xref: /openbmc/linux/drivers/net/bonding/bond_main.c (revision 90099433)
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *	Cisco 5500
11  *	Sun Trunking (Solaris)
12  *	Alteon AceDirector Trunks
13  *	Linux Bonding
14  *	and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *	will be assigned at this time.  The hw mac address will come from
20  *	the first slave bonded to the channel.  All slaves will then use
21  *	this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *	will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *	a: be used as initial mac address
29  *	b: if a hw mac address already is there, eth0's hw mac address
30  *	   will then be set from bond0.
31  *
32  */
33 
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.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/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/dma.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
65 #include <net/sock.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
69 #include <net/arp.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include "bonding.h"
80 #include "bond_3ad.h"
81 #include "bond_alb.h"
82 
83 /*---------------------------- Module parameters ----------------------------*/
84 
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV	0
87 #define BOND_LINK_ARP_INTERV	0
88 
89 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
90 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
91 static int num_peer_notif = 1;
92 static int miimon	= BOND_LINK_MON_INTERV;
93 static int updelay;
94 static int downdelay;
95 static int use_carrier	= 1;
96 static char *mode;
97 static char *primary;
98 static char *primary_reselect;
99 static char *lacp_rate;
100 static int min_links;
101 static char *ad_select;
102 static char *xmit_hash_policy;
103 static int arp_interval = BOND_LINK_ARP_INTERV;
104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
105 static char *arp_validate;
106 static char *fail_over_mac;
107 static int all_slaves_active = 0;
108 static struct bond_params bonding_defaults;
109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
110 
111 module_param(max_bonds, int, 0);
112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
113 module_param(tx_queues, int, 0);
114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
117 			       "failover event (alias of num_unsol_na)");
118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
120 			       "failover event (alias of num_grat_arp)");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
127 			    "in milliseconds");
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130 			      "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
133 		       "1 for active-backup, 2 for balance-xor, "
134 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135 		       "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
140 				   "once it comes up; "
141 				   "0 for always (default), "
142 				   "1 for only if speed of primary is "
143 				   "better, "
144 				   "2 for only on active slave "
145 				   "failure");
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
148 			    "0 for slow, 1 for fast");
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
151 			    "0 for stable (default), 1 for bandwidth, "
152 			    "2 for count");
153 module_param(min_links, int, 0);
154 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
155 
156 module_param(xmit_hash_policy, charp, 0);
157 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
158 				   "0 for layer 2 (default), 1 for layer 3+4, "
159 				   "2 for layer 2+3");
160 module_param(arp_interval, int, 0);
161 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
162 module_param_array(arp_ip_target, charp, NULL, 0);
163 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
164 module_param(arp_validate, charp, 0);
165 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
166 			       "0 for none (default), 1 for active, "
167 			       "2 for backup, 3 for all");
168 module_param(fail_over_mac, charp, 0);
169 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
170 				"the same MAC; 0 for none (default), "
171 				"1 for active, 2 for follow");
172 module_param(all_slaves_active, int, 0);
173 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
174 				     "by setting active flag for all slaves; "
175 				     "0 for never (default), 1 for always.");
176 module_param(resend_igmp, int, 0);
177 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
178 			      "link failure");
179 
180 /*----------------------------- Global variables ----------------------------*/
181 
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
184 #endif
185 
186 int bond_net_id __read_mostly;
187 
188 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
189 static int arp_ip_count;
190 static int bond_mode	= BOND_MODE_ROUNDROBIN;
191 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
192 static int lacp_fast;
193 
194 const struct bond_parm_tbl bond_lacp_tbl[] = {
195 {	"slow",		AD_LACP_SLOW},
196 {	"fast",		AD_LACP_FAST},
197 {	NULL,		-1},
198 };
199 
200 const struct bond_parm_tbl bond_mode_tbl[] = {
201 {	"balance-rr",		BOND_MODE_ROUNDROBIN},
202 {	"active-backup",	BOND_MODE_ACTIVEBACKUP},
203 {	"balance-xor",		BOND_MODE_XOR},
204 {	"broadcast",		BOND_MODE_BROADCAST},
205 {	"802.3ad",		BOND_MODE_8023AD},
206 {	"balance-tlb",		BOND_MODE_TLB},
207 {	"balance-alb",		BOND_MODE_ALB},
208 {	NULL,			-1},
209 };
210 
211 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
212 {	"layer2",		BOND_XMIT_POLICY_LAYER2},
213 {	"layer3+4",		BOND_XMIT_POLICY_LAYER34},
214 {	"layer2+3",		BOND_XMIT_POLICY_LAYER23},
215 {	NULL,			-1},
216 };
217 
218 const struct bond_parm_tbl arp_validate_tbl[] = {
219 {	"none",			BOND_ARP_VALIDATE_NONE},
220 {	"active",		BOND_ARP_VALIDATE_ACTIVE},
221 {	"backup",		BOND_ARP_VALIDATE_BACKUP},
222 {	"all",			BOND_ARP_VALIDATE_ALL},
223 {	NULL,			-1},
224 };
225 
226 const struct bond_parm_tbl fail_over_mac_tbl[] = {
227 {	"none",			BOND_FOM_NONE},
228 {	"active",		BOND_FOM_ACTIVE},
229 {	"follow",		BOND_FOM_FOLLOW},
230 {	NULL,			-1},
231 };
232 
233 const struct bond_parm_tbl pri_reselect_tbl[] = {
234 {	"always",		BOND_PRI_RESELECT_ALWAYS},
235 {	"better",		BOND_PRI_RESELECT_BETTER},
236 {	"failure",		BOND_PRI_RESELECT_FAILURE},
237 {	NULL,			-1},
238 };
239 
240 struct bond_parm_tbl ad_select_tbl[] = {
241 {	"stable",	BOND_AD_STABLE},
242 {	"bandwidth",	BOND_AD_BANDWIDTH},
243 {	"count",	BOND_AD_COUNT},
244 {	NULL,		-1},
245 };
246 
247 /*-------------------------- Forward declarations ---------------------------*/
248 
249 static int bond_init(struct net_device *bond_dev);
250 static void bond_uninit(struct net_device *bond_dev);
251 
252 /*---------------------------- General routines -----------------------------*/
253 
254 const char *bond_mode_name(int mode)
255 {
256 	static const char *names[] = {
257 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
258 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
259 		[BOND_MODE_XOR] = "load balancing (xor)",
260 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
261 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
262 		[BOND_MODE_TLB] = "transmit load balancing",
263 		[BOND_MODE_ALB] = "adaptive load balancing",
264 	};
265 
266 	if (mode < 0 || mode > BOND_MODE_ALB)
267 		return "unknown";
268 
269 	return names[mode];
270 }
271 
272 /*---------------------------------- VLAN -----------------------------------*/
273 
274 /**
275  * bond_add_vlan - add a new vlan id on bond
276  * @bond: bond that got the notification
277  * @vlan_id: the vlan id to add
278  *
279  * Returns -ENOMEM if allocation failed.
280  */
281 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
282 {
283 	struct vlan_entry *vlan;
284 
285 	pr_debug("bond: %s, vlan id %d\n",
286 		 (bond ? bond->dev->name : "None"), vlan_id);
287 
288 	vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
289 	if (!vlan)
290 		return -ENOMEM;
291 
292 	INIT_LIST_HEAD(&vlan->vlan_list);
293 	vlan->vlan_id = vlan_id;
294 
295 	write_lock_bh(&bond->lock);
296 
297 	list_add_tail(&vlan->vlan_list, &bond->vlan_list);
298 
299 	write_unlock_bh(&bond->lock);
300 
301 	pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
302 
303 	return 0;
304 }
305 
306 /**
307  * bond_del_vlan - delete a vlan id from bond
308  * @bond: bond that got the notification
309  * @vlan_id: the vlan id to delete
310  *
311  * returns -ENODEV if @vlan_id was not found in @bond.
312  */
313 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
314 {
315 	struct vlan_entry *vlan;
316 	int res = -ENODEV;
317 
318 	pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
319 
320 	block_netpoll_tx();
321 	write_lock_bh(&bond->lock);
322 
323 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
324 		if (vlan->vlan_id == vlan_id) {
325 			list_del(&vlan->vlan_list);
326 
327 			if (bond_is_lb(bond))
328 				bond_alb_clear_vlan(bond, vlan_id);
329 
330 			pr_debug("removed VLAN ID %d from bond %s\n",
331 				 vlan_id, bond->dev->name);
332 
333 			kfree(vlan);
334 
335 			res = 0;
336 			goto out;
337 		}
338 	}
339 
340 	pr_debug("couldn't find VLAN ID %d in bond %s\n",
341 		 vlan_id, bond->dev->name);
342 
343 out:
344 	write_unlock_bh(&bond->lock);
345 	unblock_netpoll_tx();
346 	return res;
347 }
348 
349 /**
350  * bond_next_vlan - safely skip to the next item in the vlans list.
351  * @bond: the bond we're working on
352  * @curr: item we're advancing from
353  *
354  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
355  * or @curr->next otherwise (even if it is @curr itself again).
356  *
357  * Caller must hold bond->lock
358  */
359 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
360 {
361 	struct vlan_entry *next, *last;
362 
363 	if (list_empty(&bond->vlan_list))
364 		return NULL;
365 
366 	if (!curr) {
367 		next = list_entry(bond->vlan_list.next,
368 				  struct vlan_entry, vlan_list);
369 	} else {
370 		last = list_entry(bond->vlan_list.prev,
371 				  struct vlan_entry, vlan_list);
372 		if (last == curr) {
373 			next = list_entry(bond->vlan_list.next,
374 					  struct vlan_entry, vlan_list);
375 		} else {
376 			next = list_entry(curr->vlan_list.next,
377 					  struct vlan_entry, vlan_list);
378 		}
379 	}
380 
381 	return next;
382 }
383 
384 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
385 
386 /**
387  * bond_dev_queue_xmit - Prepare skb for xmit.
388  *
389  * @bond: bond device that got this skb for tx.
390  * @skb: hw accel VLAN tagged skb to transmit
391  * @slave_dev: slave that is supposed to xmit this skbuff
392  */
393 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
394 			struct net_device *slave_dev)
395 {
396 	skb->dev = slave_dev;
397 
398 	skb->queue_mapping = bond_queue_mapping(skb);
399 
400 	if (unlikely(netpoll_tx_running(slave_dev)))
401 		bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
402 	else
403 		dev_queue_xmit(skb);
404 
405 	return 0;
406 }
407 
408 /*
409  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
410  * We don't protect the slave list iteration with a lock because:
411  * a. This operation is performed in IOCTL context,
412  * b. The operation is protected by the RTNL semaphore in the 8021q code,
413  * c. Holding a lock with BH disabled while directly calling a base driver
414  *    entry point is generally a BAD idea.
415  *
416  * The design of synchronization/protection for this operation in the 8021q
417  * module is good for one or more VLAN devices over a single physical device
418  * and cannot be extended for a teaming solution like bonding, so there is a
419  * potential race condition here where a net device from the vlan group might
420  * be referenced (either by a base driver or the 8021q code) while it is being
421  * removed from the system. However, it turns out we're not making matters
422  * worse, and if it works for regular VLAN usage it will work here too.
423 */
424 
425 /**
426  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
427  * @bond_dev: bonding net device that got called
428  * @vid: vlan id being added
429  */
430 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
431 {
432 	struct bonding *bond = netdev_priv(bond_dev);
433 	struct slave *slave, *stop_at;
434 	int i, res;
435 
436 	bond_for_each_slave(bond, slave, i) {
437 		res = vlan_vid_add(slave->dev, vid);
438 		if (res)
439 			goto unwind;
440 	}
441 
442 	res = bond_add_vlan(bond, vid);
443 	if (res) {
444 		pr_err("%s: Error: Failed to add vlan id %d\n",
445 		       bond_dev->name, vid);
446 		return res;
447 	}
448 
449 	return 0;
450 
451 unwind:
452 	/* unwind from head to the slave that failed */
453 	stop_at = slave;
454 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
455 		vlan_vid_del(slave->dev, vid);
456 
457 	return res;
458 }
459 
460 /**
461  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
462  * @bond_dev: bonding net device that got called
463  * @vid: vlan id being removed
464  */
465 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
466 {
467 	struct bonding *bond = netdev_priv(bond_dev);
468 	struct slave *slave;
469 	int i, res;
470 
471 	bond_for_each_slave(bond, slave, i)
472 		vlan_vid_del(slave->dev, vid);
473 
474 	res = bond_del_vlan(bond, vid);
475 	if (res) {
476 		pr_err("%s: Error: Failed to remove vlan id %d\n",
477 		       bond_dev->name, vid);
478 		return res;
479 	}
480 
481 	return 0;
482 }
483 
484 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
485 {
486 	struct vlan_entry *vlan;
487 	int res;
488 
489 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
490 		res = vlan_vid_add(slave_dev, vlan->vlan_id);
491 		if (res)
492 			pr_warning("%s: Failed to add vlan id %d to device %s\n",
493 				   bond->dev->name, vlan->vlan_id,
494 				   slave_dev->name);
495 	}
496 }
497 
498 static void bond_del_vlans_from_slave(struct bonding *bond,
499 				      struct net_device *slave_dev)
500 {
501 	struct vlan_entry *vlan;
502 
503 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
504 		if (!vlan->vlan_id)
505 			continue;
506 		vlan_vid_del(slave_dev, vlan->vlan_id);
507 	}
508 }
509 
510 /*------------------------------- Link status -------------------------------*/
511 
512 /*
513  * Set the carrier state for the master according to the state of its
514  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
515  * do special 802.3ad magic.
516  *
517  * Returns zero if carrier state does not change, nonzero if it does.
518  */
519 static int bond_set_carrier(struct bonding *bond)
520 {
521 	struct slave *slave;
522 	int i;
523 
524 	if (bond->slave_cnt == 0)
525 		goto down;
526 
527 	if (bond->params.mode == BOND_MODE_8023AD)
528 		return bond_3ad_set_carrier(bond);
529 
530 	bond_for_each_slave(bond, slave, i) {
531 		if (slave->link == BOND_LINK_UP) {
532 			if (!netif_carrier_ok(bond->dev)) {
533 				netif_carrier_on(bond->dev);
534 				return 1;
535 			}
536 			return 0;
537 		}
538 	}
539 
540 down:
541 	if (netif_carrier_ok(bond->dev)) {
542 		netif_carrier_off(bond->dev);
543 		return 1;
544 	}
545 	return 0;
546 }
547 
548 /*
549  * Get link speed and duplex from the slave's base driver
550  * using ethtool. If for some reason the call fails or the
551  * values are invalid, set speed and duplex to -1,
552  * and return.
553  */
554 static void bond_update_speed_duplex(struct slave *slave)
555 {
556 	struct net_device *slave_dev = slave->dev;
557 	struct ethtool_cmd ecmd;
558 	u32 slave_speed;
559 	int res;
560 
561 	slave->speed = SPEED_UNKNOWN;
562 	slave->duplex = DUPLEX_UNKNOWN;
563 
564 	res = __ethtool_get_settings(slave_dev, &ecmd);
565 	if (res < 0)
566 		return;
567 
568 	slave_speed = ethtool_cmd_speed(&ecmd);
569 	if (slave_speed == 0 || slave_speed == ((__u32) -1))
570 		return;
571 
572 	switch (ecmd.duplex) {
573 	case DUPLEX_FULL:
574 	case DUPLEX_HALF:
575 		break;
576 	default:
577 		return;
578 	}
579 
580 	slave->speed = slave_speed;
581 	slave->duplex = ecmd.duplex;
582 
583 	return;
584 }
585 
586 /*
587  * if <dev> supports MII link status reporting, check its link status.
588  *
589  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
590  * depending upon the setting of the use_carrier parameter.
591  *
592  * Return either BMSR_LSTATUS, meaning that the link is up (or we
593  * can't tell and just pretend it is), or 0, meaning that the link is
594  * down.
595  *
596  * If reporting is non-zero, instead of faking link up, return -1 if
597  * both ETHTOOL and MII ioctls fail (meaning the device does not
598  * support them).  If use_carrier is set, return whatever it says.
599  * It'd be nice if there was a good way to tell if a driver supports
600  * netif_carrier, but there really isn't.
601  */
602 static int bond_check_dev_link(struct bonding *bond,
603 			       struct net_device *slave_dev, int reporting)
604 {
605 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
606 	int (*ioctl)(struct net_device *, struct ifreq *, int);
607 	struct ifreq ifr;
608 	struct mii_ioctl_data *mii;
609 
610 	if (!reporting && !netif_running(slave_dev))
611 		return 0;
612 
613 	if (bond->params.use_carrier)
614 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
615 
616 	/* Try to get link status using Ethtool first. */
617 	if (slave_dev->ethtool_ops) {
618 		if (slave_dev->ethtool_ops->get_link) {
619 			u32 link;
620 
621 			link = slave_dev->ethtool_ops->get_link(slave_dev);
622 
623 			return link ? BMSR_LSTATUS : 0;
624 		}
625 	}
626 
627 	/* Ethtool can't be used, fallback to MII ioctls. */
628 	ioctl = slave_ops->ndo_do_ioctl;
629 	if (ioctl) {
630 		/* TODO: set pointer to correct ioctl on a per team member */
631 		/*       bases to make this more efficient. that is, once  */
632 		/*       we determine the correct ioctl, we will always    */
633 		/*       call it and not the others for that team          */
634 		/*       member.                                           */
635 
636 		/*
637 		 * We cannot assume that SIOCGMIIPHY will also read a
638 		 * register; not all network drivers (e.g., e100)
639 		 * support that.
640 		 */
641 
642 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
643 		strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
644 		mii = if_mii(&ifr);
645 		if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
646 			mii->reg_num = MII_BMSR;
647 			if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
648 				return mii->val_out & BMSR_LSTATUS;
649 		}
650 	}
651 
652 	/*
653 	 * If reporting, report that either there's no dev->do_ioctl,
654 	 * or both SIOCGMIIREG and get_link failed (meaning that we
655 	 * cannot report link status).  If not reporting, pretend
656 	 * we're ok.
657 	 */
658 	return reporting ? -1 : BMSR_LSTATUS;
659 }
660 
661 /*----------------------------- Multicast list ------------------------------*/
662 
663 /*
664  * Push the promiscuity flag down to appropriate slaves
665  */
666 static int bond_set_promiscuity(struct bonding *bond, int inc)
667 {
668 	int err = 0;
669 	if (USES_PRIMARY(bond->params.mode)) {
670 		/* write lock already acquired */
671 		if (bond->curr_active_slave) {
672 			err = dev_set_promiscuity(bond->curr_active_slave->dev,
673 						  inc);
674 		}
675 	} else {
676 		struct slave *slave;
677 		int i;
678 		bond_for_each_slave(bond, slave, i) {
679 			err = dev_set_promiscuity(slave->dev, inc);
680 			if (err)
681 				return err;
682 		}
683 	}
684 	return err;
685 }
686 
687 /*
688  * Push the allmulti flag down to all slaves
689  */
690 static int bond_set_allmulti(struct bonding *bond, int inc)
691 {
692 	int err = 0;
693 	if (USES_PRIMARY(bond->params.mode)) {
694 		/* write lock already acquired */
695 		if (bond->curr_active_slave) {
696 			err = dev_set_allmulti(bond->curr_active_slave->dev,
697 					       inc);
698 		}
699 	} else {
700 		struct slave *slave;
701 		int i;
702 		bond_for_each_slave(bond, slave, i) {
703 			err = dev_set_allmulti(slave->dev, inc);
704 			if (err)
705 				return err;
706 		}
707 	}
708 	return err;
709 }
710 
711 /*
712  * Add a Multicast address to slaves
713  * according to mode
714  */
715 static void bond_mc_add(struct bonding *bond, void *addr)
716 {
717 	if (USES_PRIMARY(bond->params.mode)) {
718 		/* write lock already acquired */
719 		if (bond->curr_active_slave)
720 			dev_mc_add(bond->curr_active_slave->dev, addr);
721 	} else {
722 		struct slave *slave;
723 		int i;
724 
725 		bond_for_each_slave(bond, slave, i)
726 			dev_mc_add(slave->dev, addr);
727 	}
728 }
729 
730 /*
731  * Remove a multicast address from slave
732  * according to mode
733  */
734 static void bond_mc_del(struct bonding *bond, void *addr)
735 {
736 	if (USES_PRIMARY(bond->params.mode)) {
737 		/* write lock already acquired */
738 		if (bond->curr_active_slave)
739 			dev_mc_del(bond->curr_active_slave->dev, addr);
740 	} else {
741 		struct slave *slave;
742 		int i;
743 		bond_for_each_slave(bond, slave, i) {
744 			dev_mc_del(slave->dev, addr);
745 		}
746 	}
747 }
748 
749 
750 static void __bond_resend_igmp_join_requests(struct net_device *dev)
751 {
752 	struct in_device *in_dev;
753 
754 	rcu_read_lock();
755 	in_dev = __in_dev_get_rcu(dev);
756 	if (in_dev)
757 		ip_mc_rejoin_groups(in_dev);
758 	rcu_read_unlock();
759 }
760 
761 /*
762  * Retrieve the list of registered multicast addresses for the bonding
763  * device and retransmit an IGMP JOIN request to the current active
764  * slave.
765  */
766 static void bond_resend_igmp_join_requests(struct bonding *bond)
767 {
768 	struct net_device *bond_dev, *vlan_dev, *master_dev;
769 	struct vlan_entry *vlan;
770 
771 	read_lock(&bond->lock);
772 
773 	bond_dev = bond->dev;
774 
775 	/* rejoin all groups on bond device */
776 	__bond_resend_igmp_join_requests(bond_dev);
777 
778 	/*
779 	 * if bond is enslaved to a bridge,
780 	 * then rejoin all groups on its master
781 	 */
782 	master_dev = bond_dev->master;
783 	if (master_dev)
784 		if ((master_dev->priv_flags & IFF_EBRIDGE)
785 			&& (bond_dev->priv_flags & IFF_BRIDGE_PORT))
786 			__bond_resend_igmp_join_requests(master_dev);
787 
788 	/* rejoin all groups on vlan devices */
789 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
790 		rcu_read_lock();
791 		vlan_dev = __vlan_find_dev_deep(bond_dev,
792 						vlan->vlan_id);
793 		rcu_read_unlock();
794 		if (vlan_dev)
795 			__bond_resend_igmp_join_requests(vlan_dev);
796 	}
797 
798 	if (--bond->igmp_retrans > 0)
799 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
800 
801 	read_unlock(&bond->lock);
802 }
803 
804 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
805 {
806 	struct bonding *bond = container_of(work, struct bonding,
807 					    mcast_work.work);
808 	bond_resend_igmp_join_requests(bond);
809 }
810 
811 /*
812  * flush all members of flush->mc_list from device dev->mc_list
813  */
814 static void bond_mc_list_flush(struct net_device *bond_dev,
815 			       struct net_device *slave_dev)
816 {
817 	struct bonding *bond = netdev_priv(bond_dev);
818 	struct netdev_hw_addr *ha;
819 
820 	netdev_for_each_mc_addr(ha, bond_dev)
821 		dev_mc_del(slave_dev, ha->addr);
822 
823 	if (bond->params.mode == BOND_MODE_8023AD) {
824 		/* del lacpdu mc addr from mc list */
825 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
826 
827 		dev_mc_del(slave_dev, lacpdu_multicast);
828 	}
829 }
830 
831 /*--------------------------- Active slave change ---------------------------*/
832 
833 /*
834  * Update the mc list and multicast-related flags for the new and
835  * old active slaves (if any) according to the multicast mode, and
836  * promiscuous flags unconditionally.
837  */
838 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
839 			 struct slave *old_active)
840 {
841 	struct netdev_hw_addr *ha;
842 
843 	if (!USES_PRIMARY(bond->params.mode))
844 		/* nothing to do -  mc list is already up-to-date on
845 		 * all slaves
846 		 */
847 		return;
848 
849 	if (old_active) {
850 		if (bond->dev->flags & IFF_PROMISC)
851 			dev_set_promiscuity(old_active->dev, -1);
852 
853 		if (bond->dev->flags & IFF_ALLMULTI)
854 			dev_set_allmulti(old_active->dev, -1);
855 
856 		netdev_for_each_mc_addr(ha, bond->dev)
857 			dev_mc_del(old_active->dev, ha->addr);
858 	}
859 
860 	if (new_active) {
861 		/* FIXME: Signal errors upstream. */
862 		if (bond->dev->flags & IFF_PROMISC)
863 			dev_set_promiscuity(new_active->dev, 1);
864 
865 		if (bond->dev->flags & IFF_ALLMULTI)
866 			dev_set_allmulti(new_active->dev, 1);
867 
868 		netdev_for_each_mc_addr(ha, bond->dev)
869 			dev_mc_add(new_active->dev, ha->addr);
870 	}
871 }
872 
873 /*
874  * bond_do_fail_over_mac
875  *
876  * Perform special MAC address swapping for fail_over_mac settings
877  *
878  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
879  */
880 static void bond_do_fail_over_mac(struct bonding *bond,
881 				  struct slave *new_active,
882 				  struct slave *old_active)
883 	__releases(&bond->curr_slave_lock)
884 	__releases(&bond->lock)
885 	__acquires(&bond->lock)
886 	__acquires(&bond->curr_slave_lock)
887 {
888 	u8 tmp_mac[ETH_ALEN];
889 	struct sockaddr saddr;
890 	int rv;
891 
892 	switch (bond->params.fail_over_mac) {
893 	case BOND_FOM_ACTIVE:
894 		if (new_active) {
895 			memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
896 			       new_active->dev->addr_len);
897 			write_unlock_bh(&bond->curr_slave_lock);
898 			read_unlock(&bond->lock);
899 			call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
900 			read_lock(&bond->lock);
901 			write_lock_bh(&bond->curr_slave_lock);
902 		}
903 		break;
904 	case BOND_FOM_FOLLOW:
905 		/*
906 		 * if new_active && old_active, swap them
907 		 * if just old_active, do nothing (going to no active slave)
908 		 * if just new_active, set new_active to bond's MAC
909 		 */
910 		if (!new_active)
911 			return;
912 
913 		write_unlock_bh(&bond->curr_slave_lock);
914 		read_unlock(&bond->lock);
915 
916 		if (old_active) {
917 			memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
918 			memcpy(saddr.sa_data, old_active->dev->dev_addr,
919 			       ETH_ALEN);
920 			saddr.sa_family = new_active->dev->type;
921 		} else {
922 			memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
923 			saddr.sa_family = bond->dev->type;
924 		}
925 
926 		rv = dev_set_mac_address(new_active->dev, &saddr);
927 		if (rv) {
928 			pr_err("%s: Error %d setting MAC of slave %s\n",
929 			       bond->dev->name, -rv, new_active->dev->name);
930 			goto out;
931 		}
932 
933 		if (!old_active)
934 			goto out;
935 
936 		memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
937 		saddr.sa_family = old_active->dev->type;
938 
939 		rv = dev_set_mac_address(old_active->dev, &saddr);
940 		if (rv)
941 			pr_err("%s: Error %d setting MAC of slave %s\n",
942 			       bond->dev->name, -rv, new_active->dev->name);
943 out:
944 		read_lock(&bond->lock);
945 		write_lock_bh(&bond->curr_slave_lock);
946 		break;
947 	default:
948 		pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
949 		       bond->dev->name, bond->params.fail_over_mac);
950 		break;
951 	}
952 
953 }
954 
955 static bool bond_should_change_active(struct bonding *bond)
956 {
957 	struct slave *prim = bond->primary_slave;
958 	struct slave *curr = bond->curr_active_slave;
959 
960 	if (!prim || !curr || curr->link != BOND_LINK_UP)
961 		return true;
962 	if (bond->force_primary) {
963 		bond->force_primary = false;
964 		return true;
965 	}
966 	if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
967 	    (prim->speed < curr->speed ||
968 	     (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
969 		return false;
970 	if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
971 		return false;
972 	return true;
973 }
974 
975 /**
976  * find_best_interface - select the best available slave to be the active one
977  * @bond: our bonding struct
978  *
979  * Warning: Caller must hold curr_slave_lock for writing.
980  */
981 static struct slave *bond_find_best_slave(struct bonding *bond)
982 {
983 	struct slave *new_active, *old_active;
984 	struct slave *bestslave = NULL;
985 	int mintime = bond->params.updelay;
986 	int i;
987 
988 	new_active = bond->curr_active_slave;
989 
990 	if (!new_active) { /* there were no active slaves left */
991 		if (bond->slave_cnt > 0)   /* found one slave */
992 			new_active = bond->first_slave;
993 		else
994 			return NULL; /* still no slave, return NULL */
995 	}
996 
997 	if ((bond->primary_slave) &&
998 	    bond->primary_slave->link == BOND_LINK_UP &&
999 	    bond_should_change_active(bond)) {
1000 		new_active = bond->primary_slave;
1001 	}
1002 
1003 	/* remember where to stop iterating over the slaves */
1004 	old_active = new_active;
1005 
1006 	bond_for_each_slave_from(bond, new_active, i, old_active) {
1007 		if (new_active->link == BOND_LINK_UP) {
1008 			return new_active;
1009 		} else if (new_active->link == BOND_LINK_BACK &&
1010 			   IS_UP(new_active->dev)) {
1011 			/* link up, but waiting for stabilization */
1012 			if (new_active->delay < mintime) {
1013 				mintime = new_active->delay;
1014 				bestslave = new_active;
1015 			}
1016 		}
1017 	}
1018 
1019 	return bestslave;
1020 }
1021 
1022 static bool bond_should_notify_peers(struct bonding *bond)
1023 {
1024 	struct slave *slave = bond->curr_active_slave;
1025 
1026 	pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1027 		 bond->dev->name, slave ? slave->dev->name : "NULL");
1028 
1029 	if (!slave || !bond->send_peer_notif ||
1030 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1031 		return false;
1032 
1033 	bond->send_peer_notif--;
1034 	return true;
1035 }
1036 
1037 /**
1038  * change_active_interface - change the active slave into the specified one
1039  * @bond: our bonding struct
1040  * @new: the new slave to make the active one
1041  *
1042  * Set the new slave to the bond's settings and unset them on the old
1043  * curr_active_slave.
1044  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1045  *
1046  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1047  * because it is apparently the best available slave we have, even though its
1048  * updelay hasn't timed out yet.
1049  *
1050  * If new_active is not NULL, caller must hold bond->lock for read and
1051  * curr_slave_lock for write_bh.
1052  */
1053 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1054 {
1055 	struct slave *old_active = bond->curr_active_slave;
1056 
1057 	if (old_active == new_active)
1058 		return;
1059 
1060 	if (new_active) {
1061 		new_active->jiffies = jiffies;
1062 
1063 		if (new_active->link == BOND_LINK_BACK) {
1064 			if (USES_PRIMARY(bond->params.mode)) {
1065 				pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1066 					bond->dev->name, new_active->dev->name,
1067 					(bond->params.updelay - new_active->delay) * bond->params.miimon);
1068 			}
1069 
1070 			new_active->delay = 0;
1071 			new_active->link = BOND_LINK_UP;
1072 
1073 			if (bond->params.mode == BOND_MODE_8023AD)
1074 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1075 
1076 			if (bond_is_lb(bond))
1077 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1078 		} else {
1079 			if (USES_PRIMARY(bond->params.mode)) {
1080 				pr_info("%s: making interface %s the new active one.\n",
1081 					bond->dev->name, new_active->dev->name);
1082 			}
1083 		}
1084 	}
1085 
1086 	if (USES_PRIMARY(bond->params.mode))
1087 		bond_mc_swap(bond, new_active, old_active);
1088 
1089 	if (bond_is_lb(bond)) {
1090 		bond_alb_handle_active_change(bond, new_active);
1091 		if (old_active)
1092 			bond_set_slave_inactive_flags(old_active);
1093 		if (new_active)
1094 			bond_set_slave_active_flags(new_active);
1095 	} else {
1096 		bond->curr_active_slave = new_active;
1097 	}
1098 
1099 	if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1100 		if (old_active)
1101 			bond_set_slave_inactive_flags(old_active);
1102 
1103 		if (new_active) {
1104 			bool should_notify_peers = false;
1105 
1106 			bond_set_slave_active_flags(new_active);
1107 
1108 			if (bond->params.fail_over_mac)
1109 				bond_do_fail_over_mac(bond, new_active,
1110 						      old_active);
1111 
1112 			if (netif_running(bond->dev)) {
1113 				bond->send_peer_notif =
1114 					bond->params.num_peer_notif;
1115 				should_notify_peers =
1116 					bond_should_notify_peers(bond);
1117 			}
1118 
1119 			write_unlock_bh(&bond->curr_slave_lock);
1120 			read_unlock(&bond->lock);
1121 
1122 			netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1123 			if (should_notify_peers)
1124 				netdev_bonding_change(bond->dev,
1125 						      NETDEV_NOTIFY_PEERS);
1126 
1127 			read_lock(&bond->lock);
1128 			write_lock_bh(&bond->curr_slave_lock);
1129 		}
1130 	}
1131 
1132 	/* resend IGMP joins since active slave has changed or
1133 	 * all were sent on curr_active_slave.
1134 	 * resend only if bond is brought up with the affected
1135 	 * bonding modes and the retransmission is enabled */
1136 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1137 	    ((USES_PRIMARY(bond->params.mode) && new_active) ||
1138 	     bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1139 		bond->igmp_retrans = bond->params.resend_igmp;
1140 		queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1141 	}
1142 }
1143 
1144 /**
1145  * bond_select_active_slave - select a new active slave, if needed
1146  * @bond: our bonding struct
1147  *
1148  * This functions should be called when one of the following occurs:
1149  * - The old curr_active_slave has been released or lost its link.
1150  * - The primary_slave has got its link back.
1151  * - A slave has got its link back and there's no old curr_active_slave.
1152  *
1153  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1154  */
1155 void bond_select_active_slave(struct bonding *bond)
1156 {
1157 	struct slave *best_slave;
1158 	int rv;
1159 
1160 	best_slave = bond_find_best_slave(bond);
1161 	if (best_slave != bond->curr_active_slave) {
1162 		bond_change_active_slave(bond, best_slave);
1163 		rv = bond_set_carrier(bond);
1164 		if (!rv)
1165 			return;
1166 
1167 		if (netif_carrier_ok(bond->dev)) {
1168 			pr_info("%s: first active interface up!\n",
1169 				bond->dev->name);
1170 		} else {
1171 			pr_info("%s: now running without any active interface !\n",
1172 				bond->dev->name);
1173 		}
1174 	}
1175 }
1176 
1177 /*--------------------------- slave list handling ---------------------------*/
1178 
1179 /*
1180  * This function attaches the slave to the end of list.
1181  *
1182  * bond->lock held for writing by caller.
1183  */
1184 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1185 {
1186 	if (bond->first_slave == NULL) { /* attaching the first slave */
1187 		new_slave->next = new_slave;
1188 		new_slave->prev = new_slave;
1189 		bond->first_slave = new_slave;
1190 	} else {
1191 		new_slave->next = bond->first_slave;
1192 		new_slave->prev = bond->first_slave->prev;
1193 		new_slave->next->prev = new_slave;
1194 		new_slave->prev->next = new_slave;
1195 	}
1196 
1197 	bond->slave_cnt++;
1198 }
1199 
1200 /*
1201  * This function detaches the slave from the list.
1202  * WARNING: no check is made to verify if the slave effectively
1203  * belongs to <bond>.
1204  * Nothing is freed on return, structures are just unchained.
1205  * If any slave pointer in bond was pointing to <slave>,
1206  * it should be changed by the calling function.
1207  *
1208  * bond->lock held for writing by caller.
1209  */
1210 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1211 {
1212 	if (slave->next)
1213 		slave->next->prev = slave->prev;
1214 
1215 	if (slave->prev)
1216 		slave->prev->next = slave->next;
1217 
1218 	if (bond->first_slave == slave) { /* slave is the first slave */
1219 		if (bond->slave_cnt > 1) { /* there are more slave */
1220 			bond->first_slave = slave->next;
1221 		} else {
1222 			bond->first_slave = NULL; /* slave was the last one */
1223 		}
1224 	}
1225 
1226 	slave->next = NULL;
1227 	slave->prev = NULL;
1228 	bond->slave_cnt--;
1229 }
1230 
1231 #ifdef CONFIG_NET_POLL_CONTROLLER
1232 static inline int slave_enable_netpoll(struct slave *slave)
1233 {
1234 	struct netpoll *np;
1235 	int err = 0;
1236 
1237 	np = kzalloc(sizeof(*np), GFP_KERNEL);
1238 	err = -ENOMEM;
1239 	if (!np)
1240 		goto out;
1241 
1242 	np->dev = slave->dev;
1243 	strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1244 	err = __netpoll_setup(np);
1245 	if (err) {
1246 		kfree(np);
1247 		goto out;
1248 	}
1249 	slave->np = np;
1250 out:
1251 	return err;
1252 }
1253 static inline void slave_disable_netpoll(struct slave *slave)
1254 {
1255 	struct netpoll *np = slave->np;
1256 
1257 	if (!np)
1258 		return;
1259 
1260 	slave->np = NULL;
1261 	synchronize_rcu_bh();
1262 	__netpoll_cleanup(np);
1263 	kfree(np);
1264 }
1265 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1266 {
1267 	if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1268 		return false;
1269 	if (!slave_dev->netdev_ops->ndo_poll_controller)
1270 		return false;
1271 	return true;
1272 }
1273 
1274 static void bond_poll_controller(struct net_device *bond_dev)
1275 {
1276 }
1277 
1278 static void __bond_netpoll_cleanup(struct bonding *bond)
1279 {
1280 	struct slave *slave;
1281 	int i;
1282 
1283 	bond_for_each_slave(bond, slave, i)
1284 		if (IS_UP(slave->dev))
1285 			slave_disable_netpoll(slave);
1286 }
1287 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1288 {
1289 	struct bonding *bond = netdev_priv(bond_dev);
1290 
1291 	read_lock(&bond->lock);
1292 	__bond_netpoll_cleanup(bond);
1293 	read_unlock(&bond->lock);
1294 }
1295 
1296 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1297 {
1298 	struct bonding *bond = netdev_priv(dev);
1299 	struct slave *slave;
1300 	int i, err = 0;
1301 
1302 	read_lock(&bond->lock);
1303 	bond_for_each_slave(bond, slave, i) {
1304 		err = slave_enable_netpoll(slave);
1305 		if (err) {
1306 			__bond_netpoll_cleanup(bond);
1307 			break;
1308 		}
1309 	}
1310 	read_unlock(&bond->lock);
1311 	return err;
1312 }
1313 
1314 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1315 {
1316 	return bond->dev->npinfo;
1317 }
1318 
1319 #else
1320 static inline int slave_enable_netpoll(struct slave *slave)
1321 {
1322 	return 0;
1323 }
1324 static inline void slave_disable_netpoll(struct slave *slave)
1325 {
1326 }
1327 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1328 {
1329 }
1330 #endif
1331 
1332 /*---------------------------------- IOCTL ----------------------------------*/
1333 
1334 static int bond_sethwaddr(struct net_device *bond_dev,
1335 			  struct net_device *slave_dev)
1336 {
1337 	pr_debug("bond_dev=%p\n", bond_dev);
1338 	pr_debug("slave_dev=%p\n", slave_dev);
1339 	pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1340 	memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1341 	return 0;
1342 }
1343 
1344 static netdev_features_t bond_fix_features(struct net_device *dev,
1345 	netdev_features_t features)
1346 {
1347 	struct slave *slave;
1348 	struct bonding *bond = netdev_priv(dev);
1349 	netdev_features_t mask;
1350 	int i;
1351 
1352 	read_lock(&bond->lock);
1353 
1354 	if (!bond->first_slave) {
1355 		/* Disable adding VLANs to empty bond. But why? --mq */
1356 		features |= NETIF_F_VLAN_CHALLENGED;
1357 		goto out;
1358 	}
1359 
1360 	mask = features;
1361 	features &= ~NETIF_F_ONE_FOR_ALL;
1362 	features |= NETIF_F_ALL_FOR_ALL;
1363 
1364 	bond_for_each_slave(bond, slave, i) {
1365 		features = netdev_increment_features(features,
1366 						     slave->dev->features,
1367 						     mask);
1368 	}
1369 
1370 out:
1371 	read_unlock(&bond->lock);
1372 	return features;
1373 }
1374 
1375 #define BOND_VLAN_FEATURES	(NETIF_F_ALL_CSUM | NETIF_F_SG | \
1376 				 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1377 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1378 
1379 static void bond_compute_features(struct bonding *bond)
1380 {
1381 	struct slave *slave;
1382 	struct net_device *bond_dev = bond->dev;
1383 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1384 	unsigned short max_hard_header_len = ETH_HLEN;
1385 	int i;
1386 
1387 	read_lock(&bond->lock);
1388 
1389 	if (!bond->first_slave)
1390 		goto done;
1391 
1392 	bond_for_each_slave(bond, slave, i) {
1393 		vlan_features = netdev_increment_features(vlan_features,
1394 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1395 
1396 		if (slave->dev->hard_header_len > max_hard_header_len)
1397 			max_hard_header_len = slave->dev->hard_header_len;
1398 	}
1399 
1400 done:
1401 	bond_dev->vlan_features = vlan_features;
1402 	bond_dev->hard_header_len = max_hard_header_len;
1403 
1404 	read_unlock(&bond->lock);
1405 
1406 	netdev_change_features(bond_dev);
1407 }
1408 
1409 static void bond_setup_by_slave(struct net_device *bond_dev,
1410 				struct net_device *slave_dev)
1411 {
1412 	struct bonding *bond = netdev_priv(bond_dev);
1413 
1414 	bond_dev->header_ops	    = slave_dev->header_ops;
1415 
1416 	bond_dev->type		    = slave_dev->type;
1417 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1418 	bond_dev->addr_len	    = slave_dev->addr_len;
1419 
1420 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1421 		slave_dev->addr_len);
1422 	bond->setup_by_slave = 1;
1423 }
1424 
1425 /* On bonding slaves other than the currently active slave, suppress
1426  * duplicates except for alb non-mcast/bcast.
1427  */
1428 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1429 					    struct slave *slave,
1430 					    struct bonding *bond)
1431 {
1432 	if (bond_is_slave_inactive(slave)) {
1433 		if (bond->params.mode == BOND_MODE_ALB &&
1434 		    skb->pkt_type != PACKET_BROADCAST &&
1435 		    skb->pkt_type != PACKET_MULTICAST)
1436 			return false;
1437 		return true;
1438 	}
1439 	return false;
1440 }
1441 
1442 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1443 {
1444 	struct sk_buff *skb = *pskb;
1445 	struct slave *slave;
1446 	struct bonding *bond;
1447 	void (*recv_probe)(struct sk_buff *, struct bonding *,
1448 				struct slave *);
1449 
1450 	skb = skb_share_check(skb, GFP_ATOMIC);
1451 	if (unlikely(!skb))
1452 		return RX_HANDLER_CONSUMED;
1453 
1454 	*pskb = skb;
1455 
1456 	slave = bond_slave_get_rcu(skb->dev);
1457 	bond = slave->bond;
1458 
1459 	if (bond->params.arp_interval)
1460 		slave->dev->last_rx = jiffies;
1461 
1462 	recv_probe = ACCESS_ONCE(bond->recv_probe);
1463 	if (recv_probe) {
1464 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1465 
1466 		if (likely(nskb)) {
1467 			recv_probe(nskb, bond, slave);
1468 			dev_kfree_skb(nskb);
1469 		}
1470 	}
1471 
1472 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1473 		return RX_HANDLER_EXACT;
1474 	}
1475 
1476 	skb->dev = bond->dev;
1477 
1478 	if (bond->params.mode == BOND_MODE_ALB &&
1479 	    bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1480 	    skb->pkt_type == PACKET_HOST) {
1481 
1482 		if (unlikely(skb_cow_head(skb,
1483 					  skb->data - skb_mac_header(skb)))) {
1484 			kfree_skb(skb);
1485 			return RX_HANDLER_CONSUMED;
1486 		}
1487 		memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1488 	}
1489 
1490 	return RX_HANDLER_ANOTHER;
1491 }
1492 
1493 /* enslave device <slave> to bond device <master> */
1494 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1495 {
1496 	struct bonding *bond = netdev_priv(bond_dev);
1497 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1498 	struct slave *new_slave = NULL;
1499 	struct netdev_hw_addr *ha;
1500 	struct sockaddr addr;
1501 	int link_reporting;
1502 	int res = 0;
1503 
1504 	if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1505 		slave_ops->ndo_do_ioctl == NULL) {
1506 		pr_warning("%s: Warning: no link monitoring support for %s\n",
1507 			   bond_dev->name, slave_dev->name);
1508 	}
1509 
1510 	/* already enslaved */
1511 	if (slave_dev->flags & IFF_SLAVE) {
1512 		pr_debug("Error, Device was already enslaved\n");
1513 		return -EBUSY;
1514 	}
1515 
1516 	/* vlan challenged mutual exclusion */
1517 	/* no need to lock since we're protected by rtnl_lock */
1518 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1519 		pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1520 		if (bond_vlan_used(bond)) {
1521 			pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1522 			       bond_dev->name, slave_dev->name, bond_dev->name);
1523 			return -EPERM;
1524 		} else {
1525 			pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1526 				   bond_dev->name, slave_dev->name,
1527 				   slave_dev->name, bond_dev->name);
1528 		}
1529 	} else {
1530 		pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1531 	}
1532 
1533 	/*
1534 	 * Old ifenslave binaries are no longer supported.  These can
1535 	 * be identified with moderate accuracy by the state of the slave:
1536 	 * the current ifenslave will set the interface down prior to
1537 	 * enslaving it; the old ifenslave will not.
1538 	 */
1539 	if ((slave_dev->flags & IFF_UP)) {
1540 		pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1541 		       slave_dev->name);
1542 		res = -EPERM;
1543 		goto err_undo_flags;
1544 	}
1545 
1546 	/* set bonding device ether type by slave - bonding netdevices are
1547 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1548 	 * there is a need to override some of the type dependent attribs/funcs.
1549 	 *
1550 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1551 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1552 	 */
1553 	if (bond->slave_cnt == 0) {
1554 		if (bond_dev->type != slave_dev->type) {
1555 			pr_debug("%s: change device type from %d to %d\n",
1556 				 bond_dev->name,
1557 				 bond_dev->type, slave_dev->type);
1558 
1559 			res = netdev_bonding_change(bond_dev,
1560 						    NETDEV_PRE_TYPE_CHANGE);
1561 			res = notifier_to_errno(res);
1562 			if (res) {
1563 				pr_err("%s: refused to change device type\n",
1564 				       bond_dev->name);
1565 				res = -EBUSY;
1566 				goto err_undo_flags;
1567 			}
1568 
1569 			/* Flush unicast and multicast addresses */
1570 			dev_uc_flush(bond_dev);
1571 			dev_mc_flush(bond_dev);
1572 
1573 			if (slave_dev->type != ARPHRD_ETHER)
1574 				bond_setup_by_slave(bond_dev, slave_dev);
1575 			else {
1576 				ether_setup(bond_dev);
1577 				bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1578 			}
1579 
1580 			netdev_bonding_change(bond_dev,
1581 					      NETDEV_POST_TYPE_CHANGE);
1582 		}
1583 	} else if (bond_dev->type != slave_dev->type) {
1584 		pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1585 		       slave_dev->name,
1586 		       slave_dev->type, bond_dev->type);
1587 		res = -EINVAL;
1588 		goto err_undo_flags;
1589 	}
1590 
1591 	if (slave_ops->ndo_set_mac_address == NULL) {
1592 		if (bond->slave_cnt == 0) {
1593 			pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1594 				   bond_dev->name);
1595 			bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1596 		} else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1597 			pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1598 			       bond_dev->name);
1599 			res = -EOPNOTSUPP;
1600 			goto err_undo_flags;
1601 		}
1602 	}
1603 
1604 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1605 
1606 	/* If this is the first slave, then we need to set the master's hardware
1607 	 * address to be the same as the slave's. */
1608 	if (is_zero_ether_addr(bond->dev->dev_addr))
1609 		memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1610 		       slave_dev->addr_len);
1611 
1612 
1613 	new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1614 	if (!new_slave) {
1615 		res = -ENOMEM;
1616 		goto err_undo_flags;
1617 	}
1618 
1619 	/*
1620 	 * Set the new_slave's queue_id to be zero.  Queue ID mapping
1621 	 * is set via sysfs or module option if desired.
1622 	 */
1623 	new_slave->queue_id = 0;
1624 
1625 	/* Save slave's original mtu and then set it to match the bond */
1626 	new_slave->original_mtu = slave_dev->mtu;
1627 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1628 	if (res) {
1629 		pr_debug("Error %d calling dev_set_mtu\n", res);
1630 		goto err_free;
1631 	}
1632 
1633 	/*
1634 	 * Save slave's original ("permanent") mac address for modes
1635 	 * that need it, and for restoring it upon release, and then
1636 	 * set it to the master's address
1637 	 */
1638 	memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1639 
1640 	if (!bond->params.fail_over_mac) {
1641 		/*
1642 		 * Set slave to master's mac address.  The application already
1643 		 * set the master's mac address to that of the first slave
1644 		 */
1645 		memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1646 		addr.sa_family = slave_dev->type;
1647 		res = dev_set_mac_address(slave_dev, &addr);
1648 		if (res) {
1649 			pr_debug("Error %d calling set_mac_address\n", res);
1650 			goto err_restore_mtu;
1651 		}
1652 	}
1653 
1654 	res = netdev_set_bond_master(slave_dev, bond_dev);
1655 	if (res) {
1656 		pr_debug("Error %d calling netdev_set_bond_master\n", res);
1657 		goto err_restore_mac;
1658 	}
1659 
1660 	/* open the slave since the application closed it */
1661 	res = dev_open(slave_dev);
1662 	if (res) {
1663 		pr_debug("Opening slave %s failed\n", slave_dev->name);
1664 		goto err_unset_master;
1665 	}
1666 
1667 	new_slave->bond = bond;
1668 	new_slave->dev = slave_dev;
1669 	slave_dev->priv_flags |= IFF_BONDING;
1670 
1671 	if (bond_is_lb(bond)) {
1672 		/* bond_alb_init_slave() must be called before all other stages since
1673 		 * it might fail and we do not want to have to undo everything
1674 		 */
1675 		res = bond_alb_init_slave(bond, new_slave);
1676 		if (res)
1677 			goto err_close;
1678 	}
1679 
1680 	/* If the mode USES_PRIMARY, then the new slave gets the
1681 	 * master's promisc (and mc) settings only if it becomes the
1682 	 * curr_active_slave, and that is taken care of later when calling
1683 	 * bond_change_active()
1684 	 */
1685 	if (!USES_PRIMARY(bond->params.mode)) {
1686 		/* set promiscuity level to new slave */
1687 		if (bond_dev->flags & IFF_PROMISC) {
1688 			res = dev_set_promiscuity(slave_dev, 1);
1689 			if (res)
1690 				goto err_close;
1691 		}
1692 
1693 		/* set allmulti level to new slave */
1694 		if (bond_dev->flags & IFF_ALLMULTI) {
1695 			res = dev_set_allmulti(slave_dev, 1);
1696 			if (res)
1697 				goto err_close;
1698 		}
1699 
1700 		netif_addr_lock_bh(bond_dev);
1701 		/* upload master's mc_list to new slave */
1702 		netdev_for_each_mc_addr(ha, bond_dev)
1703 			dev_mc_add(slave_dev, ha->addr);
1704 		netif_addr_unlock_bh(bond_dev);
1705 	}
1706 
1707 	if (bond->params.mode == BOND_MODE_8023AD) {
1708 		/* add lacpdu mc addr to mc list */
1709 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1710 
1711 		dev_mc_add(slave_dev, lacpdu_multicast);
1712 	}
1713 
1714 	bond_add_vlans_on_slave(bond, slave_dev);
1715 
1716 	write_lock_bh(&bond->lock);
1717 
1718 	bond_attach_slave(bond, new_slave);
1719 
1720 	new_slave->delay = 0;
1721 	new_slave->link_failure_count = 0;
1722 
1723 	write_unlock_bh(&bond->lock);
1724 
1725 	bond_compute_features(bond);
1726 
1727 	read_lock(&bond->lock);
1728 
1729 	new_slave->last_arp_rx = jiffies -
1730 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
1731 
1732 	if (bond->params.miimon && !bond->params.use_carrier) {
1733 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1734 
1735 		if ((link_reporting == -1) && !bond->params.arp_interval) {
1736 			/*
1737 			 * miimon is set but a bonded network driver
1738 			 * does not support ETHTOOL/MII and
1739 			 * arp_interval is not set.  Note: if
1740 			 * use_carrier is enabled, we will never go
1741 			 * here (because netif_carrier is always
1742 			 * supported); thus, we don't need to change
1743 			 * the messages for netif_carrier.
1744 			 */
1745 			pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1746 			       bond_dev->name, slave_dev->name);
1747 		} else if (link_reporting == -1) {
1748 			/* unable get link status using mii/ethtool */
1749 			pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1750 				   bond_dev->name, slave_dev->name);
1751 		}
1752 	}
1753 
1754 	/* check for initial state */
1755 	if (bond->params.miimon) {
1756 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1757 			if (bond->params.updelay) {
1758 				new_slave->link = BOND_LINK_BACK;
1759 				new_slave->delay = bond->params.updelay;
1760 			} else {
1761 				new_slave->link = BOND_LINK_UP;
1762 			}
1763 		} else {
1764 			new_slave->link = BOND_LINK_DOWN;
1765 		}
1766 	} else if (bond->params.arp_interval) {
1767 		new_slave->link = (netif_carrier_ok(slave_dev) ?
1768 			BOND_LINK_UP : BOND_LINK_DOWN);
1769 	} else {
1770 		new_slave->link = BOND_LINK_UP;
1771 	}
1772 
1773 	if (new_slave->link != BOND_LINK_DOWN)
1774 		new_slave->jiffies = jiffies;
1775 	pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1776 		new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1777 			(new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1778 
1779 	bond_update_speed_duplex(new_slave);
1780 
1781 	if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1782 		/* if there is a primary slave, remember it */
1783 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1784 			bond->primary_slave = new_slave;
1785 			bond->force_primary = true;
1786 		}
1787 	}
1788 
1789 	write_lock_bh(&bond->curr_slave_lock);
1790 
1791 	switch (bond->params.mode) {
1792 	case BOND_MODE_ACTIVEBACKUP:
1793 		bond_set_slave_inactive_flags(new_slave);
1794 		bond_select_active_slave(bond);
1795 		break;
1796 	case BOND_MODE_8023AD:
1797 		/* in 802.3ad mode, the internal mechanism
1798 		 * will activate the slaves in the selected
1799 		 * aggregator
1800 		 */
1801 		bond_set_slave_inactive_flags(new_slave);
1802 		/* if this is the first slave */
1803 		if (bond->slave_cnt == 1) {
1804 			SLAVE_AD_INFO(new_slave).id = 1;
1805 			/* Initialize AD with the number of times that the AD timer is called in 1 second
1806 			 * can be called only after the mac address of the bond is set
1807 			 */
1808 			bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1809 		} else {
1810 			SLAVE_AD_INFO(new_slave).id =
1811 				SLAVE_AD_INFO(new_slave->prev).id + 1;
1812 		}
1813 
1814 		bond_3ad_bind_slave(new_slave);
1815 		break;
1816 	case BOND_MODE_TLB:
1817 	case BOND_MODE_ALB:
1818 		bond_set_active_slave(new_slave);
1819 		bond_set_slave_inactive_flags(new_slave);
1820 		bond_select_active_slave(bond);
1821 		break;
1822 	default:
1823 		pr_debug("This slave is always active in trunk mode\n");
1824 
1825 		/* always active in trunk mode */
1826 		bond_set_active_slave(new_slave);
1827 
1828 		/* In trunking mode there is little meaning to curr_active_slave
1829 		 * anyway (it holds no special properties of the bond device),
1830 		 * so we can change it without calling change_active_interface()
1831 		 */
1832 		if (!bond->curr_active_slave)
1833 			bond->curr_active_slave = new_slave;
1834 
1835 		break;
1836 	} /* switch(bond_mode) */
1837 
1838 	write_unlock_bh(&bond->curr_slave_lock);
1839 
1840 	bond_set_carrier(bond);
1841 
1842 #ifdef CONFIG_NET_POLL_CONTROLLER
1843 	slave_dev->npinfo = bond_netpoll_info(bond);
1844 	if (slave_dev->npinfo) {
1845 		if (slave_enable_netpoll(new_slave)) {
1846 			read_unlock(&bond->lock);
1847 			pr_info("Error, %s: master_dev is using netpoll, "
1848 				 "but new slave device does not support netpoll.\n",
1849 				 bond_dev->name);
1850 			res = -EBUSY;
1851 			goto err_detach;
1852 		}
1853 	}
1854 #endif
1855 
1856 	read_unlock(&bond->lock);
1857 
1858 	res = bond_create_slave_symlinks(bond_dev, slave_dev);
1859 	if (res)
1860 		goto err_detach;
1861 
1862 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1863 					 new_slave);
1864 	if (res) {
1865 		pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1866 		goto err_dest_symlinks;
1867 	}
1868 
1869 	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1870 		bond_dev->name, slave_dev->name,
1871 		bond_is_active_slave(new_slave) ? "n active" : " backup",
1872 		new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1873 
1874 	/* enslave is successful */
1875 	return 0;
1876 
1877 /* Undo stages on error */
1878 err_dest_symlinks:
1879 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
1880 
1881 err_detach:
1882 	write_lock_bh(&bond->lock);
1883 	bond_detach_slave(bond, new_slave);
1884 	write_unlock_bh(&bond->lock);
1885 
1886 err_close:
1887 	dev_close(slave_dev);
1888 
1889 err_unset_master:
1890 	netdev_set_bond_master(slave_dev, NULL);
1891 
1892 err_restore_mac:
1893 	if (!bond->params.fail_over_mac) {
1894 		/* XXX TODO - fom follow mode needs to change master's
1895 		 * MAC if this slave's MAC is in use by the bond, or at
1896 		 * least print a warning.
1897 		 */
1898 		memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1899 		addr.sa_family = slave_dev->type;
1900 		dev_set_mac_address(slave_dev, &addr);
1901 	}
1902 
1903 err_restore_mtu:
1904 	dev_set_mtu(slave_dev, new_slave->original_mtu);
1905 
1906 err_free:
1907 	kfree(new_slave);
1908 
1909 err_undo_flags:
1910 	bond_compute_features(bond);
1911 
1912 	return res;
1913 }
1914 
1915 /*
1916  * Try to release the slave device <slave> from the bond device <master>
1917  * It is legal to access curr_active_slave without a lock because all the function
1918  * is write-locked.
1919  *
1920  * The rules for slave state should be:
1921  *   for Active/Backup:
1922  *     Active stays on all backups go down
1923  *   for Bonded connections:
1924  *     The first up interface should be left on and all others downed.
1925  */
1926 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1927 {
1928 	struct bonding *bond = netdev_priv(bond_dev);
1929 	struct slave *slave, *oldcurrent;
1930 	struct sockaddr addr;
1931 	netdev_features_t old_features = bond_dev->features;
1932 
1933 	/* slave is not a slave or master is not master of this slave */
1934 	if (!(slave_dev->flags & IFF_SLAVE) ||
1935 	    (slave_dev->master != bond_dev)) {
1936 		pr_err("%s: Error: cannot release %s.\n",
1937 		       bond_dev->name, slave_dev->name);
1938 		return -EINVAL;
1939 	}
1940 
1941 	block_netpoll_tx();
1942 	netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1943 	write_lock_bh(&bond->lock);
1944 
1945 	slave = bond_get_slave_by_dev(bond, slave_dev);
1946 	if (!slave) {
1947 		/* not a slave of this bond */
1948 		pr_info("%s: %s not enslaved\n",
1949 			bond_dev->name, slave_dev->name);
1950 		write_unlock_bh(&bond->lock);
1951 		unblock_netpoll_tx();
1952 		return -EINVAL;
1953 	}
1954 
1955 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
1956 	 * for this slave anymore.
1957 	 */
1958 	netdev_rx_handler_unregister(slave_dev);
1959 	write_unlock_bh(&bond->lock);
1960 	synchronize_net();
1961 	write_lock_bh(&bond->lock);
1962 
1963 	if (!bond->params.fail_over_mac) {
1964 		if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1965 		    bond->slave_cnt > 1)
1966 			pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1967 				   bond_dev->name, slave_dev->name,
1968 				   slave->perm_hwaddr,
1969 				   bond_dev->name, slave_dev->name);
1970 	}
1971 
1972 	/* Inform AD package of unbinding of slave. */
1973 	if (bond->params.mode == BOND_MODE_8023AD) {
1974 		/* must be called before the slave is
1975 		 * detached from the list
1976 		 */
1977 		bond_3ad_unbind_slave(slave);
1978 	}
1979 
1980 	pr_info("%s: releasing %s interface %s\n",
1981 		bond_dev->name,
1982 		bond_is_active_slave(slave) ? "active" : "backup",
1983 		slave_dev->name);
1984 
1985 	oldcurrent = bond->curr_active_slave;
1986 
1987 	bond->current_arp_slave = NULL;
1988 
1989 	/* release the slave from its bond */
1990 	bond_detach_slave(bond, slave);
1991 
1992 	if (bond->primary_slave == slave)
1993 		bond->primary_slave = NULL;
1994 
1995 	if (oldcurrent == slave)
1996 		bond_change_active_slave(bond, NULL);
1997 
1998 	if (bond_is_lb(bond)) {
1999 		/* Must be called only after the slave has been
2000 		 * detached from the list and the curr_active_slave
2001 		 * has been cleared (if our_slave == old_current),
2002 		 * but before a new active slave is selected.
2003 		 */
2004 		write_unlock_bh(&bond->lock);
2005 		bond_alb_deinit_slave(bond, slave);
2006 		write_lock_bh(&bond->lock);
2007 	}
2008 
2009 	if (oldcurrent == slave) {
2010 		/*
2011 		 * Note that we hold RTNL over this sequence, so there
2012 		 * is no concern that another slave add/remove event
2013 		 * will interfere.
2014 		 */
2015 		write_unlock_bh(&bond->lock);
2016 		read_lock(&bond->lock);
2017 		write_lock_bh(&bond->curr_slave_lock);
2018 
2019 		bond_select_active_slave(bond);
2020 
2021 		write_unlock_bh(&bond->curr_slave_lock);
2022 		read_unlock(&bond->lock);
2023 		write_lock_bh(&bond->lock);
2024 	}
2025 
2026 	if (bond->slave_cnt == 0) {
2027 		bond_set_carrier(bond);
2028 
2029 		/* if the last slave was removed, zero the mac address
2030 		 * of the master so it will be set by the application
2031 		 * to the mac address of the first slave
2032 		 */
2033 		memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2034 
2035 		if (bond_vlan_used(bond)) {
2036 			pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2037 				   bond_dev->name, bond_dev->name);
2038 			pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2039 				   bond_dev->name);
2040 		}
2041 	}
2042 
2043 	write_unlock_bh(&bond->lock);
2044 	unblock_netpoll_tx();
2045 
2046 	if (bond->slave_cnt == 0)
2047 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2048 
2049 	bond_compute_features(bond);
2050 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2051 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2052 		pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2053 			bond_dev->name, slave_dev->name, bond_dev->name);
2054 
2055 	/* must do this from outside any spinlocks */
2056 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
2057 
2058 	bond_del_vlans_from_slave(bond, slave_dev);
2059 
2060 	/* If the mode USES_PRIMARY, then we should only remove its
2061 	 * promisc and mc settings if it was the curr_active_slave, but that was
2062 	 * already taken care of above when we detached the slave
2063 	 */
2064 	if (!USES_PRIMARY(bond->params.mode)) {
2065 		/* unset promiscuity level from slave */
2066 		if (bond_dev->flags & IFF_PROMISC)
2067 			dev_set_promiscuity(slave_dev, -1);
2068 
2069 		/* unset allmulti level from slave */
2070 		if (bond_dev->flags & IFF_ALLMULTI)
2071 			dev_set_allmulti(slave_dev, -1);
2072 
2073 		/* flush master's mc_list from slave */
2074 		netif_addr_lock_bh(bond_dev);
2075 		bond_mc_list_flush(bond_dev, slave_dev);
2076 		netif_addr_unlock_bh(bond_dev);
2077 	}
2078 
2079 	netdev_set_bond_master(slave_dev, NULL);
2080 
2081 	slave_disable_netpoll(slave);
2082 
2083 	/* close slave before restoring its mac address */
2084 	dev_close(slave_dev);
2085 
2086 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2087 		/* restore original ("permanent") mac address */
2088 		memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2089 		addr.sa_family = slave_dev->type;
2090 		dev_set_mac_address(slave_dev, &addr);
2091 	}
2092 
2093 	dev_set_mtu(slave_dev, slave->original_mtu);
2094 
2095 	slave_dev->priv_flags &= ~IFF_BONDING;
2096 
2097 	kfree(slave);
2098 
2099 	return 0;  /* deletion OK */
2100 }
2101 
2102 /*
2103 * First release a slave and then destroy the bond if no more slaves are left.
2104 * Must be under rtnl_lock when this function is called.
2105 */
2106 static int  bond_release_and_destroy(struct net_device *bond_dev,
2107 				     struct net_device *slave_dev)
2108 {
2109 	struct bonding *bond = netdev_priv(bond_dev);
2110 	int ret;
2111 
2112 	ret = bond_release(bond_dev, slave_dev);
2113 	if ((ret == 0) && (bond->slave_cnt == 0)) {
2114 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2115 		pr_info("%s: destroying bond %s.\n",
2116 			bond_dev->name, bond_dev->name);
2117 		unregister_netdevice(bond_dev);
2118 	}
2119 	return ret;
2120 }
2121 
2122 /*
2123  * This function releases all slaves.
2124  */
2125 static int bond_release_all(struct net_device *bond_dev)
2126 {
2127 	struct bonding *bond = netdev_priv(bond_dev);
2128 	struct slave *slave;
2129 	struct net_device *slave_dev;
2130 	struct sockaddr addr;
2131 
2132 	write_lock_bh(&bond->lock);
2133 
2134 	netif_carrier_off(bond_dev);
2135 
2136 	if (bond->slave_cnt == 0)
2137 		goto out;
2138 
2139 	bond->current_arp_slave = NULL;
2140 	bond->primary_slave = NULL;
2141 	bond_change_active_slave(bond, NULL);
2142 
2143 	while ((slave = bond->first_slave) != NULL) {
2144 		/* Inform AD package of unbinding of slave
2145 		 * before slave is detached from the list.
2146 		 */
2147 		if (bond->params.mode == BOND_MODE_8023AD)
2148 			bond_3ad_unbind_slave(slave);
2149 
2150 		slave_dev = slave->dev;
2151 		bond_detach_slave(bond, slave);
2152 
2153 		/* now that the slave is detached, unlock and perform
2154 		 * all the undo steps that should not be called from
2155 		 * within a lock.
2156 		 */
2157 		write_unlock_bh(&bond->lock);
2158 
2159 		/* unregister rx_handler early so bond_handle_frame wouldn't
2160 		 * be called for this slave anymore.
2161 		 */
2162 		netdev_rx_handler_unregister(slave_dev);
2163 		synchronize_net();
2164 
2165 		if (bond_is_lb(bond)) {
2166 			/* must be called only after the slave
2167 			 * has been detached from the list
2168 			 */
2169 			bond_alb_deinit_slave(bond, slave);
2170 		}
2171 
2172 		bond_destroy_slave_symlinks(bond_dev, slave_dev);
2173 		bond_del_vlans_from_slave(bond, slave_dev);
2174 
2175 		/* If the mode USES_PRIMARY, then we should only remove its
2176 		 * promisc and mc settings if it was the curr_active_slave, but that was
2177 		 * already taken care of above when we detached the slave
2178 		 */
2179 		if (!USES_PRIMARY(bond->params.mode)) {
2180 			/* unset promiscuity level from slave */
2181 			if (bond_dev->flags & IFF_PROMISC)
2182 				dev_set_promiscuity(slave_dev, -1);
2183 
2184 			/* unset allmulti level from slave */
2185 			if (bond_dev->flags & IFF_ALLMULTI)
2186 				dev_set_allmulti(slave_dev, -1);
2187 
2188 			/* flush master's mc_list from slave */
2189 			netif_addr_lock_bh(bond_dev);
2190 			bond_mc_list_flush(bond_dev, slave_dev);
2191 			netif_addr_unlock_bh(bond_dev);
2192 		}
2193 
2194 		netdev_set_bond_master(slave_dev, NULL);
2195 
2196 		slave_disable_netpoll(slave);
2197 
2198 		/* close slave before restoring its mac address */
2199 		dev_close(slave_dev);
2200 
2201 		if (!bond->params.fail_over_mac) {
2202 			/* restore original ("permanent") mac address*/
2203 			memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2204 			addr.sa_family = slave_dev->type;
2205 			dev_set_mac_address(slave_dev, &addr);
2206 		}
2207 
2208 		kfree(slave);
2209 
2210 		/* re-acquire the lock before getting the next slave */
2211 		write_lock_bh(&bond->lock);
2212 	}
2213 
2214 	/* zero the mac address of the master so it will be
2215 	 * set by the application to the mac address of the
2216 	 * first slave
2217 	 */
2218 	memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2219 
2220 	if (bond_vlan_used(bond)) {
2221 		pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2222 			   bond_dev->name, bond_dev->name);
2223 		pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2224 			   bond_dev->name);
2225 	}
2226 
2227 	pr_info("%s: released all slaves\n", bond_dev->name);
2228 
2229 out:
2230 	write_unlock_bh(&bond->lock);
2231 
2232 	bond_compute_features(bond);
2233 
2234 	return 0;
2235 }
2236 
2237 /*
2238  * This function changes the active slave to slave <slave_dev>.
2239  * It returns -EINVAL in the following cases.
2240  *  - <slave_dev> is not found in the list.
2241  *  - There is not active slave now.
2242  *  - <slave_dev> is already active.
2243  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2244  *  - <slave_dev> is not running.
2245  * In these cases, this function does nothing.
2246  * In the other cases, current_slave pointer is changed and 0 is returned.
2247  */
2248 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2249 {
2250 	struct bonding *bond = netdev_priv(bond_dev);
2251 	struct slave *old_active = NULL;
2252 	struct slave *new_active = NULL;
2253 	int res = 0;
2254 
2255 	if (!USES_PRIMARY(bond->params.mode))
2256 		return -EINVAL;
2257 
2258 	/* Verify that master_dev is indeed the master of slave_dev */
2259 	if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2260 		return -EINVAL;
2261 
2262 	read_lock(&bond->lock);
2263 
2264 	read_lock(&bond->curr_slave_lock);
2265 	old_active = bond->curr_active_slave;
2266 	read_unlock(&bond->curr_slave_lock);
2267 
2268 	new_active = bond_get_slave_by_dev(bond, slave_dev);
2269 
2270 	/*
2271 	 * Changing to the current active: do nothing; return success.
2272 	 */
2273 	if (new_active && (new_active == old_active)) {
2274 		read_unlock(&bond->lock);
2275 		return 0;
2276 	}
2277 
2278 	if ((new_active) &&
2279 	    (old_active) &&
2280 	    (new_active->link == BOND_LINK_UP) &&
2281 	    IS_UP(new_active->dev)) {
2282 		block_netpoll_tx();
2283 		write_lock_bh(&bond->curr_slave_lock);
2284 		bond_change_active_slave(bond, new_active);
2285 		write_unlock_bh(&bond->curr_slave_lock);
2286 		unblock_netpoll_tx();
2287 	} else
2288 		res = -EINVAL;
2289 
2290 	read_unlock(&bond->lock);
2291 
2292 	return res;
2293 }
2294 
2295 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2296 {
2297 	struct bonding *bond = netdev_priv(bond_dev);
2298 
2299 	info->bond_mode = bond->params.mode;
2300 	info->miimon = bond->params.miimon;
2301 
2302 	read_lock(&bond->lock);
2303 	info->num_slaves = bond->slave_cnt;
2304 	read_unlock(&bond->lock);
2305 
2306 	return 0;
2307 }
2308 
2309 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2310 {
2311 	struct bonding *bond = netdev_priv(bond_dev);
2312 	struct slave *slave;
2313 	int i, res = -ENODEV;
2314 
2315 	read_lock(&bond->lock);
2316 
2317 	bond_for_each_slave(bond, slave, i) {
2318 		if (i == (int)info->slave_id) {
2319 			res = 0;
2320 			strcpy(info->slave_name, slave->dev->name);
2321 			info->link = slave->link;
2322 			info->state = bond_slave_state(slave);
2323 			info->link_failure_count = slave->link_failure_count;
2324 			break;
2325 		}
2326 	}
2327 
2328 	read_unlock(&bond->lock);
2329 
2330 	return res;
2331 }
2332 
2333 /*-------------------------------- Monitoring -------------------------------*/
2334 
2335 
2336 static int bond_miimon_inspect(struct bonding *bond)
2337 {
2338 	struct slave *slave;
2339 	int i, link_state, commit = 0;
2340 	bool ignore_updelay;
2341 
2342 	ignore_updelay = !bond->curr_active_slave ? true : false;
2343 
2344 	bond_for_each_slave(bond, slave, i) {
2345 		slave->new_link = BOND_LINK_NOCHANGE;
2346 
2347 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2348 
2349 		switch (slave->link) {
2350 		case BOND_LINK_UP:
2351 			if (link_state)
2352 				continue;
2353 
2354 			slave->link = BOND_LINK_FAIL;
2355 			slave->delay = bond->params.downdelay;
2356 			if (slave->delay) {
2357 				pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2358 					bond->dev->name,
2359 					(bond->params.mode ==
2360 					 BOND_MODE_ACTIVEBACKUP) ?
2361 					(bond_is_active_slave(slave) ?
2362 					 "active " : "backup ") : "",
2363 					slave->dev->name,
2364 					bond->params.downdelay * bond->params.miimon);
2365 			}
2366 			/*FALLTHRU*/
2367 		case BOND_LINK_FAIL:
2368 			if (link_state) {
2369 				/*
2370 				 * recovered before downdelay expired
2371 				 */
2372 				slave->link = BOND_LINK_UP;
2373 				slave->jiffies = jiffies;
2374 				pr_info("%s: link status up again after %d ms for interface %s.\n",
2375 					bond->dev->name,
2376 					(bond->params.downdelay - slave->delay) *
2377 					bond->params.miimon,
2378 					slave->dev->name);
2379 				continue;
2380 			}
2381 
2382 			if (slave->delay <= 0) {
2383 				slave->new_link = BOND_LINK_DOWN;
2384 				commit++;
2385 				continue;
2386 			}
2387 
2388 			slave->delay--;
2389 			break;
2390 
2391 		case BOND_LINK_DOWN:
2392 			if (!link_state)
2393 				continue;
2394 
2395 			slave->link = BOND_LINK_BACK;
2396 			slave->delay = bond->params.updelay;
2397 
2398 			if (slave->delay) {
2399 				pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2400 					bond->dev->name, slave->dev->name,
2401 					ignore_updelay ? 0 :
2402 					bond->params.updelay *
2403 					bond->params.miimon);
2404 			}
2405 			/*FALLTHRU*/
2406 		case BOND_LINK_BACK:
2407 			if (!link_state) {
2408 				slave->link = BOND_LINK_DOWN;
2409 				pr_info("%s: link status down again after %d ms for interface %s.\n",
2410 					bond->dev->name,
2411 					(bond->params.updelay - slave->delay) *
2412 					bond->params.miimon,
2413 					slave->dev->name);
2414 
2415 				continue;
2416 			}
2417 
2418 			if (ignore_updelay)
2419 				slave->delay = 0;
2420 
2421 			if (slave->delay <= 0) {
2422 				slave->new_link = BOND_LINK_UP;
2423 				commit++;
2424 				ignore_updelay = false;
2425 				continue;
2426 			}
2427 
2428 			slave->delay--;
2429 			break;
2430 		}
2431 	}
2432 
2433 	return commit;
2434 }
2435 
2436 static void bond_miimon_commit(struct bonding *bond)
2437 {
2438 	struct slave *slave;
2439 	int i;
2440 
2441 	bond_for_each_slave(bond, slave, i) {
2442 		switch (slave->new_link) {
2443 		case BOND_LINK_NOCHANGE:
2444 			continue;
2445 
2446 		case BOND_LINK_UP:
2447 			slave->link = BOND_LINK_UP;
2448 			slave->jiffies = jiffies;
2449 
2450 			if (bond->params.mode == BOND_MODE_8023AD) {
2451 				/* prevent it from being the active one */
2452 				bond_set_backup_slave(slave);
2453 			} else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2454 				/* make it immediately active */
2455 				bond_set_active_slave(slave);
2456 			} else if (slave != bond->primary_slave) {
2457 				/* prevent it from being the active one */
2458 				bond_set_backup_slave(slave);
2459 			}
2460 
2461 			bond_update_speed_duplex(slave);
2462 
2463 			pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2464 				bond->dev->name, slave->dev->name,
2465 				slave->speed, slave->duplex ? "full" : "half");
2466 
2467 			/* notify ad that the link status has changed */
2468 			if (bond->params.mode == BOND_MODE_8023AD)
2469 				bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2470 
2471 			if (bond_is_lb(bond))
2472 				bond_alb_handle_link_change(bond, slave,
2473 							    BOND_LINK_UP);
2474 
2475 			if (!bond->curr_active_slave ||
2476 			    (slave == bond->primary_slave))
2477 				goto do_failover;
2478 
2479 			continue;
2480 
2481 		case BOND_LINK_DOWN:
2482 			if (slave->link_failure_count < UINT_MAX)
2483 				slave->link_failure_count++;
2484 
2485 			slave->link = BOND_LINK_DOWN;
2486 
2487 			if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2488 			    bond->params.mode == BOND_MODE_8023AD)
2489 				bond_set_slave_inactive_flags(slave);
2490 
2491 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
2492 				bond->dev->name, slave->dev->name);
2493 
2494 			if (bond->params.mode == BOND_MODE_8023AD)
2495 				bond_3ad_handle_link_change(slave,
2496 							    BOND_LINK_DOWN);
2497 
2498 			if (bond_is_lb(bond))
2499 				bond_alb_handle_link_change(bond, slave,
2500 							    BOND_LINK_DOWN);
2501 
2502 			if (slave == bond->curr_active_slave)
2503 				goto do_failover;
2504 
2505 			continue;
2506 
2507 		default:
2508 			pr_err("%s: invalid new link %d on slave %s\n",
2509 			       bond->dev->name, slave->new_link,
2510 			       slave->dev->name);
2511 			slave->new_link = BOND_LINK_NOCHANGE;
2512 
2513 			continue;
2514 		}
2515 
2516 do_failover:
2517 		ASSERT_RTNL();
2518 		block_netpoll_tx();
2519 		write_lock_bh(&bond->curr_slave_lock);
2520 		bond_select_active_slave(bond);
2521 		write_unlock_bh(&bond->curr_slave_lock);
2522 		unblock_netpoll_tx();
2523 	}
2524 
2525 	bond_set_carrier(bond);
2526 }
2527 
2528 /*
2529  * bond_mii_monitor
2530  *
2531  * Really a wrapper that splits the mii monitor into two phases: an
2532  * inspection, then (if inspection indicates something needs to be done)
2533  * an acquisition of appropriate locks followed by a commit phase to
2534  * implement whatever link state changes are indicated.
2535  */
2536 void bond_mii_monitor(struct work_struct *work)
2537 {
2538 	struct bonding *bond = container_of(work, struct bonding,
2539 					    mii_work.work);
2540 	bool should_notify_peers = false;
2541 	unsigned long delay;
2542 
2543 	read_lock(&bond->lock);
2544 
2545 	delay = msecs_to_jiffies(bond->params.miimon);
2546 
2547 	if (bond->slave_cnt == 0)
2548 		goto re_arm;
2549 
2550 	should_notify_peers = bond_should_notify_peers(bond);
2551 
2552 	if (bond_miimon_inspect(bond)) {
2553 		read_unlock(&bond->lock);
2554 
2555 		/* Race avoidance with bond_close cancel of workqueue */
2556 		if (!rtnl_trylock()) {
2557 			read_lock(&bond->lock);
2558 			delay = 1;
2559 			should_notify_peers = false;
2560 			goto re_arm;
2561 		}
2562 
2563 		read_lock(&bond->lock);
2564 
2565 		bond_miimon_commit(bond);
2566 
2567 		read_unlock(&bond->lock);
2568 		rtnl_unlock();	/* might sleep, hold no other locks */
2569 		read_lock(&bond->lock);
2570 	}
2571 
2572 re_arm:
2573 	if (bond->params.miimon)
2574 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2575 
2576 	read_unlock(&bond->lock);
2577 
2578 	if (should_notify_peers) {
2579 		if (!rtnl_trylock()) {
2580 			read_lock(&bond->lock);
2581 			bond->send_peer_notif++;
2582 			read_unlock(&bond->lock);
2583 			return;
2584 		}
2585 		netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2586 		rtnl_unlock();
2587 	}
2588 }
2589 
2590 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2591 {
2592 	struct vlan_entry *vlan;
2593 	struct net_device *vlan_dev;
2594 
2595 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2596 		return 1;
2597 
2598 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2599 		rcu_read_lock();
2600 		vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2601 		rcu_read_unlock();
2602 		if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2603 			return 1;
2604 	}
2605 
2606 	return 0;
2607 }
2608 
2609 /*
2610  * We go to the (large) trouble of VLAN tagging ARP frames because
2611  * switches in VLAN mode (especially if ports are configured as
2612  * "native" to a VLAN) might not pass non-tagged frames.
2613  */
2614 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2615 {
2616 	struct sk_buff *skb;
2617 
2618 	pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2619 		 slave_dev->name, dest_ip, src_ip, vlan_id);
2620 
2621 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2622 			 NULL, slave_dev->dev_addr, NULL);
2623 
2624 	if (!skb) {
2625 		pr_err("ARP packet allocation failed\n");
2626 		return;
2627 	}
2628 	if (vlan_id) {
2629 		skb = vlan_put_tag(skb, vlan_id);
2630 		if (!skb) {
2631 			pr_err("failed to insert VLAN tag\n");
2632 			return;
2633 		}
2634 	}
2635 	arp_xmit(skb);
2636 }
2637 
2638 
2639 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2640 {
2641 	int i, vlan_id;
2642 	__be32 *targets = bond->params.arp_targets;
2643 	struct vlan_entry *vlan;
2644 	struct net_device *vlan_dev = NULL;
2645 	struct rtable *rt;
2646 
2647 	for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2648 		__be32 addr;
2649 		if (!targets[i])
2650 			break;
2651 		pr_debug("basa: target %x\n", targets[i]);
2652 		if (!bond_vlan_used(bond)) {
2653 			pr_debug("basa: empty vlan: arp_send\n");
2654 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2655 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2656 				      addr, 0);
2657 			continue;
2658 		}
2659 
2660 		/*
2661 		 * If VLANs are configured, we do a route lookup to
2662 		 * determine which VLAN interface would be used, so we
2663 		 * can tag the ARP with the proper VLAN tag.
2664 		 */
2665 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2666 				     RTO_ONLINK, 0);
2667 		if (IS_ERR(rt)) {
2668 			if (net_ratelimit()) {
2669 				pr_warning("%s: no route to arp_ip_target %pI4\n",
2670 					   bond->dev->name, &targets[i]);
2671 			}
2672 			continue;
2673 		}
2674 
2675 		/*
2676 		 * This target is not on a VLAN
2677 		 */
2678 		if (rt->dst.dev == bond->dev) {
2679 			ip_rt_put(rt);
2680 			pr_debug("basa: rtdev == bond->dev: arp_send\n");
2681 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2682 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2683 				      addr, 0);
2684 			continue;
2685 		}
2686 
2687 		vlan_id = 0;
2688 		list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2689 			rcu_read_lock();
2690 			vlan_dev = __vlan_find_dev_deep(bond->dev,
2691 							vlan->vlan_id);
2692 			rcu_read_unlock();
2693 			if (vlan_dev == rt->dst.dev) {
2694 				vlan_id = vlan->vlan_id;
2695 				pr_debug("basa: vlan match on %s %d\n",
2696 				       vlan_dev->name, vlan_id);
2697 				break;
2698 			}
2699 		}
2700 
2701 		if (vlan_id && vlan_dev) {
2702 			ip_rt_put(rt);
2703 			addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2704 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2705 				      addr, vlan_id);
2706 			continue;
2707 		}
2708 
2709 		if (net_ratelimit()) {
2710 			pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2711 				   bond->dev->name, &targets[i],
2712 				   rt->dst.dev ? rt->dst.dev->name : "NULL");
2713 		}
2714 		ip_rt_put(rt);
2715 	}
2716 }
2717 
2718 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2719 {
2720 	int i;
2721 	__be32 *targets = bond->params.arp_targets;
2722 
2723 	for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2724 		pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2725 			 &sip, &tip, i, &targets[i],
2726 			 bond_has_this_ip(bond, tip));
2727 		if (sip == targets[i]) {
2728 			if (bond_has_this_ip(bond, tip))
2729 				slave->last_arp_rx = jiffies;
2730 			return;
2731 		}
2732 	}
2733 }
2734 
2735 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2736 			 struct slave *slave)
2737 {
2738 	struct arphdr *arp;
2739 	unsigned char *arp_ptr;
2740 	__be32 sip, tip;
2741 
2742 	if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2743 		return;
2744 
2745 	read_lock(&bond->lock);
2746 
2747 	pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2748 		 bond->dev->name, skb->dev->name);
2749 
2750 	if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2751 		goto out_unlock;
2752 
2753 	arp = arp_hdr(skb);
2754 	if (arp->ar_hln != bond->dev->addr_len ||
2755 	    skb->pkt_type == PACKET_OTHERHOST ||
2756 	    skb->pkt_type == PACKET_LOOPBACK ||
2757 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
2758 	    arp->ar_pro != htons(ETH_P_IP) ||
2759 	    arp->ar_pln != 4)
2760 		goto out_unlock;
2761 
2762 	arp_ptr = (unsigned char *)(arp + 1);
2763 	arp_ptr += bond->dev->addr_len;
2764 	memcpy(&sip, arp_ptr, 4);
2765 	arp_ptr += 4 + bond->dev->addr_len;
2766 	memcpy(&tip, arp_ptr, 4);
2767 
2768 	pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2769 		 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2770 		 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2771 		 &sip, &tip);
2772 
2773 	/*
2774 	 * Backup slaves won't see the ARP reply, but do come through
2775 	 * here for each ARP probe (so we swap the sip/tip to validate
2776 	 * the probe).  In a "redundant switch, common router" type of
2777 	 * configuration, the ARP probe will (hopefully) travel from
2778 	 * the active, through one switch, the router, then the other
2779 	 * switch before reaching the backup.
2780 	 */
2781 	if (bond_is_active_slave(slave))
2782 		bond_validate_arp(bond, slave, sip, tip);
2783 	else
2784 		bond_validate_arp(bond, slave, tip, sip);
2785 
2786 out_unlock:
2787 	read_unlock(&bond->lock);
2788 }
2789 
2790 /*
2791  * this function is called regularly to monitor each slave's link
2792  * ensuring that traffic is being sent and received when arp monitoring
2793  * is used in load-balancing mode. if the adapter has been dormant, then an
2794  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2795  * arp monitoring in active backup mode.
2796  */
2797 void bond_loadbalance_arp_mon(struct work_struct *work)
2798 {
2799 	struct bonding *bond = container_of(work, struct bonding,
2800 					    arp_work.work);
2801 	struct slave *slave, *oldcurrent;
2802 	int do_failover = 0;
2803 	int delta_in_ticks;
2804 	int i;
2805 
2806 	read_lock(&bond->lock);
2807 
2808 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2809 
2810 	if (bond->slave_cnt == 0)
2811 		goto re_arm;
2812 
2813 	read_lock(&bond->curr_slave_lock);
2814 	oldcurrent = bond->curr_active_slave;
2815 	read_unlock(&bond->curr_slave_lock);
2816 
2817 	/* see if any of the previous devices are up now (i.e. they have
2818 	 * xmt and rcv traffic). the curr_active_slave does not come into
2819 	 * the picture unless it is null. also, slave->jiffies is not needed
2820 	 * here because we send an arp on each slave and give a slave as
2821 	 * long as it needs to get the tx/rx within the delta.
2822 	 * TODO: what about up/down delay in arp mode? it wasn't here before
2823 	 *       so it can wait
2824 	 */
2825 	bond_for_each_slave(bond, slave, i) {
2826 		unsigned long trans_start = dev_trans_start(slave->dev);
2827 
2828 		if (slave->link != BOND_LINK_UP) {
2829 			if (time_in_range(jiffies,
2830 				trans_start - delta_in_ticks,
2831 				trans_start + delta_in_ticks) &&
2832 			    time_in_range(jiffies,
2833 				slave->dev->last_rx - delta_in_ticks,
2834 				slave->dev->last_rx + delta_in_ticks)) {
2835 
2836 				slave->link  = BOND_LINK_UP;
2837 				bond_set_active_slave(slave);
2838 
2839 				/* primary_slave has no meaning in round-robin
2840 				 * mode. the window of a slave being up and
2841 				 * curr_active_slave being null after enslaving
2842 				 * is closed.
2843 				 */
2844 				if (!oldcurrent) {
2845 					pr_info("%s: link status definitely up for interface %s, ",
2846 						bond->dev->name,
2847 						slave->dev->name);
2848 					do_failover = 1;
2849 				} else {
2850 					pr_info("%s: interface %s is now up\n",
2851 						bond->dev->name,
2852 						slave->dev->name);
2853 				}
2854 			}
2855 		} else {
2856 			/* slave->link == BOND_LINK_UP */
2857 
2858 			/* not all switches will respond to an arp request
2859 			 * when the source ip is 0, so don't take the link down
2860 			 * if we don't know our ip yet
2861 			 */
2862 			if (!time_in_range(jiffies,
2863 				trans_start - delta_in_ticks,
2864 				trans_start + 2 * delta_in_ticks) ||
2865 			    !time_in_range(jiffies,
2866 				slave->dev->last_rx - delta_in_ticks,
2867 				slave->dev->last_rx + 2 * delta_in_ticks)) {
2868 
2869 				slave->link  = BOND_LINK_DOWN;
2870 				bond_set_backup_slave(slave);
2871 
2872 				if (slave->link_failure_count < UINT_MAX)
2873 					slave->link_failure_count++;
2874 
2875 				pr_info("%s: interface %s is now down.\n",
2876 					bond->dev->name,
2877 					slave->dev->name);
2878 
2879 				if (slave == oldcurrent)
2880 					do_failover = 1;
2881 			}
2882 		}
2883 
2884 		/* note: if switch is in round-robin mode, all links
2885 		 * must tx arp to ensure all links rx an arp - otherwise
2886 		 * links may oscillate or not come up at all; if switch is
2887 		 * in something like xor mode, there is nothing we can
2888 		 * do - all replies will be rx'ed on same link causing slaves
2889 		 * to be unstable during low/no traffic periods
2890 		 */
2891 		if (IS_UP(slave->dev))
2892 			bond_arp_send_all(bond, slave);
2893 	}
2894 
2895 	if (do_failover) {
2896 		block_netpoll_tx();
2897 		write_lock_bh(&bond->curr_slave_lock);
2898 
2899 		bond_select_active_slave(bond);
2900 
2901 		write_unlock_bh(&bond->curr_slave_lock);
2902 		unblock_netpoll_tx();
2903 	}
2904 
2905 re_arm:
2906 	if (bond->params.arp_interval)
2907 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2908 
2909 	read_unlock(&bond->lock);
2910 }
2911 
2912 /*
2913  * Called to inspect slaves for active-backup mode ARP monitor link state
2914  * changes.  Sets new_link in slaves to specify what action should take
2915  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2916  * to link states must be committed.
2917  *
2918  * Called with bond->lock held for read.
2919  */
2920 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2921 {
2922 	struct slave *slave;
2923 	int i, commit = 0;
2924 	unsigned long trans_start;
2925 
2926 	bond_for_each_slave(bond, slave, i) {
2927 		slave->new_link = BOND_LINK_NOCHANGE;
2928 
2929 		if (slave->link != BOND_LINK_UP) {
2930 			if (time_in_range(jiffies,
2931 				slave_last_rx(bond, slave) - delta_in_ticks,
2932 				slave_last_rx(bond, slave) + delta_in_ticks)) {
2933 
2934 				slave->new_link = BOND_LINK_UP;
2935 				commit++;
2936 			}
2937 
2938 			continue;
2939 		}
2940 
2941 		/*
2942 		 * Give slaves 2*delta after being enslaved or made
2943 		 * active.  This avoids bouncing, as the last receive
2944 		 * times need a full ARP monitor cycle to be updated.
2945 		 */
2946 		if (time_in_range(jiffies,
2947 				  slave->jiffies - delta_in_ticks,
2948 				  slave->jiffies + 2 * delta_in_ticks))
2949 			continue;
2950 
2951 		/*
2952 		 * Backup slave is down if:
2953 		 * - No current_arp_slave AND
2954 		 * - more than 3*delta since last receive AND
2955 		 * - the bond has an IP address
2956 		 *
2957 		 * Note: a non-null current_arp_slave indicates
2958 		 * the curr_active_slave went down and we are
2959 		 * searching for a new one; under this condition
2960 		 * we only take the curr_active_slave down - this
2961 		 * gives each slave a chance to tx/rx traffic
2962 		 * before being taken out
2963 		 */
2964 		if (!bond_is_active_slave(slave) &&
2965 		    !bond->current_arp_slave &&
2966 		    !time_in_range(jiffies,
2967 			slave_last_rx(bond, slave) - delta_in_ticks,
2968 			slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2969 
2970 			slave->new_link = BOND_LINK_DOWN;
2971 			commit++;
2972 		}
2973 
2974 		/*
2975 		 * Active slave is down if:
2976 		 * - more than 2*delta since transmitting OR
2977 		 * - (more than 2*delta since receive AND
2978 		 *    the bond has an IP address)
2979 		 */
2980 		trans_start = dev_trans_start(slave->dev);
2981 		if (bond_is_active_slave(slave) &&
2982 		    (!time_in_range(jiffies,
2983 			trans_start - delta_in_ticks,
2984 			trans_start + 2 * delta_in_ticks) ||
2985 		     !time_in_range(jiffies,
2986 			slave_last_rx(bond, slave) - delta_in_ticks,
2987 			slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2988 
2989 			slave->new_link = BOND_LINK_DOWN;
2990 			commit++;
2991 		}
2992 	}
2993 
2994 	return commit;
2995 }
2996 
2997 /*
2998  * Called to commit link state changes noted by inspection step of
2999  * active-backup mode ARP monitor.
3000  *
3001  * Called with RTNL and bond->lock for read.
3002  */
3003 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3004 {
3005 	struct slave *slave;
3006 	int i;
3007 	unsigned long trans_start;
3008 
3009 	bond_for_each_slave(bond, slave, i) {
3010 		switch (slave->new_link) {
3011 		case BOND_LINK_NOCHANGE:
3012 			continue;
3013 
3014 		case BOND_LINK_UP:
3015 			trans_start = dev_trans_start(slave->dev);
3016 			if ((!bond->curr_active_slave &&
3017 			     time_in_range(jiffies,
3018 					   trans_start - delta_in_ticks,
3019 					   trans_start + delta_in_ticks)) ||
3020 			    bond->curr_active_slave != slave) {
3021 				slave->link = BOND_LINK_UP;
3022 				if (bond->current_arp_slave) {
3023 					bond_set_slave_inactive_flags(
3024 						bond->current_arp_slave);
3025 					bond->current_arp_slave = NULL;
3026 				}
3027 
3028 				pr_info("%s: link status definitely up for interface %s.\n",
3029 					bond->dev->name, slave->dev->name);
3030 
3031 				if (!bond->curr_active_slave ||
3032 				    (slave == bond->primary_slave))
3033 					goto do_failover;
3034 
3035 			}
3036 
3037 			continue;
3038 
3039 		case BOND_LINK_DOWN:
3040 			if (slave->link_failure_count < UINT_MAX)
3041 				slave->link_failure_count++;
3042 
3043 			slave->link = BOND_LINK_DOWN;
3044 			bond_set_slave_inactive_flags(slave);
3045 
3046 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
3047 				bond->dev->name, slave->dev->name);
3048 
3049 			if (slave == bond->curr_active_slave) {
3050 				bond->current_arp_slave = NULL;
3051 				goto do_failover;
3052 			}
3053 
3054 			continue;
3055 
3056 		default:
3057 			pr_err("%s: impossible: new_link %d on slave %s\n",
3058 			       bond->dev->name, slave->new_link,
3059 			       slave->dev->name);
3060 			continue;
3061 		}
3062 
3063 do_failover:
3064 		ASSERT_RTNL();
3065 		block_netpoll_tx();
3066 		write_lock_bh(&bond->curr_slave_lock);
3067 		bond_select_active_slave(bond);
3068 		write_unlock_bh(&bond->curr_slave_lock);
3069 		unblock_netpoll_tx();
3070 	}
3071 
3072 	bond_set_carrier(bond);
3073 }
3074 
3075 /*
3076  * Send ARP probes for active-backup mode ARP monitor.
3077  *
3078  * Called with bond->lock held for read.
3079  */
3080 static void bond_ab_arp_probe(struct bonding *bond)
3081 {
3082 	struct slave *slave;
3083 	int i;
3084 
3085 	read_lock(&bond->curr_slave_lock);
3086 
3087 	if (bond->current_arp_slave && bond->curr_active_slave)
3088 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
3089 			bond->current_arp_slave->dev->name,
3090 			bond->curr_active_slave->dev->name);
3091 
3092 	if (bond->curr_active_slave) {
3093 		bond_arp_send_all(bond, bond->curr_active_slave);
3094 		read_unlock(&bond->curr_slave_lock);
3095 		return;
3096 	}
3097 
3098 	read_unlock(&bond->curr_slave_lock);
3099 
3100 	/* if we don't have a curr_active_slave, search for the next available
3101 	 * backup slave from the current_arp_slave and make it the candidate
3102 	 * for becoming the curr_active_slave
3103 	 */
3104 
3105 	if (!bond->current_arp_slave) {
3106 		bond->current_arp_slave = bond->first_slave;
3107 		if (!bond->current_arp_slave)
3108 			return;
3109 	}
3110 
3111 	bond_set_slave_inactive_flags(bond->current_arp_slave);
3112 
3113 	/* search for next candidate */
3114 	bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3115 		if (IS_UP(slave->dev)) {
3116 			slave->link = BOND_LINK_BACK;
3117 			bond_set_slave_active_flags(slave);
3118 			bond_arp_send_all(bond, slave);
3119 			slave->jiffies = jiffies;
3120 			bond->current_arp_slave = slave;
3121 			break;
3122 		}
3123 
3124 		/* if the link state is up at this point, we
3125 		 * mark it down - this can happen if we have
3126 		 * simultaneous link failures and
3127 		 * reselect_active_interface doesn't make this
3128 		 * one the current slave so it is still marked
3129 		 * up when it is actually down
3130 		 */
3131 		if (slave->link == BOND_LINK_UP) {
3132 			slave->link = BOND_LINK_DOWN;
3133 			if (slave->link_failure_count < UINT_MAX)
3134 				slave->link_failure_count++;
3135 
3136 			bond_set_slave_inactive_flags(slave);
3137 
3138 			pr_info("%s: backup interface %s is now down.\n",
3139 				bond->dev->name, slave->dev->name);
3140 		}
3141 	}
3142 }
3143 
3144 void bond_activebackup_arp_mon(struct work_struct *work)
3145 {
3146 	struct bonding *bond = container_of(work, struct bonding,
3147 					    arp_work.work);
3148 	bool should_notify_peers = false;
3149 	int delta_in_ticks;
3150 
3151 	read_lock(&bond->lock);
3152 
3153 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3154 
3155 	if (bond->slave_cnt == 0)
3156 		goto re_arm;
3157 
3158 	should_notify_peers = bond_should_notify_peers(bond);
3159 
3160 	if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3161 		read_unlock(&bond->lock);
3162 
3163 		/* Race avoidance with bond_close flush of workqueue */
3164 		if (!rtnl_trylock()) {
3165 			read_lock(&bond->lock);
3166 			delta_in_ticks = 1;
3167 			should_notify_peers = false;
3168 			goto re_arm;
3169 		}
3170 
3171 		read_lock(&bond->lock);
3172 
3173 		bond_ab_arp_commit(bond, delta_in_ticks);
3174 
3175 		read_unlock(&bond->lock);
3176 		rtnl_unlock();
3177 		read_lock(&bond->lock);
3178 	}
3179 
3180 	bond_ab_arp_probe(bond);
3181 
3182 re_arm:
3183 	if (bond->params.arp_interval)
3184 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3185 
3186 	read_unlock(&bond->lock);
3187 
3188 	if (should_notify_peers) {
3189 		if (!rtnl_trylock()) {
3190 			read_lock(&bond->lock);
3191 			bond->send_peer_notif++;
3192 			read_unlock(&bond->lock);
3193 			return;
3194 		}
3195 		netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3196 		rtnl_unlock();
3197 	}
3198 }
3199 
3200 /*-------------------------- netdev event handling --------------------------*/
3201 
3202 /*
3203  * Change device name
3204  */
3205 static int bond_event_changename(struct bonding *bond)
3206 {
3207 	bond_remove_proc_entry(bond);
3208 	bond_create_proc_entry(bond);
3209 
3210 	bond_debug_reregister(bond);
3211 
3212 	return NOTIFY_DONE;
3213 }
3214 
3215 static int bond_master_netdev_event(unsigned long event,
3216 				    struct net_device *bond_dev)
3217 {
3218 	struct bonding *event_bond = netdev_priv(bond_dev);
3219 
3220 	switch (event) {
3221 	case NETDEV_CHANGENAME:
3222 		return bond_event_changename(event_bond);
3223 	default:
3224 		break;
3225 	}
3226 
3227 	return NOTIFY_DONE;
3228 }
3229 
3230 static int bond_slave_netdev_event(unsigned long event,
3231 				   struct net_device *slave_dev)
3232 {
3233 	struct net_device *bond_dev = slave_dev->master;
3234 	struct bonding *bond = netdev_priv(bond_dev);
3235 	struct slave *slave = NULL;
3236 
3237 	switch (event) {
3238 	case NETDEV_UNREGISTER:
3239 		if (bond_dev) {
3240 			if (bond->setup_by_slave)
3241 				bond_release_and_destroy(bond_dev, slave_dev);
3242 			else
3243 				bond_release(bond_dev, slave_dev);
3244 		}
3245 		break;
3246 	case NETDEV_UP:
3247 	case NETDEV_CHANGE:
3248 		slave = bond_get_slave_by_dev(bond, slave_dev);
3249 		if (slave) {
3250 			u32 old_speed = slave->speed;
3251 			u8  old_duplex = slave->duplex;
3252 
3253 			bond_update_speed_duplex(slave);
3254 
3255 			if (bond->params.mode == BOND_MODE_8023AD) {
3256 				if (old_speed != slave->speed)
3257 					bond_3ad_adapter_speed_changed(slave);
3258 				if (old_duplex != slave->duplex)
3259 					bond_3ad_adapter_duplex_changed(slave);
3260 			}
3261 		}
3262 
3263 		break;
3264 	case NETDEV_DOWN:
3265 		/*
3266 		 * ... Or is it this?
3267 		 */
3268 		break;
3269 	case NETDEV_CHANGEMTU:
3270 		/*
3271 		 * TODO: Should slaves be allowed to
3272 		 * independently alter their MTU?  For
3273 		 * an active-backup bond, slaves need
3274 		 * not be the same type of device, so
3275 		 * MTUs may vary.  For other modes,
3276 		 * slaves arguably should have the
3277 		 * same MTUs. To do this, we'd need to
3278 		 * take over the slave's change_mtu
3279 		 * function for the duration of their
3280 		 * servitude.
3281 		 */
3282 		break;
3283 	case NETDEV_CHANGENAME:
3284 		/*
3285 		 * TODO: handle changing the primary's name
3286 		 */
3287 		break;
3288 	case NETDEV_FEAT_CHANGE:
3289 		bond_compute_features(bond);
3290 		break;
3291 	default:
3292 		break;
3293 	}
3294 
3295 	return NOTIFY_DONE;
3296 }
3297 
3298 /*
3299  * bond_netdev_event: handle netdev notifier chain events.
3300  *
3301  * This function receives events for the netdev chain.  The caller (an
3302  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3303  * locks for us to safely manipulate the slave devices (RTNL lock,
3304  * dev_probe_lock).
3305  */
3306 static int bond_netdev_event(struct notifier_block *this,
3307 			     unsigned long event, void *ptr)
3308 {
3309 	struct net_device *event_dev = (struct net_device *)ptr;
3310 
3311 	pr_debug("event_dev: %s, event: %lx\n",
3312 		 event_dev ? event_dev->name : "None",
3313 		 event);
3314 
3315 	if (!(event_dev->priv_flags & IFF_BONDING))
3316 		return NOTIFY_DONE;
3317 
3318 	if (event_dev->flags & IFF_MASTER) {
3319 		pr_debug("IFF_MASTER\n");
3320 		return bond_master_netdev_event(event, event_dev);
3321 	}
3322 
3323 	if (event_dev->flags & IFF_SLAVE) {
3324 		pr_debug("IFF_SLAVE\n");
3325 		return bond_slave_netdev_event(event, event_dev);
3326 	}
3327 
3328 	return NOTIFY_DONE;
3329 }
3330 
3331 static struct notifier_block bond_netdev_notifier = {
3332 	.notifier_call = bond_netdev_event,
3333 };
3334 
3335 /*---------------------------- Hashing Policies -----------------------------*/
3336 
3337 /*
3338  * Hash for the output device based upon layer 2 and layer 3 data. If
3339  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3340  */
3341 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3342 {
3343 	struct ethhdr *data = (struct ethhdr *)skb->data;
3344 	struct iphdr *iph = ip_hdr(skb);
3345 
3346 	if (skb->protocol == htons(ETH_P_IP)) {
3347 		return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3348 			(data->h_dest[5] ^ data->h_source[5])) % count;
3349 	}
3350 
3351 	return (data->h_dest[5] ^ data->h_source[5]) % count;
3352 }
3353 
3354 /*
3355  * Hash for the output device based upon layer 3 and layer 4 data. If
3356  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3357  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3358  */
3359 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3360 {
3361 	struct ethhdr *data = (struct ethhdr *)skb->data;
3362 	struct iphdr *iph = ip_hdr(skb);
3363 	__be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3364 	int layer4_xor = 0;
3365 
3366 	if (skb->protocol == htons(ETH_P_IP)) {
3367 		if (!ip_is_fragment(iph) &&
3368 		    (iph->protocol == IPPROTO_TCP ||
3369 		     iph->protocol == IPPROTO_UDP)) {
3370 			layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3371 		}
3372 		return (layer4_xor ^
3373 			((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3374 
3375 	}
3376 
3377 	return (data->h_dest[5] ^ data->h_source[5]) % count;
3378 }
3379 
3380 /*
3381  * Hash for the output device based upon layer 2 data
3382  */
3383 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3384 {
3385 	struct ethhdr *data = (struct ethhdr *)skb->data;
3386 
3387 	return (data->h_dest[5] ^ data->h_source[5]) % count;
3388 }
3389 
3390 /*-------------------------- Device entry points ----------------------------*/
3391 
3392 static int bond_open(struct net_device *bond_dev)
3393 {
3394 	struct bonding *bond = netdev_priv(bond_dev);
3395 	struct slave *slave;
3396 	int i;
3397 
3398 	/* reset slave->backup and slave->inactive */
3399 	read_lock(&bond->lock);
3400 	if (bond->slave_cnt > 0) {
3401 		read_lock(&bond->curr_slave_lock);
3402 		bond_for_each_slave(bond, slave, i) {
3403 			if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3404 				&& (slave != bond->curr_active_slave)) {
3405 				bond_set_slave_inactive_flags(slave);
3406 			} else {
3407 				bond_set_slave_active_flags(slave);
3408 			}
3409 		}
3410 		read_unlock(&bond->curr_slave_lock);
3411 	}
3412 	read_unlock(&bond->lock);
3413 
3414 	INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3415 
3416 	if (bond_is_lb(bond)) {
3417 		/* bond_alb_initialize must be called before the timer
3418 		 * is started.
3419 		 */
3420 		if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3421 			/* something went wrong - fail the open operation */
3422 			return -ENOMEM;
3423 		}
3424 
3425 		INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3426 		queue_delayed_work(bond->wq, &bond->alb_work, 0);
3427 	}
3428 
3429 	if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3430 		INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3431 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
3432 	}
3433 
3434 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3435 		if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3436 			INIT_DELAYED_WORK(&bond->arp_work,
3437 					  bond_activebackup_arp_mon);
3438 		else
3439 			INIT_DELAYED_WORK(&bond->arp_work,
3440 					  bond_loadbalance_arp_mon);
3441 
3442 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
3443 		if (bond->params.arp_validate)
3444 			bond->recv_probe = bond_arp_rcv;
3445 	}
3446 
3447 	if (bond->params.mode == BOND_MODE_8023AD) {
3448 		INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3449 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
3450 		/* register to receive LACPDUs */
3451 		bond->recv_probe = bond_3ad_lacpdu_recv;
3452 		bond_3ad_initiate_agg_selection(bond, 1);
3453 	}
3454 
3455 	return 0;
3456 }
3457 
3458 static int bond_close(struct net_device *bond_dev)
3459 {
3460 	struct bonding *bond = netdev_priv(bond_dev);
3461 
3462 	write_lock_bh(&bond->lock);
3463 
3464 	bond->send_peer_notif = 0;
3465 
3466 	write_unlock_bh(&bond->lock);
3467 
3468 	if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3469 		cancel_delayed_work_sync(&bond->mii_work);
3470 	}
3471 
3472 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3473 		cancel_delayed_work_sync(&bond->arp_work);
3474 	}
3475 
3476 	switch (bond->params.mode) {
3477 	case BOND_MODE_8023AD:
3478 		cancel_delayed_work_sync(&bond->ad_work);
3479 		break;
3480 	case BOND_MODE_TLB:
3481 	case BOND_MODE_ALB:
3482 		cancel_delayed_work_sync(&bond->alb_work);
3483 		break;
3484 	default:
3485 		break;
3486 	}
3487 
3488 	if (delayed_work_pending(&bond->mcast_work))
3489 		cancel_delayed_work_sync(&bond->mcast_work);
3490 
3491 	if (bond_is_lb(bond)) {
3492 		/* Must be called only after all
3493 		 * slaves have been released
3494 		 */
3495 		bond_alb_deinitialize(bond);
3496 	}
3497 	bond->recv_probe = NULL;
3498 
3499 	return 0;
3500 }
3501 
3502 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3503 						struct rtnl_link_stats64 *stats)
3504 {
3505 	struct bonding *bond = netdev_priv(bond_dev);
3506 	struct rtnl_link_stats64 temp;
3507 	struct slave *slave;
3508 	int i;
3509 
3510 	memset(stats, 0, sizeof(*stats));
3511 
3512 	read_lock_bh(&bond->lock);
3513 
3514 	bond_for_each_slave(bond, slave, i) {
3515 		const struct rtnl_link_stats64 *sstats =
3516 			dev_get_stats(slave->dev, &temp);
3517 
3518 		stats->rx_packets += sstats->rx_packets;
3519 		stats->rx_bytes += sstats->rx_bytes;
3520 		stats->rx_errors += sstats->rx_errors;
3521 		stats->rx_dropped += sstats->rx_dropped;
3522 
3523 		stats->tx_packets += sstats->tx_packets;
3524 		stats->tx_bytes += sstats->tx_bytes;
3525 		stats->tx_errors += sstats->tx_errors;
3526 		stats->tx_dropped += sstats->tx_dropped;
3527 
3528 		stats->multicast += sstats->multicast;
3529 		stats->collisions += sstats->collisions;
3530 
3531 		stats->rx_length_errors += sstats->rx_length_errors;
3532 		stats->rx_over_errors += sstats->rx_over_errors;
3533 		stats->rx_crc_errors += sstats->rx_crc_errors;
3534 		stats->rx_frame_errors += sstats->rx_frame_errors;
3535 		stats->rx_fifo_errors += sstats->rx_fifo_errors;
3536 		stats->rx_missed_errors += sstats->rx_missed_errors;
3537 
3538 		stats->tx_aborted_errors += sstats->tx_aborted_errors;
3539 		stats->tx_carrier_errors += sstats->tx_carrier_errors;
3540 		stats->tx_fifo_errors += sstats->tx_fifo_errors;
3541 		stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3542 		stats->tx_window_errors += sstats->tx_window_errors;
3543 	}
3544 
3545 	read_unlock_bh(&bond->lock);
3546 
3547 	return stats;
3548 }
3549 
3550 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3551 {
3552 	struct net_device *slave_dev = NULL;
3553 	struct ifbond k_binfo;
3554 	struct ifbond __user *u_binfo = NULL;
3555 	struct ifslave k_sinfo;
3556 	struct ifslave __user *u_sinfo = NULL;
3557 	struct mii_ioctl_data *mii = NULL;
3558 	int res = 0;
3559 
3560 	pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3561 
3562 	switch (cmd) {
3563 	case SIOCGMIIPHY:
3564 		mii = if_mii(ifr);
3565 		if (!mii)
3566 			return -EINVAL;
3567 
3568 		mii->phy_id = 0;
3569 		/* Fall Through */
3570 	case SIOCGMIIREG:
3571 		/*
3572 		 * We do this again just in case we were called by SIOCGMIIREG
3573 		 * instead of SIOCGMIIPHY.
3574 		 */
3575 		mii = if_mii(ifr);
3576 		if (!mii)
3577 			return -EINVAL;
3578 
3579 
3580 		if (mii->reg_num == 1) {
3581 			struct bonding *bond = netdev_priv(bond_dev);
3582 			mii->val_out = 0;
3583 			read_lock(&bond->lock);
3584 			read_lock(&bond->curr_slave_lock);
3585 			if (netif_carrier_ok(bond->dev))
3586 				mii->val_out = BMSR_LSTATUS;
3587 
3588 			read_unlock(&bond->curr_slave_lock);
3589 			read_unlock(&bond->lock);
3590 		}
3591 
3592 		return 0;
3593 	case BOND_INFO_QUERY_OLD:
3594 	case SIOCBONDINFOQUERY:
3595 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
3596 
3597 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3598 			return -EFAULT;
3599 
3600 		res = bond_info_query(bond_dev, &k_binfo);
3601 		if (res == 0 &&
3602 		    copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3603 			return -EFAULT;
3604 
3605 		return res;
3606 	case BOND_SLAVE_INFO_QUERY_OLD:
3607 	case SIOCBONDSLAVEINFOQUERY:
3608 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3609 
3610 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3611 			return -EFAULT;
3612 
3613 		res = bond_slave_info_query(bond_dev, &k_sinfo);
3614 		if (res == 0 &&
3615 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3616 			return -EFAULT;
3617 
3618 		return res;
3619 	default:
3620 		/* Go on */
3621 		break;
3622 	}
3623 
3624 	if (!capable(CAP_NET_ADMIN))
3625 		return -EPERM;
3626 
3627 	slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3628 
3629 	pr_debug("slave_dev=%p:\n", slave_dev);
3630 
3631 	if (!slave_dev)
3632 		res = -ENODEV;
3633 	else {
3634 		pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3635 		switch (cmd) {
3636 		case BOND_ENSLAVE_OLD:
3637 		case SIOCBONDENSLAVE:
3638 			res = bond_enslave(bond_dev, slave_dev);
3639 			break;
3640 		case BOND_RELEASE_OLD:
3641 		case SIOCBONDRELEASE:
3642 			res = bond_release(bond_dev, slave_dev);
3643 			break;
3644 		case BOND_SETHWADDR_OLD:
3645 		case SIOCBONDSETHWADDR:
3646 			res = bond_sethwaddr(bond_dev, slave_dev);
3647 			break;
3648 		case BOND_CHANGE_ACTIVE_OLD:
3649 		case SIOCBONDCHANGEACTIVE:
3650 			res = bond_ioctl_change_active(bond_dev, slave_dev);
3651 			break;
3652 		default:
3653 			res = -EOPNOTSUPP;
3654 		}
3655 
3656 		dev_put(slave_dev);
3657 	}
3658 
3659 	return res;
3660 }
3661 
3662 static bool bond_addr_in_mc_list(unsigned char *addr,
3663 				 struct netdev_hw_addr_list *list,
3664 				 int addrlen)
3665 {
3666 	struct netdev_hw_addr *ha;
3667 
3668 	netdev_hw_addr_list_for_each(ha, list)
3669 		if (!memcmp(ha->addr, addr, addrlen))
3670 			return true;
3671 
3672 	return false;
3673 }
3674 
3675 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3676 {
3677 	struct bonding *bond = netdev_priv(bond_dev);
3678 
3679 	if (change & IFF_PROMISC)
3680 		bond_set_promiscuity(bond,
3681 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
3682 
3683 	if (change & IFF_ALLMULTI)
3684 		bond_set_allmulti(bond,
3685 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3686 }
3687 
3688 static void bond_set_multicast_list(struct net_device *bond_dev)
3689 {
3690 	struct bonding *bond = netdev_priv(bond_dev);
3691 	struct netdev_hw_addr *ha;
3692 	bool found;
3693 
3694 	read_lock(&bond->lock);
3695 
3696 	/* looking for addresses to add to slaves' mc list */
3697 	netdev_for_each_mc_addr(ha, bond_dev) {
3698 		found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3699 					     bond_dev->addr_len);
3700 		if (!found)
3701 			bond_mc_add(bond, ha->addr);
3702 	}
3703 
3704 	/* looking for addresses to delete from slaves' list */
3705 	netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3706 		found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3707 					     bond_dev->addr_len);
3708 		if (!found)
3709 			bond_mc_del(bond, ha->addr);
3710 	}
3711 
3712 	/* save master's multicast list */
3713 	__hw_addr_flush(&bond->mc_list);
3714 	__hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3715 			       bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3716 
3717 	read_unlock(&bond->lock);
3718 }
3719 
3720 static int bond_neigh_init(struct neighbour *n)
3721 {
3722 	struct bonding *bond = netdev_priv(n->dev);
3723 	struct slave *slave = bond->first_slave;
3724 	const struct net_device_ops *slave_ops;
3725 	struct neigh_parms parms;
3726 	int ret;
3727 
3728 	if (!slave)
3729 		return 0;
3730 
3731 	slave_ops = slave->dev->netdev_ops;
3732 
3733 	if (!slave_ops->ndo_neigh_setup)
3734 		return 0;
3735 
3736 	parms.neigh_setup = NULL;
3737 	parms.neigh_cleanup = NULL;
3738 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3739 	if (ret)
3740 		return ret;
3741 
3742 	/*
3743 	 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3744 	 * after the last slave has been detached.  Assumes that all slaves
3745 	 * utilize the same neigh_cleanup (true at this writing as only user
3746 	 * is ipoib).
3747 	 */
3748 	n->parms->neigh_cleanup = parms.neigh_cleanup;
3749 
3750 	if (!parms.neigh_setup)
3751 		return 0;
3752 
3753 	return parms.neigh_setup(n);
3754 }
3755 
3756 /*
3757  * The bonding ndo_neigh_setup is called at init time beofre any
3758  * slave exists. So we must declare proxy setup function which will
3759  * be used at run time to resolve the actual slave neigh param setup.
3760  */
3761 static int bond_neigh_setup(struct net_device *dev,
3762 			    struct neigh_parms *parms)
3763 {
3764 	parms->neigh_setup   = bond_neigh_init;
3765 
3766 	return 0;
3767 }
3768 
3769 /*
3770  * Change the MTU of all of a master's slaves to match the master
3771  */
3772 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3773 {
3774 	struct bonding *bond = netdev_priv(bond_dev);
3775 	struct slave *slave, *stop_at;
3776 	int res = 0;
3777 	int i;
3778 
3779 	pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3780 		 (bond_dev ? bond_dev->name : "None"), new_mtu);
3781 
3782 	/* Can't hold bond->lock with bh disabled here since
3783 	 * some base drivers panic. On the other hand we can't
3784 	 * hold bond->lock without bh disabled because we'll
3785 	 * deadlock. The only solution is to rely on the fact
3786 	 * that we're under rtnl_lock here, and the slaves
3787 	 * list won't change. This doesn't solve the problem
3788 	 * of setting the slave's MTU while it is
3789 	 * transmitting, but the assumption is that the base
3790 	 * driver can handle that.
3791 	 *
3792 	 * TODO: figure out a way to safely iterate the slaves
3793 	 * list, but without holding a lock around the actual
3794 	 * call to the base driver.
3795 	 */
3796 
3797 	bond_for_each_slave(bond, slave, i) {
3798 		pr_debug("s %p s->p %p c_m %p\n",
3799 			 slave,
3800 			 slave->prev,
3801 			 slave->dev->netdev_ops->ndo_change_mtu);
3802 
3803 		res = dev_set_mtu(slave->dev, new_mtu);
3804 
3805 		if (res) {
3806 			/* If we failed to set the slave's mtu to the new value
3807 			 * we must abort the operation even in ACTIVE_BACKUP
3808 			 * mode, because if we allow the backup slaves to have
3809 			 * different mtu values than the active slave we'll
3810 			 * need to change their mtu when doing a failover. That
3811 			 * means changing their mtu from timer context, which
3812 			 * is probably not a good idea.
3813 			 */
3814 			pr_debug("err %d %s\n", res, slave->dev->name);
3815 			goto unwind;
3816 		}
3817 	}
3818 
3819 	bond_dev->mtu = new_mtu;
3820 
3821 	return 0;
3822 
3823 unwind:
3824 	/* unwind from head to the slave that failed */
3825 	stop_at = slave;
3826 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3827 		int tmp_res;
3828 
3829 		tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3830 		if (tmp_res) {
3831 			pr_debug("unwind err %d dev %s\n",
3832 				 tmp_res, slave->dev->name);
3833 		}
3834 	}
3835 
3836 	return res;
3837 }
3838 
3839 /*
3840  * Change HW address
3841  *
3842  * Note that many devices must be down to change the HW address, and
3843  * downing the master releases all slaves.  We can make bonds full of
3844  * bonding devices to test this, however.
3845  */
3846 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3847 {
3848 	struct bonding *bond = netdev_priv(bond_dev);
3849 	struct sockaddr *sa = addr, tmp_sa;
3850 	struct slave *slave, *stop_at;
3851 	int res = 0;
3852 	int i;
3853 
3854 	if (bond->params.mode == BOND_MODE_ALB)
3855 		return bond_alb_set_mac_address(bond_dev, addr);
3856 
3857 
3858 	pr_debug("bond=%p, name=%s\n",
3859 		 bond, bond_dev ? bond_dev->name : "None");
3860 
3861 	/*
3862 	 * If fail_over_mac is set to active, do nothing and return
3863 	 * success.  Returning an error causes ifenslave to fail.
3864 	 */
3865 	if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3866 		return 0;
3867 
3868 	if (!is_valid_ether_addr(sa->sa_data))
3869 		return -EADDRNOTAVAIL;
3870 
3871 	/* Can't hold bond->lock with bh disabled here since
3872 	 * some base drivers panic. On the other hand we can't
3873 	 * hold bond->lock without bh disabled because we'll
3874 	 * deadlock. The only solution is to rely on the fact
3875 	 * that we're under rtnl_lock here, and the slaves
3876 	 * list won't change. This doesn't solve the problem
3877 	 * of setting the slave's hw address while it is
3878 	 * transmitting, but the assumption is that the base
3879 	 * driver can handle that.
3880 	 *
3881 	 * TODO: figure out a way to safely iterate the slaves
3882 	 * list, but without holding a lock around the actual
3883 	 * call to the base driver.
3884 	 */
3885 
3886 	bond_for_each_slave(bond, slave, i) {
3887 		const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3888 		pr_debug("slave %p %s\n", slave, slave->dev->name);
3889 
3890 		if (slave_ops->ndo_set_mac_address == NULL) {
3891 			res = -EOPNOTSUPP;
3892 			pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3893 			goto unwind;
3894 		}
3895 
3896 		res = dev_set_mac_address(slave->dev, addr);
3897 		if (res) {
3898 			/* TODO: consider downing the slave
3899 			 * and retry ?
3900 			 * User should expect communications
3901 			 * breakage anyway until ARP finish
3902 			 * updating, so...
3903 			 */
3904 			pr_debug("err %d %s\n", res, slave->dev->name);
3905 			goto unwind;
3906 		}
3907 	}
3908 
3909 	/* success */
3910 	memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3911 	return 0;
3912 
3913 unwind:
3914 	memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3915 	tmp_sa.sa_family = bond_dev->type;
3916 
3917 	/* unwind from head to the slave that failed */
3918 	stop_at = slave;
3919 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3920 		int tmp_res;
3921 
3922 		tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3923 		if (tmp_res) {
3924 			pr_debug("unwind err %d dev %s\n",
3925 				 tmp_res, slave->dev->name);
3926 		}
3927 	}
3928 
3929 	return res;
3930 }
3931 
3932 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3933 {
3934 	struct bonding *bond = netdev_priv(bond_dev);
3935 	struct slave *slave, *start_at;
3936 	int i, slave_no, res = 1;
3937 	struct iphdr *iph = ip_hdr(skb);
3938 
3939 	/*
3940 	 * Start with the curr_active_slave that joined the bond as the
3941 	 * default for sending IGMP traffic.  For failover purposes one
3942 	 * needs to maintain some consistency for the interface that will
3943 	 * send the join/membership reports.  The curr_active_slave found
3944 	 * will send all of this type of traffic.
3945 	 */
3946 	if ((iph->protocol == IPPROTO_IGMP) &&
3947 	    (skb->protocol == htons(ETH_P_IP))) {
3948 
3949 		read_lock(&bond->curr_slave_lock);
3950 		slave = bond->curr_active_slave;
3951 		read_unlock(&bond->curr_slave_lock);
3952 
3953 		if (!slave)
3954 			goto out;
3955 	} else {
3956 		/*
3957 		 * Concurrent TX may collide on rr_tx_counter; we accept
3958 		 * that as being rare enough not to justify using an
3959 		 * atomic op here.
3960 		 */
3961 		slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3962 
3963 		bond_for_each_slave(bond, slave, i) {
3964 			slave_no--;
3965 			if (slave_no < 0)
3966 				break;
3967 		}
3968 	}
3969 
3970 	start_at = slave;
3971 	bond_for_each_slave_from(bond, slave, i, start_at) {
3972 		if (IS_UP(slave->dev) &&
3973 		    (slave->link == BOND_LINK_UP) &&
3974 		    bond_is_active_slave(slave)) {
3975 			res = bond_dev_queue_xmit(bond, skb, slave->dev);
3976 			break;
3977 		}
3978 	}
3979 
3980 out:
3981 	if (res) {
3982 		/* no suitable interface, frame not sent */
3983 		dev_kfree_skb(skb);
3984 	}
3985 
3986 	return NETDEV_TX_OK;
3987 }
3988 
3989 
3990 /*
3991  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3992  * the bond has a usable interface.
3993  */
3994 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3995 {
3996 	struct bonding *bond = netdev_priv(bond_dev);
3997 	int res = 1;
3998 
3999 	read_lock(&bond->curr_slave_lock);
4000 
4001 	if (bond->curr_active_slave)
4002 		res = bond_dev_queue_xmit(bond, skb,
4003 			bond->curr_active_slave->dev);
4004 
4005 	if (res)
4006 		/* no suitable interface, frame not sent */
4007 		dev_kfree_skb(skb);
4008 
4009 	read_unlock(&bond->curr_slave_lock);
4010 
4011 	return NETDEV_TX_OK;
4012 }
4013 
4014 /*
4015  * In bond_xmit_xor() , we determine the output device by using a pre-
4016  * determined xmit_hash_policy(), If the selected device is not enabled,
4017  * find the next active slave.
4018  */
4019 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4020 {
4021 	struct bonding *bond = netdev_priv(bond_dev);
4022 	struct slave *slave, *start_at;
4023 	int slave_no;
4024 	int i;
4025 	int res = 1;
4026 
4027 	slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4028 
4029 	bond_for_each_slave(bond, slave, i) {
4030 		slave_no--;
4031 		if (slave_no < 0)
4032 			break;
4033 	}
4034 
4035 	start_at = slave;
4036 
4037 	bond_for_each_slave_from(bond, slave, i, start_at) {
4038 		if (IS_UP(slave->dev) &&
4039 		    (slave->link == BOND_LINK_UP) &&
4040 		    bond_is_active_slave(slave)) {
4041 			res = bond_dev_queue_xmit(bond, skb, slave->dev);
4042 			break;
4043 		}
4044 	}
4045 
4046 	if (res) {
4047 		/* no suitable interface, frame not sent */
4048 		dev_kfree_skb(skb);
4049 	}
4050 
4051 	return NETDEV_TX_OK;
4052 }
4053 
4054 /*
4055  * in broadcast mode, we send everything to all usable interfaces.
4056  */
4057 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4058 {
4059 	struct bonding *bond = netdev_priv(bond_dev);
4060 	struct slave *slave, *start_at;
4061 	struct net_device *tx_dev = NULL;
4062 	int i;
4063 	int res = 1;
4064 
4065 	read_lock(&bond->curr_slave_lock);
4066 	start_at = bond->curr_active_slave;
4067 	read_unlock(&bond->curr_slave_lock);
4068 
4069 	if (!start_at)
4070 		goto out;
4071 
4072 	bond_for_each_slave_from(bond, slave, i, start_at) {
4073 		if (IS_UP(slave->dev) &&
4074 		    (slave->link == BOND_LINK_UP) &&
4075 		    bond_is_active_slave(slave)) {
4076 			if (tx_dev) {
4077 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4078 				if (!skb2) {
4079 					pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4080 					       bond_dev->name);
4081 					continue;
4082 				}
4083 
4084 				res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4085 				if (res) {
4086 					dev_kfree_skb(skb2);
4087 					continue;
4088 				}
4089 			}
4090 			tx_dev = slave->dev;
4091 		}
4092 	}
4093 
4094 	if (tx_dev)
4095 		res = bond_dev_queue_xmit(bond, skb, tx_dev);
4096 
4097 out:
4098 	if (res)
4099 		/* no suitable interface, frame not sent */
4100 		dev_kfree_skb(skb);
4101 
4102 	/* frame sent to all suitable interfaces */
4103 	return NETDEV_TX_OK;
4104 }
4105 
4106 /*------------------------- Device initialization ---------------------------*/
4107 
4108 static void bond_set_xmit_hash_policy(struct bonding *bond)
4109 {
4110 	switch (bond->params.xmit_policy) {
4111 	case BOND_XMIT_POLICY_LAYER23:
4112 		bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4113 		break;
4114 	case BOND_XMIT_POLICY_LAYER34:
4115 		bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4116 		break;
4117 	case BOND_XMIT_POLICY_LAYER2:
4118 	default:
4119 		bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4120 		break;
4121 	}
4122 }
4123 
4124 /*
4125  * Lookup the slave that corresponds to a qid
4126  */
4127 static inline int bond_slave_override(struct bonding *bond,
4128 				      struct sk_buff *skb)
4129 {
4130 	int i, res = 1;
4131 	struct slave *slave = NULL;
4132 	struct slave *check_slave;
4133 
4134 	if (!skb->queue_mapping)
4135 		return 1;
4136 
4137 	/* Find out if any slaves have the same mapping as this skb. */
4138 	bond_for_each_slave(bond, check_slave, i) {
4139 		if (check_slave->queue_id == skb->queue_mapping) {
4140 			slave = check_slave;
4141 			break;
4142 		}
4143 	}
4144 
4145 	/* If the slave isn't UP, use default transmit policy. */
4146 	if (slave && slave->queue_id && IS_UP(slave->dev) &&
4147 	    (slave->link == BOND_LINK_UP)) {
4148 		res = bond_dev_queue_xmit(bond, skb, slave->dev);
4149 	}
4150 
4151 	return res;
4152 }
4153 
4154 
4155 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4156 {
4157 	/*
4158 	 * This helper function exists to help dev_pick_tx get the correct
4159 	 * destination queue.  Using a helper function skips a call to
4160 	 * skb_tx_hash and will put the skbs in the queue we expect on their
4161 	 * way down to the bonding driver.
4162 	 */
4163 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4164 
4165 	/*
4166 	 * Save the original txq to restore before passing to the driver
4167 	 */
4168 	bond_queue_mapping(skb) = skb->queue_mapping;
4169 
4170 	if (unlikely(txq >= dev->real_num_tx_queues)) {
4171 		do {
4172 			txq -= dev->real_num_tx_queues;
4173 		} while (txq >= dev->real_num_tx_queues);
4174 	}
4175 	return txq;
4176 }
4177 
4178 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4179 {
4180 	struct bonding *bond = netdev_priv(dev);
4181 
4182 	if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4183 		if (!bond_slave_override(bond, skb))
4184 			return NETDEV_TX_OK;
4185 	}
4186 
4187 	switch (bond->params.mode) {
4188 	case BOND_MODE_ROUNDROBIN:
4189 		return bond_xmit_roundrobin(skb, dev);
4190 	case BOND_MODE_ACTIVEBACKUP:
4191 		return bond_xmit_activebackup(skb, dev);
4192 	case BOND_MODE_XOR:
4193 		return bond_xmit_xor(skb, dev);
4194 	case BOND_MODE_BROADCAST:
4195 		return bond_xmit_broadcast(skb, dev);
4196 	case BOND_MODE_8023AD:
4197 		return bond_3ad_xmit_xor(skb, dev);
4198 	case BOND_MODE_ALB:
4199 	case BOND_MODE_TLB:
4200 		return bond_alb_xmit(skb, dev);
4201 	default:
4202 		/* Should never happen, mode already checked */
4203 		pr_err("%s: Error: Unknown bonding mode %d\n",
4204 		       dev->name, bond->params.mode);
4205 		WARN_ON_ONCE(1);
4206 		dev_kfree_skb(skb);
4207 		return NETDEV_TX_OK;
4208 	}
4209 }
4210 
4211 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4212 {
4213 	struct bonding *bond = netdev_priv(dev);
4214 	netdev_tx_t ret = NETDEV_TX_OK;
4215 
4216 	/*
4217 	 * If we risk deadlock from transmitting this in the
4218 	 * netpoll path, tell netpoll to queue the frame for later tx
4219 	 */
4220 	if (is_netpoll_tx_blocked(dev))
4221 		return NETDEV_TX_BUSY;
4222 
4223 	read_lock(&bond->lock);
4224 
4225 	if (bond->slave_cnt)
4226 		ret = __bond_start_xmit(skb, dev);
4227 	else
4228 		dev_kfree_skb(skb);
4229 
4230 	read_unlock(&bond->lock);
4231 
4232 	return ret;
4233 }
4234 
4235 /*
4236  * set bond mode specific net device operations
4237  */
4238 void bond_set_mode_ops(struct bonding *bond, int mode)
4239 {
4240 	struct net_device *bond_dev = bond->dev;
4241 
4242 	switch (mode) {
4243 	case BOND_MODE_ROUNDROBIN:
4244 		break;
4245 	case BOND_MODE_ACTIVEBACKUP:
4246 		break;
4247 	case BOND_MODE_XOR:
4248 		bond_set_xmit_hash_policy(bond);
4249 		break;
4250 	case BOND_MODE_BROADCAST:
4251 		break;
4252 	case BOND_MODE_8023AD:
4253 		bond_set_xmit_hash_policy(bond);
4254 		break;
4255 	case BOND_MODE_ALB:
4256 		/* FALLTHRU */
4257 	case BOND_MODE_TLB:
4258 		break;
4259 	default:
4260 		/* Should never happen, mode already checked */
4261 		pr_err("%s: Error: Unknown bonding mode %d\n",
4262 		       bond_dev->name, mode);
4263 		break;
4264 	}
4265 }
4266 
4267 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4268 				    struct ethtool_drvinfo *drvinfo)
4269 {
4270 	strncpy(drvinfo->driver, DRV_NAME, 32);
4271 	strncpy(drvinfo->version, DRV_VERSION, 32);
4272 	snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4273 }
4274 
4275 static const struct ethtool_ops bond_ethtool_ops = {
4276 	.get_drvinfo		= bond_ethtool_get_drvinfo,
4277 	.get_link		= ethtool_op_get_link,
4278 };
4279 
4280 static const struct net_device_ops bond_netdev_ops = {
4281 	.ndo_init		= bond_init,
4282 	.ndo_uninit		= bond_uninit,
4283 	.ndo_open		= bond_open,
4284 	.ndo_stop		= bond_close,
4285 	.ndo_start_xmit		= bond_start_xmit,
4286 	.ndo_select_queue	= bond_select_queue,
4287 	.ndo_get_stats64	= bond_get_stats,
4288 	.ndo_do_ioctl		= bond_do_ioctl,
4289 	.ndo_change_rx_flags	= bond_change_rx_flags,
4290 	.ndo_set_rx_mode	= bond_set_multicast_list,
4291 	.ndo_change_mtu		= bond_change_mtu,
4292 	.ndo_set_mac_address	= bond_set_mac_address,
4293 	.ndo_neigh_setup	= bond_neigh_setup,
4294 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
4295 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
4296 #ifdef CONFIG_NET_POLL_CONTROLLER
4297 	.ndo_netpoll_setup	= bond_netpoll_setup,
4298 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
4299 	.ndo_poll_controller	= bond_poll_controller,
4300 #endif
4301 	.ndo_add_slave		= bond_enslave,
4302 	.ndo_del_slave		= bond_release,
4303 	.ndo_fix_features	= bond_fix_features,
4304 };
4305 
4306 static void bond_destructor(struct net_device *bond_dev)
4307 {
4308 	struct bonding *bond = netdev_priv(bond_dev);
4309 	if (bond->wq)
4310 		destroy_workqueue(bond->wq);
4311 	free_netdev(bond_dev);
4312 }
4313 
4314 static void bond_setup(struct net_device *bond_dev)
4315 {
4316 	struct bonding *bond = netdev_priv(bond_dev);
4317 
4318 	/* initialize rwlocks */
4319 	rwlock_init(&bond->lock);
4320 	rwlock_init(&bond->curr_slave_lock);
4321 
4322 	bond->params = bonding_defaults;
4323 
4324 	/* Initialize pointers */
4325 	bond->dev = bond_dev;
4326 	INIT_LIST_HEAD(&bond->vlan_list);
4327 
4328 	/* Initialize the device entry points */
4329 	ether_setup(bond_dev);
4330 	bond_dev->netdev_ops = &bond_netdev_ops;
4331 	bond_dev->ethtool_ops = &bond_ethtool_ops;
4332 	bond_set_mode_ops(bond, bond->params.mode);
4333 
4334 	bond_dev->destructor = bond_destructor;
4335 
4336 	/* Initialize the device options */
4337 	bond_dev->tx_queue_len = 0;
4338 	bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4339 	bond_dev->priv_flags |= IFF_BONDING;
4340 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4341 
4342 	/* At first, we block adding VLANs. That's the only way to
4343 	 * prevent problems that occur when adding VLANs over an
4344 	 * empty bond. The block will be removed once non-challenged
4345 	 * slaves are enslaved.
4346 	 */
4347 	bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4348 
4349 	/* don't acquire bond device's netif_tx_lock when
4350 	 * transmitting */
4351 	bond_dev->features |= NETIF_F_LLTX;
4352 
4353 	/* By default, we declare the bond to be fully
4354 	 * VLAN hardware accelerated capable. Special
4355 	 * care is taken in the various xmit functions
4356 	 * when there are slaves that are not hw accel
4357 	 * capable
4358 	 */
4359 
4360 	bond_dev->hw_features = BOND_VLAN_FEATURES |
4361 				NETIF_F_HW_VLAN_TX |
4362 				NETIF_F_HW_VLAN_RX |
4363 				NETIF_F_HW_VLAN_FILTER;
4364 
4365 	bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4366 	bond_dev->features |= bond_dev->hw_features;
4367 }
4368 
4369 static void bond_work_cancel_all(struct bonding *bond)
4370 {
4371 	if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4372 		cancel_delayed_work_sync(&bond->mii_work);
4373 
4374 	if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4375 		cancel_delayed_work_sync(&bond->arp_work);
4376 
4377 	if (bond->params.mode == BOND_MODE_ALB &&
4378 	    delayed_work_pending(&bond->alb_work))
4379 		cancel_delayed_work_sync(&bond->alb_work);
4380 
4381 	if (bond->params.mode == BOND_MODE_8023AD &&
4382 	    delayed_work_pending(&bond->ad_work))
4383 		cancel_delayed_work_sync(&bond->ad_work);
4384 
4385 	if (delayed_work_pending(&bond->mcast_work))
4386 		cancel_delayed_work_sync(&bond->mcast_work);
4387 }
4388 
4389 /*
4390 * Destroy a bonding device.
4391 * Must be under rtnl_lock when this function is called.
4392 */
4393 static void bond_uninit(struct net_device *bond_dev)
4394 {
4395 	struct bonding *bond = netdev_priv(bond_dev);
4396 	struct vlan_entry *vlan, *tmp;
4397 
4398 	bond_netpoll_cleanup(bond_dev);
4399 
4400 	/* Release the bonded slaves */
4401 	bond_release_all(bond_dev);
4402 
4403 	list_del(&bond->bond_list);
4404 
4405 	bond_work_cancel_all(bond);
4406 
4407 	bond_remove_proc_entry(bond);
4408 
4409 	bond_debug_unregister(bond);
4410 
4411 	__hw_addr_flush(&bond->mc_list);
4412 
4413 	list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4414 		list_del(&vlan->vlan_list);
4415 		kfree(vlan);
4416 	}
4417 }
4418 
4419 /*------------------------- Module initialization ---------------------------*/
4420 
4421 /*
4422  * Convert string input module parms.  Accept either the
4423  * number of the mode or its string name.  A bit complicated because
4424  * some mode names are substrings of other names, and calls from sysfs
4425  * may have whitespace in the name (trailing newlines, for example).
4426  */
4427 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4428 {
4429 	int modeint = -1, i, rv;
4430 	char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4431 
4432 	for (p = (char *)buf; *p; p++)
4433 		if (!(isdigit(*p) || isspace(*p)))
4434 			break;
4435 
4436 	if (*p)
4437 		rv = sscanf(buf, "%20s", modestr);
4438 	else
4439 		rv = sscanf(buf, "%d", &modeint);
4440 
4441 	if (!rv)
4442 		return -1;
4443 
4444 	for (i = 0; tbl[i].modename; i++) {
4445 		if (modeint == tbl[i].mode)
4446 			return tbl[i].mode;
4447 		if (strcmp(modestr, tbl[i].modename) == 0)
4448 			return tbl[i].mode;
4449 	}
4450 
4451 	return -1;
4452 }
4453 
4454 static int bond_check_params(struct bond_params *params)
4455 {
4456 	int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4457 
4458 	/*
4459 	 * Convert string parameters.
4460 	 */
4461 	if (mode) {
4462 		bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4463 		if (bond_mode == -1) {
4464 			pr_err("Error: Invalid bonding mode \"%s\"\n",
4465 			       mode == NULL ? "NULL" : mode);
4466 			return -EINVAL;
4467 		}
4468 	}
4469 
4470 	if (xmit_hash_policy) {
4471 		if ((bond_mode != BOND_MODE_XOR) &&
4472 		    (bond_mode != BOND_MODE_8023AD)) {
4473 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4474 			       bond_mode_name(bond_mode));
4475 		} else {
4476 			xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4477 							xmit_hashtype_tbl);
4478 			if (xmit_hashtype == -1) {
4479 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4480 				       xmit_hash_policy == NULL ? "NULL" :
4481 				       xmit_hash_policy);
4482 				return -EINVAL;
4483 			}
4484 		}
4485 	}
4486 
4487 	if (lacp_rate) {
4488 		if (bond_mode != BOND_MODE_8023AD) {
4489 			pr_info("lacp_rate param is irrelevant in mode %s\n",
4490 				bond_mode_name(bond_mode));
4491 		} else {
4492 			lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4493 			if (lacp_fast == -1) {
4494 				pr_err("Error: Invalid lacp rate \"%s\"\n",
4495 				       lacp_rate == NULL ? "NULL" : lacp_rate);
4496 				return -EINVAL;
4497 			}
4498 		}
4499 	}
4500 
4501 	if (ad_select) {
4502 		params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4503 		if (params->ad_select == -1) {
4504 			pr_err("Error: Invalid ad_select \"%s\"\n",
4505 			       ad_select == NULL ? "NULL" : ad_select);
4506 			return -EINVAL;
4507 		}
4508 
4509 		if (bond_mode != BOND_MODE_8023AD) {
4510 			pr_warning("ad_select param only affects 802.3ad mode\n");
4511 		}
4512 	} else {
4513 		params->ad_select = BOND_AD_STABLE;
4514 	}
4515 
4516 	if (max_bonds < 0) {
4517 		pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4518 			   max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4519 		max_bonds = BOND_DEFAULT_MAX_BONDS;
4520 	}
4521 
4522 	if (miimon < 0) {
4523 		pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4524 			   miimon, INT_MAX, BOND_LINK_MON_INTERV);
4525 		miimon = BOND_LINK_MON_INTERV;
4526 	}
4527 
4528 	if (updelay < 0) {
4529 		pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4530 			   updelay, INT_MAX);
4531 		updelay = 0;
4532 	}
4533 
4534 	if (downdelay < 0) {
4535 		pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4536 			   downdelay, INT_MAX);
4537 		downdelay = 0;
4538 	}
4539 
4540 	if ((use_carrier != 0) && (use_carrier != 1)) {
4541 		pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4542 			   use_carrier);
4543 		use_carrier = 1;
4544 	}
4545 
4546 	if (num_peer_notif < 0 || num_peer_notif > 255) {
4547 		pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4548 			   num_peer_notif);
4549 		num_peer_notif = 1;
4550 	}
4551 
4552 	/* reset values for 802.3ad */
4553 	if (bond_mode == BOND_MODE_8023AD) {
4554 		if (!miimon) {
4555 			pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4556 			pr_warning("Forcing miimon to 100msec\n");
4557 			miimon = 100;
4558 		}
4559 	}
4560 
4561 	if (tx_queues < 1 || tx_queues > 255) {
4562 		pr_warning("Warning: tx_queues (%d) should be between "
4563 			   "1 and 255, resetting to %d\n",
4564 			   tx_queues, BOND_DEFAULT_TX_QUEUES);
4565 		tx_queues = BOND_DEFAULT_TX_QUEUES;
4566 	}
4567 
4568 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4569 		pr_warning("Warning: all_slaves_active module parameter (%d), "
4570 			   "not of valid value (0/1), so it was set to "
4571 			   "0\n", all_slaves_active);
4572 		all_slaves_active = 0;
4573 	}
4574 
4575 	if (resend_igmp < 0 || resend_igmp > 255) {
4576 		pr_warning("Warning: resend_igmp (%d) should be between "
4577 			   "0 and 255, resetting to %d\n",
4578 			   resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4579 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4580 	}
4581 
4582 	/* reset values for TLB/ALB */
4583 	if ((bond_mode == BOND_MODE_TLB) ||
4584 	    (bond_mode == BOND_MODE_ALB)) {
4585 		if (!miimon) {
4586 			pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4587 			pr_warning("Forcing miimon to 100msec\n");
4588 			miimon = 100;
4589 		}
4590 	}
4591 
4592 	if (bond_mode == BOND_MODE_ALB) {
4593 		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",
4594 			  updelay);
4595 	}
4596 
4597 	if (!miimon) {
4598 		if (updelay || downdelay) {
4599 			/* just warn the user the up/down delay will have
4600 			 * no effect since miimon is zero...
4601 			 */
4602 			pr_warning("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",
4603 				   updelay, downdelay);
4604 		}
4605 	} else {
4606 		/* don't allow arp monitoring */
4607 		if (arp_interval) {
4608 			pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4609 				   miimon, arp_interval);
4610 			arp_interval = 0;
4611 		}
4612 
4613 		if ((updelay % miimon) != 0) {
4614 			pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4615 				   updelay, miimon,
4616 				   (updelay / miimon) * miimon);
4617 		}
4618 
4619 		updelay /= miimon;
4620 
4621 		if ((downdelay % miimon) != 0) {
4622 			pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4623 				   downdelay, miimon,
4624 				   (downdelay / miimon) * miimon);
4625 		}
4626 
4627 		downdelay /= miimon;
4628 	}
4629 
4630 	if (arp_interval < 0) {
4631 		pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4632 			   arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4633 		arp_interval = BOND_LINK_ARP_INTERV;
4634 	}
4635 
4636 	for (arp_ip_count = 0;
4637 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4638 	     arp_ip_count++) {
4639 		/* not complete check, but should be good enough to
4640 		   catch mistakes */
4641 		if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4642 			pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4643 				   arp_ip_target[arp_ip_count]);
4644 			arp_interval = 0;
4645 		} else {
4646 			__be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4647 			arp_target[arp_ip_count] = ip;
4648 		}
4649 	}
4650 
4651 	if (arp_interval && !arp_ip_count) {
4652 		/* don't allow arping if no arp_ip_target given... */
4653 		pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4654 			   arp_interval);
4655 		arp_interval = 0;
4656 	}
4657 
4658 	if (arp_validate) {
4659 		if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4660 			pr_err("arp_validate only supported in active-backup mode\n");
4661 			return -EINVAL;
4662 		}
4663 		if (!arp_interval) {
4664 			pr_err("arp_validate requires arp_interval\n");
4665 			return -EINVAL;
4666 		}
4667 
4668 		arp_validate_value = bond_parse_parm(arp_validate,
4669 						     arp_validate_tbl);
4670 		if (arp_validate_value == -1) {
4671 			pr_err("Error: invalid arp_validate \"%s\"\n",
4672 			       arp_validate == NULL ? "NULL" : arp_validate);
4673 			return -EINVAL;
4674 		}
4675 	} else
4676 		arp_validate_value = 0;
4677 
4678 	if (miimon) {
4679 		pr_info("MII link monitoring set to %d ms\n", miimon);
4680 	} else if (arp_interval) {
4681 		int i;
4682 
4683 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4684 			arp_interval,
4685 			arp_validate_tbl[arp_validate_value].modename,
4686 			arp_ip_count);
4687 
4688 		for (i = 0; i < arp_ip_count; i++)
4689 			pr_info(" %s", arp_ip_target[i]);
4690 
4691 		pr_info("\n");
4692 
4693 	} else if (max_bonds) {
4694 		/* miimon and arp_interval not set, we need one so things
4695 		 * work as expected, see bonding.txt for details
4696 		 */
4697 		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");
4698 	}
4699 
4700 	if (primary && !USES_PRIMARY(bond_mode)) {
4701 		/* currently, using a primary only makes sense
4702 		 * in active backup, TLB or ALB modes
4703 		 */
4704 		pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4705 			   primary, bond_mode_name(bond_mode));
4706 		primary = NULL;
4707 	}
4708 
4709 	if (primary && primary_reselect) {
4710 		primary_reselect_value = bond_parse_parm(primary_reselect,
4711 							 pri_reselect_tbl);
4712 		if (primary_reselect_value == -1) {
4713 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
4714 			       primary_reselect ==
4715 					NULL ? "NULL" : primary_reselect);
4716 			return -EINVAL;
4717 		}
4718 	} else {
4719 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4720 	}
4721 
4722 	if (fail_over_mac) {
4723 		fail_over_mac_value = bond_parse_parm(fail_over_mac,
4724 						      fail_over_mac_tbl);
4725 		if (fail_over_mac_value == -1) {
4726 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
4727 			       arp_validate == NULL ? "NULL" : arp_validate);
4728 			return -EINVAL;
4729 		}
4730 
4731 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4732 			pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4733 	} else {
4734 		fail_over_mac_value = BOND_FOM_NONE;
4735 	}
4736 
4737 	/* fill params struct with the proper values */
4738 	params->mode = bond_mode;
4739 	params->xmit_policy = xmit_hashtype;
4740 	params->miimon = miimon;
4741 	params->num_peer_notif = num_peer_notif;
4742 	params->arp_interval = arp_interval;
4743 	params->arp_validate = arp_validate_value;
4744 	params->updelay = updelay;
4745 	params->downdelay = downdelay;
4746 	params->use_carrier = use_carrier;
4747 	params->lacp_fast = lacp_fast;
4748 	params->primary[0] = 0;
4749 	params->primary_reselect = primary_reselect_value;
4750 	params->fail_over_mac = fail_over_mac_value;
4751 	params->tx_queues = tx_queues;
4752 	params->all_slaves_active = all_slaves_active;
4753 	params->resend_igmp = resend_igmp;
4754 	params->min_links = min_links;
4755 
4756 	if (primary) {
4757 		strncpy(params->primary, primary, IFNAMSIZ);
4758 		params->primary[IFNAMSIZ - 1] = 0;
4759 	}
4760 
4761 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4762 
4763 	return 0;
4764 }
4765 
4766 static struct lock_class_key bonding_netdev_xmit_lock_key;
4767 static struct lock_class_key bonding_netdev_addr_lock_key;
4768 
4769 static void bond_set_lockdep_class_one(struct net_device *dev,
4770 				       struct netdev_queue *txq,
4771 				       void *_unused)
4772 {
4773 	lockdep_set_class(&txq->_xmit_lock,
4774 			  &bonding_netdev_xmit_lock_key);
4775 }
4776 
4777 static void bond_set_lockdep_class(struct net_device *dev)
4778 {
4779 	lockdep_set_class(&dev->addr_list_lock,
4780 			  &bonding_netdev_addr_lock_key);
4781 	netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4782 }
4783 
4784 /*
4785  * Called from registration process
4786  */
4787 static int bond_init(struct net_device *bond_dev)
4788 {
4789 	struct bonding *bond = netdev_priv(bond_dev);
4790 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4791 	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4792 
4793 	pr_debug("Begin bond_init for %s\n", bond_dev->name);
4794 
4795 	/*
4796 	 * Initialize locks that may be required during
4797 	 * en/deslave operations.  All of the bond_open work
4798 	 * (of which this is part) should really be moved to
4799 	 * a phase prior to dev_open
4800 	 */
4801 	spin_lock_init(&(bond_info->tx_hashtbl_lock));
4802 	spin_lock_init(&(bond_info->rx_hashtbl_lock));
4803 
4804 	bond->wq = create_singlethread_workqueue(bond_dev->name);
4805 	if (!bond->wq)
4806 		return -ENOMEM;
4807 
4808 	bond_set_lockdep_class(bond_dev);
4809 
4810 	bond_create_proc_entry(bond);
4811 	list_add_tail(&bond->bond_list, &bn->dev_list);
4812 
4813 	bond_prepare_sysfs_group(bond);
4814 
4815 	bond_debug_register(bond);
4816 
4817 	__hw_addr_init(&bond->mc_list);
4818 	return 0;
4819 }
4820 
4821 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4822 {
4823 	if (tb[IFLA_ADDRESS]) {
4824 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4825 			return -EINVAL;
4826 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4827 			return -EADDRNOTAVAIL;
4828 	}
4829 	return 0;
4830 }
4831 
4832 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[])
4833 {
4834 	return tx_queues;
4835 }
4836 
4837 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4838 	.kind		= "bond",
4839 	.priv_size	= sizeof(struct bonding),
4840 	.setup		= bond_setup,
4841 	.validate	= bond_validate,
4842 	.get_tx_queues	= bond_get_tx_queues,
4843 };
4844 
4845 /* Create a new bond based on the specified name and bonding parameters.
4846  * If name is NULL, obtain a suitable "bond%d" name for us.
4847  * Caller must NOT hold rtnl_lock; we need to release it here before we
4848  * set up our sysfs entries.
4849  */
4850 int bond_create(struct net *net, const char *name)
4851 {
4852 	struct net_device *bond_dev;
4853 	int res;
4854 
4855 	rtnl_lock();
4856 
4857 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4858 				   name ? name : "bond%d",
4859 				   bond_setup, tx_queues);
4860 	if (!bond_dev) {
4861 		pr_err("%s: eek! can't alloc netdev!\n", name);
4862 		rtnl_unlock();
4863 		return -ENOMEM;
4864 	}
4865 
4866 	dev_net_set(bond_dev, net);
4867 	bond_dev->rtnl_link_ops = &bond_link_ops;
4868 
4869 	res = register_netdevice(bond_dev);
4870 
4871 	netif_carrier_off(bond_dev);
4872 
4873 	rtnl_unlock();
4874 	if (res < 0)
4875 		bond_destructor(bond_dev);
4876 	return res;
4877 }
4878 
4879 static int __net_init bond_net_init(struct net *net)
4880 {
4881 	struct bond_net *bn = net_generic(net, bond_net_id);
4882 
4883 	bn->net = net;
4884 	INIT_LIST_HEAD(&bn->dev_list);
4885 
4886 	bond_create_proc_dir(bn);
4887 	bond_create_sysfs(bn);
4888 
4889 	return 0;
4890 }
4891 
4892 static void __net_exit bond_net_exit(struct net *net)
4893 {
4894 	struct bond_net *bn = net_generic(net, bond_net_id);
4895 
4896 	bond_destroy_sysfs(bn);
4897 	bond_destroy_proc_dir(bn);
4898 }
4899 
4900 static struct pernet_operations bond_net_ops = {
4901 	.init = bond_net_init,
4902 	.exit = bond_net_exit,
4903 	.id   = &bond_net_id,
4904 	.size = sizeof(struct bond_net),
4905 };
4906 
4907 static int __init bonding_init(void)
4908 {
4909 	int i;
4910 	int res;
4911 
4912 	pr_info("%s", bond_version);
4913 
4914 	res = bond_check_params(&bonding_defaults);
4915 	if (res)
4916 		goto out;
4917 
4918 	res = register_pernet_subsys(&bond_net_ops);
4919 	if (res)
4920 		goto out;
4921 
4922 	res = rtnl_link_register(&bond_link_ops);
4923 	if (res)
4924 		goto err_link;
4925 
4926 	bond_create_debugfs();
4927 
4928 	for (i = 0; i < max_bonds; i++) {
4929 		res = bond_create(&init_net, NULL);
4930 		if (res)
4931 			goto err;
4932 	}
4933 
4934 	register_netdevice_notifier(&bond_netdev_notifier);
4935 out:
4936 	return res;
4937 err:
4938 	rtnl_link_unregister(&bond_link_ops);
4939 err_link:
4940 	unregister_pernet_subsys(&bond_net_ops);
4941 	goto out;
4942 
4943 }
4944 
4945 static void __exit bonding_exit(void)
4946 {
4947 	unregister_netdevice_notifier(&bond_netdev_notifier);
4948 
4949 	bond_destroy_debugfs();
4950 
4951 	rtnl_link_unregister(&bond_link_ops);
4952 	unregister_pernet_subsys(&bond_net_ops);
4953 
4954 #ifdef CONFIG_NET_POLL_CONTROLLER
4955 	/*
4956 	 * Make sure we don't have an imbalance on our netpoll blocking
4957 	 */
4958 	WARN_ON(atomic_read(&netpoll_block_tx));
4959 #endif
4960 }
4961 
4962 module_init(bonding_init);
4963 module_exit(bonding_exit);
4964 MODULE_LICENSE("GPL");
4965 MODULE_VERSION(DRV_VERSION);
4966 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4967 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4968 MODULE_ALIAS_RTNL_LINK("bond");
4969