167	"2 for count");
168	module_param(min_links, int, 0);
169	MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
170
171	module_param(xmit_hash_policy, charp, 0);
172	MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
173	"0 for layer 2 (default), 1 for layer 3+4, "
174	"2 for layer 2+3, 3 for encap layer 2+3, "
175	"4 for encap layer 3+4, 5 for vlan+srcmac");
176	module_param(arp_interval, int, 0);
177	MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178	module_param_array(arp_ip_target, charp, NULL, 0);
179	MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180	module_param(arp_validate, charp, 0);
181	MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
182	"0 for none (default), 1 for active, "
183	"2 for backup, 3 for all");
184	module_param(arp_all_targets, charp, 0);
185	MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186	module_param(fail_over_mac, charp, 0);
187	MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
188	"the same MAC; 0 for none (default), "
189	"1 for active, 2 for follow");
190	module_param(all_slaves_active, int, 0);
191	MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
192	"by setting active flag for all slaves; "
193	"0 for never (default), 1 for always.");
194	module_param(resend_igmp, int, 0);
195	MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
196	"link failure");
197	module_param(packets_per_slave, int, 0);
198	MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
199	"mode; 0 for a random slave, 1 packet per "
200	"slave (default), >1 packets per slave.");
201	module_param(lp_interval, uint, 0);
202	MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
203	"the bonding driver sends learning packets to "
204	"each slaves peer switch. The default is 1.");
205
206	/*----------------------------- Global variables ----------------------------*/
207
208	#ifdef CONFIG_NET_POLL_CONTROLLER
209	atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210	#endif
211
212	unsigned int bond_net_id __read_mostly;
213
214	DEFINE_STATIC_KEY_FALSE(bond_bcast_neigh_enabled);
215
216	static const struct flow_dissector_key flow_keys_bonding_keys[] = {
217	{
218	.key_id = FLOW_DISSECTOR_KEY_CONTROL,
219	.offset = offsetof(struct flow_keys, control),
220	},
221	{
222	.key_id = FLOW_DISSECTOR_KEY_BASIC,
223	.offset = offsetof(struct flow_keys, basic),
224	},
225	{
226	.key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
227	.offset = offsetof(struct flow_keys, addrs.v4addrs),
228	},
229	{
230	.key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
231	.offset = offsetof(struct flow_keys, addrs.v6addrs),
232	},
233	{
234	.key_id = FLOW_DISSECTOR_KEY_TIPC,
235	.offset = offsetof(struct flow_keys, addrs.tipckey),
236	},
237	{
238	.key_id = FLOW_DISSECTOR_KEY_PORTS,
239	.offset = offsetof(struct flow_keys, ports),
240	},
241	{
242	.key_id = FLOW_DISSECTOR_KEY_ICMP,
243	.offset = offsetof(struct flow_keys, icmp),
244	},
245	{
246	.key_id = FLOW_DISSECTOR_KEY_VLAN,
247	.offset = offsetof(struct flow_keys, vlan),
248	},
249	{
250	.key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
251	.offset = offsetof(struct flow_keys, tags),
252	},
253	{
254	.key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
255	.offset = offsetof(struct flow_keys, keyid),
256	},
257	};
258
259	static struct flow_dissector flow_keys_bonding __read_mostly;
260
261	/*-------------------------- Forward declarations ---------------------------*/
262
263	static int bond_init(struct net_device *bond_dev);
264	static void bond_uninit(struct net_device *bond_dev);
265	static void bond_get_stats(struct net_device *bond_dev,
266	struct rtnl_link_stats64 *stats);
267	static void bond_slave_arr_handler(struct work_struct *work);
268	static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
269	int mod);
270	static void bond_netdev_notify_work(struct work_struct *work);
271
272	/*---------------------------- General routines -----------------------------*/
273
274	const char *bond_mode_name(int mode)
275	{
276	static const char *names[] = {
277	[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
278	[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
279	[BOND_MODE_XOR] = "load balancing (xor)",
280	[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
281	[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
282	[BOND_MODE_TLB] = "transmit load balancing",
283	[BOND_MODE_ALB] = "adaptive load balancing",
284	};
285
286	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
287	return "unknown";
288
289	return names[mode];
290	}
291
292	/**
293	* bond_dev_queue_xmit - Prepare skb for xmit.
294	*
295	* @bond: bond device that got this skb for tx.
296	* @skb: hw accel VLAN tagged skb to transmit
297	* @slave_dev: slave that is supposed to xmit this skbuff
298	*/
299	netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
300	struct net_device *slave_dev)
301	{
302	skb->dev = slave_dev;
303
304	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
305	sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
306	skb_set_queue_mapping(skb, queue_mapping: qdisc_skb_cb(skb)->slave_dev_queue_mapping);
307
308	if (unlikely(netpoll_tx_running(bond->dev)))
309	return bond_netpoll_send_skb(slave: bond_get_slave_by_dev(bond, slave_dev), skb);
310
311	return dev_queue_xmit(skb);
312	}
313
314	static bool bond_sk_check(struct bonding *bond)
315	{
316	switch (BOND_MODE(bond)) {
317	case BOND_MODE_8023AD:
318	case BOND_MODE_XOR:
319	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
320	return true;
321	fallthrough;
322	default:
323	return false;
324	}
325	}
326
327	bool bond_xdp_check(struct bonding *bond, int mode)
328	{
329	switch (mode) {
330	case BOND_MODE_ROUNDROBIN:
331	case BOND_MODE_ACTIVEBACKUP:
332	return true;
333	case BOND_MODE_8023AD:
334	case BOND_MODE_XOR:
335	/* vlan+srcmac is not supported with XDP as in most cases the 802.1q
336	* payload is not in the packet due to hardware offload.
337	*/
338	if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
339	return true;
340	fallthrough;
341	default:
342	return false;
343	}
344	}
345
346	/*---------------------------------- VLAN -----------------------------------*/
347
348	/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
349	* We don't protect the slave list iteration with a lock because:
350	* a. This operation is performed in IOCTL context,
351	* b. The operation is protected by the RTNL semaphore in the 8021q code,
352	* c. Holding a lock with BH disabled while directly calling a base driver
353	* entry point is generally a BAD idea.
354	*
355	* The design of synchronization/protection for this operation in the 8021q
356	* module is good for one or more VLAN devices over a single physical device
357	* and cannot be extended for a teaming solution like bonding, so there is a
358	* potential race condition here where a net device from the vlan group might
359	* be referenced (either by a base driver or the 8021q code) while it is being
360	* removed from the system. However, it turns out we're not making matters
361	* worse, and if it works for regular VLAN usage it will work here too.
362	*/
363
364	/**
365	* bond_vlan_rx_add_vid - Propagates adding an id to slaves
366	* @bond_dev: bonding net device that got called
367	* @proto: network protocol ID
368	* @vid: vlan id being added
369	*/
370	static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
371	__be16 proto, u16 vid)
372	{
373	struct bonding *bond = netdev_priv(dev: bond_dev);
374	struct slave *slave, *rollback_slave;
375	struct list_head *iter;
376	int res;
377
378	bond_for_each_slave(bond, slave, iter) {
379	res = vlan_vid_add(dev: slave->dev, proto, vid);
380	if (res)
381	goto unwind;
382	}
383
384	return 0;
385
386	unwind:
387	/* unwind to the slave that failed */
388	bond_for_each_slave(bond, rollback_slave, iter) {
389	if (rollback_slave == slave)
390	break;
391
392	vlan_vid_del(dev: rollback_slave->dev, proto, vid);
393	}
394
395	return res;
396	}
397
398	/**
399	* bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
400	* @bond_dev: bonding net device that got called
401	* @proto: network protocol ID
402	* @vid: vlan id being removed
403	*/
404	static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
405	__be16 proto, u16 vid)
406	{
407	struct bonding *bond = netdev_priv(dev: bond_dev);
408	struct list_head *iter;
409	struct slave *slave;
410
411	bond_for_each_slave(bond, slave, iter)
412	vlan_vid_del(dev: slave->dev, proto, vid);
413
414	if (bond_is_lb(bond))
415	bond_alb_clear_vlan(bond, vlan_id: vid);
416
417	return 0;
418	}
419
420	/*---------------------------------- XFRM -----------------------------------*/
421
422	#ifdef CONFIG_XFRM_OFFLOAD
423	/**
424	* bond_ipsec_dev - Get active device for IPsec offload
425	* @xs: pointer to transformer state struct
426	*
427	* Context: caller must hold rcu_read_lock.
428	*
429	* Return: the device for ipsec offload, or NULL if not exist.
430	**/
431	static struct net_device *bond_ipsec_dev(struct xfrm_state *xs)
432	{
433	struct net_device *bond_dev = xs->xso.dev;
434	struct bonding *bond;
435	struct slave *slave;
436
437	bond = netdev_priv(dev: bond_dev);
438	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
439	return NULL;
440
441	slave = rcu_dereference(bond->curr_active_slave);
442	if (!slave)
443	return NULL;
444
445	if (!xs->xso.real_dev)
446	return NULL;
447
448	if (xs->xso.real_dev != slave->dev)
449	pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n",
450	bond_dev->name, slave->dev->name, xs->xso.real_dev->name);
451
452	return slave->dev;
453	}
454
455	/**
456	* bond_ipsec_add_sa - program device with a security association
457	* @bond_dev: pointer to the bond net device
458	* @xs: pointer to transformer state struct
459	* @extack: extack point to fill failure reason
460	**/
461	static int bond_ipsec_add_sa(struct net_device *bond_dev,
462	struct xfrm_state *xs,
463	struct netlink_ext_ack *extack)
464	{
465	struct net_device *real_dev;
466	netdevice_tracker tracker;
467	struct bond_ipsec *ipsec;
468	struct bonding *bond;
469	struct slave *slave;
470	int err;
471
472	if (!bond_dev)
473	return -EINVAL;
474
475	rcu_read_lock();
476	bond = netdev_priv(dev: bond_dev);
477	slave = rcu_dereference(bond->curr_active_slave);
478	real_dev = slave ? slave->dev : NULL;
479	netdev_hold(dev: real_dev, tracker: &tracker, GFP_ATOMIC);
480	rcu_read_unlock();
481	if (!real_dev) {
482	err = -ENODEV;
483	goto out;
484	}
485
486	if (!real_dev->xfrmdev_ops ||
487	!real_dev->xfrmdev_ops->xdo_dev_state_add ||
488	netif_is_bond_master(dev: real_dev)) {
489	NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
490	err = -EINVAL;
491	goto out;
492	}
493
494	ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL);
495	if (!ipsec) {
496	err = -ENOMEM;
497	goto out;
498	}
499
500	err = real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev, xs, extack);
501	if (!err) {
502	xs->xso.real_dev = real_dev;
503	ipsec->xs = xs;
504	INIT_LIST_HEAD(list: &ipsec->list);
505	mutex_lock(&bond->ipsec_lock);
506	list_add(new: &ipsec->list, head: &bond->ipsec_list);
507	mutex_unlock(lock: &bond->ipsec_lock);
508	} else {
509	kfree(objp: ipsec);
510	}
511	out:
512	netdev_put(dev: real_dev, tracker: &tracker);
513	return err;
514	}
515
516	static void bond_ipsec_add_sa_all(struct bonding *bond)
517	{
518	struct net_device *bond_dev = bond->dev;
519	struct net_device *real_dev;
520	struct bond_ipsec *ipsec;
521	struct slave *slave;
522
523	slave = rtnl_dereference(bond->curr_active_slave);
524	real_dev = slave ? slave->dev : NULL;
525	if (!real_dev)
526	return;
527
528	mutex_lock(&bond->ipsec_lock);
529	if (!real_dev->xfrmdev_ops ||
530	!real_dev->xfrmdev_ops->xdo_dev_state_add ||
531	netif_is_bond_master(dev: real_dev)) {
532	if (!list_empty(head: &bond->ipsec_list))
533	slave_warn(bond_dev, real_dev,
534	"%s: no slave xdo_dev_state_add\n",
535	__func__);
536	goto out;
537	}
538
539	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
540	/* If new state is added before ipsec_lock acquired */
541	if (ipsec->xs->xso.real_dev == real_dev)
542	continue;
543
544	if (real_dev->xfrmdev_ops->xdo_dev_state_add(real_dev,
545	ipsec->xs, NULL)) {
546	slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__);
547	continue;
548	}
549
550	spin_lock_bh(lock: &ipsec->xs->lock);
551	/* xs might have been killed by the user during the migration
552	* to the new dev, but bond_ipsec_del_sa() should have done
553	* nothing, as xso.real_dev is NULL.
554	* Delete it from the device we just added it to. The pending
555	* bond_ipsec_free_sa() call will do the rest of the cleanup.
556	*/
557	if (ipsec->xs->km.state == XFRM_STATE_DEAD &&
558	real_dev->xfrmdev_ops->xdo_dev_state_delete)
559	real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev,
560	ipsec->xs);
561	ipsec->xs->xso.real_dev = real_dev;
562	spin_unlock_bh(lock: &ipsec->xs->lock);
563	}
564	out:
565	mutex_unlock(lock: &bond->ipsec_lock);
566	}
567
568	/**
569	* bond_ipsec_del_sa - clear out this specific SA
570	* @bond_dev: pointer to the bond net device
571	* @xs: pointer to transformer state struct
572	**/
573	static void bond_ipsec_del_sa(struct net_device *bond_dev,
574	struct xfrm_state *xs)
575	{
576	struct net_device *real_dev;
577
578	if (!bond_dev || !xs->xso.real_dev)
579	return;
580
581	real_dev = xs->xso.real_dev;
582
583	if (!real_dev->xfrmdev_ops ||
584	!real_dev->xfrmdev_ops->xdo_dev_state_delete ||
585	netif_is_bond_master(dev: real_dev)) {
586	slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__);
587	return;
588	}
589
590	real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev, xs);
591	}
592
593	static void bond_ipsec_del_sa_all(struct bonding *bond)
594	{
595	struct net_device *bond_dev = bond->dev;
596	struct net_device *real_dev;
597	struct bond_ipsec *ipsec;
598	struct slave *slave;
599
600	slave = rtnl_dereference(bond->curr_active_slave);
601	real_dev = slave ? slave->dev : NULL;
602	if (!real_dev)
603	return;
604
605	mutex_lock(&bond->ipsec_lock);
606	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
607	if (!ipsec->xs->xso.real_dev)
608	continue;
609
610	if (!real_dev->xfrmdev_ops ||
611	!real_dev->xfrmdev_ops->xdo_dev_state_delete ||
612	netif_is_bond_master(dev: real_dev)) {
613	slave_warn(bond_dev, real_dev,
614	"%s: no slave xdo_dev_state_delete\n",
615	__func__);
616	continue;
617	}
618
619	spin_lock_bh(lock: &ipsec->xs->lock);
620	ipsec->xs->xso.real_dev = NULL;
621	/* Don't double delete states killed by the user. */
622	if (ipsec->xs->km.state != XFRM_STATE_DEAD)
623	real_dev->xfrmdev_ops->xdo_dev_state_delete(real_dev,
624	ipsec->xs);
625	spin_unlock_bh(lock: &ipsec->xs->lock);
626
627	if (real_dev->xfrmdev_ops->xdo_dev_state_free)
628	real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev,
629	ipsec->xs);
630	}
631	mutex_unlock(lock: &bond->ipsec_lock);
632	}
633
634	static void bond_ipsec_free_sa(struct net_device *bond_dev,
635	struct xfrm_state *xs)
636	{
637	struct net_device *real_dev;
638	struct bond_ipsec *ipsec;
639	struct bonding *bond;
640
641	if (!bond_dev)
642	return;
643
644	bond = netdev_priv(dev: bond_dev);
645
646	mutex_lock(&bond->ipsec_lock);
647	if (!xs->xso.real_dev)
648	goto out;
649
650	real_dev = xs->xso.real_dev;
651
652	xs->xso.real_dev = NULL;
653	if (real_dev->xfrmdev_ops &&
654	real_dev->xfrmdev_ops->xdo_dev_state_free)
655	real_dev->xfrmdev_ops->xdo_dev_state_free(real_dev, xs);
656	out:
657	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
658	if (ipsec->xs == xs) {
659	list_del(entry: &ipsec->list);
660	kfree(objp: ipsec);
661	break;
662	}
663	}
664	mutex_unlock(lock: &bond->ipsec_lock);
665	}
666
667	/**
668	* bond_ipsec_offload_ok - can this packet use the xfrm hw offload
669	* @skb: current data packet
670	* @xs: pointer to transformer state struct
671	**/
672	static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
673	{
674	struct net_device *real_dev;
675
676	rcu_read_lock();
677	real_dev = bond_ipsec_dev(xs);
678	if (!real_dev || netif_is_bond_master(dev: real_dev)) {
679	rcu_read_unlock();
680	return false;
681	}
682
683	rcu_read_unlock();
684	return true;
685	}
686
687	/**
688	* bond_advance_esn_state - ESN support for IPSec HW offload
689	* @xs: pointer to transformer state struct
690	**/
691	static void bond_advance_esn_state(struct xfrm_state *xs)
692	{
693	struct net_device *real_dev;
694
695	rcu_read_lock();
696	real_dev = bond_ipsec_dev(xs);
697	if (!real_dev)
698	goto out;
699
700	if (!real_dev->xfrmdev_ops ||
701	!real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
702	pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name);
703	goto out;
704	}
705
706	real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs);
707	out:
708	rcu_read_unlock();
709	}
710
711	/**
712	* bond_xfrm_update_stats - Update xfrm state
713	* @xs: pointer to transformer state struct
714	**/
715	static void bond_xfrm_update_stats(struct xfrm_state *xs)
716	{
717	struct net_device *real_dev;
718
719	rcu_read_lock();
720	real_dev = bond_ipsec_dev(xs);
721	if (!real_dev)
722	goto out;
723
724	if (!real_dev->xfrmdev_ops ||
725	!real_dev->xfrmdev_ops->xdo_dev_state_update_stats) {
726	pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name);
727	goto out;
728	}
729
730	real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs);
731	out:
732	rcu_read_unlock();
733	}
734
735	static const struct xfrmdev_ops bond_xfrmdev_ops = {
736	.xdo_dev_state_add = bond_ipsec_add_sa,
737	.xdo_dev_state_delete = bond_ipsec_del_sa,
738	.xdo_dev_state_free = bond_ipsec_free_sa,
739	.xdo_dev_offload_ok = bond_ipsec_offload_ok,
740	.xdo_dev_state_advance_esn = bond_advance_esn_state,
741	.xdo_dev_state_update_stats = bond_xfrm_update_stats,
742	};
743	#endif /* CONFIG_XFRM_OFFLOAD */
744
745	/*------------------------------- Link status -------------------------------*/
746
747	/* Set the carrier state for the master according to the state of its
748	* slaves. If any slaves are up, the master is up. In 802.3ad mode,
749	* do special 802.3ad magic.
750	*
751	* Returns zero if carrier state does not change, nonzero if it does.
752	*/
753	int bond_set_carrier(struct bonding *bond)
754	{
755	struct list_head *iter;
756	struct slave *slave;
757
758	if (!bond_has_slaves(bond))
759	goto down;
760
761	if (BOND_MODE(bond) == BOND_MODE_8023AD)
762	return bond_3ad_set_carrier(bond);
763
764	bond_for_each_slave(bond, slave, iter) {
765	if (slave->link == BOND_LINK_UP) {
766	if (!netif_carrier_ok(dev: bond->dev)) {
767	netif_carrier_on(dev: bond->dev);
768	return 1;
769	}
770	return 0;
771	}
772	}
773
774	down:
775	if (netif_carrier_ok(dev: bond->dev)) {
776	netif_carrier_off(dev: bond->dev);
777	return 1;
778	}
779	return 0;
780	}
781
782	/* Get link speed and duplex from the slave's base driver
783	* using ethtool. If for some reason the call fails or the
784	* values are invalid, set speed and duplex to -1,
785	* and return. Return 1 if speed or duplex settings are
786	* UNKNOWN; 0 otherwise.
787	*/
788	static int bond_update_speed_duplex(struct slave *slave)
789	{
790	struct net_device *slave_dev = slave->dev;
791	struct ethtool_link_ksettings ecmd;
792	int res;
793
794	slave->speed = SPEED_UNKNOWN;
795	slave->duplex = DUPLEX_UNKNOWN;
796
797	res = __ethtool_get_link_ksettings(dev: slave_dev, link_ksettings: &ecmd);
798	if (res < 0)
799	return 1;
800	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
801	return 1;
802	switch (ecmd.base.duplex) {
803	case DUPLEX_FULL:
804	case DUPLEX_HALF:
805	break;
806	default:
807	return 1;
808	}
809
810	slave->speed = ecmd.base.speed;
811	slave->duplex = ecmd.base.duplex;
812
813	return 0;
814	}
815
816	const char *bond_slave_link_status(s8 link)
817	{
818	switch (link) {
819	case BOND_LINK_UP:
820	return "up";
821	case BOND_LINK_FAIL:
822	return "going down";
823	case BOND_LINK_DOWN:
824	return "down";
825	case BOND_LINK_BACK:
826	return "going back";
827	default:
828	return "unknown";
829	}
830	}
831
832	/*----------------------------- Multicast list ------------------------------*/
833
834	/* Push the promiscuity flag down to appropriate slaves */
835	static int bond_set_promiscuity(struct bonding *bond, int inc)
836	{
837	struct list_head *iter;
838	int err = 0;
839
840	if (bond_uses_primary(bond)) {
841	struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
842
843	if (curr_active)
844	err = dev_set_promiscuity(dev: curr_active->dev, inc);
845	} else {
846	struct slave *slave;
847
848	bond_for_each_slave(bond, slave, iter) {
849	err = dev_set_promiscuity(dev: slave->dev, inc);
850	if (err)
851	return err;
852	}
853	}
854	return err;
855	}
856
857	/* Push the allmulti flag down to all slaves */
858	static int bond_set_allmulti(struct bonding *bond, int inc)
859	{
860	struct list_head *iter;
861	int err = 0;
862
863	if (bond_uses_primary(bond)) {
864	struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
865
866	if (curr_active)
867	err = dev_set_allmulti(dev: curr_active->dev, inc);
868	} else {
869	struct slave *slave;
870
871	bond_for_each_slave(bond, slave, iter) {
872	err = dev_set_allmulti(dev: slave->dev, inc);
873	if (err)
874	return err;
875	}
876	}
877	return err;
878	}
879
880	/* Retrieve the list of registered multicast addresses for the bonding
881	* device and retransmit an IGMP JOIN request to the current active
882	* slave.
883	*/
884	static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
885	{
886	struct bonding *bond = container_of(work, struct bonding,
887	mcast_work.work);
888
889	if (!rtnl_trylock()) {
890	queue_delayed_work(wq: bond->wq, dwork: &bond->mcast_work, delay: 1);
891	return;
892	}
893	call_netdevice_notifiers(val: NETDEV_RESEND_IGMP, dev: bond->dev);
894
895	if (bond->igmp_retrans > 1) {
896	bond->igmp_retrans--;
897	queue_delayed_work(wq: bond->wq, dwork: &bond->mcast_work, HZ/5);
898	}
899	rtnl_unlock();
900	}
901
902	/* Flush bond's hardware addresses from slave */
903	static void bond_hw_addr_flush(struct net_device *bond_dev,
904	struct net_device *slave_dev)
905	{
906	struct bonding *bond = netdev_priv(dev: bond_dev);
907
908	dev_uc_unsync(to: slave_dev, from: bond_dev);
909	dev_mc_unsync(to: slave_dev, from: bond_dev);
910
911	if (BOND_MODE(bond) == BOND_MODE_8023AD)
912	dev_mc_del(dev: slave_dev, addr: lacpdu_mcast_addr);
913	}
914
915	/*--------------------------- Active slave change ---------------------------*/
916
917	/* Update the hardware address list and promisc/allmulti for the new and
918	* old active slaves (if any). Modes that are not using primary keep all
919	* slaves up date at all times; only the modes that use primary need to call
920	* this function to swap these settings during a failover.
921	*/
922	static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
923	struct slave *old_active)
924	{
925	if (old_active) {
926	if (bond->dev->flags & IFF_PROMISC)
927	dev_set_promiscuity(dev: old_active->dev, inc: -1);
928
929	if (bond->dev->flags & IFF_ALLMULTI)
930	dev_set_allmulti(dev: old_active->dev, inc: -1);
931
932	if (bond->dev->flags & IFF_UP)
933	bond_hw_addr_flush(bond_dev: bond->dev, slave_dev: old_active->dev);
934
935	bond_slave_ns_maddrs_add(bond, slave: old_active);
936	}
937
938	if (new_active) {
939	/* FIXME: Signal errors upstream. */
940	if (bond->dev->flags & IFF_PROMISC)
941	dev_set_promiscuity(dev: new_active->dev, inc: 1);
942
943	if (bond->dev->flags & IFF_ALLMULTI)
944	dev_set_allmulti(dev: new_active->dev, inc: 1);
945
946	if (bond->dev->flags & IFF_UP) {
947	netif_addr_lock_bh(dev: bond->dev);
948	dev_uc_sync(to: new_active->dev, from: bond->dev);
949	dev_mc_sync(to: new_active->dev, from: bond->dev);
950	netif_addr_unlock_bh(dev: bond->dev);
951	}
952
953	bond_slave_ns_maddrs_del(bond, slave: new_active);
954	}
955	}
956
957	/**
958	* bond_set_dev_addr - clone slave's address to bond
959	* @bond_dev: bond net device
960	* @slave_dev: slave net device
961	*
962	* Should be called with RTNL held.
963	*/
964	static int bond_set_dev_addr(struct net_device *bond_dev,
965	struct net_device *slave_dev)
966	{
967	int err;
968
969	slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
970	bond_dev, slave_dev, slave_dev->addr_len);
971	err = netif_pre_changeaddr_notify(dev: bond_dev, addr: slave_dev->dev_addr, NULL);
972	if (err)
973	return err;
974
975	__dev_addr_set(dev: bond_dev, addr: slave_dev->dev_addr, len: slave_dev->addr_len);
976	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
977	call_netdevice_notifiers(val: NETDEV_CHANGEADDR, dev: bond_dev);
978	return 0;
979	}
980
981	static struct slave *bond_get_old_active(struct bonding *bond,
982	struct slave *new_active)
983	{
984	struct slave *slave;
985	struct list_head *iter;
986
987	bond_for_each_slave(bond, slave, iter) {
988	if (slave == new_active)
989	continue;
990
991	if (ether_addr_equal(addr1: bond->dev->dev_addr, addr2: slave->dev->dev_addr))
992	return slave;
993	}
994
995	return NULL;
996	}
997
998	/* bond_do_fail_over_mac
999	*
1000	* Perform special MAC address swapping for fail_over_mac settings
1001	*
1002	* Called with RTNL
1003	*/
1004	static void bond_do_fail_over_mac(struct bonding *bond,
1005	struct slave *new_active,
1006	struct slave *old_active)
1007	{
1008	u8 tmp_mac[MAX_ADDR_LEN];
1009	struct sockaddr_storage ss;
1010	int rv;
1011
1012	switch (bond->params.fail_over_mac) {
1013	case BOND_FOM_ACTIVE:
1014	if (new_active) {
1015	rv = bond_set_dev_addr(bond_dev: bond->dev, slave_dev: new_active->dev);
1016	if (rv)
1017	slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
1018	-rv);
1019	}
1020	break;
1021	case BOND_FOM_FOLLOW:
1022	/* if new_active && old_active, swap them
1023	* if just old_active, do nothing (going to no active slave)
1024	* if just new_active, set new_active to bond's MAC
1025	*/
1026	if (!new_active)
1027	return;
1028
1029	if (!old_active)
1030	old_active = bond_get_old_active(bond, new_active);
1031
1032	if (old_active) {
1033	bond_hw_addr_copy(dst: tmp_mac, src: new_active->dev->dev_addr,
1034	len: new_active->dev->addr_len);
1035	bond_hw_addr_copy(dst: ss.__data,
1036	src: old_active->dev->dev_addr,
1037	len: old_active->dev->addr_len);
1038	ss.ss_family = new_active->dev->type;
1039	} else {
1040	bond_hw_addr_copy(dst: ss.__data, src: bond->dev->dev_addr,
1041	len: bond->dev->addr_len);
1042	ss.ss_family = bond->dev->type;
1043	}
1044
1045	rv = dev_set_mac_address(dev: new_active->dev, ss: &ss, NULL);
1046	if (rv) {
1047	slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1048	-rv);
1049	goto out;
1050	}
1051
1052	if (!old_active)
1053	goto out;
1054
1055	bond_hw_addr_copy(dst: ss.__data, src: tmp_mac,
1056	len: new_active->dev->addr_len);
1057	ss.ss_family = old_active->dev->type;
1058
1059	rv = dev_set_mac_address(dev: old_active->dev, ss: &ss, NULL);
1060	if (rv)
1061	slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1062	-rv);
1063	out:
1064	break;
1065	default:
1066	netdev_err(dev: bond->dev, format: "bond_do_fail_over_mac impossible: bad policy %d\n",
1067	bond->params.fail_over_mac);
1068	break;
1069	}
1070
1071	}
1072
1073	/**
1074	* bond_choose_primary_or_current - select the primary or high priority slave
1075	* @bond: our bonding struct
1076	*
1077	* - Check if there is a primary link. If the primary link was set and is up,
1078	* go on and do link reselection.
1079	*
1080	* - If primary link is not set or down, find the highest priority link.
1081	* If the highest priority link is not current slave, set it as primary
1082	* link and do link reselection.
1083	*/
1084	static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1085	{
1086	struct slave *prim = rtnl_dereference(bond->primary_slave);
1087	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1088	struct slave *slave, *hprio = NULL;
1089	struct list_head *iter;
1090
1091	if (!prim || prim->link != BOND_LINK_UP) {
1092	bond_for_each_slave(bond, slave, iter) {
1093	if (slave->link == BOND_LINK_UP) {
1094	hprio = hprio ?: slave;
1095	if (slave->prio > hprio->prio)
1096	hprio = slave;
1097	}
1098	}
1099
1100	if (hprio && hprio != curr) {
1101	prim = hprio;
1102	goto link_reselect;
1103	}
1104
1105	if (!curr || curr->link != BOND_LINK_UP)
1106	return NULL;
1107	return curr;
1108	}
1109
1110	if (bond->force_primary) {
1111	bond->force_primary = false;
1112	return prim;
1113	}
1114
1115	link_reselect:
1116	if (!curr || curr->link != BOND_LINK_UP)
1117	return prim;
1118
1119	/* At this point, prim and curr are both up */
1120	switch (bond->params.primary_reselect) {
1121	case BOND_PRI_RESELECT_ALWAYS:
1122	return prim;
1123	case BOND_PRI_RESELECT_BETTER:
1124	if (prim->speed < curr->speed)
1125	return curr;
1126	if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1127	return curr;
1128	return prim;
1129	case BOND_PRI_RESELECT_FAILURE:
1130	return curr;
1131	default:
1132	netdev_err(dev: bond->dev, format: "impossible primary_reselect %d\n",
1133	bond->params.primary_reselect);
1134	return curr;
1135	}
1136	}
1137
1138	/**
1139	* bond_find_best_slave - select the best available slave to be the active one
1140	* @bond: our bonding struct
1141	*/
1142	static struct slave *bond_find_best_slave(struct bonding *bond)
1143	{
1144	struct slave *slave, *bestslave = NULL;
1145	struct list_head *iter;
1146	int mintime = bond->params.updelay;
1147
1148	slave = bond_choose_primary_or_current(bond);
1149	if (slave)
1150	return slave;
1151
1152	bond_for_each_slave(bond, slave, iter) {
1153	if (slave->link == BOND_LINK_UP)
1154	return slave;
1155	if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1156	slave->delay < mintime) {
1157	mintime = slave->delay;
1158	bestslave = slave;
1159	}
1160	}
1161
1162	return bestslave;
1163	}
1164
1165	/* must be called in RCU critical section or with RTNL held */
1166	static bool bond_should_notify_peers(struct bonding *bond)
1167	{
1168	struct bond_up_slave *usable;
1169	struct slave *slave = NULL;
1170
1171	if (!bond->send_peer_notif ||
1172	bond->send_peer_notif %
1173	max(1, bond->params.peer_notif_delay) != 0 ||
1174	!netif_carrier_ok(dev: bond->dev))
1175	return false;
1176
1177	/* The send_peer_notif is set by active-backup or 8023ad
1178	* mode, and cleared in bond_close() when changing mode.
1179	* It is safe to only check bond mode here.
1180	*/
1181	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1182	usable = rcu_dereference_rtnl(bond->usable_slaves);
1183	if (!usable || !READ_ONCE(usable->count))
1184	return false;
1185	} else {
1186	slave = rcu_dereference_rtnl(bond->curr_active_slave);
1187	if (!slave || test_bit(__LINK_STATE_LINKWATCH_PENDING,
1188	&slave->dev->state))
1189	return false;
1190	}
1191
1192	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1193	slave ? slave->dev->name : "all");
1194
1195	return true;
1196	}
1197
1198	/**
1199	* bond_change_active_slave - change the active slave into the specified one
1200	* @bond: our bonding struct
1201	* @new_active: the new slave to make the active one
1202	*
1203	* Set the new slave to the bond's settings and unset them on the old
1204	* curr_active_slave.
1205	* Setting include flags, mc-list, promiscuity, allmulti, etc.
1206	*
1207	* If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1208	* because it is apparently the best available slave we have, even though its
1209	* updelay hasn't timed out yet.
1210	*
1211	* Caller must hold RTNL.
1212	*/
1213	void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1214	{
1215	struct slave *old_active;
1216
1217	ASSERT_RTNL();
1218
1219	old_active = rtnl_dereference(bond->curr_active_slave);
1220
1221	if (old_active == new_active)
1222	return;
1223
1224	#ifdef CONFIG_XFRM_OFFLOAD
1225	bond_ipsec_del_sa_all(bond);
1226	#endif /* CONFIG_XFRM_OFFLOAD */
1227
1228	if (new_active) {
1229	new_active->last_link_up = jiffies;
1230
1231	if (new_active->link == BOND_LINK_BACK) {
1232	if (bond_uses_primary(bond)) {
1233	slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1234	(bond->params.updelay - new_active->delay) * bond->params.miimon);
1235	}
1236
1237	new_active->delay = 0;
1238	bond_set_slave_link_state(slave: new_active, BOND_LINK_UP,
1239	BOND_SLAVE_NOTIFY_NOW);
1240
1241	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1242	bond_3ad_handle_link_change(slave: new_active, BOND_LINK_UP);
1243
1244	if (bond_is_lb(bond))
1245	bond_alb_handle_link_change(bond, slave: new_active, BOND_LINK_UP);
1246	} else {
1247	if (bond_uses_primary(bond))
1248	slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1249	}
1250	}
1251
1252	if (bond_uses_primary(bond))
1253	bond_hw_addr_swap(bond, new_active, old_active);
1254
1255	if (bond_is_lb(bond)) {
1256	bond_alb_handle_active_change(bond, new_slave: new_active);
1257	if (old_active)
1258	bond_set_slave_inactive_flags(slave: old_active,
1259	BOND_SLAVE_NOTIFY_NOW);
1260	if (new_active)
1261	bond_set_slave_active_flags(slave: new_active,
1262	BOND_SLAVE_NOTIFY_NOW);
1263	} else {
1264	rcu_assign_pointer(bond->curr_active_slave, new_active);
1265	}
1266
1267	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1268	if (old_active)
1269	bond_set_slave_inactive_flags(slave: old_active,
1270	BOND_SLAVE_NOTIFY_NOW);
1271
1272	if (new_active) {
1273	bool should_notify_peers = false;
1274
1275	bond_set_slave_active_flags(slave: new_active,
1276	BOND_SLAVE_NOTIFY_NOW);
1277
1278	if (bond->params.fail_over_mac)
1279	bond_do_fail_over_mac(bond, new_active,
1280	old_active);
1281
1282	if (netif_running(dev: bond->dev)) {
1283	bond->send_peer_notif =
1284	bond->params.num_peer_notif *
1285	max(1, bond->params.peer_notif_delay);
1286	should_notify_peers =
1287	bond_should_notify_peers(bond);
1288	}
1289
1290	call_netdevice_notifiers(val: NETDEV_BONDING_FAILOVER, dev: bond->dev);
1291	if (should_notify_peers) {
1292	bond->send_peer_notif--;
1293	call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS,
1294	dev: bond->dev);
1295	}
1296	}
1297	}
1298
1299	#ifdef CONFIG_XFRM_OFFLOAD
1300	bond_ipsec_add_sa_all(bond);
1301	#endif /* CONFIG_XFRM_OFFLOAD */
1302
1303	/* resend IGMP joins since active slave has changed or
1304	* all were sent on curr_active_slave.
1305	* resend only if bond is brought up with the affected
1306	* bonding modes and the retransmission is enabled
1307	*/
1308	if (netif_running(dev: bond->dev) && (bond->params.resend_igmp > 0) &&
1309	((bond_uses_primary(bond) && new_active) ||
1310	BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1311	bond->igmp_retrans = bond->params.resend_igmp;
1312	queue_delayed_work(wq: bond->wq, dwork: &bond->mcast_work, delay: 1);
1313	}
1314	}
1315
1316	/**
1317	* bond_select_active_slave - select a new active slave, if needed
1318	* @bond: our bonding struct
1319	*
1320	* This functions should be called when one of the following occurs:
1321	* - The old curr_active_slave has been released or lost its link.
1322	* - The primary_slave has got its link back.
1323	* - A slave has got its link back and there's no old curr_active_slave.
1324	*
1325	* Caller must hold RTNL.
1326	*/
1327	void bond_select_active_slave(struct bonding *bond)
1328	{
1329	struct slave *best_slave;
1330	int rv;
1331
1332	ASSERT_RTNL();
1333
1334	best_slave = bond_find_best_slave(bond);
1335	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1336	bond_change_active_slave(bond, new_active: best_slave);
1337	rv = bond_set_carrier(bond);
1338	if (!rv)
1339	return;
1340
1341	if (netif_carrier_ok(dev: bond->dev))
1342	netdev_info(dev: bond->dev, format: "active interface up!\n");
1343	else
1344	netdev_info(dev: bond->dev, format: "now running without any active interface!\n");
1345	}
1346	}
1347
1348	#ifdef CONFIG_NET_POLL_CONTROLLER
1349	static inline int slave_enable_netpoll(struct slave *slave)
1350	{
1351	struct netpoll *np;
1352	int err = 0;
1353
1354	np = kzalloc(sizeof(*np), GFP_KERNEL);
1355	err = -ENOMEM;
1356	if (!np)
1357	goto out;
1358
1359	err = __netpoll_setup(np, ndev: slave->dev);
1360	if (err) {
1361	kfree(objp: np);
1362	goto out;
1363	}
1364	slave->np = np;
1365	out:
1366	return err;
1367	}
1368	static inline void slave_disable_netpoll(struct slave *slave)
1369	{
1370	struct netpoll *np = slave->np;
1371
1372	if (!np)
1373	return;
1374
1375	slave->np = NULL;
1376
1377	__netpoll_free(np);
1378	}
1379
1380	static void bond_poll_controller(struct net_device *bond_dev)
1381	{
1382	struct bonding *bond = netdev_priv(dev: bond_dev);
1383	struct slave *slave = NULL;
1384	struct list_head *iter;
1385	struct ad_info ad_info;
1386
1387	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1388	if (bond_3ad_get_active_agg_info(bond, ad_info: &ad_info))
1389	return;
1390
1391	bond_for_each_slave_rcu(bond, slave, iter) {
1392	if (!bond_slave_is_up(slave))
1393	continue;
1394
1395	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1396	struct aggregator *agg =
1397	SLAVE_AD_INFO(slave)->port.aggregator;
1398
1399	if (agg &&
1400	agg->aggregator_identifier != ad_info.aggregator_id)
1401	continue;
1402	}
1403
1404	netpoll_poll_dev(dev: slave->dev);
1405	}
1406	}
1407
1408	static void bond_netpoll_cleanup(struct net_device *bond_dev)
1409	{
1410	struct bonding *bond = netdev_priv(dev: bond_dev);
1411	struct list_head *iter;
1412	struct slave *slave;
1413
1414	bond_for_each_slave(bond, slave, iter)
1415	if (bond_slave_is_up(slave))
1416	slave_disable_netpoll(slave);
1417	}
1418
1419	static int bond_netpoll_setup(struct net_device *dev)
1420	{
1421	struct bonding *bond = netdev_priv(dev);
1422	struct list_head *iter;
1423	struct slave *slave;
1424	int err = 0;
1425
1426	bond_for_each_slave(bond, slave, iter) {
1427	err = slave_enable_netpoll(slave);
1428	if (err) {
1429	bond_netpoll_cleanup(bond_dev: dev);
1430	break;
1431	}
1432	}
1433	return err;
1434	}
1435	#else
1436	static inline int slave_enable_netpoll(struct slave *slave)
1437	{
1438	return 0;
1439	}
1440	static inline void slave_disable_netpoll(struct slave *slave)
1441	{
1442	}
1443	static void bond_netpoll_cleanup(struct net_device *bond_dev)
1444	{
1445	}
1446	#endif
1447
1448	/*---------------------------------- IOCTL ----------------------------------*/
1449
1450	static netdev_features_t bond_fix_features(struct net_device *dev,
1451	netdev_features_t features)
1452	{
1453	struct bonding *bond = netdev_priv(dev);
1454	struct list_head *iter;
1455	netdev_features_t mask;
1456	struct slave *slave;
1457
1458	mask = features;
1459	features = netdev_base_features(features);
1460
1461	bond_for_each_slave(bond, slave, iter) {
1462	features = netdev_increment_features(all: features,
1463	one: slave->dev->features,
1464	mask);
1465	}
1466	features = netdev_add_tso_features(features, mask);
1467
1468	return features;
1469	}
1470
1471	static void bond_setup_by_slave(struct net_device *bond_dev,
1472	struct net_device *slave_dev)
1473	{
1474	bool was_up = !!(bond_dev->flags & IFF_UP);
1475
1476	dev_close(dev: bond_dev);
1477
1478	bond_dev->header_ops = slave_dev->header_ops;
1479
1480	bond_dev->type = slave_dev->type;
1481	bond_dev->hard_header_len = slave_dev->hard_header_len;
1482	bond_dev->needed_headroom = slave_dev->needed_headroom;
1483	bond_dev->addr_len = slave_dev->addr_len;
1484
1485	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1486	slave_dev->addr_len);
1487
1488	if (slave_dev->flags & IFF_POINTOPOINT) {
1489	bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1490	bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1491	}
1492	if (was_up)
1493	dev_open(dev: bond_dev, NULL);
1494	}
1495
1496	/* On bonding slaves other than the currently active slave, suppress
1497	* duplicates except for alb non-mcast/bcast.
1498	*/
1499	static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1500	struct slave *slave,
1501	struct bonding *bond)
1502	{
1503	if (bond_is_slave_inactive(slave)) {
1504	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1505	skb->pkt_type != PACKET_BROADCAST &&
1506	skb->pkt_type != PACKET_MULTICAST)
1507	return false;
1508	return true;
1509	}
1510	return false;
1511	}
1512
1513	static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1514	{
1515	struct sk_buff *skb = *pskb;
1516	struct slave *slave;
1517	struct bonding *bond;
1518	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1519	struct slave *);
1520	int ret = RX_HANDLER_ANOTHER;
1521
1522	skb = skb_share_check(skb, GFP_ATOMIC);
1523	if (unlikely(!skb))
1524	return RX_HANDLER_CONSUMED;
1525
1526	*pskb = skb;
1527
1528	slave = bond_slave_get_rcu(skb->dev);
1529	bond = slave->bond;
1530
1531	recv_probe = READ_ONCE(bond->recv_probe);
1532	if (recv_probe) {
1533	ret = recv_probe(skb, bond, slave);
1534	if (ret == RX_HANDLER_CONSUMED) {
1535	consume_skb(skb);
1536	return ret;
1537	}
1538	}
1539
1540	/*
1541	* For packets determined by bond_should_deliver_exact_match() call to
1542	* be suppressed we want to make an exception for link-local packets.
1543	* This is necessary for e.g. LLDP daemons to be able to monitor
1544	* inactive slave links without being forced to bind to them
1545	* explicitly.
1546	*
1547	* At the same time, packets that are passed to the bonding master
1548	* (including link-local ones) can have their originating interface
1549	* determined via PACKET_ORIGDEV socket option.
1550	*/
1551	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1552	if (is_link_local_ether_addr(addr: eth_hdr(skb)->h_dest))
1553	return RX_HANDLER_PASS;
1554	return RX_HANDLER_EXACT;
1555	}
1556
1557	skb->dev = bond->dev;
1558
1559	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1560	netif_is_bridge_port(dev: bond->dev) &&
1561	skb->pkt_type == PACKET_HOST) {
1562
1563	if (unlikely(skb_cow_head(skb,
1564	skb->data - skb_mac_header(skb)))) {
1565	kfree_skb(skb);
1566	return RX_HANDLER_CONSUMED;
1567	}
1568	bond_hw_addr_copy(dst: eth_hdr(skb)->h_dest, src: bond->dev->dev_addr,
1569	len: bond->dev->addr_len);
1570	}
1571
1572	return ret;
1573	}
1574
1575	static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1576	{
1577	switch (BOND_MODE(bond)) {
1578	case BOND_MODE_ROUNDROBIN:
1579	return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1580	case BOND_MODE_ACTIVEBACKUP:
1581	return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1582	case BOND_MODE_BROADCAST:
1583	return NETDEV_LAG_TX_TYPE_BROADCAST;
1584	case BOND_MODE_XOR:
1585	case BOND_MODE_8023AD:
1586	return NETDEV_LAG_TX_TYPE_HASH;
1587	default:
1588	return NETDEV_LAG_TX_TYPE_UNKNOWN;
1589	}
1590	}
1591
1592	static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1593	enum netdev_lag_tx_type type)
1594	{
1595	if (type != NETDEV_LAG_TX_TYPE_HASH)
1596	return NETDEV_LAG_HASH_NONE;
1597
1598	switch (bond->params.xmit_policy) {
1599	case BOND_XMIT_POLICY_LAYER2:
1600	return NETDEV_LAG_HASH_L2;
1601	case BOND_XMIT_POLICY_LAYER34:
1602	return NETDEV_LAG_HASH_L34;
1603	case BOND_XMIT_POLICY_LAYER23:
1604	return NETDEV_LAG_HASH_L23;
1605	case BOND_XMIT_POLICY_ENCAP23:
1606	return NETDEV_LAG_HASH_E23;
1607	case BOND_XMIT_POLICY_ENCAP34:
1608	return NETDEV_LAG_HASH_E34;
1609	case BOND_XMIT_POLICY_VLAN_SRCMAC:
1610	return NETDEV_LAG_HASH_VLAN_SRCMAC;
1611	default:
1612	return NETDEV_LAG_HASH_UNKNOWN;
1613	}
1614	}
1615
1616	static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1617	struct netlink_ext_ack *extack)
1618	{
1619	struct netdev_lag_upper_info lag_upper_info;
1620	enum netdev_lag_tx_type type;
1621	int err;
1622
1623	type = bond_lag_tx_type(bond);
1624	lag_upper_info.tx_type = type;
1625	lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1626
1627	err = netdev_master_upper_dev_link(dev: slave->dev, upper_dev: bond->dev, upper_priv: slave,
1628	upper_info: &lag_upper_info, extack);
1629	if (err)
1630	return err;
1631
1632	slave->dev->flags |= IFF_SLAVE;
1633	return 0;
1634	}
1635
1636	static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1637	{
1638	netdev_upper_dev_unlink(dev: slave->dev, upper_dev: bond->dev);
1639	slave->dev->flags &= ~IFF_SLAVE;
1640	}
1641
1642	static void slave_kobj_release(struct kobject *kobj)
1643	{
1644	struct slave *slave = to_slave(kobj);
1645	struct bonding *bond = bond_get_bond_by_slave(slave);
1646
1647	cancel_delayed_work_sync(dwork: &slave->notify_work);
1648	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1649	kfree(SLAVE_AD_INFO(slave));
1650
1651	kfree(objp: slave);
1652	}
1653
1654	static struct kobj_type slave_ktype = {
1655	.release = slave_kobj_release,
1656	#ifdef CONFIG_SYSFS
1657	.sysfs_ops = &slave_sysfs_ops,
1658	#endif
1659	};
1660
1661	static int bond_kobj_init(struct slave *slave)
1662	{
1663	int err;
1664
1665	err = kobject_init_and_add(kobj: &slave->kobj, ktype: &slave_ktype,
1666	parent: &(slave->dev->dev.kobj), fmt: "bonding_slave");
1667	if (err)
1668	kobject_put(kobj: &slave->kobj);
1669
1670	return err;
1671	}
1672
1673	static struct slave *bond_alloc_slave(struct bonding *bond,
1674	struct net_device *slave_dev)
1675	{
1676	struct slave *slave = NULL;
1677
1678	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1679	if (!slave)
1680	return NULL;
1681
1682	slave->bond = bond;
1683	slave->dev = slave_dev;
1684	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1685
1686	if (bond_kobj_init(slave))
1687	return NULL;
1688
1689	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1690	SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1691	GFP_KERNEL);
1692	if (!SLAVE_AD_INFO(slave)) {
1693	kobject_put(kobj: &slave->kobj);
1694	return NULL;
1695	}
1696	}
1697
1698	return slave;
1699	}
1700
1701	static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1702	{
1703	info->bond_mode = BOND_MODE(bond);
1704	info->miimon = bond->params.miimon;
1705	info->num_slaves = bond->slave_cnt;
1706	}
1707
1708	static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1709	{
1710	strcpy(p: info->slave_name, q: slave->dev->name);
1711	info->link = slave->link;
1712	info->state = bond_slave_state(slave);
1713	info->link_failure_count = slave->link_failure_count;
1714	}
1715
1716	static void bond_netdev_notify_work(struct work_struct *_work)
1717	{
1718	struct slave *slave = container_of(_work, struct slave,
1719	notify_work.work);
1720
1721	if (rtnl_trylock()) {
1722	struct netdev_bonding_info binfo;
1723
1724	bond_fill_ifslave(slave, info: &binfo.slave);
1725	bond_fill_ifbond(bond: slave->bond, info: &binfo.master);
1726	netdev_bonding_info_change(dev: slave->dev, bonding_info: &binfo);
1727	rtnl_unlock();
1728	} else {
1729	queue_delayed_work(wq: slave->bond->wq, dwork: &slave->notify_work, delay: 1);
1730	}
1731	}
1732
1733	void bond_queue_slave_event(struct slave *slave)
1734	{
1735	queue_delayed_work(wq: slave->bond->wq, dwork: &slave->notify_work, delay: 0);
1736	}
1737
1738	void bond_lower_state_changed(struct slave *slave)
1739	{
1740	struct netdev_lag_lower_state_info info;
1741
1742	info.link_up = slave->link == BOND_LINK_UP ||
1743	slave->link == BOND_LINK_FAIL;
1744	info.tx_enabled = bond_is_active_slave(slave);
1745	netdev_lower_state_changed(lower_dev: slave->dev, lower_state_info: &info);
1746	}
1747
1748	#define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1749	if (extack) \
1750	NL_SET_ERR_MSG(extack, errmsg); \
1751	else \
1752	netdev_err(bond_dev, "Error: %s\n", errmsg); \
1753	} while (0)
1754
1755	#define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1756	if (extack) \
1757	NL_SET_ERR_MSG(extack, errmsg); \
1758	else \
1759	slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1760	} while (0)
1761
1762	/* The bonding driver uses ether_setup() to convert a master bond device
1763	* to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1764	* have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1765	* if they were set
1766	*/
1767	static void bond_ether_setup(struct net_device *bond_dev)
1768	{
1769	unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1770
1771	ether_setup(dev: bond_dev);
1772	bond_dev->flags |= IFF_MASTER | flags;
1773	bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1774	}
1775
1776	void bond_xdp_set_features(struct net_device *bond_dev)
1777	{
1778	struct bonding *bond = netdev_priv(dev: bond_dev);
1779	xdp_features_t val = NETDEV_XDP_ACT_MASK;
1780	struct list_head *iter;
1781	struct slave *slave;
1782
1783	ASSERT_RTNL();
1784
1785	if (!bond_xdp_check(bond, BOND_MODE(bond)) || !bond_has_slaves(bond)) {
1786	xdp_clear_features_flag(dev: bond_dev);
1787	return;
1788	}
1789
1790	bond_for_each_slave(bond, slave, iter)
1791	val &= slave->dev->xdp_features;
1792
1793	val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1794
1795	xdp_set_features_flag(dev: bond_dev, val);
1796	}
1797
1798	/* enslave device <slave> to bond device <master> */
1799	int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1800	struct netlink_ext_ack *extack)
1801	{
1802	struct bonding *bond = netdev_priv(dev: bond_dev);
1803	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1804	struct slave *new_slave = NULL, *prev_slave;
1805	struct sockaddr_storage ss;
1806	int res = 0, i;
1807
1808	if (slave_dev->flags & IFF_MASTER &&
1809	!netif_is_bond_master(dev: slave_dev)) {
1810	BOND_NL_ERR(bond_dev, extack,
1811	"Device type (master device) cannot be enslaved");
1812	return -EPERM;
1813	}
1814
1815	/* already in-use? */
1816	if (netdev_is_rx_handler_busy(dev: slave_dev)) {
1817	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1818	"Device is in use and cannot be enslaved");
1819	return -EBUSY;
1820	}
1821
1822	if (bond_dev == slave_dev) {
1823	BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1824	return -EPERM;
1825	}
1826
1827	/* vlan challenged mutual exclusion */
1828	/* no need to lock since we're protected by rtnl_lock */
1829	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1830	slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1831	if (vlan_uses_dev(dev: bond_dev)) {
1832	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1833	"Can not enslave VLAN challenged device to VLAN enabled bond");
1834	return -EPERM;
1835	} else {
1836	slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1837	}
1838	} else {
1839	slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1840	}
1841
1842	if (slave_dev->features & NETIF_F_HW_ESP)
1843	slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1844
1845	/* Old ifenslave binaries are no longer supported. These can
1846	* be identified with moderate accuracy by the state of the slave:
1847	* the current ifenslave will set the interface down prior to
1848	* enslaving it; the old ifenslave will not.
1849	*/
1850	if (slave_dev->flags & IFF_UP) {
1851	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1852	"Device can not be enslaved while up");
1853	return -EPERM;
1854	}
1855
1856	/* set bonding device ether type by slave - bonding netdevices are
1857	* created with ether_setup, so when the slave type is not ARPHRD_ETHER
1858	* there is a need to override some of the type dependent attribs/funcs.
1859	*
1860	* bond ether type mutual exclusion - don't allow slaves of dissimilar
1861	* ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1862	*/
1863	if (!bond_has_slaves(bond)) {
1864	if (bond_dev->type != slave_dev->type) {
1865	if (slave_dev->type != ARPHRD_ETHER &&
1866	BOND_MODE(bond) == BOND_MODE_8023AD) {
1867	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1868	"8023AD mode requires Ethernet devices");
1869	return -EINVAL;
1870	}
1871	slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1872	bond_dev->type, slave_dev->type);
1873
1874	res = call_netdevice_notifiers(val: NETDEV_PRE_TYPE_CHANGE,
1875	dev: bond_dev);
1876	res = notifier_to_errno(ret: res);
1877	if (res) {
1878	slave_err(bond_dev, slave_dev, "refused to change device type\n");
1879	return -EBUSY;
1880	}
1881
1882	/* Flush unicast and multicast addresses */
1883	dev_uc_flush(dev: bond_dev);
1884	dev_mc_flush(dev: bond_dev);
1885
1886	if (slave_dev->type != ARPHRD_ETHER)
1887	bond_setup_by_slave(bond_dev, slave_dev);
1888	else
1889	bond_ether_setup(bond_dev);
1890
1891	call_netdevice_notifiers(val: NETDEV_POST_TYPE_CHANGE,
1892	dev: bond_dev);
1893	}
1894	} else if (bond_dev->type != slave_dev->type) {
1895	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1896	"Device type is different from other slaves");
1897	return -EINVAL;
1898	}
1899
1900	if (slave_dev->type == ARPHRD_INFINIBAND &&
1901	BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1902	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1903	"Only active-backup mode is supported for infiniband slaves");
1904	res = -EOPNOTSUPP;
1905	goto err_undo_flags;
1906	}
1907
1908	if (!slave_ops->ndo_set_mac_address ||
1909	slave_dev->type == ARPHRD_INFINIBAND) {
1910	slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1911	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1912	bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1913	if (!bond_has_slaves(bond)) {
1914	bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1915	slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1916	} else {
1917	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1918	"Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1919	res = -EOPNOTSUPP;
1920	goto err_undo_flags;
1921	}
1922	}
1923	}
1924
1925	call_netdevice_notifiers(val: NETDEV_JOIN, dev: slave_dev);
1926
1927	/* If this is the first slave, then we need to set the master's hardware
1928	* address to be the same as the slave's.
1929	*/
1930	if (!bond_has_slaves(bond) &&
1931	bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1932	res = bond_set_dev_addr(bond_dev: bond->dev, slave_dev);
1933	if (res)
1934	goto err_undo_flags;
1935	}
1936
1937	new_slave = bond_alloc_slave(bond, slave_dev);
1938	if (!new_slave) {
1939	res = -ENOMEM;
1940	goto err_undo_flags;
1941	}
1942
1943	/* Set the new_slave's queue_id to be zero. Queue ID mapping
1944	* is set via sysfs or module option if desired.
1945	*/
1946	new_slave->queue_id = 0;
1947
1948	/* Save slave's original mtu and then set it to match the bond */
1949	new_slave->original_mtu = slave_dev->mtu;
1950	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1951	if (res) {
1952	slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1953	goto err_free;
1954	}
1955
1956	/* Save slave's original ("permanent") mac address for modes
1957	* that need it, and for restoring it upon release, and then
1958	* set it to the master's address
1959	*/
1960	bond_hw_addr_copy(dst: new_slave->perm_hwaddr, src: slave_dev->dev_addr,
1961	len: slave_dev->addr_len);
1962
1963	if (!bond->params.fail_over_mac ||
1964	BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1965	/* Set slave to master's mac address. The application already
1966	* set the master's mac address to that of the first slave
1967	*/
1968	memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1969	} else if (bond->params.fail_over_mac == BOND_FOM_FOLLOW &&
1970	BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1971	bond_has_slaves(bond) &&
1972	memcmp(p: slave_dev->dev_addr, q: bond_dev->dev_addr, size: bond_dev->addr_len) == 0) {
1973	/* Set slave to random address to avoid duplicate mac
1974	* address in later fail over.
1975	*/
1976	eth_random_addr(addr: ss.__data);
1977	} else {
1978	goto skip_mac_set;
1979	}
1980
1981	ss.ss_family = slave_dev->type;
1982	res = dev_set_mac_address(dev: slave_dev, ss: &ss, extack);
1983	if (res) {
1984	slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1985	goto err_restore_mtu;
1986	}
1987
1988	skip_mac_set:
1989
1990	/* set no_addrconf flag before open to prevent IPv6 addrconf */
1991	slave_dev->priv_flags |= IFF_NO_ADDRCONF;
1992
1993	/* open the slave since the application closed it */
1994	res = dev_open(dev: slave_dev, extack);
1995	if (res) {
1996	slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1997	goto err_restore_mac;
1998	}
1999
2000	slave_dev->priv_flags |= IFF_BONDING;
2001	/* initialize slave stats */
2002	dev_get_stats(dev: new_slave->dev, storage: &new_slave->slave_stats);
2003
2004	if (bond_is_lb(bond)) {
2005	/* bond_alb_init_slave() must be called before all other stages since
2006	* it might fail and we do not want to have to undo everything
2007	*/
2008	res = bond_alb_init_slave(bond, slave: new_slave);
2009	if (res)
2010	goto err_close;
2011	}
2012
2013	res = vlan_vids_add_by_dev(dev: slave_dev, by_dev: bond_dev);
2014	if (res) {
2015	slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2016	goto err_close;
2017	}
2018
2019	prev_slave = bond_last_slave(bond);
2020
2021	new_slave->delay = 0;
2022	new_slave->link_failure_count = 0;
2023
2024	if (bond_update_speed_duplex(slave: new_slave) &&
2025	bond_needs_speed_duplex(bond))
2026	new_slave->link = BOND_LINK_DOWN;
2027
2028	new_slave->last_rx = jiffies -
2029	(msecs_to_jiffies(m: bond->params.arp_interval) + 1);
2030	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2031	new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2032
2033	new_slave->last_tx = new_slave->last_rx;
2034
2035	/* check for initial state */
2036	new_slave->link = BOND_LINK_NOCHANGE;
2037	if (bond->params.miimon) {
2038	if (netif_running(dev: slave_dev) && netif_carrier_ok(dev: slave_dev)) {
2039	if (bond->params.updelay) {
2040	bond_set_slave_link_state(slave: new_slave,
2041	BOND_LINK_BACK,
2042	BOND_SLAVE_NOTIFY_NOW);
2043	new_slave->delay = bond->params.updelay;
2044	} else {
2045	bond_set_slave_link_state(slave: new_slave,
2046	BOND_LINK_UP,
2047	BOND_SLAVE_NOTIFY_NOW);
2048	}
2049	} else {
2050	bond_set_slave_link_state(slave: new_slave, BOND_LINK_DOWN,
2051	BOND_SLAVE_NOTIFY_NOW);
2052	}
2053	} else if (bond->params.arp_interval) {
2054	bond_set_slave_link_state(slave: new_slave,
2055	state: (netif_carrier_ok(dev: slave_dev) ?
2056	BOND_LINK_UP : BOND_LINK_DOWN),
2057	BOND_SLAVE_NOTIFY_NOW);
2058	} else {
2059	bond_set_slave_link_state(slave: new_slave, BOND_LINK_UP,
2060	BOND_SLAVE_NOTIFY_NOW);
2061	}
2062
2063	if (new_slave->link != BOND_LINK_DOWN)
2064	new_slave->last_link_up = jiffies;
2065	slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2066	new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2067	(new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2068
2069	if (bond_uses_primary(bond) && bond->params.primary[0]) {
2070	/* if there is a primary slave, remember it */
2071	if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2072	rcu_assign_pointer(bond->primary_slave, new_slave);
2073	bond->force_primary = true;
2074	}
2075	}
2076
2077	switch (BOND_MODE(bond)) {
2078	case BOND_MODE_ACTIVEBACKUP:
2079	bond_set_slave_inactive_flags(slave: new_slave,
2080	BOND_SLAVE_NOTIFY_NOW);
2081	break;
2082	case BOND_MODE_8023AD:
2083	/* in 802.3ad mode, the internal mechanism
2084	* will activate the slaves in the selected
2085	* aggregator
2086	*/
2087	bond_set_slave_inactive_flags(slave: new_slave, BOND_SLAVE_NOTIFY_NOW);
2088	/* if this is the first slave */
2089	if (!prev_slave) {
2090	SLAVE_AD_INFO(new_slave)->id = 1;
2091	/* Initialize AD with the number of times that the AD timer is called in 1 second
2092	* can be called only after the mac address of the bond is set
2093	*/
2094	bond_3ad_initialize(bond);
2095	} else {
2096	SLAVE_AD_INFO(new_slave)->id =
2097	SLAVE_AD_INFO(prev_slave)->id + 1;
2098	}
2099
2100	bond_3ad_bind_slave(slave: new_slave);
2101	break;
2102	case BOND_MODE_TLB:
2103	case BOND_MODE_ALB:
2104	bond_set_active_slave(slave: new_slave);
2105	bond_set_slave_inactive_flags(slave: new_slave, BOND_SLAVE_NOTIFY_NOW);
2106	break;
2107	default:
2108	slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2109
2110	/* always active in trunk mode */
2111	bond_set_active_slave(slave: new_slave);
2112
2113	/* In trunking mode there is little meaning to curr_active_slave
2114	* anyway (it holds no special properties of the bond device),
2115	* so we can change it without calling change_active_interface()
2116	*/
2117	if (!rcu_access_pointer(bond->curr_active_slave) &&
2118	new_slave->link == BOND_LINK_UP)
2119	rcu_assign_pointer(bond->curr_active_slave, new_slave);
2120
2121	break;
2122	} /* switch(bond_mode) */
2123
2124	#ifdef CONFIG_NET_POLL_CONTROLLER
2125	if (bond->dev->npinfo) {
2126	if (slave_enable_netpoll(slave: new_slave)) {
2127	slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2128	res = -EBUSY;
2129	goto err_detach;
2130	}
2131	}
2132	#endif
2133
2134	if (!(bond_dev->features & NETIF_F_LRO))
2135	dev_disable_lro(dev: slave_dev);
2136
2137	res = netdev_rx_handler_register(dev: slave_dev, rx_handler: bond_handle_frame,
2138	rx_handler_data: new_slave);
2139	if (res) {
2140	slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2141	goto err_detach;
2142	}
2143
2144	res = bond_master_upper_dev_link(bond, slave: new_slave, extack);
2145	if (res) {
2146	slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2147	goto err_unregister;
2148	}
2149
2150	bond_lower_state_changed(slave: new_slave);
2151
2152	res = bond_sysfs_slave_add(slave: new_slave);
2153	if (res) {
2154	slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2155	goto err_upper_unlink;
2156	}
2157
2158	/* If the mode uses primary, then the following is handled by
2159	* bond_change_active_slave().
2160	*/
2161	if (!bond_uses_primary(bond)) {
2162	/* set promiscuity level to new slave */
2163	if (bond_dev->flags & IFF_PROMISC) {
2164	res = dev_set_promiscuity(dev: slave_dev, inc: 1);
2165	if (res)
2166	goto err_sysfs_del;
2167	}
2168
2169	/* set allmulti level to new slave */
2170	if (bond_dev->flags & IFF_ALLMULTI) {
2171	res = dev_set_allmulti(dev: slave_dev, inc: 1);
2172	if (res) {
2173	if (bond_dev->flags & IFF_PROMISC)
2174	dev_set_promiscuity(dev: slave_dev, inc: -1);
2175	goto err_sysfs_del;
2176	}
2177	}
2178
2179	if (bond_dev->flags & IFF_UP) {
2180	netif_addr_lock_bh(dev: bond_dev);
2181	dev_mc_sync_multiple(to: slave_dev, from: bond_dev);
2182	dev_uc_sync_multiple(to: slave_dev, from: bond_dev);
2183	netif_addr_unlock_bh(dev: bond_dev);
2184
2185	if (BOND_MODE(bond) == BOND_MODE_8023AD)
2186	dev_mc_add(dev: slave_dev, addr: lacpdu_mcast_addr);
2187	}
2188	}
2189
2190	bond->slave_cnt++;
2191	netdev_compute_master_upper_features(dev: bond->dev, update_header: true);
2192	bond_set_carrier(bond);
2193
2194	/* Needs to be called before bond_select_active_slave(), which will
2195	* remove the maddrs if the slave is selected as active slave.
2196	*/
2197	bond_slave_ns_maddrs_add(bond, slave: new_slave);
2198
2199	if (bond_uses_primary(bond)) {
2200	block_netpoll_tx();
2201	bond_select_active_slave(bond);
2202	unblock_netpoll_tx();
2203	}
2204
2205	if (!slave_dev->netdev_ops->ndo_bpf ||
2206	!slave_dev->netdev_ops->ndo_xdp_xmit) {
2207	if (bond->xdp_prog) {
2208	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2209	"Slave does not support XDP");
2210	res = -EOPNOTSUPP;
2211	goto err_sysfs_del;
2212	}
2213	} else if (bond->xdp_prog) {
2214	struct netdev_bpf xdp = {
2215	.command = XDP_SETUP_PROG,
2216	.flags = 0,
2217	.prog = bond->xdp_prog,
2218	.extack = extack,
2219	};
2220
2221	if (dev_xdp_prog_count(dev: slave_dev) > 0) {
2222	SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2223	"Slave has XDP program loaded, please unload before enslaving");
2224	res = -EOPNOTSUPP;
2225	goto err_sysfs_del;
2226	}
2227
2228	res = dev_xdp_propagate(dev: slave_dev, bpf: &xdp);
2229	if (res < 0) {
2230	/* ndo_bpf() sets extack error message */
2231	slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2232	goto err_sysfs_del;
2233	}
2234	if (bond->xdp_prog)
2235	bpf_prog_inc(prog: bond->xdp_prog);
2236	}
2237
2238	/* broadcast mode uses the all_slaves to loop through slaves. */
2239	if (bond_mode_can_use_xmit_hash(bond) ||
2240	BOND_MODE(bond) == BOND_MODE_BROADCAST)
2241	bond_update_slave_arr(bond, NULL);
2242
2243	bond_xdp_set_features(bond_dev);
2244
2245	slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2246	bond_is_active_slave(new_slave) ? "an active" : "a backup",
2247	new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2248
2249	/* enslave is successful */
2250	bond_queue_slave_event(slave: new_slave);
2251	return 0;
2252
2253	/* Undo stages on error */
2254	err_sysfs_del:
2255	bond_sysfs_slave_del(slave: new_slave);
2256
2257	err_upper_unlink:
2258	bond_upper_dev_unlink(bond, slave: new_slave);
2259
2260	err_unregister:
2261	netdev_rx_handler_unregister(dev: slave_dev);
2262
2263	err_detach:
2264	vlan_vids_del_by_dev(dev: slave_dev, by_dev: bond_dev);
2265	if (rcu_access_pointer(bond->primary_slave) == new_slave)
2266	RCU_INIT_POINTER(bond->primary_slave, NULL);
2267	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2268	block_netpoll_tx();
2269	bond_change_active_slave(bond, NULL);
2270	bond_select_active_slave(bond);
2271	unblock_netpoll_tx();
2272	}
2273	/* either primary_slave or curr_active_slave might've changed */
2274	synchronize_rcu();
2275	slave_disable_netpoll(slave: new_slave);
2276
2277	err_close:
2278	if (!netif_is_bond_master(dev: slave_dev))
2279	slave_dev->priv_flags &= ~IFF_BONDING;
2280	dev_close(dev: slave_dev);
2281
2282	err_restore_mac:
2283	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2284	if (!bond->params.fail_over_mac ||
2285	BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2286	/* XXX TODO - fom follow mode needs to change master's
2287	* MAC if this slave's MAC is in use by the bond, or at
2288	* least print a warning.
2289	*/
2290	bond_hw_addr_copy(dst: ss.__data, src: new_slave->perm_hwaddr,
2291	len: new_slave->dev->addr_len);
2292	ss.ss_family = slave_dev->type;
2293	dev_set_mac_address(dev: slave_dev, ss: &ss, NULL);
2294	}
2295
2296	err_restore_mtu:
2297	dev_set_mtu(slave_dev, new_slave->original_mtu);
2298
2299	err_free:
2300	kobject_put(kobj: &new_slave->kobj);
2301
2302	err_undo_flags:
2303	/* Enslave of first slave has failed and we need to fix master's mac */
2304	if (!bond_has_slaves(bond)) {
2305	if (ether_addr_equal_64bits(addr1: bond_dev->dev_addr,
2306	addr2: slave_dev->dev_addr))
2307	eth_hw_addr_random(dev: bond_dev);
2308	if (bond_dev->type != ARPHRD_ETHER) {
2309	dev_close(dev: bond_dev);
2310	bond_ether_setup(bond_dev);
2311	}
2312	}
2313
2314	return res;
2315	}
2316
2317	/* Try to release the slave device <slave> from the bond device <master>
2318	* It is legal to access curr_active_slave without a lock because all the function
2319	* is RTNL-locked. If "all" is true it means that the function is being called
2320	* while destroying a bond interface and all slaves are being released.
2321	*
2322	* The rules for slave state should be:
2323	* for Active/Backup:
2324	* Active stays on all backups go down
2325	* for Bonded connections:
2326	* The first up interface should be left on and all others downed.
2327	*/
2328	static int __bond_release_one(struct net_device *bond_dev,
2329	struct net_device *slave_dev,
2330	bool all, bool unregister)
2331	{
2332	struct bonding *bond = netdev_priv(dev: bond_dev);
2333	struct slave *slave, *oldcurrent;
2334	struct sockaddr_storage ss;
2335	int old_flags = bond_dev->flags;
2336	netdev_features_t old_features = bond_dev->features;
2337
2338	/* slave is not a slave or master is not master of this slave */
2339	if (!(slave_dev->flags & IFF_SLAVE) ||
2340	!netdev_has_upper_dev(dev: slave_dev, upper_dev: bond_dev)) {
2341	slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2342	return -EINVAL;
2343	}
2344
2345	block_netpoll_tx();
2346
2347	slave = bond_get_slave_by_dev(bond, slave_dev);
2348	if (!slave) {
2349	/* not a slave of this bond */
2350	slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2351	unblock_netpoll_tx();
2352	return -EINVAL;
2353	}
2354
2355	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2356
2357	bond_sysfs_slave_del(slave);
2358
2359	/* recompute stats just before removing the slave */
2360	bond_get_stats(bond_dev: bond->dev, stats: &bond->bond_stats);
2361
2362	if (bond->xdp_prog) {
2363	struct netdev_bpf xdp = {
2364	.command = XDP_SETUP_PROG,
2365	.flags = 0,
2366	.prog = NULL,
2367	.extack = NULL,
2368	};
2369	if (dev_xdp_propagate(dev: slave_dev, bpf: &xdp))
2370	slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2371	}
2372
2373	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2374	* for this slave anymore.
2375	*/
2376	netdev_rx_handler_unregister(dev: slave_dev);
2377
2378	if (BOND_MODE(bond) == BOND_MODE_8023AD)
2379	bond_3ad_unbind_slave(slave);
2380
2381	bond_upper_dev_unlink(bond, slave);
2382
2383	if (bond_mode_can_use_xmit_hash(bond) ||
2384	BOND_MODE(bond) == BOND_MODE_BROADCAST)
2385	bond_update_slave_arr(bond, skipslave: slave);
2386
2387	slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2388	bond_is_active_slave(slave) ? "active" : "backup");
2389
2390	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2391
2392	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2393
2394	if (!all && (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2395	BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2396	if (ether_addr_equal_64bits(addr1: bond_dev->dev_addr, addr2: slave->perm_hwaddr) &&
2397	bond_has_slaves(bond))
2398	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",
2399	slave->perm_hwaddr);
2400	}
2401
2402	if (rtnl_dereference(bond->primary_slave) == slave)
2403	RCU_INIT_POINTER(bond->primary_slave, NULL);
2404
2405	if (oldcurrent == slave)
2406	bond_change_active_slave(bond, NULL);
2407
2408	/* Must be called after bond_change_active_slave () as the slave
2409	* might change from an active slave to a backup slave. Then it is
2410	* necessary to clear the maddrs on the backup slave.
2411	*/
2412	bond_slave_ns_maddrs_del(bond, slave);
2413
2414	if (bond_is_lb(bond)) {
2415	/* Must be called only after the slave has been
2416	* detached from the list and the curr_active_slave
2417	* has been cleared (if our_slave == old_current),
2418	* but before a new active slave is selected.
2419	*/
2420	bond_alb_deinit_slave(bond, slave);
2421	}
2422
2423	if (all) {
2424	RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2425	} else if (oldcurrent == slave) {
2426	/* Note that we hold RTNL over this sequence, so there
2427	* is no concern that another slave add/remove event
2428	* will interfere.
2429	*/
2430	bond_select_active_slave(bond);
2431	}
2432
2433	bond_set_carrier(bond);
2434	if (!bond_has_slaves(bond))
2435	eth_hw_addr_random(dev: bond_dev);
2436
2437	unblock_netpoll_tx();
2438	synchronize_rcu();
2439	bond->slave_cnt--;
2440
2441	if (!bond_has_slaves(bond)) {
2442	call_netdevice_notifiers(val: NETDEV_CHANGEADDR, dev: bond->dev);
2443	call_netdevice_notifiers(val: NETDEV_RELEASE, dev: bond->dev);
2444	}
2445
2446	netdev_compute_master_upper_features(dev: bond->dev, update_header: true);
2447	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2448	(old_features & NETIF_F_VLAN_CHALLENGED))
2449	slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2450
2451	vlan_vids_del_by_dev(dev: slave_dev, by_dev: bond_dev);
2452
2453	/* If the mode uses primary, then this case was handled above by
2454	* bond_change_active_slave(..., NULL)
2455	*/
2456	if (!bond_uses_primary(bond)) {
2457	/* unset promiscuity level from slave
2458	* NOTE: The NETDEV_CHANGEADDR call above may change the value
2459	* of the IFF_PROMISC flag in the bond_dev, but we need the
2460	* value of that flag before that change, as that was the value
2461	* when this slave was attached, so we cache at the start of the
2462	* function and use it here. Same goes for ALLMULTI below
2463	*/
2464	if (old_flags & IFF_PROMISC)
2465	dev_set_promiscuity(dev: slave_dev, inc: -1);
2466
2467	/* unset allmulti level from slave */
2468	if (old_flags & IFF_ALLMULTI)
2469	dev_set_allmulti(dev: slave_dev, inc: -1);
2470
2471	if (old_flags & IFF_UP)
2472	bond_hw_addr_flush(bond_dev, slave_dev);
2473	}
2474
2475	slave_disable_netpoll(slave);
2476
2477	/* close slave before restoring its mac address */
2478	dev_close(dev: slave_dev);
2479
2480	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2481
2482	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2483	BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2484	/* restore original ("permanent") mac address */
2485	bond_hw_addr_copy(dst: ss.__data, src: slave->perm_hwaddr,
2486	len: slave->dev->addr_len);
2487	ss.ss_family = slave_dev->type;
2488	dev_set_mac_address(dev: slave_dev, ss: &ss, NULL);
2489	}
2490
2491	if (unregister) {
2492	netdev_lock_ops(dev: slave_dev);
2493	__netif_set_mtu(dev: slave_dev, new_mtu: slave->original_mtu);
2494	netdev_unlock_ops(dev: slave_dev);
2495	} else {
2496	dev_set_mtu(slave_dev, slave->original_mtu);
2497	}
2498
2499	if (!netif_is_bond_master(dev: slave_dev))
2500	slave_dev->priv_flags &= ~IFF_BONDING;
2501
2502	bond_xdp_set_features(bond_dev);
2503	kobject_put(kobj: &slave->kobj);
2504
2505	return 0;
2506	}
2507
2508	/* A wrapper used because of ndo_del_link */
2509	int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2510	{
2511	return __bond_release_one(bond_dev, slave_dev, all: false, unregister: false);
2512	}
2513
2514	/* First release a slave and then destroy the bond if no more slaves are left.
2515	* Must be under rtnl_lock when this function is called.
2516	*/
2517	static int bond_release_and_destroy(struct net_device *bond_dev,
2518	struct net_device *slave_dev)
2519	{
2520	struct bonding *bond = netdev_priv(dev: bond_dev);
2521	int ret;
2522
2523	ret = __bond_release_one(bond_dev, slave_dev, all: false, unregister: true);
2524	if (ret == 0 && !bond_has_slaves(bond) &&
2525	bond_dev->reg_state != NETREG_UNREGISTERING) {
2526	bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2527	netdev_info(dev: bond_dev, format: "Destroying bond\n");
2528	bond_remove_proc_entry(bond);
2529	unregister_netdevice(dev: bond_dev);
2530	}
2531	return ret;
2532	}
2533
2534	static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2535	{
2536	struct bonding *bond = netdev_priv(dev: bond_dev);
2537
2538	bond_fill_ifbond(bond, info);
2539	}
2540
2541	static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2542	{
2543	struct bonding *bond = netdev_priv(dev: bond_dev);
2544	struct list_head *iter;
2545	int i = 0, res = -ENODEV;
2546	struct slave *slave;
2547
2548	bond_for_each_slave(bond, slave, iter) {
2549	if (i++ == (int)info->slave_id) {
2550	res = 0;
2551	bond_fill_ifslave(slave, info);
2552	break;
2553	}
2554	}
2555
2556	return res;
2557	}
2558
2559	/*-------------------------------- Monitoring -------------------------------*/
2560
2561	/* called with rcu_read_lock() */
2562	static int bond_miimon_inspect(struct bonding *bond)
2563	{
2564	bool ignore_updelay = false;
2565	int link_state, commit = 0;
2566	struct list_head *iter;
2567	struct slave *slave;
2568
2569	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2570	ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2571	} else {
2572	struct bond_up_slave *usable_slaves;
2573
2574	usable_slaves = rcu_dereference(bond->usable_slaves);
2575
2576	if (usable_slaves && usable_slaves->count == 0)
2577	ignore_updelay = true;
2578	}
2579
2580	bond_for_each_slave_rcu(bond, slave, iter) {
2581	bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2582
2583	link_state = netif_running(dev: slave->dev) &&
2584	netif_carrier_ok(dev: slave->dev);
2585
2586	switch (slave->link) {
2587	case BOND_LINK_UP:
2588	if (link_state)
2589	continue;
2590
2591	bond_propose_link_state(slave, BOND_LINK_FAIL);
2592	commit++;
2593	slave->delay = bond->params.downdelay;
2594	if (slave->delay && net_ratelimit()) {
2595	slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2596	(BOND_MODE(bond) ==
2597	BOND_MODE_ACTIVEBACKUP) ?
2598	(bond_is_active_slave(slave) ?
2599	"active " : "backup ") : "",
2600	bond->params.downdelay * bond->params.miimon);
2601	}
2602	fallthrough;
2603	case BOND_LINK_FAIL:
2604	if (link_state) {
2605	/* recovered before downdelay expired */
2606	bond_propose_link_state(slave, BOND_LINK_UP);
2607	slave->last_link_up = jiffies;
2608	if (net_ratelimit())
2609	slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2610	(bond->params.downdelay - slave->delay) *
2611	bond->params.miimon);
2612	commit++;
2613	continue;
2614	}
2615
2616	if (slave->delay <= 0) {
2617	bond_propose_link_state(slave, BOND_LINK_DOWN);
2618	commit++;
2619	continue;
2620	}
2621
2622	slave->delay--;
2623	break;
2624
2625	case BOND_LINK_DOWN:
2626	if (!link_state)
2627	continue;
2628
2629	bond_propose_link_state(slave, BOND_LINK_BACK);
2630	commit++;
2631	slave->delay = bond->params.updelay;
2632
2633	if (slave->delay && net_ratelimit()) {
2634	slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2635	ignore_updelay ? 0 :
2636	bond->params.updelay *
2637	bond->params.miimon);
2638	}
2639	fallthrough;
2640	case BOND_LINK_BACK:
2641	if (!link_state) {
2642	bond_propose_link_state(slave, BOND_LINK_DOWN);
2643	if (net_ratelimit())
2644	slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2645	(bond->params.updelay - slave->delay) *
2646	bond->params.miimon);
2647	commit++;
2648	continue;
2649	}
2650
2651	if (ignore_updelay)
2652	slave->delay = 0;
2653
2654	if (slave->delay <= 0) {
2655	bond_propose_link_state(slave, BOND_LINK_UP);
2656	commit++;
2657	ignore_updelay = false;
2658	continue;
2659	}
2660
2661	slave->delay--;
2662	break;
2663	}
2664	}
2665
2666	return commit;
2667	}
2668
2669	static void bond_miimon_link_change(struct bonding *bond,
2670	struct slave *slave,
2671	char link)
2672	{
2673	switch (BOND_MODE(bond)) {
2674	case BOND_MODE_8023AD:
2675	bond_3ad_handle_link_change(slave, link);
2676	break;
2677	case BOND_MODE_TLB:
2678	case BOND_MODE_ALB:
2679	bond_alb_handle_link_change(bond, slave, link);
2680	break;
2681	case BOND_MODE_XOR:
2682	bond_update_slave_arr(bond, NULL);
2683	break;
2684	}
2685	}
2686
2687	static void bond_miimon_commit(struct bonding *bond)
2688	{
2689	struct slave *slave, *primary, *active;
2690	bool do_failover = false;
2691	struct list_head *iter;
2692
2693	ASSERT_RTNL();
2694
2695	bond_for_each_slave(bond, slave, iter) {
2696	switch (slave->link_new_state) {
2697	case BOND_LINK_NOCHANGE:
2698	/* For 802.3ad mode, check current slave speed and
2699	* duplex again in case its port was disabled after
2700	* invalid speed/duplex reporting but recovered before
2701	* link monitoring could make a decision on the actual
2702	* link status
2703	*/
2704	if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2705	slave->link == BOND_LINK_UP)
2706	bond_3ad_adapter_speed_duplex_changed(slave);
2707	continue;
2708
2709	case BOND_LINK_UP:
2710	if (bond_update_speed_duplex(slave) &&
2711	bond_needs_speed_duplex(bond)) {
2712	slave->link = BOND_LINK_DOWN;
2713	if (net_ratelimit())
2714	slave_warn(bond->dev, slave->dev,
2715	"failed to get link speed/duplex\n");
2716	continue;
2717	}
2718	bond_set_slave_link_state(slave, BOND_LINK_UP,
2719	BOND_SLAVE_NOTIFY_NOW);
2720	slave->last_link_up = jiffies;
2721
2722	primary = rtnl_dereference(bond->primary_slave);
2723	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2724	/* prevent it from being the active one */
2725	bond_set_backup_slave(slave);
2726	} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2727	/* make it immediately active */
2728	bond_set_active_slave(slave);
2729	}
2730
2731	slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2732	slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2733	slave->duplex ? "full" : "half");
2734
2735	bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2736
2737	active = rtnl_dereference(bond->curr_active_slave);
2738	if (!active || slave == primary || slave->prio > active->prio)
2739	do_failover = true;
2740
2741	continue;
2742
2743	case BOND_LINK_DOWN:
2744	if (slave->link_failure_count < UINT_MAX)
2745	slave->link_failure_count++;
2746
2747	bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2748	BOND_SLAVE_NOTIFY_NOW);
2749
2750	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2751	BOND_MODE(bond) == BOND_MODE_8023AD)
2752	bond_set_slave_inactive_flags(slave,
2753	BOND_SLAVE_NOTIFY_NOW);
2754
2755	slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2756
2757	bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2758
2759	if (slave == rcu_access_pointer(bond->curr_active_slave))
2760	do_failover = true;
2761
2762	continue;
2763
2764	default:
2765	slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2766	slave->link_new_state);
2767	bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2768
2769	continue;
2770	}
2771	}
2772
2773	if (do_failover) {
2774	block_netpoll_tx();
2775	bond_select_active_slave(bond);
2776	unblock_netpoll_tx();
2777	}
2778
2779	bond_set_carrier(bond);
2780	}
2781
2782	/* bond_mii_monitor
2783	*
2784	* Really a wrapper that splits the mii monitor into two phases: an
2785	* inspection, then (if inspection indicates something needs to be done)
2786	* an acquisition of appropriate locks followed by a commit phase to
2787	* implement whatever link state changes are indicated.
2788	*/
2789	static void bond_mii_monitor(struct work_struct *work)
2790	{
2791	struct bonding *bond = container_of(work, struct bonding,
2792	mii_work.work);
2793	bool should_notify_peers;
2794	bool commit;
2795	unsigned long delay;
2796	struct slave *slave;
2797	struct list_head *iter;
2798
2799	delay = msecs_to_jiffies(m: bond->params.miimon);
2800
2801	if (!bond_has_slaves(bond))
2802	goto re_arm;
2803
2804	rcu_read_lock();
2805
2806	should_notify_peers = bond_should_notify_peers(bond);
2807	commit = !!bond_miimon_inspect(bond);
2808
2809	rcu_read_unlock();
2810
2811	if (commit || bond->send_peer_notif) {
2812	/* Race avoidance with bond_close cancel of workqueue */
2813	if (!rtnl_trylock()) {
2814	delay = 1;
2815	goto re_arm;
2816	}
2817
2818	if (commit) {
2819	bond_for_each_slave(bond, slave, iter) {
2820	bond_commit_link_state(slave,
2821	BOND_SLAVE_NOTIFY_LATER);
2822	}
2823	bond_miimon_commit(bond);
2824	}
2825
2826	if (bond->send_peer_notif) {
2827	bond->send_peer_notif--;
2828	if (should_notify_peers)
2829	call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS,
2830	dev: bond->dev);
2831	}
2832
2833	rtnl_unlock(); /* might sleep, hold no other locks */
2834	}
2835
2836	re_arm:
2837	if (bond->params.miimon)
2838	queue_delayed_work(wq: bond->wq, dwork: &bond->mii_work, delay);
2839	}
2840
2841	static int bond_upper_dev_walk(struct net_device *upper,
2842	struct netdev_nested_priv *priv)
2843	{
2844	__be32 ip = *(__be32 *)priv->data;
2845
2846	return ip == bond_confirm_addr(dev: upper, dst: 0, local: ip);
2847	}
2848
2849	static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2850	{
2851	struct netdev_nested_priv priv = {
2852	.data = (void *)&ip,
2853	};
2854	bool ret = false;
2855
2856	if (ip == bond_confirm_addr(dev: bond->dev, dst: 0, local: ip))
2857	return true;
2858
2859	rcu_read_lock();
2860	if (netdev_walk_all_upper_dev_rcu(dev: bond->dev, fn: bond_upper_dev_walk, priv: &priv))
2861	ret = true;
2862	rcu_read_unlock();
2863
2864	return ret;
2865	}
2866
2867	#define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
2868
2869	static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2870	struct sk_buff *skb)
2871	{
2872	struct net_device *bond_dev = slave->bond->dev;
2873	struct net_device *slave_dev = slave->dev;
2874	struct bond_vlan_tag *outer_tag = tags;
2875
2876	if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
2877	return true;
2878
2879	tags++;
2880
2881	/* Go through all the tags backwards and add them to the packet */
2882	while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
2883	if (!tags->vlan_id) {
2884	tags++;
2885	continue;
2886	}
2887
2888	slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2889	ntohs(outer_tag->vlan_proto), tags->vlan_id);
2890	skb = vlan_insert_tag_set_proto(skb, vlan_proto: tags->vlan_proto,
2891	vlan_tci: tags->vlan_id);
2892	if (!skb) {
2893	net_err_ratelimited("failed to insert inner VLAN tag\n");
2894	return false;
2895	}
2896
2897	tags++;
2898	}
2899	/* Set the outer tag */
2900	if (outer_tag->vlan_id) {
2901	slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2902	ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2903	__vlan_hwaccel_put_tag(skb, vlan_proto: outer_tag->vlan_proto,
2904	vlan_tci: outer_tag->vlan_id);
2905	}
2906
2907	return true;
2908	}
2909
2910	/* We go to the (large) trouble of VLAN tagging ARP frames because
2911	* switches in VLAN mode (especially if ports are configured as
2912	* "native" to a VLAN) might not pass non-tagged frames.
2913	*/
2914	static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2915	__be32 src_ip, struct bond_vlan_tag *tags)
2916	{
2917	struct net_device *bond_dev = slave->bond->dev;
2918	struct net_device *slave_dev = slave->dev;
2919	struct sk_buff *skb;
2920
2921	slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2922	arp_op, &dest_ip, &src_ip);
2923
2924	skb = arp_create(type: arp_op, ETH_P_ARP, dest_ip, dev: slave_dev, src_ip,
2925	NULL, src_hw: slave_dev->dev_addr, NULL);
2926
2927	if (!skb) {
2928	net_err_ratelimited("ARP packet allocation failed\n");
2929	return;
2930	}
2931
2932	if (bond_handle_vlan(slave, tags, skb)) {
2933	slave_update_last_tx(slave);
2934	arp_xmit(skb);
2935	}
2936
2937	return;
2938	}
2939
2940	/* Validate the device path between the @start_dev and the @end_dev.
2941	* The path is valid if the @end_dev is reachable through device
2942	* stacking.
2943	* When the path is validated, collect any vlan information in the
2944	* path.
2945	*/
2946	struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2947	struct net_device *end_dev,
2948	int level)
2949	{
2950	struct bond_vlan_tag *tags;
2951	struct net_device *upper;
2952	struct list_head *iter;
2953
2954	if (start_dev == end_dev) {
2955	tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2956	if (!tags)
2957	return ERR_PTR(error: -ENOMEM);
2958	tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
2959	return tags;
2960	}
2961
2962	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2963	tags = bond_verify_device_path(start_dev: upper, end_dev, level: level + 1);
2964	if (IS_ERR_OR_NULL(ptr: tags)) {
2965	if (IS_ERR(ptr: tags))
2966	return tags;
2967	continue;
2968	}
2969	if (is_vlan_dev(dev: upper)) {
2970	tags[level].vlan_proto = vlan_dev_vlan_proto(dev: upper);
2971	tags[level].vlan_id = vlan_dev_vlan_id(dev: upper);
2972	}
2973
2974	return tags;
2975	}
2976
2977	return NULL;
2978	}
2979
2980	static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2981	{
2982	struct rtable *rt;
2983	struct bond_vlan_tag *tags;
2984	__be32 *targets = bond->params.arp_targets, addr;
2985	int i;
2986
2987	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2988	slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2989	__func__, &targets[i]);
2990	tags = NULL;
2991
2992	/* Find out through which dev should the packet go */
2993	rt = ip_route_output(net: dev_net(dev: bond->dev), daddr: targets[i], saddr: 0, dscp: 0, oif: 0,
2994	scope: RT_SCOPE_LINK);
2995	if (IS_ERR(ptr: rt)) {
2996	/* there's no route to target - try to send arp
2997	* probe to generate any traffic (arp_validate=0)
2998	*/
2999	if (bond->params.arp_validate)
3000	pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3001	bond->dev->name,
3002	&targets[i]);
3003	bond_arp_send(slave, ARPOP_REQUEST, dest_ip: targets[i],
3004	src_ip: 0, tags);
3005	continue;
3006	}
3007
3008	/* bond device itself */
3009	if (rt->dst.dev == bond->dev)
3010	goto found;
3011
3012	rcu_read_lock();
3013	tags = bond_verify_device_path(start_dev: bond->dev, end_dev: rt->dst.dev, level: 0);
3014	rcu_read_unlock();
3015
3016	if (!IS_ERR_OR_NULL(ptr: tags))
3017	goto found;
3018
3019	/* Not our device - skip */
3020	slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3021	&targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3022
3023	ip_rt_put(rt);
3024	continue;
3025
3026	found:
3027	addr = bond_confirm_addr(dev: rt->dst.dev, dst: targets[i], local: 0);
3028	ip_rt_put(rt);
3029	bond_arp_send(slave, ARPOP_REQUEST, dest_ip: targets[i], src_ip: addr, tags);
3030	kfree(objp: tags);
3031	}
3032	}
3033
3034	static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3035	{
3036	int i;
3037
3038	if (!sip || !bond_has_this_ip(bond, ip: tip)) {
3039	slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3040	__func__, &sip, &tip);
3041	return;
3042	}
3043
3044	i = bond_get_targets_ip(targets: bond->params.arp_targets, ip: sip);
3045	if (i == -1) {
3046	slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3047	__func__, &sip);
3048	return;
3049	}
3050	WRITE_ONCE(slave->last_rx, jiffies);
3051	WRITE_ONCE(slave->target_last_arp_rx[i], jiffies);
3052	}
3053
3054	static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3055	struct slave *slave)
3056	{
3057	struct arphdr *arp = (struct arphdr *)skb->data;
3058	struct slave *curr_active_slave, *curr_arp_slave;
3059	unsigned char *arp_ptr;
3060	__be32 sip, tip;
3061	unsigned int alen;
3062
3063	alen = arp_hdr_len(dev: bond->dev);
3064
3065	if (alen > skb_headlen(skb)) {
3066	arp = kmalloc(alen, GFP_ATOMIC);
3067	if (!arp)
3068	goto out_unlock;
3069	if (skb_copy_bits(skb, offset: 0, to: arp, len: alen) < 0)
3070	goto out_unlock;
3071	}
3072
3073	if (arp->ar_hln != bond->dev->addr_len ||
3074	skb->pkt_type == PACKET_OTHERHOST ||
3075	skb->pkt_type == PACKET_LOOPBACK ||
3076	arp->ar_hrd != htons(ARPHRD_ETHER) ||
3077	arp->ar_pro != htons(ETH_P_IP) ||
3078	arp->ar_pln != 4)
3079	goto out_unlock;
3080
3081	arp_ptr = (unsigned char *)(arp + 1);
3082	arp_ptr += bond->dev->addr_len;
3083	memcpy(&sip, arp_ptr, 4);
3084	arp_ptr += 4 + bond->dev->addr_len;
3085	memcpy(&tip, arp_ptr, 4);
3086
3087	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3088	__func__, slave->dev->name, bond_slave_state(slave),
3089	bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3090	&sip, &tip);
3091
3092	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3093	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3094
3095	/* We 'trust' the received ARP enough to validate it if:
3096	*
3097	* (a) the slave receiving the ARP is active (which includes the
3098	* current ARP slave, if any), or
3099	*
3100	* (b) the receiving slave isn't active, but there is a currently
3101	* active slave and it received valid arp reply(s) after it became
3102	* the currently active slave, or
3103	*
3104	* (c) there is an ARP slave that sent an ARP during the prior ARP
3105	* interval, and we receive an ARP reply on any slave. We accept
3106	* these because switch FDB update delays may deliver the ARP
3107	* reply to a slave other than the sender of the ARP request.
3108	*
3109	* Note: for (b), backup slaves are receiving the broadcast ARP
3110	* request, not a reply. This request passes from the sending
3111	* slave through the L2 switch(es) to the receiving slave. Since
3112	* this is checking the request, sip/tip are swapped for
3113	* validation.
3114	*
3115	* This is done to avoid endless looping when we can't reach the
3116	* arp_ip_target and fool ourselves with our own arp requests.
3117	*/
3118	if (bond_is_active_slave(slave))
3119	bond_validate_arp(bond, slave, sip, tip);
3120	else if (curr_active_slave &&
3121	time_after(slave_last_rx(bond, curr_active_slave),
3122	curr_active_slave->last_link_up))
3123	bond_validate_arp(bond, slave, sip: tip, tip: sip);
3124	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3125	bond_time_in_interval(bond, last_act: slave_last_tx(slave: curr_arp_slave), mod: 1))
3126	bond_validate_arp(bond, slave, sip, tip);
3127
3128	out_unlock:
3129	if (arp != (struct arphdr *)skb->data)
3130	kfree(objp: arp);
3131	return RX_HANDLER_ANOTHER;
3132	}
3133
3134	#if IS_ENABLED(CONFIG_IPV6)
3135	static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3136	const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3137	{
3138	struct net_device *bond_dev = slave->bond->dev;
3139	struct net_device *slave_dev = slave->dev;
3140	struct in6_addr mcaddr;
3141	struct sk_buff *skb;
3142
3143	slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3144	daddr, saddr);
3145
3146	skb = ndisc_ns_create(dev: slave_dev, solicit: daddr, saddr, nonce: 0);
3147	if (!skb) {
3148	net_err_ratelimited("NS packet allocation failed\n");
3149	return;
3150	}
3151
3152	addrconf_addr_solict_mult(addr: daddr, solicited: &mcaddr);
3153	if (bond_handle_vlan(slave, tags, skb)) {
3154	slave_update_last_tx(slave);
3155	ndisc_send_skb(skb, daddr: &mcaddr, saddr);
3156	}
3157	}
3158
3159	static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3160	{
3161	struct in6_addr *targets = bond->params.ns_targets;
3162	struct bond_vlan_tag *tags;
3163	struct dst_entry *dst;
3164	struct in6_addr saddr;
3165	struct flowi6 fl6;
3166	int i;
3167
3168	for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(a: &targets[i]); i++) {
3169	slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3170	__func__, &targets[i]);
3171	tags = NULL;
3172
3173	/* Find out through which dev should the packet go */
3174	memset(&fl6, 0, sizeof(struct flowi6));
3175	fl6.daddr = targets[i];
3176
3177	dst = ip6_route_output(net: dev_net(dev: bond->dev), NULL, fl6: &fl6);
3178	if (dst->error) {
3179	dst_release(dst);
3180	/* there's no route to target - try to send arp
3181	* probe to generate any traffic (arp_validate=0)
3182	*/
3183	if (bond->params.arp_validate)
3184	pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3185	bond->dev->name,
3186	&targets[i]);
3187	bond_ns_send(slave, daddr: &targets[i], saddr: &in6addr_any, tags);
3188	continue;
3189	}
3190
3191	/* bond device itself */
3192	if (dst->dev == bond->dev)
3193	goto found;
3194
3195	rcu_read_lock();
3196	tags = bond_verify_device_path(start_dev: bond->dev, end_dev: dst->dev, level: 0);
3197	rcu_read_unlock();
3198
3199	if (!IS_ERR_OR_NULL(ptr: tags))
3200	goto found;
3201
3202	/* Not our device - skip */
3203	slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3204	&targets[i], dst->dev ? dst->dev->name : "NULL");
3205
3206	dst_release(dst);
3207	continue;
3208
3209	found:
3210	if (!ipv6_dev_get_saddr(net: dev_net(dev: dst->dev), dev: dst->dev, daddr: &targets[i], srcprefs: 0, saddr: &saddr))
3211	bond_ns_send(slave, daddr: &targets[i], saddr: &saddr, tags);
3212	else
3213	bond_ns_send(slave, daddr: &targets[i], saddr: &in6addr_any, tags);
3214
3215	dst_release(dst);
3216	kfree(objp: tags);
3217	}
3218	}
3219
3220	static int bond_confirm_addr6(struct net_device *dev,
3221	struct netdev_nested_priv *priv)
3222	{
3223	struct in6_addr *addr = (struct in6_addr *)priv->data;
3224
3225	return ipv6_chk_addr(net: dev_net(dev), addr, dev, strict: 0);
3226	}
3227
3228	static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3229	{
3230	struct netdev_nested_priv priv = {
3231	.data = addr,
3232	};
3233	int ret = false;
3234
3235	if (bond_confirm_addr6(dev: bond->dev, priv: &priv))
3236	return true;
3237
3238	rcu_read_lock();
3239	if (netdev_walk_all_upper_dev_rcu(dev: bond->dev, fn: bond_confirm_addr6, priv: &priv))
3240	ret = true;
3241	rcu_read_unlock();
3242
3243	return ret;
3244	}
3245
3246	static void bond_validate_na(struct bonding *bond, struct slave *slave,
3247	struct in6_addr *saddr, struct in6_addr *daddr)
3248	{
3249	int i;
3250
3251	/* Ignore NAs that:
3252	* 1. Source address is unspecified address.
3253	* 2. Dest address is neither all-nodes multicast address nor
3254	* exist on bond interface.
3255	*/
3256	if (ipv6_addr_any(a: saddr) ||
3257	(!ipv6_addr_equal(a1: daddr, a2: &in6addr_linklocal_allnodes) &&
3258	!bond_has_this_ip6(bond, addr: daddr))) {
3259	slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3260	__func__, saddr, daddr);
3261	return;
3262	}
3263
3264	i = bond_get_targets_ip6(targets: bond->params.ns_targets, ip: saddr);
3265	if (i == -1) {
3266	slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3267	__func__, saddr);
3268	return;
3269	}
3270	WRITE_ONCE(slave->last_rx, jiffies);
3271	WRITE_ONCE(slave->target_last_arp_rx[i], jiffies);
3272	}
3273
3274	static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3275	struct slave *slave)
3276	{
3277	struct slave *curr_active_slave, *curr_arp_slave;
3278	struct in6_addr *saddr, *daddr;
3279	struct {
3280	struct ipv6hdr ip6;
3281	struct icmp6hdr icmp6;
3282	} *combined, _combined;
3283
3284	if (skb->pkt_type == PACKET_OTHERHOST ||
3285	skb->pkt_type == PACKET_LOOPBACK)
3286	goto out;
3287
3288	combined = skb_header_pointer(skb, offset: 0, len: sizeof(_combined), buffer: &_combined);
3289	if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3290	(combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3291	combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3292	goto out;
3293
3294	saddr = &combined->ip6.saddr;
3295	daddr = &combined->ip6.daddr;
3296
3297	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3298	__func__, slave->dev->name, bond_slave_state(slave),
3299	bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3300	saddr, daddr);
3301
3302	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3303	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3304
3305	/* We 'trust' the received ARP enough to validate it if:
3306	* see bond_arp_rcv().
3307	*/
3308	if (bond_is_active_slave(slave))
3309	bond_validate_na(bond, slave, saddr, daddr);
3310	else if (curr_active_slave &&
3311	time_after(slave_last_rx(bond, curr_active_slave),
3312	curr_active_slave->last_link_up))
3313	bond_validate_na(bond, slave, saddr: daddr, daddr: saddr);
3314	else if (curr_arp_slave &&
3315	bond_time_in_interval(bond, last_act: slave_last_tx(slave: curr_arp_slave), mod: 1))
3316	bond_validate_na(bond, slave, saddr, daddr);
3317
3318	out:
3319	return RX_HANDLER_ANOTHER;
3320	}
3321	#endif
3322
3323	int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3324	struct slave *slave)
3325	{
3326	#if IS_ENABLED(CONFIG_IPV6)
3327	bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3328	#endif
3329	bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3330
3331	slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3332	__func__, skb->dev->name);
3333
3334	/* Use arp validate logic for both ARP and NS */
3335	if (!slave_do_arp_validate(bond, slave)) {
3336	if ((slave_do_arp_validate_only(bond) && is_arp) ||
3337	#if IS_ENABLED(CONFIG_IPV6)
3338	(slave_do_arp_validate_only(bond) && is_ipv6) ||
3339	#endif
3340	!slave_do_arp_validate_only(bond))
3341	WRITE_ONCE(slave->last_rx, jiffies);
3342	return RX_HANDLER_ANOTHER;
3343	} else if (is_arp) {
3344	return bond_arp_rcv(skb, bond, slave);
3345	#if IS_ENABLED(CONFIG_IPV6)
3346	} else if (is_ipv6) {
3347	return bond_na_rcv(skb, bond, slave);
3348	#endif
3349	} else {
3350	return RX_HANDLER_ANOTHER;
3351	}
3352	}
3353
3354	static void bond_send_validate(struct bonding *bond, struct slave *slave)
3355	{
3356	bond_arp_send_all(bond, slave);
3357	#if IS_ENABLED(CONFIG_IPV6)
3358	bond_ns_send_all(bond, slave);
3359	#endif
3360	}
3361
3362	/* function to verify if we're in the arp_interval timeslice, returns true if
3363	* (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3364	* arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3365	*/
3366	static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3367	int mod)
3368	{
3369	int delta_in_ticks = msecs_to_jiffies(m: bond->params.arp_interval);
3370
3371	return time_in_range(jiffies,
3372	last_act - delta_in_ticks,
3373	last_act + mod * delta_in_ticks + delta_in_ticks/2);
3374	}
3375
3376	/* This function is called regularly to monitor each slave's link
3377	* ensuring that traffic is being sent and received when arp monitoring
3378	* is used in load-balancing mode. if the adapter has been dormant, then an
3379	* arp is transmitted to generate traffic. see activebackup_arp_monitor for
3380	* arp monitoring in active backup mode.
3381	*/
3382	static void bond_loadbalance_arp_mon(struct bonding *bond)
3383	{
3384	struct slave *slave, *oldcurrent;
3385	struct list_head *iter;
3386	int do_failover = 0, slave_state_changed = 0;
3387
3388	if (!bond_has_slaves(bond))
3389	goto re_arm;
3390
3391	rcu_read_lock();
3392
3393	oldcurrent = rcu_dereference(bond->curr_active_slave);
3394	/* see if any of the previous devices are up now (i.e. they have
3395	* xmt and rcv traffic). the curr_active_slave does not come into
3396	* the picture unless it is null. also, slave->last_link_up is not
3397	* needed here because we send an arp on each slave and give a slave
3398	* as long as it needs to get the tx/rx within the delta.
3399	* TODO: what about up/down delay in arp mode? it wasn't here before
3400	* so it can wait
3401	*/
3402	bond_for_each_slave_rcu(bond, slave, iter) {
3403	unsigned long last_tx = slave_last_tx(slave);
3404
3405	bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3406
3407	if (slave->link != BOND_LINK_UP) {
3408	if (bond_time_in_interval(bond, last_act: last_tx, mod: 1) &&
3409	bond_time_in_interval(bond, READ_ONCE(slave->last_rx), mod: 1)) {
3410
3411	bond_propose_link_state(slave, BOND_LINK_UP);
3412	slave_state_changed = 1;
3413
3414	/* primary_slave has no meaning in round-robin
3415	* mode. the window of a slave being up and
3416	* curr_active_slave being null after enslaving
3417	* is closed.
3418	*/
3419	if (!oldcurrent) {
3420	slave_info(bond->dev, slave->dev, "link status definitely up\n");
3421	do_failover = 1;
3422	} else {
3423	slave_info(bond->dev, slave->dev, "interface is now up\n");
3424	}
3425	}
3426	} else {
3427	/* slave->link == BOND_LINK_UP */
3428
3429	/* not all switches will respond to an arp request
3430	* when the source ip is 0, so don't take the link down
3431	* if we don't know our ip yet
3432	*/
3433	if (!bond_time_in_interval(bond, last_act: last_tx,
3434	mod: bond->params.missed_max) ||
3435	!bond_time_in_interval(bond, READ_ONCE(slave->last_rx),
3436	mod: bond->params.missed_max)) {
3437
3438	bond_propose_link_state(slave, BOND_LINK_DOWN);
3439	slave_state_changed = 1;
3440
3441	if (slave->link_failure_count < UINT_MAX)
3442	slave->link_failure_count++;
3443
3444	slave_info(bond->dev, slave->dev, "interface is now down\n");
3445
3446	if (slave == oldcurrent)
3447	do_failover = 1;
3448	}
3449	}
3450
3451	/* note: if switch is in round-robin mode, all links
3452	* must tx arp to ensure all links rx an arp - otherwise
3453	* links may oscillate or not come up at all; if switch is
3454	* in something like xor mode, there is nothing we can
3455	* do - all replies will be rx'ed on same link causing slaves
3456	* to be unstable during low/no traffic periods
3457	*/
3458	if (bond_slave_is_up(slave))
3459	bond_send_validate(bond, slave);
3460	}
3461
3462	rcu_read_unlock();
3463
3464	if (do_failover || slave_state_changed) {
3465	if (!rtnl_trylock())
3466	goto re_arm;
3467
3468	bond_for_each_slave(bond, slave, iter) {
3469	if (slave->link_new_state != BOND_LINK_NOCHANGE)
3470	slave->link = slave->link_new_state;
3471	}
3472
3473	if (slave_state_changed) {
3474	bond_slave_state_change(bond);
3475	if (BOND_MODE(bond) == BOND_MODE_XOR)
3476	bond_update_slave_arr(bond, NULL);
3477	}
3478	if (do_failover) {
3479	block_netpoll_tx();
3480	bond_select_active_slave(bond);
3481	unblock_netpoll_tx();
3482	}
3483	rtnl_unlock();
3484	}
3485
3486	re_arm:
3487	if (bond->params.arp_interval)
3488	queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work,
3489	delay: msecs_to_jiffies(m: bond->params.arp_interval));
3490	}
3491
3492	/* Called to inspect slaves for active-backup mode ARP monitor link state
3493	* changes. Sets proposed link state in slaves to specify what action
3494	* should take place for the slave. Returns 0 if no changes are found, >0
3495	* if changes to link states must be committed.
3496	*
3497	* Called with rcu_read_lock held.
3498	*/
3499	static int bond_ab_arp_inspect(struct bonding *bond)
3500	{
3501	unsigned long last_tx, last_rx;
3502	struct list_head *iter;
3503	struct slave *slave;
3504	int commit = 0;
3505
3506	bond_for_each_slave_rcu(bond, slave, iter) {
3507	bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3508	last_rx = slave_last_rx(bond, slave);
3509
3510	if (slave->link != BOND_LINK_UP) {
3511	if (bond_time_in_interval(bond, last_act: last_rx, mod: 1)) {
3512	bond_propose_link_state(slave, BOND_LINK_UP);
3513	commit++;
3514	} else if (slave->link == BOND_LINK_BACK) {
3515	bond_propose_link_state(slave, BOND_LINK_FAIL);
3516	commit++;
3517	}
3518	continue;
3519	}
3520
3521	/* Give slaves 2*delta after being enslaved or made
3522	* active. This avoids bouncing, as the last receive
3523	* times need a full ARP monitor cycle to be updated.
3524	*/
3525	if (bond_time_in_interval(bond, last_act: slave->last_link_up, mod: 2))
3526	continue;
3527
3528	/* Backup slave is down if:
3529	* - No current_arp_slave AND
3530	* - more than (missed_max+1)*delta since last receive AND
3531	* - the bond has an IP address
3532	*
3533	* Note: a non-null current_arp_slave indicates
3534	* the curr_active_slave went down and we are
3535	* searching for a new one; under this condition
3536	* we only take the curr_active_slave down - this
3537	* gives each slave a chance to tx/rx traffic
3538	* before being taken out
3539	*/
3540	if (!bond_is_active_slave(slave) &&
3541	!rcu_access_pointer(bond->current_arp_slave) &&
3542	!bond_time_in_interval(bond, last_act: last_rx, mod: bond->params.missed_max + 1)) {
3543	bond_propose_link_state(slave, BOND_LINK_DOWN);
3544	commit++;
3545	}
3546
3547	/* Active slave is down if:
3548	* - more than missed_max*delta since transmitting OR
3549	* - (more than missed_max*delta since receive AND
3550	* the bond has an IP address)
3551	*/
3552	last_tx = slave_last_tx(slave);
3553	if (bond_is_active_slave(slave) &&
3554	(!bond_time_in_interval(bond, last_act: last_tx, mod: bond->params.missed_max) ||
3555	!bond_time_in_interval(bond, last_act: last_rx, mod: bond->params.missed_max))) {
3556	bond_propose_link_state(slave, BOND_LINK_DOWN);
3557	commit++;
3558	}
3559	}
3560
3561	return commit;
3562	}
3563
3564	/* Called to commit link state changes noted by inspection step of
3565	* active-backup mode ARP monitor.
3566	*
3567	* Called with RTNL hold.
3568	*/
3569	static void bond_ab_arp_commit(struct bonding *bond)
3570	{
3571	bool do_failover = false;
3572	struct list_head *iter;
3573	unsigned long last_tx;
3574	struct slave *slave;
3575
3576	bond_for_each_slave(bond, slave, iter) {
3577	switch (slave->link_new_state) {
3578	case BOND_LINK_NOCHANGE:
3579	continue;
3580
3581	case BOND_LINK_UP:
3582	last_tx = slave_last_tx(slave);
3583	if (rtnl_dereference(bond->curr_active_slave) != slave ||
3584	(!rtnl_dereference(bond->curr_active_slave) &&
3585	bond_time_in_interval(bond, last_act: last_tx, mod: 1))) {
3586	struct slave *current_arp_slave;
3587
3588	current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3589	bond_set_slave_link_state(slave, BOND_LINK_UP,
3590	BOND_SLAVE_NOTIFY_NOW);
3591	if (current_arp_slave) {
3592	bond_set_slave_inactive_flags(
3593	slave: current_arp_slave,
3594	BOND_SLAVE_NOTIFY_NOW);
3595	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3596	}
3597
3598	slave_info(bond->dev, slave->dev, "link status definitely up\n");
3599
3600	if (!rtnl_dereference(bond->curr_active_slave) ||
3601	slave == rtnl_dereference(bond->primary_slave) ||
3602	slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3603	do_failover = true;
3604
3605	}
3606
3607	continue;
3608
3609	case BOND_LINK_DOWN:
3610	if (slave->link_failure_count < UINT_MAX)
3611	slave->link_failure_count++;
3612
3613	bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3614	BOND_SLAVE_NOTIFY_NOW);
3615	bond_set_slave_inactive_flags(slave,
3616	BOND_SLAVE_NOTIFY_NOW);
3617
3618	slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3619
3620	if (slave == rtnl_dereference(bond->curr_active_slave)) {
3621	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3622	do_failover = true;
3623	}
3624
3625	continue;
3626
3627	case BOND_LINK_FAIL:
3628	bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3629	BOND_SLAVE_NOTIFY_NOW);
3630	bond_set_slave_inactive_flags(slave,
3631	BOND_SLAVE_NOTIFY_NOW);
3632
3633	/* A slave has just been enslaved and has become
3634	* the current active slave.
3635	*/
3636	if (rtnl_dereference(bond->curr_active_slave))
3637	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3638	continue;
3639
3640	default:
3641	slave_err(bond->dev, slave->dev,
3642	"impossible: link_new_state %d on slave\n",
3643	slave->link_new_state);
3644	continue;
3645	}
3646	}
3647
3648	if (do_failover) {
3649	block_netpoll_tx();
3650	bond_select_active_slave(bond);
3651	unblock_netpoll_tx();
3652	}
3653
3654	bond_set_carrier(bond);
3655	}
3656
3657	/* Send ARP probes for active-backup mode ARP monitor.
3658	*
3659	* Called with rcu_read_lock held.
3660	*/
3661	static bool bond_ab_arp_probe(struct bonding *bond)
3662	{
3663	struct slave *slave, *before = NULL, *new_slave = NULL,
3664	*curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3665	*curr_active_slave = rcu_dereference(bond->curr_active_slave);
3666	struct list_head *iter;
3667	bool found = false;
3668	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3669
3670	if (curr_arp_slave && curr_active_slave)
3671	netdev_info(dev: bond->dev, format: "PROBE: c_arp %s && cas %s BAD\n",
3672	curr_arp_slave->dev->name,
3673	curr_active_slave->dev->name);
3674
3675	if (curr_active_slave) {
3676	bond_send_validate(bond, slave: curr_active_slave);
3677	return should_notify_rtnl;
3678	}
3679
3680	/* if we don't have a curr_active_slave, search for the next available
3681	* backup slave from the current_arp_slave and make it the candidate
3682	* for becoming the curr_active_slave
3683	*/
3684
3685	if (!curr_arp_slave) {
3686	curr_arp_slave = bond_first_slave_rcu(bond);
3687	if (!curr_arp_slave)
3688	return should_notify_rtnl;
3689	}
3690
3691	bond_for_each_slave_rcu(bond, slave, iter) {
3692	if (!found && !before && bond_slave_is_up(slave))
3693	before = slave;
3694
3695	if (found && !new_slave && bond_slave_is_up(slave))
3696	new_slave = slave;
3697	/* if the link state is up at this point, we
3698	* mark it down - this can happen if we have
3699	* simultaneous link failures and
3700	* reselect_active_interface doesn't make this
3701	* one the current slave so it is still marked
3702	* up when it is actually down
3703	*/
3704	if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3705	bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3706	BOND_SLAVE_NOTIFY_LATER);
3707	if (slave->link_failure_count < UINT_MAX)
3708	slave->link_failure_count++;
3709
3710	bond_set_slave_inactive_flags(slave,
3711	BOND_SLAVE_NOTIFY_LATER);
3712
3713	slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3714	}
3715	if (slave == curr_arp_slave)
3716	found = true;
3717	}
3718
3719	if (!new_slave && before)
3720	new_slave = before;
3721
3722	if (!new_slave)
3723	goto check_state;
3724
3725	bond_set_slave_link_state(slave: new_slave, BOND_LINK_BACK,
3726	BOND_SLAVE_NOTIFY_LATER);
3727	bond_set_slave_active_flags(slave: new_slave, BOND_SLAVE_NOTIFY_LATER);
3728	bond_send_validate(bond, slave: new_slave);
3729	new_slave->last_link_up = jiffies;
3730	rcu_assign_pointer(bond->current_arp_slave, new_slave);
3731
3732	check_state:
3733	bond_for_each_slave_rcu(bond, slave, iter) {
3734	if (slave->should_notify || slave->should_notify_link) {
3735	should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3736	break;
3737	}
3738	}
3739	return should_notify_rtnl;
3740	}
3741
3742	static void bond_activebackup_arp_mon(struct bonding *bond)
3743	{
3744	bool should_notify_peers = false;
3745	bool should_notify_rtnl = false;
3746	int delta_in_ticks;
3747
3748	delta_in_ticks = msecs_to_jiffies(m: bond->params.arp_interval);
3749
3750	if (!bond_has_slaves(bond))
3751	goto re_arm;
3752
3753	rcu_read_lock();
3754
3755	should_notify_peers = bond_should_notify_peers(bond);
3756
3757	if (bond_ab_arp_inspect(bond)) {
3758	rcu_read_unlock();
3759
3760	/* Race avoidance with bond_close flush of workqueue */
3761	if (!rtnl_trylock()) {
3762	delta_in_ticks = 1;
3763	should_notify_peers = false;
3764	goto re_arm;
3765	}
3766
3767	bond_ab_arp_commit(bond);
3768
3769	rtnl_unlock();
3770	rcu_read_lock();
3771	}
3772
3773	should_notify_rtnl = bond_ab_arp_probe(bond);
3774	rcu_read_unlock();
3775
3776	re_arm:
3777	if (bond->params.arp_interval)
3778	queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work, delay: delta_in_ticks);
3779
3780	if (should_notify_peers || should_notify_rtnl) {
3781	if (!rtnl_trylock())
3782	return;
3783
3784	if (should_notify_peers) {
3785	bond->send_peer_notif--;
3786	call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS,
3787	dev: bond->dev);
3788	}
3789	if (should_notify_rtnl) {
3790	bond_slave_state_notify(bond);
3791	bond_slave_link_notify(bond);
3792	}
3793
3794	rtnl_unlock();
3795	}
3796	}
3797
3798	static void bond_arp_monitor(struct work_struct *work)
3799	{
3800	struct bonding *bond = container_of(work, struct bonding,
3801	arp_work.work);
3802
3803	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3804	bond_activebackup_arp_mon(bond);
3805	else
3806	bond_loadbalance_arp_mon(bond);
3807	}
3808
3809	/*-------------------------- netdev event handling --------------------------*/
3810
3811	/* Change device name */
3812	static int bond_event_changename(struct bonding *bond)
3813	{
3814	bond_remove_proc_entry(bond);
3815	bond_create_proc_entry(bond);
3816
3817	bond_debug_reregister(bond);
3818
3819	return NOTIFY_DONE;
3820	}
3821
3822	static int bond_master_netdev_event(unsigned long event,
3823	struct net_device *bond_dev)
3824	{
3825	struct bonding *event_bond = netdev_priv(dev: bond_dev);
3826
3827	netdev_dbg(bond_dev, "%s called\n", __func__);
3828
3829	switch (event) {
3830	case NETDEV_CHANGENAME:
3831	return bond_event_changename(bond: event_bond);
3832	case NETDEV_UNREGISTER:
3833	bond_remove_proc_entry(bond: event_bond);
3834	#ifdef CONFIG_XFRM_OFFLOAD
3835	xfrm_dev_state_flush(net: dev_net(dev: bond_dev), dev: bond_dev, task_valid: true);
3836	#endif /* CONFIG_XFRM_OFFLOAD */
3837	break;
3838	case NETDEV_REGISTER:
3839	bond_create_proc_entry(bond: event_bond);
3840	break;
3841	default:
3842	break;
3843	}
3844
3845	return NOTIFY_DONE;
3846	}
3847
3848	static int bond_slave_netdev_event(unsigned long event,
3849	struct net_device *slave_dev)
3850	{
3851	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3852	struct bonding *bond;
3853	struct net_device *bond_dev;
3854
3855	/* A netdev event can be generated while enslaving a device
3856	* before netdev_rx_handler_register is called in which case
3857	* slave will be NULL
3858	*/
3859	if (!slave) {
3860	netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3861	return NOTIFY_DONE;
3862	}
3863
3864	bond_dev = slave->bond->dev;
3865	bond = slave->bond;
3866	primary = rtnl_dereference(bond->primary_slave);
3867
3868	slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3869
3870	switch (event) {
3871	case NETDEV_UNREGISTER:
3872	if (bond_dev->type != ARPHRD_ETHER)
3873	bond_release_and_destroy(bond_dev, slave_dev);
3874	else
3875	__bond_release_one(bond_dev, slave_dev, all: false, unregister: true);
3876	break;
3877	case NETDEV_UP:
3878	case NETDEV_CHANGE:
3879	/* For 802.3ad mode only:
3880	* Getting invalid Speed/Duplex values here will put slave
3881	* in weird state. Mark it as link-fail if the link was
3882	* previously up or link-down if it hasn't yet come up, and
3883	* let link-monitoring (miimon) set it right when correct
3884	* speeds/duplex are available.
3885	*/
3886	if (bond_update_speed_duplex(slave) &&
3887	BOND_MODE(bond) == BOND_MODE_8023AD) {
3888	if (slave->last_link_up)
3889	slave->link = BOND_LINK_FAIL;
3890	else
3891	slave->link = BOND_LINK_DOWN;
3892	}
3893
3894	if (BOND_MODE(bond) == BOND_MODE_8023AD)
3895	bond_3ad_adapter_speed_duplex_changed(slave);
3896	fallthrough;
3897	case NETDEV_DOWN:
3898	/* Refresh slave-array if applicable!
3899	* If the setup does not use miimon or arpmon (mode-specific!),
3900	* then these events will not cause the slave-array to be
3901	* refreshed. This will cause xmit to use a slave that is not
3902	* usable. Avoid such situation by refeshing the array at these
3903	* events. If these (miimon/arpmon) parameters are configured
3904	* then array gets refreshed twice and that should be fine!
3905	*/
3906	if (bond_mode_can_use_xmit_hash(bond))
3907	bond_update_slave_arr(bond, NULL);
3908	break;
3909	case NETDEV_CHANGEMTU:
3910	/* TODO: Should slaves be allowed to
3911	* independently alter their MTU? For
3912	* an active-backup bond, slaves need
3913	* not be the same type of device, so
3914	* MTUs may vary. For other modes,
3915	* slaves arguably should have the
3916	* same MTUs. To do this, we'd need to
3917	* take over the slave's change_mtu
3918	* function for the duration of their
3919	* servitude.
3920	*/
3921	break;
3922	case NETDEV_CHANGENAME:
3923	/* we don't care if we don't have primary set */
3924	if (!bond_uses_primary(bond) ||
3925	!bond->params.primary[0])
3926	break;
3927
3928	if (slave == primary) {
3929	/* slave's name changed - he's no longer primary */
3930	RCU_INIT_POINTER(bond->primary_slave, NULL);
3931	} else if (!strcmp(slave_dev->name, bond->params.primary)) {
3932	/* we have a new primary slave */
3933	rcu_assign_pointer(bond->primary_slave, slave);
3934	} else { /* we didn't change primary - exit */
3935	break;
3936	}
3937
3938	netdev_info(dev: bond->dev, format: "Primary slave changed to %s, reselecting active slave\n",
3939	primary ? slave_dev->name : "none");
3940
3941	block_netpoll_tx();
3942	bond_select_active_slave(bond);
3943	unblock_netpoll_tx();
3944	break;
3945	case NETDEV_FEAT_CHANGE:
3946	if (!bond->notifier_ctx) {
3947	bond->notifier_ctx = true;
3948	netdev_compute_master_upper_features(dev: bond->dev, update_header: true);
3949	bond->notifier_ctx = false;
3950	}
3951	break;
3952	case NETDEV_RESEND_IGMP:
3953	/* Propagate to master device */
3954	call_netdevice_notifiers(val: event, dev: slave->bond->dev);
3955	break;
3956	case NETDEV_XDP_FEAT_CHANGE:
3957	bond_xdp_set_features(bond_dev);
3958	break;
3959	default:
3960	break;
3961	}
3962
3963	return NOTIFY_DONE;
3964	}
3965
3966	/* bond_netdev_event: handle netdev notifier chain events.
3967	*
3968	* This function receives events for the netdev chain. The caller (an
3969	* ioctl handler calling blocking_notifier_call_chain) holds the necessary
3970	* locks for us to safely manipulate the slave devices (RTNL lock,
3971	* dev_probe_lock).
3972	*/
3973	static int bond_netdev_event(struct notifier_block *this,
3974	unsigned long event, void *ptr)
3975	{
3976	struct net_device *event_dev = netdev_notifier_info_to_dev(info: ptr);
3977
3978	netdev_dbg(event_dev, "%s received %s\n",
3979	__func__, netdev_cmd_to_name(event));
3980
3981	if (!(event_dev->priv_flags & IFF_BONDING))
3982	return NOTIFY_DONE;
3983
3984	if (event_dev->flags & IFF_MASTER) {
3985	int ret;
3986
3987	ret = bond_master_netdev_event(event, bond_dev: event_dev);
3988	if (ret != NOTIFY_DONE)
3989	return ret;
3990	}
3991
3992	if (event_dev->flags & IFF_SLAVE)
3993	return bond_slave_netdev_event(event, slave_dev: event_dev);
3994
3995	return NOTIFY_DONE;
3996	}
3997
3998	static struct notifier_block bond_netdev_notifier = {
3999	.notifier_call = bond_netdev_event,
4000	};
4001
4002	/*---------------------------- Hashing Policies -----------------------------*/
4003
4004	/* Helper to access data in a packet, with or without a backing skb.
4005	* If skb is given the data is linearized if necessary via pskb_may_pull.
4006	*/
4007	static inline const void *bond_pull_data(struct sk_buff *skb,
4008	const void *data, int hlen, int n)
4009	{
4010	if (likely(n <= hlen))
4011	return data;
4012	else if (skb && likely(pskb_may_pull(skb, n)))
4013	return skb->data;
4014
4015	return NULL;
4016	}
4017
4018	/* L2 hash helper */
4019	static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4020	{
4021	struct ethhdr *ep;
4022
4023	data = bond_pull_data(skb, data, hlen, n: mhoff + sizeof(struct ethhdr));
4024	if (!data)
4025	return 0;
4026
4027	ep = (struct ethhdr *)(data + mhoff);
4028	return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4029	}
4030
4031	static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4032	int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4033	{
4034	const struct ipv6hdr *iph6;
4035	const struct iphdr *iph;
4036
4037	if (l2_proto == htons(ETH_P_IP)) {
4038	data = bond_pull_data(skb, data, hlen, n: *nhoff + sizeof(*iph));
4039	if (!data)
4040	return false;
4041
4042	iph = (const struct iphdr *)(data + *nhoff);
4043	iph_to_flow_copy_v4addrs(flow: fk, iph);
4044	*nhoff += iph->ihl << 2;
4045	if (!ip_is_fragment(iph))
4046	*ip_proto = iph->protocol;
4047	} else if (l2_proto == htons(ETH_P_IPV6)) {
4048	data = bond_pull_data(skb, data, hlen, n: *nhoff + sizeof(*iph6));
4049	if (!data)
4050	return false;
4051
4052	iph6 = (const struct ipv6hdr *)(data + *nhoff);
4053	iph_to_flow_copy_v6addrs(flow: fk, iph: iph6);
4054	*nhoff += sizeof(*iph6);
4055	*ip_proto = iph6->nexthdr;
4056	} else {
4057	return false;
4058	}
4059
4060	if (l34 && *ip_proto >= 0)
4061	fk->ports.ports = skb_flow_get_ports(skb, thoff: *nhoff, ip_proto: *ip_proto, data, hlen_proto: hlen);
4062
4063	return true;
4064	}
4065
4066	static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4067	{
4068	u32 srcmac_vendor = 0, srcmac_dev = 0;
4069	struct ethhdr *mac_hdr;
4070	u16 vlan = 0;
4071	int i;
4072
4073	data = bond_pull_data(skb, data, hlen, n: mhoff + sizeof(struct ethhdr));
4074	if (!data)
4075	return 0;
4076	mac_hdr = (struct ethhdr *)(data + mhoff);
4077
4078	for (i = 0; i < 3; i++)
4079	srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4080
4081	for (i = 3; i < ETH_ALEN; i++)
4082	srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4083
4084	if (skb && skb_vlan_tag_present(skb))
4085	vlan = skb_vlan_tag_get(skb);
4086
4087	return vlan ^ srcmac_vendor ^ srcmac_dev;
4088	}
4089
4090	/* Extract the appropriate headers based on bond's xmit policy */
4091	static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4092	__be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4093	{
4094	bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4095	int ip_proto = -1;
4096
4097	switch (bond->params.xmit_policy) {
4098	case BOND_XMIT_POLICY_ENCAP23:
4099	case BOND_XMIT_POLICY_ENCAP34:
4100	memset(fk, 0, sizeof(*fk));
4101	return __skb_flow_dissect(net: dev_net(dev: bond->dev), skb,
4102	flow_dissector: &flow_keys_bonding, target_container: fk, data,
4103	proto: l2_proto, nhoff, hlen, flags: 0);
4104	default:
4105	break;
4106	}
4107
4108	fk->ports.ports = 0;
4109	memset(&fk->icmp, 0, sizeof(fk->icmp));
4110	if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, nhoff: &nhoff, ip_proto: &ip_proto, l34))
4111	return false;
4112
4113	/* ICMP error packets contains at least 8 bytes of the header
4114	* of the packet which generated the error. Use this information
4115	* to correlate ICMP error packets within the same flow which
4116	* generated the error.
4117	*/
4118	if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4119	skb_flow_get_icmp_tci(skb, key_icmp: &fk->icmp, data, thoff: nhoff, hlen);
4120	if (ip_proto == IPPROTO_ICMP) {
4121	if (!icmp_is_err(type: fk->icmp.type))
4122	return true;
4123
4124	nhoff += sizeof(struct icmphdr);
4125	} else if (ip_proto == IPPROTO_ICMPV6) {
4126	if (!icmpv6_is_err(type: fk->icmp.type))
4127	return true;
4128
4129	nhoff += sizeof(struct icmp6hdr);
4130	}
4131	return bond_flow_ip(skb, fk, data, hlen, l2_proto, nhoff: &nhoff, ip_proto: &ip_proto, l34);
4132	}
4133
4134	return true;
4135	}
4136
4137	static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4138	{
4139	hash ^= (__force u32)flow_get_u32_dst(flow) ^
4140	(__force u32)flow_get_u32_src(flow);
4141	hash ^= (hash >> 16);
4142	hash ^= (hash >> 8);
4143
4144	/* discard lowest hash bit to deal with the common even ports pattern */
4145	if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4146	xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4147	return hash >> 1;
4148
4149	return hash;
4150	}
4151
4152	/* Generate hash based on xmit policy. If @skb is given it is used to linearize
4153	* the data as required, but this function can be used without it if the data is
4154	* known to be linear (e.g. with xdp_buff).
4155	*/
4156	static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4157	__be16 l2_proto, int mhoff, int nhoff, int hlen)
4158	{
4159	struct flow_keys flow;
4160	u32 hash;
4161
4162	if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4163	return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4164
4165	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4166	!bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, fk: &flow))
4167	return bond_eth_hash(skb, data, mhoff, hlen);
4168
4169	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4170	bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4171	hash = bond_eth_hash(skb, data, mhoff, hlen);
4172	} else {
4173	if (flow.icmp.id)
4174	memcpy(&hash, &flow.icmp, sizeof(hash));
4175	else
4176	memcpy(&hash, &flow.ports.ports, sizeof(hash));
4177	}
4178
4179	return bond_ip_hash(hash, flow: &flow, xmit_policy: bond->params.xmit_policy);
4180	}
4181
4182	/**
4183	* bond_xmit_hash - generate a hash value based on the xmit policy
4184	* @bond: bonding device
4185	* @skb: buffer to use for headers
4186	*
4187	* This function will extract the necessary headers from the skb buffer and use
4188	* them to generate a hash based on the xmit_policy set in the bonding device
4189	*/
4190	u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4191	{
4192	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4193	skb->l4_hash)
4194	return skb->hash;
4195
4196	return __bond_xmit_hash(bond, skb, data: skb->data, l2_proto: skb->protocol,
4197	mhoff: 0, nhoff: skb_network_offset(skb),
4198	hlen: skb_headlen(skb));
4199	}
4200
4201	/**
4202	* bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4203	* @bond: bonding device
4204	* @xdp: buffer to use for headers
4205	*
4206	* The XDP variant of bond_xmit_hash.
4207	*/
4208	static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4209	{
4210	struct ethhdr *eth;
4211
4212	if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4213	return 0;
4214
4215	eth = (struct ethhdr *)xdp->data;
4216
4217	return __bond_xmit_hash(bond, NULL, data: xdp->data, l2_proto: eth->h_proto, mhoff: 0,
4218	nhoff: sizeof(struct ethhdr), hlen: xdp->data_end - xdp->data);
4219	}
4220
4221	/*-------------------------- Device entry points ----------------------------*/
4222
4223	void bond_work_init_all(struct bonding *bond)
4224	{
4225	INIT_DELAYED_WORK(&bond->mcast_work,
4226	bond_resend_igmp_join_requests_delayed);
4227	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4228	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4229	INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4230	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4231	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4232	}
4233
4234	void bond_work_cancel_all(struct bonding *bond)
4235	{
4236	cancel_delayed_work_sync(dwork: &bond->mii_work);
4237	cancel_delayed_work_sync(dwork: &bond->arp_work);
4238	cancel_delayed_work_sync(dwork: &bond->alb_work);
4239	cancel_delayed_work_sync(dwork: &bond->ad_work);
4240	cancel_delayed_work_sync(dwork: &bond->mcast_work);
4241	cancel_delayed_work_sync(dwork: &bond->slave_arr_work);
4242	}
4243
4244	static int bond_open(struct net_device *bond_dev)
4245	{
4246	struct bonding *bond = netdev_priv(dev: bond_dev);
4247	struct list_head *iter;
4248	struct slave *slave;
4249
4250	if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4251	bond->rr_tx_counter = alloc_percpu(u32);
4252	if (!bond->rr_tx_counter)
4253	return -ENOMEM;
4254	}
4255
4256	/* reset slave->backup and slave->inactive */
4257	if (bond_has_slaves(bond)) {
4258	bond_for_each_slave(bond, slave, iter) {
4259	if (bond_uses_primary(bond) &&
4260	slave != rcu_access_pointer(bond->curr_active_slave)) {
4261	bond_set_slave_inactive_flags(slave,
4262	BOND_SLAVE_NOTIFY_NOW);
4263	} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4264	bond_set_slave_active_flags(slave,
4265	BOND_SLAVE_NOTIFY_NOW);
4266	}
4267	}
4268	}
4269
4270	if (bond_is_lb(bond)) {
4271	/* bond_alb_initialize must be called before the timer
4272	* is started.
4273	*/
4274	if (bond_alb_initialize(bond, rlb_enabled: (BOND_MODE(bond) == BOND_MODE_ALB)))
4275	return -ENOMEM;
4276	if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4277	queue_delayed_work(wq: bond->wq, dwork: &bond->alb_work, delay: 0);
4278	}
4279
4280	if (bond->params.miimon) /* link check interval, in milliseconds. */
4281	queue_delayed_work(wq: bond->wq, dwork: &bond->mii_work, delay: 0);
4282
4283	if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4284	queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work, delay: 0);
4285	bond->recv_probe = bond_rcv_validate;
4286	}
4287
4288	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4289	queue_delayed_work(wq: bond->wq, dwork: &bond->ad_work, delay: 0);
4290	/* register to receive LACPDUs */
4291	bond->recv_probe = bond_3ad_lacpdu_recv;
4292	bond_3ad_initiate_agg_selection(bond, timeout: 1);
4293
4294	bond_for_each_slave(bond, slave, iter)
4295	dev_mc_add(dev: slave->dev, addr: lacpdu_mcast_addr);
4296
4297	if (bond->params.broadcast_neighbor)
4298	static_branch_inc(&bond_bcast_neigh_enabled);
4299	}
4300
4301	if (bond_mode_can_use_xmit_hash(bond))
4302	bond_update_slave_arr(bond, NULL);
4303
4304	return 0;
4305	}
4306
4307	static int bond_close(struct net_device *bond_dev)
4308	{
4309	struct bonding *bond = netdev_priv(dev: bond_dev);
4310	struct slave *slave;
4311
4312	bond_work_cancel_all(bond);
4313	bond->send_peer_notif = 0;
4314	if (bond_is_lb(bond))
4315	bond_alb_deinitialize(bond);
4316	bond->recv_probe = NULL;
4317
4318	if (BOND_MODE(bond) == BOND_MODE_8023AD &&
4319	bond->params.broadcast_neighbor)
4320	static_branch_dec(&bond_bcast_neigh_enabled);
4321
4322	if (bond_uses_primary(bond)) {
4323	rcu_read_lock();
4324	slave = rcu_dereference(bond->curr_active_slave);
4325	if (slave)
4326	bond_hw_addr_flush(bond_dev, slave_dev: slave->dev);
4327	rcu_read_unlock();
4328	} else {
4329	struct list_head *iter;
4330
4331	bond_for_each_slave(bond, slave, iter)
4332	bond_hw_addr_flush(bond_dev, slave_dev: slave->dev);
4333	}
4334
4335	return 0;
4336	}
4337
4338	/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4339	* that some drivers can provide 32bit values only.
4340	*/
4341	static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4342	const struct rtnl_link_stats64 *_new,
4343	const struct rtnl_link_stats64 *_old)
4344	{
4345	const u64 *new = (const u64 *)_new;
4346	const u64 *old = (const u64 *)_old;
4347	u64 *res = (u64 *)_res;
4348	int i;
4349
4350	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4351	u64 nv = new[i];
4352	u64 ov = old[i];
4353	s64 delta = nv - ov;
4354
4355	/* detects if this particular field is 32bit only */
4356	if (((nv | ov) >> 32) == 0)
4357	delta = (s64)(s32)((u32)nv - (u32)ov);
4358
4359	/* filter anomalies, some drivers reset their stats
4360	* at down/up events.
4361	*/
4362	if (delta > 0)
4363	res[i] += delta;
4364	}
4365	}
4366
4367	#ifdef CONFIG_LOCKDEP
4368	static int bond_get_lowest_level_rcu(struct net_device *dev)
4369	{
4370	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4371	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4372	int cur = 0, max = 0;
4373
4374	now = dev;
4375	iter = &dev->adj_list.lower;
4376
4377	while (1) {
4378	next = NULL;
4379	while (1) {
4380	ldev = netdev_next_lower_dev_rcu(dev: now, iter: &iter);
4381	if (!ldev)
4382	break;
4383
4384	next = ldev;
4385	niter = &ldev->adj_list.lower;
4386	dev_stack[cur] = now;
4387	iter_stack[cur++] = iter;
4388	if (max <= cur)
4389	max = cur;
4390	break;
4391	}
4392
4393	if (!next) {
4394	if (!cur)
4395	return max;
4396	next = dev_stack[--cur];
4397	niter = iter_stack[cur];
4398	}
4399
4400	now = next;
4401	iter = niter;
4402	}
4403
4404	return max;
4405	}
4406	#endif
4407
4408	static void bond_get_stats(struct net_device *bond_dev,
4409	struct rtnl_link_stats64 *stats)
4410	{
4411	struct bonding *bond = netdev_priv(dev: bond_dev);
4412	struct rtnl_link_stats64 temp;
4413	struct list_head *iter;
4414	struct slave *slave;
4415	int nest_level = 0;
4416
4417
4418	rcu_read_lock();
4419	#ifdef CONFIG_LOCKDEP
4420	nest_level = bond_get_lowest_level_rcu(dev: bond_dev);
4421	#endif
4422
4423	spin_lock_nested(&bond->stats_lock, nest_level);
4424	memcpy(stats, &bond->bond_stats, sizeof(*stats));
4425
4426	bond_for_each_slave_rcu(bond, slave, iter) {
4427	const struct rtnl_link_stats64 *new =
4428	dev_get_stats(dev: slave->dev, storage: &temp);
4429
4430	bond_fold_stats(res: stats, new: new, old: &slave->slave_stats);
4431
4432	/* save off the slave stats for the next run */
4433	memcpy(&slave->slave_stats, new, sizeof(*new));
4434	}
4435
4436	memcpy(&bond->bond_stats, stats, sizeof(*stats));
4437	spin_unlock(lock: &bond->stats_lock);
4438	rcu_read_unlock();
4439	}
4440
4441	static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4442	{
4443	struct bonding *bond = netdev_priv(dev: bond_dev);
4444	struct mii_ioctl_data *mii = NULL;
4445
4446	netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4447
4448	switch (cmd) {
4449	case SIOCGMIIPHY:
4450	mii = if_mii(rq: ifr);
4451	if (!mii)
4452	return -EINVAL;
4453
4454	mii->phy_id = 0;
4455	fallthrough;
4456	case SIOCGMIIREG:
4457	/* We do this again just in case we were called by SIOCGMIIREG
4458	* instead of SIOCGMIIPHY.
4459	*/
4460	mii = if_mii(rq: ifr);
4461	if (!mii)
4462	return -EINVAL;
4463
4464	if (mii->reg_num == 1) {
4465	mii->val_out = 0;
4466	if (netif_carrier_ok(dev: bond->dev))
4467	mii->val_out = BMSR_LSTATUS;
4468	}
4469
4470	break;
4471	default:
4472	return -EOPNOTSUPP;
4473	}
4474
4475	return 0;
4476	}
4477
4478	static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4479	{
4480	struct bonding *bond = netdev_priv(dev: bond_dev);
4481	struct net_device *slave_dev = NULL;
4482	struct ifbond k_binfo;
4483	struct ifbond __user *u_binfo = NULL;
4484	struct ifslave k_sinfo;
4485	struct ifslave __user *u_sinfo = NULL;
4486	struct bond_opt_value newval;
4487	struct net *net;
4488	int res = 0;
4489
4490	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4491
4492	switch (cmd) {
4493	case SIOCBONDINFOQUERY:
4494	u_binfo = (struct ifbond __user *)ifr->ifr_data;
4495
4496	if (copy_from_user(to: &k_binfo, from: u_binfo, n: sizeof(ifbond)))
4497	return -EFAULT;
4498
4499	bond_info_query(bond_dev, info: &k_binfo);
4500	if (copy_to_user(to: u_binfo, from: &k_binfo, n: sizeof(ifbond)))
4501	return -EFAULT;
4502
4503	return 0;
4504	case SIOCBONDSLAVEINFOQUERY:
4505	u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4506
4507	if (copy_from_user(to: &k_sinfo, from: u_sinfo, n: sizeof(ifslave)))
4508	return -EFAULT;
4509
4510	res = bond_slave_info_query(bond_dev, info: &k_sinfo);
4511	if (res == 0 &&
4512	copy_to_user(to: u_sinfo, from: &k_sinfo, n: sizeof(ifslave)))
4513	return -EFAULT;
4514
4515	return res;
4516	default:
4517	break;
4518	}
4519
4520	net = dev_net(dev: bond_dev);
4521
4522	if (!ns_capable(ns: net->user_ns, CAP_NET_ADMIN))
4523	return -EPERM;
4524
4525	slave_dev = __dev_get_by_name(net, name: ifr->ifr_slave);
4526
4527	slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4528
4529	if (!slave_dev)
4530	return -ENODEV;
4531
4532	switch (cmd) {
4533	case SIOCBONDENSLAVE:
4534	res = bond_enslave(bond_dev, slave_dev, NULL);
4535	break;
4536	case SIOCBONDRELEASE:
4537	res = bond_release(bond_dev, slave_dev);
4538	break;
4539	case SIOCBONDSETHWADDR:
4540	res = bond_set_dev_addr(bond_dev, slave_dev);
4541	break;
4542	case SIOCBONDCHANGEACTIVE:
4543	bond_opt_initstr(&newval, slave_dev->name);
4544	res = __bond_opt_set_notify(bond, option: BOND_OPT_ACTIVE_SLAVE,
4545	val: &newval);
4546	break;
4547	default:
4548	res = -EOPNOTSUPP;
4549	}
4550
4551	return res;
4552	}
4553
4554	static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4555	void __user *data, int cmd)
4556	{
4557	struct ifreq ifrdata = { .ifr_data = data };
4558
4559	switch (cmd) {
4560	case BOND_INFO_QUERY_OLD:
4561	return bond_do_ioctl(bond_dev, ifr: &ifrdata, SIOCBONDINFOQUERY);
4562	case BOND_SLAVE_INFO_QUERY_OLD:
4563	return bond_do_ioctl(bond_dev, ifr: &ifrdata, SIOCBONDSLAVEINFOQUERY);
4564	case BOND_ENSLAVE_OLD:
4565	return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4566	case BOND_RELEASE_OLD:
4567	return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4568	case BOND_SETHWADDR_OLD:
4569	return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4570	case BOND_CHANGE_ACTIVE_OLD:
4571	return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4572	}
4573
4574	return -EOPNOTSUPP;
4575	}
4576
4577	static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4578	{
4579	struct bonding *bond = netdev_priv(dev: bond_dev);
4580
4581	if (change & IFF_PROMISC)
4582	bond_set_promiscuity(bond,
4583	inc: bond_dev->flags & IFF_PROMISC ? 1 : -1);
4584
4585	if (change & IFF_ALLMULTI)
4586	bond_set_allmulti(bond,
4587	inc: bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4588	}
4589
4590	static void bond_set_rx_mode(struct net_device *bond_dev)
4591	{
4592	struct bonding *bond = netdev_priv(dev: bond_dev);
4593	struct list_head *iter;
4594	struct slave *slave;
4595
4596	rcu_read_lock();
4597	if (bond_uses_primary(bond)) {
4598	slave = rcu_dereference(bond->curr_active_slave);
4599	if (slave) {
4600	dev_uc_sync(to: slave->dev, from: bond_dev);
4601	dev_mc_sync(to: slave->dev, from: bond_dev);
4602	}
4603	} else {
4604	bond_for_each_slave_rcu(bond, slave, iter) {
4605	dev_uc_sync_multiple(to: slave->dev, from: bond_dev);
4606	dev_mc_sync_multiple(to: slave->dev, from: bond_dev);
4607	}
4608	}
4609	rcu_read_unlock();
4610	}
4611
4612	static int bond_neigh_init(struct neighbour *n)
4613	{
4614	struct bonding *bond = netdev_priv(dev: n->dev);
4615	const struct net_device_ops *slave_ops;
4616	struct neigh_parms parms;
4617	struct slave *slave;
4618	int ret = 0;
4619
4620	rcu_read_lock();
4621	slave = bond_first_slave_rcu(bond);
4622	if (!slave)
4623	goto out;
4624	slave_ops = slave->dev->netdev_ops;
4625	if (!slave_ops->ndo_neigh_setup)
4626	goto out;
4627
4628	/* TODO: find another way [1] to implement this.
4629	* Passing a zeroed structure is fragile,
4630	* but at least we do not pass garbage.
4631	*
4632	* [1] One way would be that ndo_neigh_setup() never touch
4633	* struct neigh_parms, but propagate the new neigh_setup()
4634	* back to ___neigh_create() / neigh_parms_alloc()
4635	*/
4636	memset(&parms, 0, sizeof(parms));
4637	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4638
4639	if (ret)
4640	goto out;
4641
4642	if (parms.neigh_setup)
4643	ret = parms.neigh_setup(n);
4644	out:
4645	rcu_read_unlock();
4646	return ret;
4647	}
4648
4649	/* The bonding ndo_neigh_setup is called at init time beofre any
4650	* slave exists. So we must declare proxy setup function which will
4651	* be used at run time to resolve the actual slave neigh param setup.
4652	*
4653	* It's also called by master devices (such as vlans) to setup their
4654	* underlying devices. In that case - do nothing, we're already set up from
4655	* our init.
4656	*/
4657	static int bond_neigh_setup(struct net_device *dev,
4658	struct neigh_parms *parms)
4659	{
4660	/* modify only our neigh_parms */
4661	if (parms->dev == dev)
4662	parms->neigh_setup = bond_neigh_init;
4663
4664	return 0;
4665	}
4666
4667	/* Change the MTU of all of a master's slaves to match the master */
4668	static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4669	{
4670	struct bonding *bond = netdev_priv(dev: bond_dev);
4671	struct slave *slave, *rollback_slave;
4672	struct list_head *iter;
4673	int res = 0;
4674
4675	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4676
4677	bond_for_each_slave(bond, slave, iter) {
4678	slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4679	slave, slave->dev->netdev_ops->ndo_change_mtu);
4680
4681	res = dev_set_mtu(slave->dev, new_mtu);
4682
4683	if (res) {
4684	/* If we failed to set the slave's mtu to the new value
4685	* we must abort the operation even in ACTIVE_BACKUP
4686	* mode, because if we allow the backup slaves to have
4687	* different mtu values than the active slave we'll
4688	* need to change their mtu when doing a failover. That
4689	* means changing their mtu from timer context, which
4690	* is probably not a good idea.
4691	*/
4692	slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4693	res, new_mtu);
4694	goto unwind;
4695	}
4696	}
4697
4698	WRITE_ONCE(bond_dev->mtu, new_mtu);
4699
4700	return 0;
4701
4702	unwind:
4703	/* unwind from head to the slave that failed */
4704	bond_for_each_slave(bond, rollback_slave, iter) {
4705	int tmp_res;
4706
4707	if (rollback_slave == slave)
4708	break;
4709
4710	tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4711	if (tmp_res)
4712	slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4713	tmp_res);
4714	}
4715
4716	return res;
4717	}
4718
4719	/* Change HW address
4720	*
4721	* Note that many devices must be down to change the HW address, and
4722	* downing the master releases all slaves. We can make bonds full of
4723	* bonding devices to test this, however.
4724	*/
4725	static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4726	{
4727	struct bonding *bond = netdev_priv(dev: bond_dev);
4728	struct slave *slave, *rollback_slave;
4729	struct sockaddr_storage *ss = addr, tmp_ss;
4730	struct list_head *iter;
4731	int res = 0;
4732
4733	if (BOND_MODE(bond) == BOND_MODE_ALB)
4734	return bond_alb_set_mac_address(bond_dev, addr);
4735
4736
4737	netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4738
4739	/* If fail_over_mac is enabled, do nothing and return success.
4740	* Returning an error causes ifenslave to fail.
4741	*/
4742	if (bond->params.fail_over_mac &&
4743	BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4744	return 0;
4745
4746	if (!is_valid_ether_addr(addr: ss->__data))
4747	return -EADDRNOTAVAIL;
4748
4749	bond_for_each_slave(bond, slave, iter) {
4750	slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4751	__func__, slave);
4752	res = dev_set_mac_address(dev: slave->dev, ss: addr, NULL);
4753	if (res) {
4754	/* TODO: consider downing the slave
4755	* and retry ?
4756	* User should expect communications
4757	* breakage anyway until ARP finish
4758	* updating, so...
4759	*/
4760	slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4761	__func__, res);
4762	goto unwind;
4763	}
4764	}
4765
4766	/* success */
4767	dev_addr_set(dev: bond_dev, addr: ss->__data);
4768	return 0;
4769
4770	unwind:
4771	memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4772	tmp_ss.ss_family = bond_dev->type;
4773
4774	/* unwind from head to the slave that failed */
4775	bond_for_each_slave(bond, rollback_slave, iter) {
4776	int tmp_res;
4777
4778	if (rollback_slave == slave)
4779	break;
4780
4781	tmp_res = dev_set_mac_address(dev: rollback_slave->dev, ss: &tmp_ss, NULL);
4782	if (tmp_res) {
4783	slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4784	__func__, tmp_res);
4785	}
4786	}
4787
4788	return res;
4789	}
4790
4791	/**
4792	* bond_get_slave_by_id - get xmit slave with slave_id
4793	* @bond: bonding device that is transmitting
4794	* @slave_id: slave id up to slave_cnt-1 through which to transmit
4795	*
4796	* This function tries to get slave with slave_id but in case
4797	* it fails, it tries to find the first available slave for transmission.
4798	*/
4799	static struct slave *bond_get_slave_by_id(struct bonding *bond,
4800	int slave_id)
4801	{
4802	struct list_head *iter;
4803	struct slave *slave;
4804	int i = slave_id;
4805
4806	/* Here we start from the slave with slave_id */
4807	bond_for_each_slave_rcu(bond, slave, iter) {
4808	if (--i < 0) {
4809	if (bond_slave_can_tx(slave))
4810	return slave;
4811	}
4812	}
4813
4814	/* Here we start from the first slave up to slave_id */
4815	i = slave_id;
4816	bond_for_each_slave_rcu(bond, slave, iter) {
4817	if (--i < 0)
4818	break;
4819	if (bond_slave_can_tx(slave))
4820	return slave;
4821	}
4822	/* no slave that can tx has been found */
4823	return NULL;
4824	}
4825
4826	/**
4827	* bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4828	* @bond: bonding device to use
4829	*
4830	* Based on the value of the bonding device's packets_per_slave parameter
4831	* this function generates a slave id, which is usually used as the next
4832	* slave to transmit through.
4833	*/
4834	static u32 bond_rr_gen_slave_id(struct bonding *bond)
4835	{
4836	u32 slave_id;
4837	struct reciprocal_value reciprocal_packets_per_slave;
4838	int packets_per_slave = bond->params.packets_per_slave;
4839
4840	switch (packets_per_slave) {
4841	case 0:
4842	slave_id = get_random_u32();
4843	break;
4844	case 1:
4845	slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4846	break;
4847	default:
4848	reciprocal_packets_per_slave =
4849	bond->params.reciprocal_packets_per_slave;
4850	slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4851	slave_id = reciprocal_divide(a: slave_id,
4852	R: reciprocal_packets_per_slave);
4853	break;
4854	}
4855
4856	return slave_id;
4857	}
4858
4859	static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4860	struct sk_buff *skb)
4861	{
4862	struct slave *slave;
4863	int slave_cnt;
4864	u32 slave_id;
4865
4866	/* Start with the curr_active_slave that joined the bond as the
4867	* default for sending IGMP traffic. For failover purposes one
4868	* needs to maintain some consistency for the interface that will
4869	* send the join/membership reports. The curr_active_slave found
4870	* will send all of this type of traffic.
4871	*/
4872	if (skb->protocol == htons(ETH_P_IP)) {
4873	int noff = skb_network_offset(skb);
4874	struct iphdr *iph;
4875
4876	if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4877	goto non_igmp;
4878
4879	iph = ip_hdr(skb);
4880	if (iph->protocol == IPPROTO_IGMP) {
4881	slave = rcu_dereference(bond->curr_active_slave);
4882	if (slave)
4883	return slave;
4884	return bond_get_slave_by_id(bond, slave_id: 0);
4885	}
4886	}
4887
4888	non_igmp:
4889	slave_cnt = READ_ONCE(bond->slave_cnt);
4890	if (likely(slave_cnt)) {
4891	slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4892	return bond_get_slave_by_id(bond, slave_id);
4893	}
4894	return NULL;
4895	}
4896
4897	static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4898	struct xdp_buff *xdp)
4899	{
4900	struct slave *slave;
4901	int slave_cnt;
4902	u32 slave_id;
4903	const struct ethhdr *eth;
4904	void *data = xdp->data;
4905
4906	if (data + sizeof(struct ethhdr) > xdp->data_end)
4907	goto non_igmp;
4908
4909	eth = (struct ethhdr *)data;
4910	data += sizeof(struct ethhdr);
4911
4912	/* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4913	if (eth->h_proto == htons(ETH_P_IP)) {
4914	const struct iphdr *iph;
4915
4916	if (data + sizeof(struct iphdr) > xdp->data_end)
4917	goto non_igmp;
4918
4919	iph = (struct iphdr *)data;
4920
4921	if (iph->protocol == IPPROTO_IGMP) {
4922	slave = rcu_dereference(bond->curr_active_slave);
4923	if (slave)
4924	return slave;
4925	return bond_get_slave_by_id(bond, slave_id: 0);
4926	}
4927	}
4928
4929	non_igmp:
4930	slave_cnt = READ_ONCE(bond->slave_cnt);
4931	if (likely(slave_cnt)) {
4932	slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4933	return bond_get_slave_by_id(bond, slave_id);
4934	}
4935	return NULL;
4936	}
4937
4938	static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4939	struct net_device *bond_dev)
4940	{
4941	struct bonding *bond = netdev_priv(dev: bond_dev);
4942	struct slave *slave;
4943
4944	slave = bond_xmit_roundrobin_slave_get(bond, skb);
4945	if (likely(slave))
4946	return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
4947
4948	return bond_tx_drop(dev: bond_dev, skb);
4949	}
4950
4951	static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4952	{
4953	return rcu_dereference(bond->curr_active_slave);
4954	}
4955
4956	/* In active-backup mode, we know that bond->curr_active_slave is always valid if
4957	* the bond has a usable interface.
4958	*/
4959	static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4960	struct net_device *bond_dev)
4961	{
4962	struct bonding *bond = netdev_priv(dev: bond_dev);
4963	struct slave *slave;
4964
4965	slave = bond_xmit_activebackup_slave_get(bond);
4966	if (slave)
4967	return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
4968
4969	return bond_tx_drop(dev: bond_dev, skb);
4970	}
4971
4972	/* Use this to update slave_array when (a) it's not appropriate to update
4973	* slave_array right away (note that update_slave_array() may sleep)
4974	* and / or (b) RTNL is not held.
4975	*/
4976	void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4977	{
4978	queue_delayed_work(wq: bond->wq, dwork: &bond->slave_arr_work, delay);
4979	}
4980
4981	/* Slave array work handler. Holds only RTNL */
4982	static void bond_slave_arr_handler(struct work_struct *work)
4983	{
4984	struct bonding *bond = container_of(work, struct bonding,
4985	slave_arr_work.work);
4986	int ret;
4987
4988	if (!rtnl_trylock())
4989	goto err;
4990
4991	ret = bond_update_slave_arr(bond, NULL);
4992	rtnl_unlock();
4993	if (ret) {
4994	pr_warn_ratelimited("Failed to update slave array from WT\n");
4995	goto err;
4996	}
4997	return;
4998
4999	err:
5000	bond_slave_arr_work_rearm(bond, delay: 1);
5001	}
5002
5003	static void bond_skip_slave(struct bond_up_slave *slaves,
5004	struct slave *skipslave)
5005	{
5006	int idx;
5007
5008	/* Rare situation where caller has asked to skip a specific
5009	* slave but allocation failed (most likely!). BTW this is
5010	* only possible when the call is initiated from
5011	* __bond_release_one(). In this situation; overwrite the
5012	* skipslave entry in the array with the last entry from the
5013	* array to avoid a situation where the xmit path may choose
5014	* this to-be-skipped slave to send a packet out.
5015	*/
5016	for (idx = 0; slaves && idx < slaves->count; idx++) {
5017	if (skipslave == slaves->arr[idx]) {
5018	slaves->arr[idx] =
5019	slaves->arr[slaves->count - 1];
5020	slaves->count--;
5021	break;
5022	}
5023	}
5024	}
5025
5026	static void bond_set_slave_arr(struct bonding *bond,
5027	struct bond_up_slave *usable_slaves,
5028	struct bond_up_slave *all_slaves)
5029	{
5030	struct bond_up_slave *usable, *all;
5031
5032	usable = rtnl_dereference(bond->usable_slaves);
5033	rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5034	kfree_rcu(usable, rcu);
5035
5036	all = rtnl_dereference(bond->all_slaves);
5037	rcu_assign_pointer(bond->all_slaves, all_slaves);
5038	kfree_rcu(all, rcu);
5039	}
5040
5041	static void bond_reset_slave_arr(struct bonding *bond)
5042	{
5043	bond_set_slave_arr(bond, NULL, NULL);
5044	}
5045
5046	/* Build the usable slaves array in control path for modes that use xmit-hash
5047	* to determine the slave interface -
5048	* (a) BOND_MODE_8023AD
5049	* (b) BOND_MODE_XOR
5050	* (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5051	*
5052	* The caller is expected to hold RTNL only and NO other lock!
5053	*/
5054	int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5055	{
5056	struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5057	struct slave *slave;
5058	struct list_head *iter;
5059	int agg_id = 0;
5060	int ret = 0;
5061
5062	might_sleep();
5063
5064	usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5065	bond->slave_cnt), GFP_KERNEL);
5066	all_slaves = kzalloc(struct_size(all_slaves, arr,
5067	bond->slave_cnt), GFP_KERNEL);
5068	if (!usable_slaves || !all_slaves) {
5069	ret = -ENOMEM;
5070	goto out;
5071	}
5072	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5073	struct ad_info ad_info;
5074
5075	spin_lock_bh(lock: &bond->mode_lock);
5076	if (bond_3ad_get_active_agg_info(bond, ad_info: &ad_info)) {
5077	spin_unlock_bh(lock: &bond->mode_lock);
5078	pr_debug("bond_3ad_get_active_agg_info failed\n");
5079	/* No active aggragator means it's not safe to use
5080	* the previous array.
5081	*/
5082	bond_reset_slave_arr(bond);
5083	goto out;
5084	}
5085	spin_unlock_bh(lock: &bond->mode_lock);
5086	agg_id = ad_info.aggregator_id;
5087	}
5088	bond_for_each_slave(bond, slave, iter) {
5089	if (skipslave == slave)
5090	continue;
5091
5092	all_slaves->arr[all_slaves->count++] = slave;
5093	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5094	struct aggregator *agg;
5095
5096	agg = SLAVE_AD_INFO(slave)->port.aggregator;
5097	if (!agg || agg->aggregator_identifier != agg_id)
5098	continue;
5099	}
5100	if (!bond_slave_can_tx(slave))
5101	continue;
5102
5103	slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5104	usable_slaves->count);
5105
5106	usable_slaves->arr[usable_slaves->count++] = slave;
5107	}
5108
5109	bond_set_slave_arr(bond, usable_slaves, all_slaves);
5110	return ret;
5111	out:
5112	if (ret != 0 && skipslave) {
5113	bond_skip_slave(rtnl_dereference(bond->all_slaves),
5114	skipslave);
5115	bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5116	skipslave);
5117	}
5118	kfree_rcu(all_slaves, rcu);
5119	kfree_rcu(usable_slaves, rcu);
5120
5121	return ret;
5122	}
5123
5124	static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5125	struct sk_buff *skb,
5126	struct bond_up_slave *slaves)
5127	{
5128	struct slave *slave;
5129	unsigned int count;
5130	u32 hash;
5131
5132	hash = bond_xmit_hash(bond, skb);
5133	count = slaves ? READ_ONCE(slaves->count) : 0;
5134	if (unlikely(!count))
5135	return NULL;
5136
5137	slave = slaves->arr[hash % count];
5138	return slave;
5139	}
5140
5141	static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5142	struct xdp_buff *xdp)
5143	{
5144	struct bond_up_slave *slaves;
5145	unsigned int count;
5146	u32 hash;
5147
5148	hash = bond_xmit_hash_xdp(bond, xdp);
5149	slaves = rcu_dereference(bond->usable_slaves);
5150	count = slaves ? READ_ONCE(slaves->count) : 0;
5151	if (unlikely(!count))
5152	return NULL;
5153
5154	return slaves->arr[hash % count];
5155	}
5156
5157	static bool bond_should_broadcast_neighbor(struct sk_buff *skb,
5158	struct net_device *dev)
5159	{
5160	struct bonding *bond = netdev_priv(dev);
5161	struct {
5162	struct ipv6hdr ip6;
5163	struct icmp6hdr icmp6;
5164	} *combined, _combined;
5165
5166	if (!static_branch_unlikely(&bond_bcast_neigh_enabled))
5167	return false;
5168
5169	if (!bond->params.broadcast_neighbor)
5170	return false;
5171
5172	if (skb->protocol == htons(ETH_P_ARP))
5173	return true;
5174
5175	if (skb->protocol == htons(ETH_P_IPV6)) {
5176	combined = skb_header_pointer(skb, offset: skb_mac_header_len(skb),
5177	len: sizeof(_combined),
5178	buffer: &_combined);
5179	if (combined && combined->ip6.nexthdr == NEXTHDR_ICMP &&
5180	(combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION ||
5181	combined->icmp6.icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT))
5182	return true;
5183	}
5184
5185	return false;
5186	}
5187
5188	/* Use this Xmit function for 3AD as well as XOR modes. The current
5189	* usable slave array is formed in the control path. The xmit function
5190	* just calculates hash and sends the packet out.
5191	*/
5192	static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5193	struct net_device *dev)
5194	{
5195	struct bonding *bond = netdev_priv(dev);
5196	struct bond_up_slave *slaves;
5197	struct slave *slave;
5198
5199	slaves = rcu_dereference(bond->usable_slaves);
5200	slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5201	if (likely(slave))
5202	return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
5203
5204	return bond_tx_drop(dev, skb);
5205	}
5206
5207	/* in broadcast mode, we send everything to all or usable slave interfaces.
5208	* under rcu_read_lock when this function is called.
5209	*/
5210	static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5211	struct net_device *bond_dev,
5212	bool all_slaves)
5213	{
5214	struct bonding *bond = netdev_priv(dev: bond_dev);
5215	struct bond_up_slave *slaves;
5216	bool xmit_suc = false;
5217	bool skb_used = false;
5218	int slaves_count, i;
5219
5220	if (all_slaves)
5221	slaves = rcu_dereference(bond->all_slaves);
5222	else
5223	slaves = rcu_dereference(bond->usable_slaves);
5224
5225	slaves_count = slaves ? READ_ONCE(slaves->count) : 0;
5226	for (i = 0; i < slaves_count; i++) {
5227	struct slave *slave = slaves->arr[i];
5228	struct sk_buff *skb2;
5229
5230	if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5231	continue;
5232
5233	if (bond_is_last_slave(bond, slave)) {
5234	skb2 = skb;
5235	skb_used = true;
5236	} else {
5237	skb2 = skb_clone(skb, GFP_ATOMIC);
5238	if (!skb2) {
5239	net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5240	bond_dev->name, __func__);
5241	continue;
5242	}
5243	}
5244
5245	if (bond_dev_queue_xmit(bond, skb: skb2, slave_dev: slave->dev) == NETDEV_TX_OK)
5246	xmit_suc = true;
5247	}
5248
5249	if (!skb_used)
5250	dev_kfree_skb_any(skb);
5251
5252	if (xmit_suc)
5253	return NETDEV_TX_OK;
5254
5255	dev_core_stats_tx_dropped_inc(dev: bond_dev);
5256	return NET_XMIT_DROP;
5257	}
5258
5259	/*------------------------- Device initialization ---------------------------*/
5260
5261	/* Lookup the slave that corresponds to a qid */
5262	static inline int bond_slave_override(struct bonding *bond,
5263	struct sk_buff *skb)
5264	{
5265	struct slave *slave = NULL;
5266	struct list_head *iter;
5267
5268	if (!skb_rx_queue_recorded(skb))
5269	return 1;
5270
5271	/* Find out if any slaves have the same mapping as this skb. */
5272	bond_for_each_slave_rcu(bond, slave, iter) {
5273	if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) {
5274	if (bond_slave_is_up(slave) &&
5275	slave->link == BOND_LINK_UP) {
5276	bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
5277	return 0;
5278	}
5279	/* If the slave isn't UP, use default transmit policy. */
5280	break;
5281	}
5282	}
5283
5284	return 1;
5285	}
5286
5287
5288	static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5289	struct net_device *sb_dev)
5290	{
5291	/* This helper function exists to help dev_pick_tx get the correct
5292	* destination queue. Using a helper function skips a call to
5293	* skb_tx_hash and will put the skbs in the queue we expect on their
5294	* way down to the bonding driver.
5295	*/
5296	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5297
5298	/* Save the original txq to restore before passing to the driver */
5299	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5300
5301	if (unlikely(txq >= dev->real_num_tx_queues)) {
5302	do {
5303	txq -= dev->real_num_tx_queues;
5304	} while (txq >= dev->real_num_tx_queues);
5305	}
5306	return txq;
5307	}
5308
5309	static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5310	struct sk_buff *skb,
5311	bool all_slaves)
5312	{
5313	struct bonding *bond = netdev_priv(dev: master_dev);
5314	struct bond_up_slave *slaves;
5315	struct slave *slave = NULL;
5316
5317	switch (BOND_MODE(bond)) {
5318	case BOND_MODE_ROUNDROBIN:
5319	slave = bond_xmit_roundrobin_slave_get(bond, skb);
5320	break;
5321	case BOND_MODE_ACTIVEBACKUP:
5322	slave = bond_xmit_activebackup_slave_get(bond);
5323	break;
5324	case BOND_MODE_8023AD:
5325	case BOND_MODE_XOR:
5326	if (all_slaves)
5327	slaves = rcu_dereference(bond->all_slaves);
5328	else
5329	slaves = rcu_dereference(bond->usable_slaves);
5330	slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5331	break;
5332	case BOND_MODE_BROADCAST:
5333	break;
5334	case BOND_MODE_ALB:
5335	slave = bond_xmit_alb_slave_get(bond, skb);
5336	break;
5337	case BOND_MODE_TLB:
5338	slave = bond_xmit_tlb_slave_get(bond, skb);
5339	break;
5340	default:
5341	/* Should never happen, mode already checked */
5342	WARN_ONCE(true, "Unknown bonding mode");
5343	break;
5344	}
5345
5346	if (slave)
5347	return slave->dev;
5348	return NULL;
5349	}
5350
5351	static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5352	{
5353	switch (sk->sk_family) {
5354	#if IS_ENABLED(CONFIG_IPV6)
5355	case AF_INET6:
5356	if (ipv6_only_sock(sk) ||
5357	ipv6_addr_type(addr: &sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5358	flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5359	flow->addrs.v6addrs.src = inet6_sk(sk: sk)->saddr;
5360	flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5361	break;
5362	}
5363	fallthrough;
5364	#endif
5365	default: /* AF_INET */
5366	flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5367	flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5368	flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5369	break;
5370	}
5371
5372	flow->ports.src = inet_sk(sk)->inet_sport;
5373	flow->ports.dst = inet_sk(sk)->inet_dport;
5374	}
5375
5376	/**
5377	* bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5378	* @sk: socket to use for headers
5379	*
5380	* This function will extract the necessary field from the socket and use
5381	* them to generate a hash based on the LAYER34 xmit_policy.
5382	* Assumes that sk is a TCP or UDP socket.
5383	*/
5384	static u32 bond_sk_hash_l34(struct sock *sk)
5385	{
5386	struct flow_keys flow;
5387	u32 hash;
5388
5389	bond_sk_to_flow(sk, flow: &flow);
5390
5391	/* L4 */
5392	memcpy(&hash, &flow.ports.ports, sizeof(hash));
5393	/* L3 */
5394	return bond_ip_hash(hash, flow: &flow, BOND_XMIT_POLICY_LAYER34);
5395	}
5396
5397	static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5398	struct sock *sk)
5399	{
5400	struct bond_up_slave *slaves;
5401	struct slave *slave;
5402	unsigned int count;
5403	u32 hash;
5404
5405	slaves = rcu_dereference(bond->usable_slaves);
5406	count = slaves ? READ_ONCE(slaves->count) : 0;
5407	if (unlikely(!count))
5408	return NULL;
5409
5410	hash = bond_sk_hash_l34(sk);
5411	slave = slaves->arr[hash % count];
5412
5413	return slave->dev;
5414	}
5415
5416	static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5417	struct sock *sk)
5418	{
5419	struct bonding *bond = netdev_priv(dev);
5420	struct net_device *lower = NULL;
5421
5422	rcu_read_lock();
5423	if (bond_sk_check(bond))
5424	lower = __bond_sk_get_lower_dev(bond, sk);
5425	rcu_read_unlock();
5426
5427	return lower;
5428	}
5429
5430	#if IS_ENABLED(CONFIG_TLS_DEVICE)
5431	static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5432	struct net_device *dev)
5433	{
5434	struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5435
5436	/* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5437	* was true, if tls_device_down is running in parallel, but it's OK,
5438	* because bond_get_slave_by_dev has a NULL check.
5439	*/
5440	if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5441	return bond_dev_queue_xmit(bond, skb, slave_dev: tls_netdev);
5442	return bond_tx_drop(dev, skb);
5443	}
5444	#endif
5445
5446	static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5447	{
5448	struct bonding *bond = netdev_priv(dev);
5449
5450	if (bond_should_override_tx_queue(bond) &&
5451	!bond_slave_override(bond, skb))
5452	return NETDEV_TX_OK;
5453
5454	#if IS_ENABLED(CONFIG_TLS_DEVICE)
5455	if (tls_is_skb_tx_device_offloaded(skb))
5456	return bond_tls_device_xmit(bond, skb, dev);
5457	#endif
5458
5459	switch (BOND_MODE(bond)) {
5460	case BOND_MODE_ROUNDROBIN:
5461	return bond_xmit_roundrobin(skb, bond_dev: dev);
5462	case BOND_MODE_ACTIVEBACKUP:
5463	return bond_xmit_activebackup(skb, bond_dev: dev);
5464	case BOND_MODE_8023AD:
5465	if (bond_should_broadcast_neighbor(skb, dev))
5466	return bond_xmit_broadcast(skb, bond_dev: dev, all_slaves: false);
5467	fallthrough;
5468	case BOND_MODE_XOR:
5469	return bond_3ad_xor_xmit(skb, dev);
5470	case BOND_MODE_BROADCAST:
5471	return bond_xmit_broadcast(skb, bond_dev: dev, all_slaves: true);
5472	case BOND_MODE_ALB:
5473	return bond_alb_xmit(skb, bond_dev: dev);
5474	case BOND_MODE_TLB:
5475	return bond_tlb_xmit(skb, bond_dev: dev);
5476	default:
5477	/* Should never happen, mode already checked */
5478	netdev_err(dev, format: "Unknown bonding mode %d\n", BOND_MODE(bond));
5479	WARN_ON_ONCE(1);
5480	return bond_tx_drop(dev, skb);
5481	}
5482	}
5483
5484	static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5485	{
5486	struct bonding *bond = netdev_priv(dev);
5487	netdev_tx_t ret = NETDEV_TX_OK;
5488
5489	/* If we risk deadlock from transmitting this in the
5490	* netpoll path, tell netpoll to queue the frame for later tx
5491	*/
5492	if (unlikely(is_netpoll_tx_blocked(dev)))
5493	return NETDEV_TX_BUSY;
5494
5495	rcu_read_lock();
5496	if (bond_has_slaves(bond))
5497	ret = __bond_start_xmit(skb, dev);
5498	else
5499	ret = bond_tx_drop(dev, skb);
5500	rcu_read_unlock();
5501
5502	return ret;
5503	}
5504
5505	static struct net_device *
5506	bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5507	{
5508	struct bonding *bond = netdev_priv(dev: bond_dev);
5509	struct slave *slave;
5510
5511	/* Caller needs to hold rcu_read_lock() */
5512
5513	switch (BOND_MODE(bond)) {
5514	case BOND_MODE_ROUNDROBIN:
5515	slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5516	break;
5517
5518	case BOND_MODE_ACTIVEBACKUP:
5519	slave = bond_xmit_activebackup_slave_get(bond);
5520	break;
5521
5522	case BOND_MODE_8023AD:
5523	case BOND_MODE_XOR:
5524	slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5525	break;
5526
5527	default:
5528	if (net_ratelimit())
5529	netdev_err(dev: bond_dev, format: "Unknown bonding mode %d for xdp xmit\n",
5530	BOND_MODE(bond));
5531	return NULL;
5532	}
5533
5534	if (slave)
5535	return slave->dev;
5536
5537	return NULL;
5538	}
5539
5540	static int bond_xdp_xmit(struct net_device *bond_dev,
5541	int n, struct xdp_frame **frames, u32 flags)
5542	{
5543	int nxmit, err = -ENXIO;
5544
5545	rcu_read_lock();
5546
5547	for (nxmit = 0; nxmit < n; nxmit++) {
5548	struct xdp_frame *frame = frames[nxmit];
5549	struct xdp_frame *frames1[] = {frame};
5550	struct net_device *slave_dev;
5551	struct xdp_buff xdp;
5552
5553	xdp_convert_frame_to_buff(frame, xdp: &xdp);
5554
5555	slave_dev = bond_xdp_get_xmit_slave(bond_dev, xdp: &xdp);
5556	if (!slave_dev) {
5557	err = -ENXIO;
5558	break;
5559	}
5560
5561	err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5562	if (err < 1)
5563	break;
5564	}
5565
5566	rcu_read_unlock();
5567
5568	/* If error happened on the first frame then we can pass the error up, otherwise
5569	* report the number of frames that were xmitted.
5570	*/
5571	if (err < 0)
5572	return (nxmit == 0 ? err : nxmit);
5573
5574	return nxmit;
5575	}
5576
5577	static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5578	struct netlink_ext_ack *extack)
5579	{
5580	struct bonding *bond = netdev_priv(dev);
5581	struct list_head *iter;
5582	struct slave *slave, *rollback_slave;
5583	struct bpf_prog *old_prog;
5584	struct netdev_bpf xdp = {
5585	.command = XDP_SETUP_PROG,
5586	.flags = 0,
5587	.prog = prog,
5588	.extack = extack,
5589	};
5590	int err;
5591
5592	ASSERT_RTNL();
5593
5594	if (!bond_xdp_check(bond, BOND_MODE(bond))) {
5595	BOND_NL_ERR(dev, extack,
5596	"No native XDP support for the current bonding mode");
5597	return -EOPNOTSUPP;
5598	}
5599
5600	old_prog = bond->xdp_prog;
5601	bond->xdp_prog = prog;
5602
5603	bond_for_each_slave(bond, slave, iter) {
5604	struct net_device *slave_dev = slave->dev;
5605
5606	if (!slave_dev->netdev_ops->ndo_bpf ||
5607	!slave_dev->netdev_ops->ndo_xdp_xmit) {
5608	SLAVE_NL_ERR(dev, slave_dev, extack,
5609	"Slave device does not support XDP");
5610	err = -EOPNOTSUPP;
5611	goto err;
5612	}
5613
5614	if (dev_xdp_prog_count(dev: slave_dev) > 0) {
5615	SLAVE_NL_ERR(dev, slave_dev, extack,
5616	"Slave has XDP program loaded, please unload before enslaving");
5617	err = -EOPNOTSUPP;
5618	goto err;
5619	}
5620
5621	err = dev_xdp_propagate(dev: slave_dev, bpf: &xdp);
5622	if (err < 0) {
5623	/* ndo_bpf() sets extack error message */
5624	slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5625	goto err;
5626	}
5627	if (prog)
5628	bpf_prog_inc(prog);
5629	}
5630
5631	if (prog) {
5632	static_branch_inc(&bpf_master_redirect_enabled_key);
5633	} else if (old_prog) {
5634	bpf_prog_put(prog: old_prog);
5635	static_branch_dec(&bpf_master_redirect_enabled_key);
5636	}
5637
5638	return 0;
5639
5640	err:
5641	/* unwind the program changes */
5642	bond->xdp_prog = old_prog;
5643	xdp.prog = old_prog;
5644	xdp.extack = NULL; /* do not overwrite original error */
5645
5646	bond_for_each_slave(bond, rollback_slave, iter) {
5647	struct net_device *slave_dev = rollback_slave->dev;
5648	int err_unwind;
5649
5650	if (slave == rollback_slave)
5651	break;
5652
5653	err_unwind = dev_xdp_propagate(dev: slave_dev, bpf: &xdp);
5654	if (err_unwind < 0)
5655	slave_err(dev, slave_dev,
5656	"Error %d when unwinding XDP program change\n", err_unwind);
5657	else if (xdp.prog)
5658	bpf_prog_inc(prog: xdp.prog);
5659	}
5660	return err;
5661	}
5662
5663	static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5664	{
5665	switch (xdp->command) {
5666	case XDP_SETUP_PROG:
5667	return bond_xdp_set(dev, prog: xdp->prog, extack: xdp->extack);
5668	default:
5669	return -EINVAL;
5670	}
5671	}
5672
5673	static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5674	{
5675	if (speed == 0 || speed == SPEED_UNKNOWN)
5676	speed = slave->speed;
5677	else
5678	speed = min(speed, slave->speed);
5679
5680	return speed;
5681	}
5682
5683	/* Set the BOND_PHC_INDEX flag to notify user space */
5684	static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5685	{
5686	struct ifreq *ifr = kernel_cfg->ifr;
5687	struct hwtstamp_config cfg;
5688
5689	if (kernel_cfg->copied_to_user) {
5690	/* Lower device has a legacy implementation */
5691	if (copy_from_user(to: &cfg, from: ifr->ifr_data, n: sizeof(cfg)))
5692	return -EFAULT;
5693
5694	cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5695	if (copy_to_user(to: ifr->ifr_data, from: &cfg, n: sizeof(cfg)))
5696	return -EFAULT;
5697	} else {
5698	kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5699	}
5700
5701	return 0;
5702	}
5703
5704	static int bond_hwtstamp_get(struct net_device *dev,
5705	struct kernel_hwtstamp_config *cfg)
5706	{
5707	struct bonding *bond = netdev_priv(dev);
5708	struct net_device *real_dev;
5709	int err;
5710
5711	real_dev = bond_option_active_slave_get_rcu(bond);
5712	if (!real_dev)
5713	return -EOPNOTSUPP;
5714
5715	err = generic_hwtstamp_get_lower(dev: real_dev, kernel_cfg: cfg);
5716	if (err)
5717	return err;
5718
5719	return bond_set_phc_index_flag(kernel_cfg: cfg);
5720	}
5721
5722	static int bond_hwtstamp_set(struct net_device *dev,
5723	struct kernel_hwtstamp_config *cfg,
5724	struct netlink_ext_ack *extack)
5725	{
5726	struct bonding *bond = netdev_priv(dev);
5727	struct net_device *real_dev;
5728	int err;
5729
5730	if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5731	return -EOPNOTSUPP;
5732
5733	real_dev = bond_option_active_slave_get_rcu(bond);
5734	if (!real_dev)
5735	return -EOPNOTSUPP;
5736
5737	err = generic_hwtstamp_set_lower(dev: real_dev, kernel_cfg: cfg, extack);
5738	if (err)
5739	return err;
5740
5741	return bond_set_phc_index_flag(kernel_cfg: cfg);
5742	}
5743
5744	static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5745	struct ethtool_link_ksettings *cmd)
5746	{
5747	struct bonding *bond = netdev_priv(dev: bond_dev);
5748	struct list_head *iter;
5749	struct slave *slave;
5750	u32 speed = 0;
5751
5752	cmd->base.duplex = DUPLEX_UNKNOWN;
5753	cmd->base.port = PORT_OTHER;
5754
5755	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5756	* do not need to check mode. Though link speed might not represent
5757	* the true receive or transmit bandwidth (not all modes are symmetric)
5758	* this is an accurate maximum.
5759	*/
5760	bond_for_each_slave(bond, slave, iter) {
5761	if (bond_slave_can_tx(slave)) {
5762	bond_update_speed_duplex(slave);
5763	if (slave->speed != SPEED_UNKNOWN) {
5764	if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5765	speed = bond_mode_bcast_speed(slave,
5766	speed);
5767	else
5768	speed += slave->speed;
5769	}
5770	if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5771	slave->duplex != DUPLEX_UNKNOWN)
5772	cmd->base.duplex = slave->duplex;
5773	}
5774	}
5775	cmd->base.speed = speed ? : SPEED_UNKNOWN;
5776
5777	return 0;
5778	}
5779
5780	static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5781	struct ethtool_drvinfo *drvinfo)
5782	{
5783	strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5784	snprintf(buf: drvinfo->fw_version, size: sizeof(drvinfo->fw_version), fmt: "%d",
5785	BOND_ABI_VERSION);
5786	}
5787
5788	static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5789	struct kernel_ethtool_ts_info *info)
5790	{
5791	struct bonding *bond = netdev_priv(dev: bond_dev);
5792	struct kernel_ethtool_ts_info ts_info;
5793	struct net_device *real_dev;
5794	bool sw_tx_support = false;
5795	struct list_head *iter;
5796	struct slave *slave;
5797	int ret = 0;
5798
5799	rcu_read_lock();
5800	real_dev = bond_option_active_slave_get_rcu(bond);
5801	dev_hold(dev: real_dev);
5802	rcu_read_unlock();
5803
5804	if (real_dev) {
5805	ret = ethtool_get_ts_info_by_layer(dev: real_dev, info);
5806	} else {
5807	/* Check if all slaves support software tx timestamping */
5808	rcu_read_lock();
5809	bond_for_each_slave_rcu(bond, slave, iter) {
5810	ret = ethtool_get_ts_info_by_layer(dev: slave->dev, info: &ts_info);
5811	if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5812	sw_tx_support = true;
5813	continue;
5814	}
5815
5816	sw_tx_support = false;
5817	break;
5818	}
5819	rcu_read_unlock();
5820	}
5821
5822	if (sw_tx_support)
5823	info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5824
5825	dev_put(dev: real_dev);
5826	return ret;
5827	}
5828
5829	static const struct ethtool_ops bond_ethtool_ops = {
5830	.get_drvinfo = bond_ethtool_get_drvinfo,
5831	.get_link = ethtool_op_get_link,
5832	.get_link_ksettings = bond_ethtool_get_link_ksettings,
5833	.get_ts_info = bond_ethtool_get_ts_info,
5834	};
5835
5836	static const struct net_device_ops bond_netdev_ops = {
5837	.ndo_init = bond_init,
5838	.ndo_uninit = bond_uninit,
5839	.ndo_open = bond_open,
5840	.ndo_stop = bond_close,
5841	.ndo_start_xmit = bond_start_xmit,
5842	.ndo_select_queue = bond_select_queue,
5843	.ndo_get_stats64 = bond_get_stats,
5844	.ndo_eth_ioctl = bond_eth_ioctl,
5845	.ndo_siocbond = bond_do_ioctl,
5846	.ndo_siocdevprivate = bond_siocdevprivate,
5847	.ndo_change_rx_flags = bond_change_rx_flags,
5848	.ndo_set_rx_mode = bond_set_rx_mode,
5849	.ndo_change_mtu = bond_change_mtu,
5850	.ndo_set_mac_address = bond_set_mac_address,
5851	.ndo_neigh_setup = bond_neigh_setup,
5852	.ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5853	.ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5854	#ifdef CONFIG_NET_POLL_CONTROLLER
5855	.ndo_netpoll_setup = bond_netpoll_setup,
5856	.ndo_netpoll_cleanup = bond_netpoll_cleanup,
5857	.ndo_poll_controller = bond_poll_controller,
5858	#endif
5859	.ndo_add_slave = bond_enslave,
5860	.ndo_del_slave = bond_release,
5861	.ndo_fix_features = bond_fix_features,
5862	.ndo_features_check = passthru_features_check,
5863	.ndo_get_xmit_slave = bond_xmit_get_slave,
5864	.ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5865	.ndo_bpf = bond_xdp,
5866	.ndo_xdp_xmit = bond_xdp_xmit,
5867	.ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5868	.ndo_hwtstamp_get = bond_hwtstamp_get,
5869	.ndo_hwtstamp_set = bond_hwtstamp_set,
5870	};
5871
5872	static const struct device_type bond_type = {
5873	.name = "bond",
5874	};
5875
5876	static void bond_destructor(struct net_device *bond_dev)
5877	{
5878	struct bonding *bond = netdev_priv(dev: bond_dev);
5879
5880	if (bond->wq)
5881	destroy_workqueue(wq: bond->wq);
5882
5883	free_percpu(pdata: bond->rr_tx_counter);
5884	}
5885
5886	void bond_setup(struct net_device *bond_dev)
5887	{
5888	struct bonding *bond = netdev_priv(dev: bond_dev);
5889
5890	spin_lock_init(&bond->mode_lock);
5891	bond->params = bonding_defaults;
5892
5893	/* Initialize pointers */
5894	bond->dev = bond_dev;
5895
5896	/* Initialize the device entry points */
5897	ether_setup(dev: bond_dev);
5898	bond_dev->max_mtu = ETH_MAX_MTU;
5899	bond_dev->netdev_ops = &bond_netdev_ops;
5900	bond_dev->ethtool_ops = &bond_ethtool_ops;
5901
5902	bond_dev->needs_free_netdev = true;
5903	bond_dev->priv_destructor = bond_destructor;
5904
5905	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5906
5907	/* Initialize the device options */
5908	bond_dev->flags |= IFF_MASTER;
5909	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5910	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5911
5912	#ifdef CONFIG_XFRM_OFFLOAD
5913	/* set up xfrm device ops (only supported in active-backup right now) */
5914	bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5915	INIT_LIST_HEAD(list: &bond->ipsec_list);
5916	mutex_init(&bond->ipsec_lock);
5917	#endif /* CONFIG_XFRM_OFFLOAD */
5918
5919	/* don't acquire bond device's netif_tx_lock when transmitting */
5920	bond_dev->lltx = true;
5921
5922	/* Don't allow bond devices to change network namespaces. */
5923	bond_dev->netns_immutable = true;
5924
5925	/* By default, we declare the bond to be fully
5926	* VLAN hardware accelerated capable. Special
5927	* care is taken in the various xmit functions
5928	* when there are slaves that are not hw accel
5929	* capable
5930	*/
5931
5932	bond_dev->hw_features = MASTER_UPPER_DEV_VLAN_FEATURES |
5933	NETIF_F_HW_VLAN_CTAG_RX |
5934	NETIF_F_HW_VLAN_CTAG_FILTER |
5935	NETIF_F_HW_VLAN_STAG_RX |
5936	NETIF_F_HW_VLAN_STAG_FILTER;
5937
5938	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5939	bond_dev->features |= bond_dev->hw_features;
5940	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5941	bond_dev->features |= NETIF_F_GSO_PARTIAL;
5942	#ifdef CONFIG_XFRM_OFFLOAD
5943	bond_dev->hw_features |= BOND_XFRM_FEATURES;
5944	/* Only enable XFRM features if this is an active-backup config */
5945	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5946	bond_dev->features |= BOND_XFRM_FEATURES;
5947	#endif /* CONFIG_XFRM_OFFLOAD */
5948	}
5949
5950	/* Destroy a bonding device.
5951	* Must be under rtnl_lock when this function is called.
5952	*/
5953	static void bond_uninit(struct net_device *bond_dev)
5954	{
5955	struct bonding *bond = netdev_priv(dev: bond_dev);
5956	struct list_head *iter;
5957	struct slave *slave;
5958
5959	bond_netpoll_cleanup(bond_dev);
5960
5961	/* Release the bonded slaves */
5962	bond_for_each_slave(bond, slave, iter)
5963	__bond_release_one(bond_dev, slave_dev: slave->dev, all: true, unregister: true);
5964	netdev_info(dev: bond_dev, format: "Released all slaves\n");
5965
5966	#ifdef CONFIG_XFRM_OFFLOAD
5967	mutex_destroy(lock: &bond->ipsec_lock);
5968	#endif /* CONFIG_XFRM_OFFLOAD */
5969
5970	bond_set_slave_arr(bond, NULL, NULL);
5971
5972	list_del_rcu(entry: &bond->bond_list);
5973
5974	bond_debug_unregister(bond);
5975	}
5976
5977	/*------------------------- Module initialization ---------------------------*/
5978
5979	static int __init bond_check_params(struct bond_params *params)
5980	{
5981	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5982	struct bond_opt_value newval;
5983	const struct bond_opt_value *valptr;
5984	int arp_all_targets_value = 0;
5985	u16 ad_actor_sys_prio = 0;
5986	u16 ad_user_port_key = 0;
5987	__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5988	int arp_ip_count;
5989	int bond_mode = BOND_MODE_ROUNDROBIN;
5990	int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5991	int lacp_fast = 0;
5992	int tlb_dynamic_lb;
5993
5994	/* Convert string parameters. */
5995	if (mode) {
5996	bond_opt_initstr(&newval, mode);
5997	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_MODE), val: &newval);
5998	if (!valptr) {
5999	pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
6000	return -EINVAL;
6001	}
6002	bond_mode = valptr->value;
6003	}
6004
6005	if (xmit_hash_policy) {
6006	if (bond_mode == BOND_MODE_ROUNDROBIN ||
6007	bond_mode == BOND_MODE_ACTIVEBACKUP ||
6008	bond_mode == BOND_MODE_BROADCAST) {
6009	pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
6010	bond_mode_name(bond_mode));
6011	} else {
6012	bond_opt_initstr(&newval, xmit_hash_policy);
6013	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_XMIT_HASH),
6014	val: &newval);
6015	if (!valptr) {
6016	pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
6017	xmit_hash_policy);
6018	return -EINVAL;
6019	}
6020	xmit_hashtype = valptr->value;
6021	}
6022	}
6023
6024	if (lacp_rate) {
6025	if (bond_mode != BOND_MODE_8023AD) {
6026	pr_info("lacp_rate param is irrelevant in mode %s\n",
6027	bond_mode_name(bond_mode));
6028	} else {
6029	bond_opt_initstr(&newval, lacp_rate);
6030	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_LACP_RATE),
6031	val: &newval);
6032	if (!valptr) {
6033	pr_err("Error: Invalid lacp rate \"%s\"\n",
6034	lacp_rate);
6035	return -EINVAL;
6036	}
6037	lacp_fast = valptr->value;
6038	}
6039	}
6040
6041	if (ad_select) {
6042	bond_opt_initstr(&newval, ad_select);
6043	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_AD_SELECT),
6044	val: &newval);
6045	if (!valptr) {
6046	pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6047	return -EINVAL;
6048	}
6049	params->ad_select = valptr->value;
6050	if (bond_mode != BOND_MODE_8023AD)
6051	pr_warn("ad_select param only affects 802.3ad mode\n");
6052	} else {
6053	params->ad_select = BOND_AD_STABLE;
6054	}
6055
6056	if (max_bonds < 0) {
6057	pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6058	max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6059	max_bonds = BOND_DEFAULT_MAX_BONDS;
6060	}
6061
6062	if (miimon < 0) {
6063	pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6064	miimon, INT_MAX);
6065	miimon = 0;
6066	}
6067
6068	if (updelay < 0) {
6069	pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6070	updelay, INT_MAX);
6071	updelay = 0;
6072	}
6073
6074	if (downdelay < 0) {
6075	pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6076	downdelay, INT_MAX);
6077	downdelay = 0;
6078	}
6079
6080	if (use_carrier != 1) {
6081	pr_err("Error: invalid use_carrier parameter (%d)\n",
6082	use_carrier);
6083	return -EINVAL;
6084	}
6085
6086	if (num_peer_notif < 0 || num_peer_notif > 255) {
6087	pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6088	num_peer_notif);
6089	num_peer_notif = 1;
6090	}
6091
6092	/* reset values for 802.3ad/TLB/ALB */
6093	if (!bond_mode_uses_arp(mode: bond_mode)) {
6094	if (!miimon) {
6095	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");
6096	pr_warn("Forcing miimon to 100msec\n");
6097	miimon = BOND_DEFAULT_MIIMON;
6098	}
6099	}
6100
6101	if (tx_queues < 1 || tx_queues > 255) {
6102	pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6103	tx_queues, BOND_DEFAULT_TX_QUEUES);
6104	tx_queues = BOND_DEFAULT_TX_QUEUES;
6105	}
6106
6107	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6108	pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6109	all_slaves_active);
6110	all_slaves_active = 0;
6111	}
6112
6113	if (resend_igmp < 0 || resend_igmp > 255) {
6114	pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6115	resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6116	resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6117	}
6118
6119	bond_opt_initval(&newval, packets_per_slave);
6120	if (!bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_PACKETS_PER_SLAVE), val: &newval)) {
6121	pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6122	packets_per_slave, USHRT_MAX);
6123	packets_per_slave = 1;
6124	}
6125
6126	if (bond_mode == BOND_MODE_ALB) {
6127	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",
6128	updelay);
6129	}
6130
6131	if (!miimon) {
6132	if (updelay || downdelay) {
6133	/* just warn the user the up/down delay will have
6134	* no effect since miimon is zero...
6135	*/
6136	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",
6137	updelay, downdelay);
6138	}
6139	} else {
6140	/* don't allow arp monitoring */
6141	if (arp_interval) {
6142	pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6143	miimon, arp_interval);
6144	arp_interval = 0;
6145	}
6146
6147	if ((updelay % miimon) != 0) {
6148	pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6149	updelay, miimon, (updelay / miimon) * miimon);
6150	}
6151
6152	updelay /= miimon;
6153
6154	if ((downdelay % miimon) != 0) {
6155	pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6156	downdelay, miimon,
6157	(downdelay / miimon) * miimon);
6158	}
6159
6160	downdelay /= miimon;
6161	}
6162
6163	if (arp_interval < 0) {
6164	pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6165	arp_interval, INT_MAX);
6166	arp_interval = 0;
6167	}
6168
6169	for (arp_ip_count = 0, i = 0;
6170	(arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6171	__be32 ip;
6172
6173	/* not a complete check, but good enough to catch mistakes */
6174	if (!in4_pton(src: arp_ip_target[i], srclen: -1, dst: (u8 *)&ip, delim: -1, NULL) ||
6175	!bond_is_ip_target_ok(addr: ip)) {
6176	pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6177	arp_ip_target[i]);
6178	arp_interval = 0;
6179	} else {
6180	if (bond_get_targets_ip(targets: arp_target, ip) == -1)
6181	arp_target[arp_ip_count++] = ip;
6182	else
6183	pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6184	&ip);
6185	}
6186	}
6187
6188	if (arp_interval && !arp_ip_count) {
6189	/* don't allow arping if no arp_ip_target given... */
6190	pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6191	arp_interval);
6192	arp_interval = 0;
6193	}
6194
6195	if (arp_validate) {
6196	if (!arp_interval) {
6197	pr_err("arp_validate requires arp_interval\n");
6198	return -EINVAL;
6199	}
6200
6201	bond_opt_initstr(&newval, arp_validate);
6202	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_ARP_VALIDATE),
6203	val: &newval);
6204	if (!valptr) {
6205	pr_err("Error: invalid arp_validate \"%s\"\n",
6206	arp_validate);
6207	return -EINVAL;
6208	}
6209	arp_validate_value = valptr->value;
6210	} else {
6211	arp_validate_value = 0;
6212	}
6213
6214	if (arp_all_targets) {
6215	bond_opt_initstr(&newval, arp_all_targets);
6216	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_ARP_ALL_TARGETS),
6217	val: &newval);
6218	if (!valptr) {
6219	pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6220	arp_all_targets);
6221	arp_all_targets_value = 0;
6222	} else {
6223	arp_all_targets_value = valptr->value;
6224	}
6225	}
6226
6227	if (miimon) {
6228	pr_info("MII link monitoring set to %d ms\n", miimon);
6229	} else if (arp_interval) {
6230	valptr = bond_opt_get_val(option: BOND_OPT_ARP_VALIDATE,
6231	val: arp_validate_value);
6232	pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6233	arp_interval, valptr->string, arp_ip_count);
6234
6235	for (i = 0; i < arp_ip_count; i++)
6236	pr_cont(" %s", arp_ip_target[i]);
6237
6238	pr_cont("\n");
6239
6240	} else if (max_bonds) {
6241	/* miimon and arp_interval not set, we need one so things
6242	* work as expected, see bonding.txt for details
6243	*/
6244	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");
6245	}
6246
6247	if (primary && !bond_mode_uses_primary(mode: bond_mode)) {
6248	/* currently, using a primary only makes sense
6249	* in active backup, TLB or ALB modes
6250	*/
6251	pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6252	primary, bond_mode_name(bond_mode));
6253	primary = NULL;
6254	}
6255
6256	if (primary && primary_reselect) {
6257	bond_opt_initstr(&newval, primary_reselect);
6258	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_PRIMARY_RESELECT),
6259	val: &newval);
6260	if (!valptr) {
6261	pr_err("Error: Invalid primary_reselect \"%s\"\n",
6262	primary_reselect);
6263	return -EINVAL;
6264	}
6265	primary_reselect_value = valptr->value;
6266	} else {
6267	primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6268	}
6269
6270	if (fail_over_mac) {
6271	bond_opt_initstr(&newval, fail_over_mac);
6272	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_FAIL_OVER_MAC),
6273	val: &newval);
6274	if (!valptr) {
6275	pr_err("Error: invalid fail_over_mac \"%s\"\n",
6276	fail_over_mac);
6277	return -EINVAL;
6278	}
6279	fail_over_mac_value = valptr->value;
6280	if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6281	pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6282	} else {
6283	fail_over_mac_value = BOND_FOM_NONE;
6284	}
6285
6286	bond_opt_initstr(&newval, "default");
6287	valptr = bond_opt_parse(
6288	opt: bond_opt_get(option: BOND_OPT_AD_ACTOR_SYS_PRIO),
6289	val: &newval);
6290	if (!valptr) {
6291	pr_err("Error: No ad_actor_sys_prio default value");
6292	return -EINVAL;
6293	}
6294	ad_actor_sys_prio = valptr->value;
6295
6296	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_AD_USER_PORT_KEY),
6297	val: &newval);
6298	if (!valptr) {
6299	pr_err("Error: No ad_user_port_key default value");
6300	return -EINVAL;
6301	}
6302	ad_user_port_key = valptr->value;
6303
6304	bond_opt_initstr(&newval, "default");
6305	valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_TLB_DYNAMIC_LB), val: &newval);
6306	if (!valptr) {
6307	pr_err("Error: No tlb_dynamic_lb default value");
6308	return -EINVAL;
6309	}
6310	tlb_dynamic_lb = valptr->value;
6311
6312	if (lp_interval == 0) {
6313	pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6314	INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6315	lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6316	}
6317
6318	/* fill params struct with the proper values */
6319	params->mode = bond_mode;
6320	params->xmit_policy = xmit_hashtype;
6321	params->miimon = miimon;
6322	params->num_peer_notif = num_peer_notif;
6323	params->arp_interval = arp_interval;
6324	params->arp_validate = arp_validate_value;
6325	params->arp_all_targets = arp_all_targets_value;
6326	params->missed_max = 2;
6327	params->updelay = updelay;
6328	params->downdelay = downdelay;
6329	params->peer_notif_delay = 0;
6330	params->lacp_active = 1;
6331	params->lacp_fast = lacp_fast;
6332	params->primary[0] = 0;
6333	params->primary_reselect = primary_reselect_value;
6334	params->fail_over_mac = fail_over_mac_value;
6335	params->tx_queues = tx_queues;
6336	params->all_slaves_active = all_slaves_active;
6337	params->resend_igmp = resend_igmp;
6338	params->min_links = min_links;
6339	params->lp_interval = lp_interval;
6340	params->packets_per_slave = packets_per_slave;
6341	params->tlb_dynamic_lb = tlb_dynamic_lb;
6342	params->ad_actor_sys_prio = ad_actor_sys_prio;
6343	eth_zero_addr(addr: params->ad_actor_system);
6344	params->ad_user_port_key = ad_user_port_key;
6345	params->coupled_control = 1;
6346	params->broadcast_neighbor = 0;
6347	if (packets_per_slave > 0) {
6348	params->reciprocal_packets_per_slave =
6349	reciprocal_value(d: packets_per_slave);
6350	} else {
6351	/* reciprocal_packets_per_slave is unused if
6352	* packets_per_slave is 0 or 1, just initialize it
6353	*/
6354	params->reciprocal_packets_per_slave =
6355	(struct reciprocal_value) { 0 };
6356	}
6357
6358	if (primary)
6359	strscpy_pad(params->primary, primary, sizeof(params->primary));
6360
6361	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6362	#if IS_ENABLED(CONFIG_IPV6)
6363	memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6364	#endif
6365
6366	return 0;
6367	}
6368
6369	/* Called from registration process */
6370	static int bond_init(struct net_device *bond_dev)
6371	{
6372	struct bonding *bond = netdev_priv(dev: bond_dev);
6373	struct bond_net *bn = net_generic(net: dev_net(dev: bond_dev), id: bond_net_id);
6374
6375	netdev_dbg(bond_dev, "Begin bond_init\n");
6376
6377	bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM,
6378	bond_dev->name);
6379	if (!bond->wq)
6380	return -ENOMEM;
6381
6382	bond->notifier_ctx = false;
6383
6384	spin_lock_init(&bond->stats_lock);
6385	netdev_lockdep_set_classes(bond_dev);
6386
6387	list_add_tail_rcu(new: &bond->bond_list, head: &bn->dev_list);
6388
6389	bond_prepare_sysfs_group(bond);
6390
6391	bond_debug_register(bond);
6392
6393	/* Ensure valid dev_addr */
6394	if (is_zero_ether_addr(addr: bond_dev->dev_addr) &&
6395	bond_dev->addr_assign_type == NET_ADDR_PERM)
6396	eth_hw_addr_random(dev: bond_dev);
6397
6398	return 0;
6399	}
6400
6401	unsigned int bond_get_num_tx_queues(void)
6402	{
6403	return tx_queues;
6404	}
6405
6406	/* Create a new bond based on the specified name and bonding parameters.
6407	* If name is NULL, obtain a suitable "bond%d" name for us.
6408	* Caller must NOT hold rtnl_lock; we need to release it here before we
6409	* set up our sysfs entries.
6410	*/
6411	int bond_create(struct net *net, const char *name)
6412	{
6413	struct net_device *bond_dev;
6414	struct bonding *bond;
6415	int res = -ENOMEM;
6416
6417	rtnl_lock();
6418
6419	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6420	name ? name : "bond%d", NET_NAME_UNKNOWN,
6421	bond_setup, tx_queues);
6422	if (!bond_dev)
6423	goto out;
6424
6425	bond = netdev_priv(dev: bond_dev);
6426	dev_net_set(dev: bond_dev, net);
6427	bond_dev->rtnl_link_ops = &bond_link_ops;
6428
6429	res = register_netdevice(dev: bond_dev);
6430	if (res < 0) {
6431	free_netdev(dev: bond_dev);
6432	goto out;
6433	}
6434
6435	netif_carrier_off(dev: bond_dev);
6436
6437	bond_work_init_all(bond);
6438
6439	out:
6440	rtnl_unlock();
6441	return res;
6442	}
6443
6444	static int __net_init bond_net_init(struct net *net)
6445	{
6446	struct bond_net *bn = net_generic(net, id: bond_net_id);
6447
6448	bn->net = net;
6449	INIT_LIST_HEAD(list: &bn->dev_list);
6450
6451	bond_create_proc_dir(bn);
6452	bond_create_sysfs(net: bn);
6453
6454	return 0;
6455	}
6456
6457	/* According to commit 69b0216ac255 ("bonding: fix bonding_masters
6458	* race condition in bond unloading") we need to remove sysfs files
6459	* before we remove our devices (done later in bond_net_exit_rtnl())
6460	*/
6461	static void __net_exit bond_net_pre_exit(struct net *net)
6462	{
6463	struct bond_net *bn = net_generic(net, id: bond_net_id);
6464
6465	bond_destroy_sysfs(net: bn);
6466	}
6467
6468	static void __net_exit bond_net_exit_rtnl(struct net *net,
6469	struct list_head *dev_kill_list)
6470	{
6471	struct bond_net *bn = net_generic(net, id: bond_net_id);
6472	struct bonding *bond, *tmp_bond;
6473
6474	/* Kill off any bonds created after unregistering bond rtnl ops */
6475	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6476	unregister_netdevice_queue(dev: bond->dev, head: dev_kill_list);
6477	}
6478
6479	/* According to commit 23fa5c2caae0 ("bonding: destroy proc directory
6480	* only after all bonds are gone") bond_destroy_proc_dir() is called
6481	* after bond_net_exit_rtnl() has completed.
6482	*/
6483	static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6484	{
6485	struct bond_net *bn;
6486	struct net *net;
6487
6488	list_for_each_entry(net, net_list, exit_list) {
6489	bn = net_generic(net, id: bond_net_id);
6490	bond_destroy_proc_dir(bn);
6491	}
6492	}
6493
6494	static struct pernet_operations bond_net_ops = {
6495	.init = bond_net_init,
6496	.pre_exit = bond_net_pre_exit,
6497	.exit_rtnl = bond_net_exit_rtnl,
6498	.exit_batch = bond_net_exit_batch,
6499	.id = &bond_net_id,
6500	.size = sizeof(struct bond_net),
6501	};
6502
6503	static int __init bonding_init(void)
6504	{
6505	int i;
6506	int res;
6507
6508	res = bond_check_params(params: &bonding_defaults);
6509	if (res)
6510	goto out;
6511
6512	bond_create_debugfs();
6513
6514	res = register_pernet_subsys(&bond_net_ops);
6515	if (res)
6516	goto err_net_ops;
6517
6518	res = bond_netlink_init();
6519	if (res)
6520	goto err_link;
6521
6522	for (i = 0; i < max_bonds; i++) {
6523	res = bond_create(net: &init_net, NULL);
6524	if (res)
6525	goto err;
6526	}
6527
6528	skb_flow_dissector_init(flow_dissector: &flow_keys_bonding,
6529	key: flow_keys_bonding_keys,
6530	ARRAY_SIZE(flow_keys_bonding_keys));
6531
6532	register_netdevice_notifier(nb: &bond_netdev_notifier);
6533	out:
6534	return res;
6535	err:
6536	bond_netlink_fini();
6537	err_link:
6538	unregister_pernet_subsys(&bond_net_ops);
6539	err_net_ops:
6540	bond_destroy_debugfs();
6541	goto out;
6542
6543	}
6544
6545	static void __exit bonding_exit(void)
6546	{
6547	unregister_netdevice_notifier(nb: &bond_netdev_notifier);
6548
6549	bond_netlink_fini();
6550	unregister_pernet_subsys(&bond_net_ops);
6551
6552	bond_destroy_debugfs();
6553
6554	#ifdef CONFIG_NET_POLL_CONTROLLER
6555	/* Make sure we don't have an imbalance on our netpoll blocking */
6556	WARN_ON(atomic_read(&netpoll_block_tx));
6557	#endif
6558	}
6559
6560	module_init(bonding_init);
6561	module_exit(bonding_exit);
6562	MODULE_LICENSE("GPL");
6563	MODULE_DESCRIPTION(DRV_DESCRIPTION);
6564	MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
6565	MODULE_IMPORT_NS("NETDEV_INTERNAL");
6566