xref: /openbmc/linux/drivers/net/bonding/bond_main.c (revision f35e839a)
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 <net/pkt_sched.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83 
84 /*---------------------------- Module parameters ----------------------------*/
85 
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV	0
88 #define BOND_LINK_ARP_INTERV	0
89 
90 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon	= BOND_LINK_MON_INTERV;
94 static int updelay;
95 static int downdelay;
96 static int use_carrier	= 1;
97 static char *mode;
98 static char *primary;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111 
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 			       "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 			       "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 			    "in milliseconds");
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 			      "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134 		       "1 for active-backup, 2 for balance-xor, "
135 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 		       "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 				   "once it comes up; "
142 				   "0 for always (default), "
143 				   "1 for only if speed of primary is "
144 				   "better, "
145 				   "2 for only on active slave "
146 				   "failure");
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 			    "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 			    "0 for stable (default), 1 for bandwidth, "
153 			    "2 for count");
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156 
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 				   "0 for layer 2 (default), 1 for layer 3+4, "
160 				   "2 for layer 2+3");
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 			       "0 for none (default), 1 for active, "
168 			       "2 for backup, 3 for all");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 				"the same MAC; 0 for none (default), "
172 				"1 for active, 2 for follow");
173 module_param(all_slaves_active, int, 0);
174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175 				     "by setting active flag for all slaves; "
176 				     "0 for never (default), 1 for always.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179 			      "link failure");
180 
181 /*----------------------------- Global variables ----------------------------*/
182 
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
185 #endif
186 
187 int bond_net_id __read_mostly;
188 
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode	= BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
194 
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 {	"slow",		AD_LACP_SLOW},
197 {	"fast",		AD_LACP_FAST},
198 {	NULL,		-1},
199 };
200 
201 const struct bond_parm_tbl bond_mode_tbl[] = {
202 {	"balance-rr",		BOND_MODE_ROUNDROBIN},
203 {	"active-backup",	BOND_MODE_ACTIVEBACKUP},
204 {	"balance-xor",		BOND_MODE_XOR},
205 {	"broadcast",		BOND_MODE_BROADCAST},
206 {	"802.3ad",		BOND_MODE_8023AD},
207 {	"balance-tlb",		BOND_MODE_TLB},
208 {	"balance-alb",		BOND_MODE_ALB},
209 {	NULL,			-1},
210 };
211 
212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213 {	"layer2",		BOND_XMIT_POLICY_LAYER2},
214 {	"layer3+4",		BOND_XMIT_POLICY_LAYER34},
215 {	"layer2+3",		BOND_XMIT_POLICY_LAYER23},
216 {	NULL,			-1},
217 };
218 
219 const struct bond_parm_tbl arp_validate_tbl[] = {
220 {	"none",			BOND_ARP_VALIDATE_NONE},
221 {	"active",		BOND_ARP_VALIDATE_ACTIVE},
222 {	"backup",		BOND_ARP_VALIDATE_BACKUP},
223 {	"all",			BOND_ARP_VALIDATE_ALL},
224 {	NULL,			-1},
225 };
226 
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 {	"none",			BOND_FOM_NONE},
229 {	"active",		BOND_FOM_ACTIVE},
230 {	"follow",		BOND_FOM_FOLLOW},
231 {	NULL,			-1},
232 };
233 
234 const struct bond_parm_tbl pri_reselect_tbl[] = {
235 {	"always",		BOND_PRI_RESELECT_ALWAYS},
236 {	"better",		BOND_PRI_RESELECT_BETTER},
237 {	"failure",		BOND_PRI_RESELECT_FAILURE},
238 {	NULL,			-1},
239 };
240 
241 struct bond_parm_tbl ad_select_tbl[] = {
242 {	"stable",	BOND_AD_STABLE},
243 {	"bandwidth",	BOND_AD_BANDWIDTH},
244 {	"count",	BOND_AD_COUNT},
245 {	NULL,		-1},
246 };
247 
248 /*-------------------------- Forward declarations ---------------------------*/
249 
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
252 
253 /*---------------------------- General routines -----------------------------*/
254 
255 const char *bond_mode_name(int mode)
256 {
257 	static const char *names[] = {
258 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 		[BOND_MODE_XOR] = "load balancing (xor)",
261 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263 		[BOND_MODE_TLB] = "transmit load balancing",
264 		[BOND_MODE_ALB] = "adaptive load balancing",
265 	};
266 
267 	if (mode < 0 || mode > BOND_MODE_ALB)
268 		return "unknown";
269 
270 	return names[mode];
271 }
272 
273 /*---------------------------------- VLAN -----------------------------------*/
274 
275 /**
276  * bond_add_vlan - add a new vlan id on bond
277  * @bond: bond that got the notification
278  * @vlan_id: the vlan id to add
279  *
280  * Returns -ENOMEM if allocation failed.
281  */
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283 {
284 	struct vlan_entry *vlan;
285 
286 	pr_debug("bond: %s, vlan id %d\n",
287 		 (bond ? bond->dev->name : "None"), vlan_id);
288 
289 	vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
290 	if (!vlan)
291 		return -ENOMEM;
292 
293 	INIT_LIST_HEAD(&vlan->vlan_list);
294 	vlan->vlan_id = vlan_id;
295 
296 	write_lock_bh(&bond->lock);
297 
298 	list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299 
300 	write_unlock_bh(&bond->lock);
301 
302 	pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
303 
304 	return 0;
305 }
306 
307 /**
308  * bond_del_vlan - delete a vlan id from bond
309  * @bond: bond that got the notification
310  * @vlan_id: the vlan id to delete
311  *
312  * returns -ENODEV if @vlan_id was not found in @bond.
313  */
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315 {
316 	struct vlan_entry *vlan;
317 	int res = -ENODEV;
318 
319 	pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
320 
321 	block_netpoll_tx();
322 	write_lock_bh(&bond->lock);
323 
324 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 		if (vlan->vlan_id == vlan_id) {
326 			list_del(&vlan->vlan_list);
327 
328 			if (bond_is_lb(bond))
329 				bond_alb_clear_vlan(bond, vlan_id);
330 
331 			pr_debug("removed VLAN ID %d from bond %s\n",
332 				 vlan_id, bond->dev->name);
333 
334 			kfree(vlan);
335 
336 			res = 0;
337 			goto out;
338 		}
339 	}
340 
341 	pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 		 vlan_id, bond->dev->name);
343 
344 out:
345 	write_unlock_bh(&bond->lock);
346 	unblock_netpoll_tx();
347 	return res;
348 }
349 
350 /**
351  * bond_next_vlan - safely skip to the next item in the vlans list.
352  * @bond: the bond we're working on
353  * @curr: item we're advancing from
354  *
355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356  * or @curr->next otherwise (even if it is @curr itself again).
357  *
358  * Caller must hold bond->lock
359  */
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361 {
362 	struct vlan_entry *next, *last;
363 
364 	if (list_empty(&bond->vlan_list))
365 		return NULL;
366 
367 	if (!curr) {
368 		next = list_entry(bond->vlan_list.next,
369 				  struct vlan_entry, vlan_list);
370 	} else {
371 		last = list_entry(bond->vlan_list.prev,
372 				  struct vlan_entry, vlan_list);
373 		if (last == curr) {
374 			next = list_entry(bond->vlan_list.next,
375 					  struct vlan_entry, vlan_list);
376 		} else {
377 			next = list_entry(curr->vlan_list.next,
378 					  struct vlan_entry, vlan_list);
379 		}
380 	}
381 
382 	return next;
383 }
384 
385 /**
386  * bond_dev_queue_xmit - Prepare skb for xmit.
387  *
388  * @bond: bond device that got this skb for tx.
389  * @skb: hw accel VLAN tagged skb to transmit
390  * @slave_dev: slave that is supposed to xmit this skbuff
391  */
392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
393 			struct net_device *slave_dev)
394 {
395 	skb->dev = slave_dev;
396 
397 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
398 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
399 	skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
400 
401 	if (unlikely(netpoll_tx_running(bond->dev)))
402 		bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
403 	else
404 		dev_queue_xmit(skb);
405 
406 	return 0;
407 }
408 
409 /*
410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411  * We don't protect the slave list iteration with a lock because:
412  * a. This operation is performed in IOCTL context,
413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
414  * c. Holding a lock with BH disabled while directly calling a base driver
415  *    entry point is generally a BAD idea.
416  *
417  * The design of synchronization/protection for this operation in the 8021q
418  * module is good for one or more VLAN devices over a single physical device
419  * and cannot be extended for a teaming solution like bonding, so there is a
420  * potential race condition here where a net device from the vlan group might
421  * be referenced (either by a base driver or the 8021q code) while it is being
422  * removed from the system. However, it turns out we're not making matters
423  * worse, and if it works for regular VLAN usage it will work here too.
424 */
425 
426 /**
427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428  * @bond_dev: bonding net device that got called
429  * @vid: vlan id being added
430  */
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
432 				__be16 proto, u16 vid)
433 {
434 	struct bonding *bond = netdev_priv(bond_dev);
435 	struct slave *slave, *stop_at;
436 	int i, res;
437 
438 	bond_for_each_slave(bond, slave, i) {
439 		res = vlan_vid_add(slave->dev, proto, vid);
440 		if (res)
441 			goto unwind;
442 	}
443 
444 	res = bond_add_vlan(bond, vid);
445 	if (res) {
446 		pr_err("%s: Error: Failed to add vlan id %d\n",
447 		       bond_dev->name, vid);
448 		return res;
449 	}
450 
451 	return 0;
452 
453 unwind:
454 	/* unwind from head to the slave that failed */
455 	stop_at = slave;
456 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
457 		vlan_vid_del(slave->dev, proto, vid);
458 
459 	return res;
460 }
461 
462 /**
463  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464  * @bond_dev: bonding net device that got called
465  * @vid: vlan id being removed
466  */
467 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
468 				 __be16 proto, u16 vid)
469 {
470 	struct bonding *bond = netdev_priv(bond_dev);
471 	struct slave *slave;
472 	int i, res;
473 
474 	bond_for_each_slave(bond, slave, i)
475 		vlan_vid_del(slave->dev, proto, vid);
476 
477 	res = bond_del_vlan(bond, vid);
478 	if (res) {
479 		pr_err("%s: Error: Failed to remove vlan id %d\n",
480 		       bond_dev->name, vid);
481 		return res;
482 	}
483 
484 	return 0;
485 }
486 
487 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
488 {
489 	struct vlan_entry *vlan;
490 	int res;
491 
492 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
493 		res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
494 				   vlan->vlan_id);
495 		if (res)
496 			pr_warning("%s: Failed to add vlan id %d to device %s\n",
497 				   bond->dev->name, vlan->vlan_id,
498 				   slave_dev->name);
499 	}
500 }
501 
502 static void bond_del_vlans_from_slave(struct bonding *bond,
503 				      struct net_device *slave_dev)
504 {
505 	struct vlan_entry *vlan;
506 
507 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
508 		if (!vlan->vlan_id)
509 			continue;
510 		vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
511 	}
512 }
513 
514 /*------------------------------- Link status -------------------------------*/
515 
516 /*
517  * Set the carrier state for the master according to the state of its
518  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
519  * do special 802.3ad magic.
520  *
521  * Returns zero if carrier state does not change, nonzero if it does.
522  */
523 static int bond_set_carrier(struct bonding *bond)
524 {
525 	struct slave *slave;
526 	int i;
527 
528 	if (bond->slave_cnt == 0)
529 		goto down;
530 
531 	if (bond->params.mode == BOND_MODE_8023AD)
532 		return bond_3ad_set_carrier(bond);
533 
534 	bond_for_each_slave(bond, slave, i) {
535 		if (slave->link == BOND_LINK_UP) {
536 			if (!netif_carrier_ok(bond->dev)) {
537 				netif_carrier_on(bond->dev);
538 				return 1;
539 			}
540 			return 0;
541 		}
542 	}
543 
544 down:
545 	if (netif_carrier_ok(bond->dev)) {
546 		netif_carrier_off(bond->dev);
547 		return 1;
548 	}
549 	return 0;
550 }
551 
552 /*
553  * Get link speed and duplex from the slave's base driver
554  * using ethtool. If for some reason the call fails or the
555  * values are invalid, set speed and duplex to -1,
556  * and return.
557  */
558 static void bond_update_speed_duplex(struct slave *slave)
559 {
560 	struct net_device *slave_dev = slave->dev;
561 	struct ethtool_cmd ecmd;
562 	u32 slave_speed;
563 	int res;
564 
565 	slave->speed = SPEED_UNKNOWN;
566 	slave->duplex = DUPLEX_UNKNOWN;
567 
568 	res = __ethtool_get_settings(slave_dev, &ecmd);
569 	if (res < 0)
570 		return;
571 
572 	slave_speed = ethtool_cmd_speed(&ecmd);
573 	if (slave_speed == 0 || slave_speed == ((__u32) -1))
574 		return;
575 
576 	switch (ecmd.duplex) {
577 	case DUPLEX_FULL:
578 	case DUPLEX_HALF:
579 		break;
580 	default:
581 		return;
582 	}
583 
584 	slave->speed = slave_speed;
585 	slave->duplex = ecmd.duplex;
586 
587 	return;
588 }
589 
590 /*
591  * if <dev> supports MII link status reporting, check its link status.
592  *
593  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
594  * depending upon the setting of the use_carrier parameter.
595  *
596  * Return either BMSR_LSTATUS, meaning that the link is up (or we
597  * can't tell and just pretend it is), or 0, meaning that the link is
598  * down.
599  *
600  * If reporting is non-zero, instead of faking link up, return -1 if
601  * both ETHTOOL and MII ioctls fail (meaning the device does not
602  * support them).  If use_carrier is set, return whatever it says.
603  * It'd be nice if there was a good way to tell if a driver supports
604  * netif_carrier, but there really isn't.
605  */
606 static int bond_check_dev_link(struct bonding *bond,
607 			       struct net_device *slave_dev, int reporting)
608 {
609 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
610 	int (*ioctl)(struct net_device *, struct ifreq *, int);
611 	struct ifreq ifr;
612 	struct mii_ioctl_data *mii;
613 
614 	if (!reporting && !netif_running(slave_dev))
615 		return 0;
616 
617 	if (bond->params.use_carrier)
618 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
619 
620 	/* Try to get link status using Ethtool first. */
621 	if (slave_dev->ethtool_ops->get_link)
622 		return slave_dev->ethtool_ops->get_link(slave_dev) ?
623 			BMSR_LSTATUS : 0;
624 
625 	/* Ethtool can't be used, fallback to MII ioctls. */
626 	ioctl = slave_ops->ndo_do_ioctl;
627 	if (ioctl) {
628 		/* TODO: set pointer to correct ioctl on a per team member */
629 		/*       bases to make this more efficient. that is, once  */
630 		/*       we determine the correct ioctl, we will always    */
631 		/*       call it and not the others for that team          */
632 		/*       member.                                           */
633 
634 		/*
635 		 * We cannot assume that SIOCGMIIPHY will also read a
636 		 * register; not all network drivers (e.g., e100)
637 		 * support that.
638 		 */
639 
640 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
641 		strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
642 		mii = if_mii(&ifr);
643 		if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
644 			mii->reg_num = MII_BMSR;
645 			if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
646 				return mii->val_out & BMSR_LSTATUS;
647 		}
648 	}
649 
650 	/*
651 	 * If reporting, report that either there's no dev->do_ioctl,
652 	 * or both SIOCGMIIREG and get_link failed (meaning that we
653 	 * cannot report link status).  If not reporting, pretend
654 	 * we're ok.
655 	 */
656 	return reporting ? -1 : BMSR_LSTATUS;
657 }
658 
659 /*----------------------------- Multicast list ------------------------------*/
660 
661 /*
662  * Push the promiscuity flag down to appropriate slaves
663  */
664 static int bond_set_promiscuity(struct bonding *bond, int inc)
665 {
666 	int err = 0;
667 	if (USES_PRIMARY(bond->params.mode)) {
668 		/* write lock already acquired */
669 		if (bond->curr_active_slave) {
670 			err = dev_set_promiscuity(bond->curr_active_slave->dev,
671 						  inc);
672 		}
673 	} else {
674 		struct slave *slave;
675 		int i;
676 		bond_for_each_slave(bond, slave, i) {
677 			err = dev_set_promiscuity(slave->dev, inc);
678 			if (err)
679 				return err;
680 		}
681 	}
682 	return err;
683 }
684 
685 /*
686  * Push the allmulti flag down to all slaves
687  */
688 static int bond_set_allmulti(struct bonding *bond, int inc)
689 {
690 	int err = 0;
691 	if (USES_PRIMARY(bond->params.mode)) {
692 		/* write lock already acquired */
693 		if (bond->curr_active_slave) {
694 			err = dev_set_allmulti(bond->curr_active_slave->dev,
695 					       inc);
696 		}
697 	} else {
698 		struct slave *slave;
699 		int i;
700 		bond_for_each_slave(bond, slave, i) {
701 			err = dev_set_allmulti(slave->dev, inc);
702 			if (err)
703 				return err;
704 		}
705 	}
706 	return err;
707 }
708 
709 /*
710  * Add a Multicast address to slaves
711  * according to mode
712  */
713 static void bond_mc_add(struct bonding *bond, void *addr)
714 {
715 	if (USES_PRIMARY(bond->params.mode)) {
716 		/* write lock already acquired */
717 		if (bond->curr_active_slave)
718 			dev_mc_add(bond->curr_active_slave->dev, addr);
719 	} else {
720 		struct slave *slave;
721 		int i;
722 
723 		bond_for_each_slave(bond, slave, i)
724 			dev_mc_add(slave->dev, addr);
725 	}
726 }
727 
728 /*
729  * Remove a multicast address from slave
730  * according to mode
731  */
732 static void bond_mc_del(struct bonding *bond, void *addr)
733 {
734 	if (USES_PRIMARY(bond->params.mode)) {
735 		/* write lock already acquired */
736 		if (bond->curr_active_slave)
737 			dev_mc_del(bond->curr_active_slave->dev, addr);
738 	} else {
739 		struct slave *slave;
740 		int i;
741 		bond_for_each_slave(bond, slave, i) {
742 			dev_mc_del(slave->dev, addr);
743 		}
744 	}
745 }
746 
747 
748 static void __bond_resend_igmp_join_requests(struct net_device *dev)
749 {
750 	struct in_device *in_dev;
751 
752 	in_dev = __in_dev_get_rcu(dev);
753 	if (in_dev)
754 		ip_mc_rejoin_groups(in_dev);
755 }
756 
757 /*
758  * Retrieve the list of registered multicast addresses for the bonding
759  * device and retransmit an IGMP JOIN request to the current active
760  * slave.
761  */
762 static void bond_resend_igmp_join_requests(struct bonding *bond)
763 {
764 	struct net_device *bond_dev, *vlan_dev, *upper_dev;
765 	struct vlan_entry *vlan;
766 
767 	rcu_read_lock();
768 	read_lock(&bond->lock);
769 
770 	bond_dev = bond->dev;
771 
772 	/* rejoin all groups on bond device */
773 	__bond_resend_igmp_join_requests(bond_dev);
774 
775 	/*
776 	 * if bond is enslaved to a bridge,
777 	 * then rejoin all groups on its master
778 	 */
779 	upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
780 	if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
781 		__bond_resend_igmp_join_requests(upper_dev);
782 
783 	/* rejoin all groups on vlan devices */
784 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
785 		vlan_dev = __vlan_find_dev_deep(bond_dev, htons(ETH_P_8021Q),
786 						vlan->vlan_id);
787 		if (vlan_dev)
788 			__bond_resend_igmp_join_requests(vlan_dev);
789 	}
790 
791 	if (--bond->igmp_retrans > 0)
792 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
793 
794 	read_unlock(&bond->lock);
795 	rcu_read_unlock();
796 }
797 
798 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
799 {
800 	struct bonding *bond = container_of(work, struct bonding,
801 					    mcast_work.work);
802 
803 	bond_resend_igmp_join_requests(bond);
804 }
805 
806 /*
807  * flush all members of flush->mc_list from device dev->mc_list
808  */
809 static void bond_mc_list_flush(struct net_device *bond_dev,
810 			       struct net_device *slave_dev)
811 {
812 	struct bonding *bond = netdev_priv(bond_dev);
813 	struct netdev_hw_addr *ha;
814 
815 	netdev_for_each_mc_addr(ha, bond_dev)
816 		dev_mc_del(slave_dev, ha->addr);
817 
818 	if (bond->params.mode == BOND_MODE_8023AD) {
819 		/* del lacpdu mc addr from mc list */
820 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
821 
822 		dev_mc_del(slave_dev, lacpdu_multicast);
823 	}
824 }
825 
826 /*--------------------------- Active slave change ---------------------------*/
827 
828 /*
829  * Update the mc list and multicast-related flags for the new and
830  * old active slaves (if any) according to the multicast mode, and
831  * promiscuous flags unconditionally.
832  */
833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
834 			 struct slave *old_active)
835 {
836 	struct netdev_hw_addr *ha;
837 
838 	if (!USES_PRIMARY(bond->params.mode))
839 		/* nothing to do -  mc list is already up-to-date on
840 		 * all slaves
841 		 */
842 		return;
843 
844 	if (old_active) {
845 		if (bond->dev->flags & IFF_PROMISC)
846 			dev_set_promiscuity(old_active->dev, -1);
847 
848 		if (bond->dev->flags & IFF_ALLMULTI)
849 			dev_set_allmulti(old_active->dev, -1);
850 
851 		netif_addr_lock_bh(bond->dev);
852 		netdev_for_each_mc_addr(ha, bond->dev)
853 			dev_mc_del(old_active->dev, ha->addr);
854 		netif_addr_unlock_bh(bond->dev);
855 	}
856 
857 	if (new_active) {
858 		/* FIXME: Signal errors upstream. */
859 		if (bond->dev->flags & IFF_PROMISC)
860 			dev_set_promiscuity(new_active->dev, 1);
861 
862 		if (bond->dev->flags & IFF_ALLMULTI)
863 			dev_set_allmulti(new_active->dev, 1);
864 
865 		netif_addr_lock_bh(bond->dev);
866 		netdev_for_each_mc_addr(ha, bond->dev)
867 			dev_mc_add(new_active->dev, ha->addr);
868 		netif_addr_unlock_bh(bond->dev);
869 	}
870 }
871 
872 /*
873  * bond_do_fail_over_mac
874  *
875  * Perform special MAC address swapping for fail_over_mac settings
876  *
877  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
878  */
879 static void bond_do_fail_over_mac(struct bonding *bond,
880 				  struct slave *new_active,
881 				  struct slave *old_active)
882 	__releases(&bond->curr_slave_lock)
883 	__releases(&bond->lock)
884 	__acquires(&bond->lock)
885 	__acquires(&bond->curr_slave_lock)
886 {
887 	u8 tmp_mac[ETH_ALEN];
888 	struct sockaddr saddr;
889 	int rv;
890 
891 	switch (bond->params.fail_over_mac) {
892 	case BOND_FOM_ACTIVE:
893 		if (new_active) {
894 			memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
895 			       new_active->dev->addr_len);
896 			write_unlock_bh(&bond->curr_slave_lock);
897 			read_unlock(&bond->lock);
898 			call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
899 			read_lock(&bond->lock);
900 			write_lock_bh(&bond->curr_slave_lock);
901 		}
902 		break;
903 	case BOND_FOM_FOLLOW:
904 		/*
905 		 * if new_active && old_active, swap them
906 		 * if just old_active, do nothing (going to no active slave)
907 		 * if just new_active, set new_active to bond's MAC
908 		 */
909 		if (!new_active)
910 			return;
911 
912 		write_unlock_bh(&bond->curr_slave_lock);
913 		read_unlock(&bond->lock);
914 
915 		if (old_active) {
916 			memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
917 			memcpy(saddr.sa_data, old_active->dev->dev_addr,
918 			       ETH_ALEN);
919 			saddr.sa_family = new_active->dev->type;
920 		} else {
921 			memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
922 			saddr.sa_family = bond->dev->type;
923 		}
924 
925 		rv = dev_set_mac_address(new_active->dev, &saddr);
926 		if (rv) {
927 			pr_err("%s: Error %d setting MAC of slave %s\n",
928 			       bond->dev->name, -rv, new_active->dev->name);
929 			goto out;
930 		}
931 
932 		if (!old_active)
933 			goto out;
934 
935 		memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
936 		saddr.sa_family = old_active->dev->type;
937 
938 		rv = dev_set_mac_address(old_active->dev, &saddr);
939 		if (rv)
940 			pr_err("%s: Error %d setting MAC of slave %s\n",
941 			       bond->dev->name, -rv, new_active->dev->name);
942 out:
943 		read_lock(&bond->lock);
944 		write_lock_bh(&bond->curr_slave_lock);
945 		break;
946 	default:
947 		pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
948 		       bond->dev->name, bond->params.fail_over_mac);
949 		break;
950 	}
951 
952 }
953 
954 static bool bond_should_change_active(struct bonding *bond)
955 {
956 	struct slave *prim = bond->primary_slave;
957 	struct slave *curr = bond->curr_active_slave;
958 
959 	if (!prim || !curr || curr->link != BOND_LINK_UP)
960 		return true;
961 	if (bond->force_primary) {
962 		bond->force_primary = false;
963 		return true;
964 	}
965 	if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
966 	    (prim->speed < curr->speed ||
967 	     (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
968 		return false;
969 	if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
970 		return false;
971 	return true;
972 }
973 
974 /**
975  * find_best_interface - select the best available slave to be the active one
976  * @bond: our bonding struct
977  *
978  * Warning: Caller must hold curr_slave_lock for writing.
979  */
980 static struct slave *bond_find_best_slave(struct bonding *bond)
981 {
982 	struct slave *new_active, *old_active;
983 	struct slave *bestslave = NULL;
984 	int mintime = bond->params.updelay;
985 	int i;
986 
987 	new_active = bond->curr_active_slave;
988 
989 	if (!new_active) { /* there were no active slaves left */
990 		if (bond->slave_cnt > 0)   /* found one slave */
991 			new_active = bond->first_slave;
992 		else
993 			return NULL; /* still no slave, return NULL */
994 	}
995 
996 	if ((bond->primary_slave) &&
997 	    bond->primary_slave->link == BOND_LINK_UP &&
998 	    bond_should_change_active(bond)) {
999 		new_active = bond->primary_slave;
1000 	}
1001 
1002 	/* remember where to stop iterating over the slaves */
1003 	old_active = new_active;
1004 
1005 	bond_for_each_slave_from(bond, new_active, i, old_active) {
1006 		if (new_active->link == BOND_LINK_UP) {
1007 			return new_active;
1008 		} else if (new_active->link == BOND_LINK_BACK &&
1009 			   IS_UP(new_active->dev)) {
1010 			/* link up, but waiting for stabilization */
1011 			if (new_active->delay < mintime) {
1012 				mintime = new_active->delay;
1013 				bestslave = new_active;
1014 			}
1015 		}
1016 	}
1017 
1018 	return bestslave;
1019 }
1020 
1021 static bool bond_should_notify_peers(struct bonding *bond)
1022 {
1023 	struct slave *slave = bond->curr_active_slave;
1024 
1025 	pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1026 		 bond->dev->name, slave ? slave->dev->name : "NULL");
1027 
1028 	if (!slave || !bond->send_peer_notif ||
1029 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1030 		return false;
1031 
1032 	bond->send_peer_notif--;
1033 	return true;
1034 }
1035 
1036 /**
1037  * change_active_interface - change the active slave into the specified one
1038  * @bond: our bonding struct
1039  * @new: the new slave to make the active one
1040  *
1041  * Set the new slave to the bond's settings and unset them on the old
1042  * curr_active_slave.
1043  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1044  *
1045  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1046  * because it is apparently the best available slave we have, even though its
1047  * updelay hasn't timed out yet.
1048  *
1049  * If new_active is not NULL, caller must hold bond->lock for read and
1050  * curr_slave_lock for write_bh.
1051  */
1052 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1053 {
1054 	struct slave *old_active = bond->curr_active_slave;
1055 
1056 	if (old_active == new_active)
1057 		return;
1058 
1059 	if (new_active) {
1060 		new_active->jiffies = jiffies;
1061 
1062 		if (new_active->link == BOND_LINK_BACK) {
1063 			if (USES_PRIMARY(bond->params.mode)) {
1064 				pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1065 					bond->dev->name, new_active->dev->name,
1066 					(bond->params.updelay - new_active->delay) * bond->params.miimon);
1067 			}
1068 
1069 			new_active->delay = 0;
1070 			new_active->link = BOND_LINK_UP;
1071 
1072 			if (bond->params.mode == BOND_MODE_8023AD)
1073 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1074 
1075 			if (bond_is_lb(bond))
1076 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1077 		} else {
1078 			if (USES_PRIMARY(bond->params.mode)) {
1079 				pr_info("%s: making interface %s the new active one.\n",
1080 					bond->dev->name, new_active->dev->name);
1081 			}
1082 		}
1083 	}
1084 
1085 	if (USES_PRIMARY(bond->params.mode))
1086 		bond_mc_swap(bond, new_active, old_active);
1087 
1088 	if (bond_is_lb(bond)) {
1089 		bond_alb_handle_active_change(bond, new_active);
1090 		if (old_active)
1091 			bond_set_slave_inactive_flags(old_active);
1092 		if (new_active)
1093 			bond_set_slave_active_flags(new_active);
1094 	} else {
1095 		bond->curr_active_slave = new_active;
1096 	}
1097 
1098 	if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1099 		if (old_active)
1100 			bond_set_slave_inactive_flags(old_active);
1101 
1102 		if (new_active) {
1103 			bool should_notify_peers = false;
1104 
1105 			bond_set_slave_active_flags(new_active);
1106 
1107 			if (bond->params.fail_over_mac)
1108 				bond_do_fail_over_mac(bond, new_active,
1109 						      old_active);
1110 
1111 			if (netif_running(bond->dev)) {
1112 				bond->send_peer_notif =
1113 					bond->params.num_peer_notif;
1114 				should_notify_peers =
1115 					bond_should_notify_peers(bond);
1116 			}
1117 
1118 			write_unlock_bh(&bond->curr_slave_lock);
1119 			read_unlock(&bond->lock);
1120 
1121 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1122 			if (should_notify_peers)
1123 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1124 							 bond->dev);
1125 
1126 			read_lock(&bond->lock);
1127 			write_lock_bh(&bond->curr_slave_lock);
1128 		}
1129 	}
1130 
1131 	/* resend IGMP joins since active slave has changed or
1132 	 * all were sent on curr_active_slave.
1133 	 * resend only if bond is brought up with the affected
1134 	 * bonding modes and the retransmission is enabled */
1135 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1136 	    ((USES_PRIMARY(bond->params.mode) && new_active) ||
1137 	     bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1138 		bond->igmp_retrans = bond->params.resend_igmp;
1139 		queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1140 	}
1141 }
1142 
1143 /**
1144  * bond_select_active_slave - select a new active slave, if needed
1145  * @bond: our bonding struct
1146  *
1147  * This functions should be called when one of the following occurs:
1148  * - The old curr_active_slave has been released or lost its link.
1149  * - The primary_slave has got its link back.
1150  * - A slave has got its link back and there's no old curr_active_slave.
1151  *
1152  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1153  */
1154 void bond_select_active_slave(struct bonding *bond)
1155 {
1156 	struct slave *best_slave;
1157 	int rv;
1158 
1159 	best_slave = bond_find_best_slave(bond);
1160 	if (best_slave != bond->curr_active_slave) {
1161 		bond_change_active_slave(bond, best_slave);
1162 		rv = bond_set_carrier(bond);
1163 		if (!rv)
1164 			return;
1165 
1166 		if (netif_carrier_ok(bond->dev)) {
1167 			pr_info("%s: first active interface up!\n",
1168 				bond->dev->name);
1169 		} else {
1170 			pr_info("%s: now running without any active interface !\n",
1171 				bond->dev->name);
1172 		}
1173 	}
1174 }
1175 
1176 /*--------------------------- slave list handling ---------------------------*/
1177 
1178 /*
1179  * This function attaches the slave to the end of list.
1180  *
1181  * bond->lock held for writing by caller.
1182  */
1183 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1184 {
1185 	if (bond->first_slave == NULL) { /* attaching the first slave */
1186 		new_slave->next = new_slave;
1187 		new_slave->prev = new_slave;
1188 		bond->first_slave = new_slave;
1189 	} else {
1190 		new_slave->next = bond->first_slave;
1191 		new_slave->prev = bond->first_slave->prev;
1192 		new_slave->next->prev = new_slave;
1193 		new_slave->prev->next = new_slave;
1194 	}
1195 
1196 	bond->slave_cnt++;
1197 }
1198 
1199 /*
1200  * This function detaches the slave from the list.
1201  * WARNING: no check is made to verify if the slave effectively
1202  * belongs to <bond>.
1203  * Nothing is freed on return, structures are just unchained.
1204  * If any slave pointer in bond was pointing to <slave>,
1205  * it should be changed by the calling function.
1206  *
1207  * bond->lock held for writing by caller.
1208  */
1209 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1210 {
1211 	if (slave->next)
1212 		slave->next->prev = slave->prev;
1213 
1214 	if (slave->prev)
1215 		slave->prev->next = slave->next;
1216 
1217 	if (bond->first_slave == slave) { /* slave is the first slave */
1218 		if (bond->slave_cnt > 1) { /* there are more slave */
1219 			bond->first_slave = slave->next;
1220 		} else {
1221 			bond->first_slave = NULL; /* slave was the last one */
1222 		}
1223 	}
1224 
1225 	slave->next = NULL;
1226 	slave->prev = NULL;
1227 	bond->slave_cnt--;
1228 }
1229 
1230 #ifdef CONFIG_NET_POLL_CONTROLLER
1231 static inline int slave_enable_netpoll(struct slave *slave)
1232 {
1233 	struct netpoll *np;
1234 	int err = 0;
1235 
1236 	np = kzalloc(sizeof(*np), GFP_ATOMIC);
1237 	err = -ENOMEM;
1238 	if (!np)
1239 		goto out;
1240 
1241 	err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1242 	if (err) {
1243 		kfree(np);
1244 		goto out;
1245 	}
1246 	slave->np = np;
1247 out:
1248 	return err;
1249 }
1250 static inline void slave_disable_netpoll(struct slave *slave)
1251 {
1252 	struct netpoll *np = slave->np;
1253 
1254 	if (!np)
1255 		return;
1256 
1257 	slave->np = NULL;
1258 	__netpoll_free_async(np);
1259 }
1260 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1261 {
1262 	if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1263 		return false;
1264 	if (!slave_dev->netdev_ops->ndo_poll_controller)
1265 		return false;
1266 	return true;
1267 }
1268 
1269 static void bond_poll_controller(struct net_device *bond_dev)
1270 {
1271 }
1272 
1273 static void __bond_netpoll_cleanup(struct bonding *bond)
1274 {
1275 	struct slave *slave;
1276 	int i;
1277 
1278 	bond_for_each_slave(bond, slave, i)
1279 		if (IS_UP(slave->dev))
1280 			slave_disable_netpoll(slave);
1281 }
1282 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1283 {
1284 	struct bonding *bond = netdev_priv(bond_dev);
1285 
1286 	read_lock(&bond->lock);
1287 	__bond_netpoll_cleanup(bond);
1288 	read_unlock(&bond->lock);
1289 }
1290 
1291 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1292 {
1293 	struct bonding *bond = netdev_priv(dev);
1294 	struct slave *slave;
1295 	int i, err = 0;
1296 
1297 	read_lock(&bond->lock);
1298 	bond_for_each_slave(bond, slave, i) {
1299 		err = slave_enable_netpoll(slave);
1300 		if (err) {
1301 			__bond_netpoll_cleanup(bond);
1302 			break;
1303 		}
1304 	}
1305 	read_unlock(&bond->lock);
1306 	return err;
1307 }
1308 
1309 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1310 {
1311 	return bond->dev->npinfo;
1312 }
1313 
1314 #else
1315 static inline int slave_enable_netpoll(struct slave *slave)
1316 {
1317 	return 0;
1318 }
1319 static inline void slave_disable_netpoll(struct slave *slave)
1320 {
1321 }
1322 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1323 {
1324 }
1325 #endif
1326 
1327 /*---------------------------------- IOCTL ----------------------------------*/
1328 
1329 static void bond_set_dev_addr(struct net_device *bond_dev,
1330 			      struct net_device *slave_dev)
1331 {
1332 	pr_debug("bond_dev=%p\n", bond_dev);
1333 	pr_debug("slave_dev=%p\n", slave_dev);
1334 	pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1335 	memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1336 	bond_dev->addr_assign_type = NET_ADDR_SET;
1337 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1338 }
1339 
1340 static netdev_features_t bond_fix_features(struct net_device *dev,
1341 	netdev_features_t features)
1342 {
1343 	struct slave *slave;
1344 	struct bonding *bond = netdev_priv(dev);
1345 	netdev_features_t mask;
1346 	int i;
1347 
1348 	read_lock(&bond->lock);
1349 
1350 	if (!bond->first_slave) {
1351 		/* Disable adding VLANs to empty bond. But why? --mq */
1352 		features |= NETIF_F_VLAN_CHALLENGED;
1353 		goto out;
1354 	}
1355 
1356 	mask = features;
1357 	features &= ~NETIF_F_ONE_FOR_ALL;
1358 	features |= NETIF_F_ALL_FOR_ALL;
1359 
1360 	bond_for_each_slave(bond, slave, i) {
1361 		features = netdev_increment_features(features,
1362 						     slave->dev->features,
1363 						     mask);
1364 	}
1365 
1366 out:
1367 	read_unlock(&bond->lock);
1368 	return features;
1369 }
1370 
1371 #define BOND_VLAN_FEATURES	(NETIF_F_ALL_CSUM | NETIF_F_SG | \
1372 				 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1373 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1374 
1375 static void bond_compute_features(struct bonding *bond)
1376 {
1377 	struct slave *slave;
1378 	struct net_device *bond_dev = bond->dev;
1379 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1380 	unsigned short max_hard_header_len = ETH_HLEN;
1381 	unsigned int gso_max_size = GSO_MAX_SIZE;
1382 	u16 gso_max_segs = GSO_MAX_SEGS;
1383 	int i;
1384 	unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1385 
1386 	read_lock(&bond->lock);
1387 
1388 	if (!bond->first_slave)
1389 		goto done;
1390 
1391 	bond_for_each_slave(bond, slave, i) {
1392 		vlan_features = netdev_increment_features(vlan_features,
1393 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1394 
1395 		dst_release_flag &= slave->dev->priv_flags;
1396 		if (slave->dev->hard_header_len > max_hard_header_len)
1397 			max_hard_header_len = slave->dev->hard_header_len;
1398 
1399 		gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1400 		gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1401 	}
1402 
1403 done:
1404 	bond_dev->vlan_features = vlan_features;
1405 	bond_dev->hard_header_len = max_hard_header_len;
1406 	bond_dev->gso_max_segs = gso_max_segs;
1407 	netif_set_gso_max_size(bond_dev, gso_max_size);
1408 
1409 	flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1410 	bond_dev->priv_flags = flags | dst_release_flag;
1411 
1412 	read_unlock(&bond->lock);
1413 
1414 	netdev_change_features(bond_dev);
1415 }
1416 
1417 static void bond_setup_by_slave(struct net_device *bond_dev,
1418 				struct net_device *slave_dev)
1419 {
1420 	struct bonding *bond = netdev_priv(bond_dev);
1421 
1422 	bond_dev->header_ops	    = slave_dev->header_ops;
1423 
1424 	bond_dev->type		    = slave_dev->type;
1425 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1426 	bond_dev->addr_len	    = slave_dev->addr_len;
1427 
1428 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1429 		slave_dev->addr_len);
1430 	bond->setup_by_slave = 1;
1431 }
1432 
1433 /* On bonding slaves other than the currently active slave, suppress
1434  * duplicates except for alb non-mcast/bcast.
1435  */
1436 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1437 					    struct slave *slave,
1438 					    struct bonding *bond)
1439 {
1440 	if (bond_is_slave_inactive(slave)) {
1441 		if (bond->params.mode == BOND_MODE_ALB &&
1442 		    skb->pkt_type != PACKET_BROADCAST &&
1443 		    skb->pkt_type != PACKET_MULTICAST)
1444 			return false;
1445 		return true;
1446 	}
1447 	return false;
1448 }
1449 
1450 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1451 {
1452 	struct sk_buff *skb = *pskb;
1453 	struct slave *slave;
1454 	struct bonding *bond;
1455 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1456 			  struct slave *);
1457 	int ret = RX_HANDLER_ANOTHER;
1458 
1459 	skb = skb_share_check(skb, GFP_ATOMIC);
1460 	if (unlikely(!skb))
1461 		return RX_HANDLER_CONSUMED;
1462 
1463 	*pskb = skb;
1464 
1465 	slave = bond_slave_get_rcu(skb->dev);
1466 	bond = slave->bond;
1467 
1468 	if (bond->params.arp_interval)
1469 		slave->dev->last_rx = jiffies;
1470 
1471 	recv_probe = ACCESS_ONCE(bond->recv_probe);
1472 	if (recv_probe) {
1473 		ret = recv_probe(skb, bond, slave);
1474 		if (ret == RX_HANDLER_CONSUMED) {
1475 			consume_skb(skb);
1476 			return ret;
1477 		}
1478 	}
1479 
1480 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1481 		return RX_HANDLER_EXACT;
1482 	}
1483 
1484 	skb->dev = bond->dev;
1485 
1486 	if (bond->params.mode == BOND_MODE_ALB &&
1487 	    bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1488 	    skb->pkt_type == PACKET_HOST) {
1489 
1490 		if (unlikely(skb_cow_head(skb,
1491 					  skb->data - skb_mac_header(skb)))) {
1492 			kfree_skb(skb);
1493 			return RX_HANDLER_CONSUMED;
1494 		}
1495 		memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1496 	}
1497 
1498 	return ret;
1499 }
1500 
1501 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1502 				      struct net_device *slave_dev)
1503 {
1504 	int err;
1505 
1506 	err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1507 	if (err)
1508 		return err;
1509 	slave_dev->flags |= IFF_SLAVE;
1510 	rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1511 	return 0;
1512 }
1513 
1514 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1515 				  struct net_device *slave_dev)
1516 {
1517 	netdev_upper_dev_unlink(slave_dev, bond_dev);
1518 	slave_dev->flags &= ~IFF_SLAVE;
1519 	rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1520 }
1521 
1522 /* enslave device <slave> to bond device <master> */
1523 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1524 {
1525 	struct bonding *bond = netdev_priv(bond_dev);
1526 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1527 	struct slave *new_slave = NULL;
1528 	struct netdev_hw_addr *ha;
1529 	struct sockaddr addr;
1530 	int link_reporting;
1531 	int res = 0;
1532 
1533 	if (!bond->params.use_carrier &&
1534 	    slave_dev->ethtool_ops->get_link == NULL &&
1535 	    slave_ops->ndo_do_ioctl == NULL) {
1536 		pr_warning("%s: Warning: no link monitoring support for %s\n",
1537 			   bond_dev->name, slave_dev->name);
1538 	}
1539 
1540 	/* already enslaved */
1541 	if (slave_dev->flags & IFF_SLAVE) {
1542 		pr_debug("Error, Device was already enslaved\n");
1543 		return -EBUSY;
1544 	}
1545 
1546 	/* vlan challenged mutual exclusion */
1547 	/* no need to lock since we're protected by rtnl_lock */
1548 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1549 		pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1550 		if (vlan_uses_dev(bond_dev)) {
1551 			pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1552 			       bond_dev->name, slave_dev->name, bond_dev->name);
1553 			return -EPERM;
1554 		} else {
1555 			pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1556 				   bond_dev->name, slave_dev->name,
1557 				   slave_dev->name, bond_dev->name);
1558 		}
1559 	} else {
1560 		pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1561 	}
1562 
1563 	/*
1564 	 * Old ifenslave binaries are no longer supported.  These can
1565 	 * be identified with moderate accuracy by the state of the slave:
1566 	 * the current ifenslave will set the interface down prior to
1567 	 * enslaving it; the old ifenslave will not.
1568 	 */
1569 	if ((slave_dev->flags & IFF_UP)) {
1570 		pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1571 		       slave_dev->name);
1572 		res = -EPERM;
1573 		goto err_undo_flags;
1574 	}
1575 
1576 	/* set bonding device ether type by slave - bonding netdevices are
1577 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1578 	 * there is a need to override some of the type dependent attribs/funcs.
1579 	 *
1580 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1581 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1582 	 */
1583 	if (bond->slave_cnt == 0) {
1584 		if (bond_dev->type != slave_dev->type) {
1585 			pr_debug("%s: change device type from %d to %d\n",
1586 				 bond_dev->name,
1587 				 bond_dev->type, slave_dev->type);
1588 
1589 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1590 						       bond_dev);
1591 			res = notifier_to_errno(res);
1592 			if (res) {
1593 				pr_err("%s: refused to change device type\n",
1594 				       bond_dev->name);
1595 				res = -EBUSY;
1596 				goto err_undo_flags;
1597 			}
1598 
1599 			/* Flush unicast and multicast addresses */
1600 			dev_uc_flush(bond_dev);
1601 			dev_mc_flush(bond_dev);
1602 
1603 			if (slave_dev->type != ARPHRD_ETHER)
1604 				bond_setup_by_slave(bond_dev, slave_dev);
1605 			else {
1606 				ether_setup(bond_dev);
1607 				bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1608 			}
1609 
1610 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1611 						 bond_dev);
1612 		}
1613 	} else if (bond_dev->type != slave_dev->type) {
1614 		pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1615 		       slave_dev->name,
1616 		       slave_dev->type, bond_dev->type);
1617 		res = -EINVAL;
1618 		goto err_undo_flags;
1619 	}
1620 
1621 	if (slave_ops->ndo_set_mac_address == NULL) {
1622 		if (bond->slave_cnt == 0) {
1623 			pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1624 				   bond_dev->name);
1625 			bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1626 		} else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1627 			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",
1628 			       bond_dev->name);
1629 			res = -EOPNOTSUPP;
1630 			goto err_undo_flags;
1631 		}
1632 	}
1633 
1634 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1635 
1636 	/* If this is the first slave, then we need to set the master's hardware
1637 	 * address to be the same as the slave's. */
1638 	if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1639 		bond_set_dev_addr(bond->dev, slave_dev);
1640 
1641 	new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1642 	if (!new_slave) {
1643 		res = -ENOMEM;
1644 		goto err_undo_flags;
1645 	}
1646 
1647 	/*
1648 	 * Set the new_slave's queue_id to be zero.  Queue ID mapping
1649 	 * is set via sysfs or module option if desired.
1650 	 */
1651 	new_slave->queue_id = 0;
1652 
1653 	/* Save slave's original mtu and then set it to match the bond */
1654 	new_slave->original_mtu = slave_dev->mtu;
1655 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1656 	if (res) {
1657 		pr_debug("Error %d calling dev_set_mtu\n", res);
1658 		goto err_free;
1659 	}
1660 
1661 	/*
1662 	 * Save slave's original ("permanent") mac address for modes
1663 	 * that need it, and for restoring it upon release, and then
1664 	 * set it to the master's address
1665 	 */
1666 	memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1667 
1668 	if (!bond->params.fail_over_mac) {
1669 		/*
1670 		 * Set slave to master's mac address.  The application already
1671 		 * set the master's mac address to that of the first slave
1672 		 */
1673 		memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1674 		addr.sa_family = slave_dev->type;
1675 		res = dev_set_mac_address(slave_dev, &addr);
1676 		if (res) {
1677 			pr_debug("Error %d calling set_mac_address\n", res);
1678 			goto err_restore_mtu;
1679 		}
1680 	}
1681 
1682 	res = bond_master_upper_dev_link(bond_dev, slave_dev);
1683 	if (res) {
1684 		pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1685 		goto err_restore_mac;
1686 	}
1687 
1688 	/* open the slave since the application closed it */
1689 	res = dev_open(slave_dev);
1690 	if (res) {
1691 		pr_debug("Opening slave %s failed\n", slave_dev->name);
1692 		goto err_unset_master;
1693 	}
1694 
1695 	new_slave->bond = bond;
1696 	new_slave->dev = slave_dev;
1697 	slave_dev->priv_flags |= IFF_BONDING;
1698 
1699 	if (bond_is_lb(bond)) {
1700 		/* bond_alb_init_slave() must be called before all other stages since
1701 		 * it might fail and we do not want to have to undo everything
1702 		 */
1703 		res = bond_alb_init_slave(bond, new_slave);
1704 		if (res)
1705 			goto err_close;
1706 	}
1707 
1708 	/* If the mode USES_PRIMARY, then the new slave gets the
1709 	 * master's promisc (and mc) settings only if it becomes the
1710 	 * curr_active_slave, and that is taken care of later when calling
1711 	 * bond_change_active()
1712 	 */
1713 	if (!USES_PRIMARY(bond->params.mode)) {
1714 		/* set promiscuity level to new slave */
1715 		if (bond_dev->flags & IFF_PROMISC) {
1716 			res = dev_set_promiscuity(slave_dev, 1);
1717 			if (res)
1718 				goto err_close;
1719 		}
1720 
1721 		/* set allmulti level to new slave */
1722 		if (bond_dev->flags & IFF_ALLMULTI) {
1723 			res = dev_set_allmulti(slave_dev, 1);
1724 			if (res)
1725 				goto err_close;
1726 		}
1727 
1728 		netif_addr_lock_bh(bond_dev);
1729 		/* upload master's mc_list to new slave */
1730 		netdev_for_each_mc_addr(ha, bond_dev)
1731 			dev_mc_add(slave_dev, ha->addr);
1732 		netif_addr_unlock_bh(bond_dev);
1733 	}
1734 
1735 	if (bond->params.mode == BOND_MODE_8023AD) {
1736 		/* add lacpdu mc addr to mc list */
1737 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1738 
1739 		dev_mc_add(slave_dev, lacpdu_multicast);
1740 	}
1741 
1742 	bond_add_vlans_on_slave(bond, slave_dev);
1743 
1744 	write_lock_bh(&bond->lock);
1745 
1746 	bond_attach_slave(bond, new_slave);
1747 
1748 	new_slave->delay = 0;
1749 	new_slave->link_failure_count = 0;
1750 
1751 	write_unlock_bh(&bond->lock);
1752 
1753 	bond_compute_features(bond);
1754 
1755 	bond_update_speed_duplex(new_slave);
1756 
1757 	read_lock(&bond->lock);
1758 
1759 	new_slave->last_arp_rx = jiffies -
1760 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
1761 
1762 	if (bond->params.miimon && !bond->params.use_carrier) {
1763 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1764 
1765 		if ((link_reporting == -1) && !bond->params.arp_interval) {
1766 			/*
1767 			 * miimon is set but a bonded network driver
1768 			 * does not support ETHTOOL/MII and
1769 			 * arp_interval is not set.  Note: if
1770 			 * use_carrier is enabled, we will never go
1771 			 * here (because netif_carrier is always
1772 			 * supported); thus, we don't need to change
1773 			 * the messages for netif_carrier.
1774 			 */
1775 			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",
1776 			       bond_dev->name, slave_dev->name);
1777 		} else if (link_reporting == -1) {
1778 			/* unable get link status using mii/ethtool */
1779 			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",
1780 				   bond_dev->name, slave_dev->name);
1781 		}
1782 	}
1783 
1784 	/* check for initial state */
1785 	if (bond->params.miimon) {
1786 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1787 			if (bond->params.updelay) {
1788 				new_slave->link = BOND_LINK_BACK;
1789 				new_slave->delay = bond->params.updelay;
1790 			} else {
1791 				new_slave->link = BOND_LINK_UP;
1792 			}
1793 		} else {
1794 			new_slave->link = BOND_LINK_DOWN;
1795 		}
1796 	} else if (bond->params.arp_interval) {
1797 		new_slave->link = (netif_carrier_ok(slave_dev) ?
1798 			BOND_LINK_UP : BOND_LINK_DOWN);
1799 	} else {
1800 		new_slave->link = BOND_LINK_UP;
1801 	}
1802 
1803 	if (new_slave->link != BOND_LINK_DOWN)
1804 		new_slave->jiffies = jiffies;
1805 	pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1806 		new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1807 			(new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1808 
1809 	if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1810 		/* if there is a primary slave, remember it */
1811 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1812 			bond->primary_slave = new_slave;
1813 			bond->force_primary = true;
1814 		}
1815 	}
1816 
1817 	write_lock_bh(&bond->curr_slave_lock);
1818 
1819 	switch (bond->params.mode) {
1820 	case BOND_MODE_ACTIVEBACKUP:
1821 		bond_set_slave_inactive_flags(new_slave);
1822 		bond_select_active_slave(bond);
1823 		break;
1824 	case BOND_MODE_8023AD:
1825 		/* in 802.3ad mode, the internal mechanism
1826 		 * will activate the slaves in the selected
1827 		 * aggregator
1828 		 */
1829 		bond_set_slave_inactive_flags(new_slave);
1830 		/* if this is the first slave */
1831 		if (bond->slave_cnt == 1) {
1832 			SLAVE_AD_INFO(new_slave).id = 1;
1833 			/* Initialize AD with the number of times that the AD timer is called in 1 second
1834 			 * can be called only after the mac address of the bond is set
1835 			 */
1836 			bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1837 		} else {
1838 			SLAVE_AD_INFO(new_slave).id =
1839 				SLAVE_AD_INFO(new_slave->prev).id + 1;
1840 		}
1841 
1842 		bond_3ad_bind_slave(new_slave);
1843 		break;
1844 	case BOND_MODE_TLB:
1845 	case BOND_MODE_ALB:
1846 		bond_set_active_slave(new_slave);
1847 		bond_set_slave_inactive_flags(new_slave);
1848 		bond_select_active_slave(bond);
1849 		break;
1850 	default:
1851 		pr_debug("This slave is always active in trunk mode\n");
1852 
1853 		/* always active in trunk mode */
1854 		bond_set_active_slave(new_slave);
1855 
1856 		/* In trunking mode there is little meaning to curr_active_slave
1857 		 * anyway (it holds no special properties of the bond device),
1858 		 * so we can change it without calling change_active_interface()
1859 		 */
1860 		if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1861 			bond->curr_active_slave = new_slave;
1862 
1863 		break;
1864 	} /* switch(bond_mode) */
1865 
1866 	write_unlock_bh(&bond->curr_slave_lock);
1867 
1868 	bond_set_carrier(bond);
1869 
1870 #ifdef CONFIG_NET_POLL_CONTROLLER
1871 	slave_dev->npinfo = bond_netpoll_info(bond);
1872 	if (slave_dev->npinfo) {
1873 		if (slave_enable_netpoll(new_slave)) {
1874 			read_unlock(&bond->lock);
1875 			pr_info("Error, %s: master_dev is using netpoll, "
1876 				 "but new slave device does not support netpoll.\n",
1877 				 bond_dev->name);
1878 			res = -EBUSY;
1879 			goto err_detach;
1880 		}
1881 	}
1882 #endif
1883 
1884 	read_unlock(&bond->lock);
1885 
1886 	res = bond_create_slave_symlinks(bond_dev, slave_dev);
1887 	if (res)
1888 		goto err_detach;
1889 
1890 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1891 					 new_slave);
1892 	if (res) {
1893 		pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1894 		goto err_dest_symlinks;
1895 	}
1896 
1897 	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1898 		bond_dev->name, slave_dev->name,
1899 		bond_is_active_slave(new_slave) ? "n active" : " backup",
1900 		new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1901 
1902 	/* enslave is successful */
1903 	return 0;
1904 
1905 /* Undo stages on error */
1906 err_dest_symlinks:
1907 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
1908 
1909 err_detach:
1910 	if (!USES_PRIMARY(bond->params.mode)) {
1911 		netif_addr_lock_bh(bond_dev);
1912 		bond_mc_list_flush(bond_dev, slave_dev);
1913 		netif_addr_unlock_bh(bond_dev);
1914 	}
1915 	bond_del_vlans_from_slave(bond, slave_dev);
1916 	write_lock_bh(&bond->lock);
1917 	bond_detach_slave(bond, new_slave);
1918 	if (bond->primary_slave == new_slave)
1919 		bond->primary_slave = NULL;
1920 	if (bond->curr_active_slave == new_slave) {
1921 		bond_change_active_slave(bond, NULL);
1922 		write_unlock_bh(&bond->lock);
1923 		read_lock(&bond->lock);
1924 		write_lock_bh(&bond->curr_slave_lock);
1925 		bond_select_active_slave(bond);
1926 		write_unlock_bh(&bond->curr_slave_lock);
1927 		read_unlock(&bond->lock);
1928 	} else {
1929 		write_unlock_bh(&bond->lock);
1930 	}
1931 	slave_disable_netpoll(new_slave);
1932 
1933 err_close:
1934 	slave_dev->priv_flags &= ~IFF_BONDING;
1935 	dev_close(slave_dev);
1936 
1937 err_unset_master:
1938 	bond_upper_dev_unlink(bond_dev, slave_dev);
1939 
1940 err_restore_mac:
1941 	if (!bond->params.fail_over_mac) {
1942 		/* XXX TODO - fom follow mode needs to change master's
1943 		 * MAC if this slave's MAC is in use by the bond, or at
1944 		 * least print a warning.
1945 		 */
1946 		memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1947 		addr.sa_family = slave_dev->type;
1948 		dev_set_mac_address(slave_dev, &addr);
1949 	}
1950 
1951 err_restore_mtu:
1952 	dev_set_mtu(slave_dev, new_slave->original_mtu);
1953 
1954 err_free:
1955 	kfree(new_slave);
1956 
1957 err_undo_flags:
1958 	bond_compute_features(bond);
1959 
1960 	return res;
1961 }
1962 
1963 /*
1964  * Try to release the slave device <slave> from the bond device <master>
1965  * It is legal to access curr_active_slave without a lock because all the function
1966  * is write-locked. If "all" is true it means that the function is being called
1967  * while destroying a bond interface and all slaves are being released.
1968  *
1969  * The rules for slave state should be:
1970  *   for Active/Backup:
1971  *     Active stays on all backups go down
1972  *   for Bonded connections:
1973  *     The first up interface should be left on and all others downed.
1974  */
1975 static int __bond_release_one(struct net_device *bond_dev,
1976 			      struct net_device *slave_dev,
1977 			      bool all)
1978 {
1979 	struct bonding *bond = netdev_priv(bond_dev);
1980 	struct slave *slave, *oldcurrent;
1981 	struct sockaddr addr;
1982 	netdev_features_t old_features = bond_dev->features;
1983 
1984 	/* slave is not a slave or master is not master of this slave */
1985 	if (!(slave_dev->flags & IFF_SLAVE) ||
1986 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
1987 		pr_err("%s: Error: cannot release %s.\n",
1988 		       bond_dev->name, slave_dev->name);
1989 		return -EINVAL;
1990 	}
1991 
1992 	block_netpoll_tx();
1993 	write_lock_bh(&bond->lock);
1994 
1995 	slave = bond_get_slave_by_dev(bond, slave_dev);
1996 	if (!slave) {
1997 		/* not a slave of this bond */
1998 		pr_info("%s: %s not enslaved\n",
1999 			bond_dev->name, slave_dev->name);
2000 		write_unlock_bh(&bond->lock);
2001 		unblock_netpoll_tx();
2002 		return -EINVAL;
2003 	}
2004 
2005 	write_unlock_bh(&bond->lock);
2006 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2007 	 * for this slave anymore.
2008 	 */
2009 	netdev_rx_handler_unregister(slave_dev);
2010 	write_lock_bh(&bond->lock);
2011 
2012 	if (!all && !bond->params.fail_over_mac) {
2013 		if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2014 		    bond->slave_cnt > 1)
2015 			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",
2016 				   bond_dev->name, slave_dev->name,
2017 				   slave->perm_hwaddr,
2018 				   bond_dev->name, slave_dev->name);
2019 	}
2020 
2021 	/* Inform AD package of unbinding of slave. */
2022 	if (bond->params.mode == BOND_MODE_8023AD) {
2023 		/* must be called before the slave is
2024 		 * detached from the list
2025 		 */
2026 		bond_3ad_unbind_slave(slave);
2027 	}
2028 
2029 	pr_info("%s: releasing %s interface %s\n",
2030 		bond_dev->name,
2031 		bond_is_active_slave(slave) ? "active" : "backup",
2032 		slave_dev->name);
2033 
2034 	oldcurrent = bond->curr_active_slave;
2035 
2036 	bond->current_arp_slave = NULL;
2037 
2038 	/* release the slave from its bond */
2039 	bond_detach_slave(bond, slave);
2040 
2041 	if (bond->primary_slave == slave)
2042 		bond->primary_slave = NULL;
2043 
2044 	if (oldcurrent == slave)
2045 		bond_change_active_slave(bond, NULL);
2046 
2047 	if (bond_is_lb(bond)) {
2048 		/* Must be called only after the slave has been
2049 		 * detached from the list and the curr_active_slave
2050 		 * has been cleared (if our_slave == old_current),
2051 		 * but before a new active slave is selected.
2052 		 */
2053 		write_unlock_bh(&bond->lock);
2054 		bond_alb_deinit_slave(bond, slave);
2055 		write_lock_bh(&bond->lock);
2056 	}
2057 
2058 	if (all) {
2059 		bond->curr_active_slave = NULL;
2060 	} else if (oldcurrent == slave) {
2061 		/*
2062 		 * Note that we hold RTNL over this sequence, so there
2063 		 * is no concern that another slave add/remove event
2064 		 * will interfere.
2065 		 */
2066 		write_unlock_bh(&bond->lock);
2067 		read_lock(&bond->lock);
2068 		write_lock_bh(&bond->curr_slave_lock);
2069 
2070 		bond_select_active_slave(bond);
2071 
2072 		write_unlock_bh(&bond->curr_slave_lock);
2073 		read_unlock(&bond->lock);
2074 		write_lock_bh(&bond->lock);
2075 	}
2076 
2077 	if (bond->slave_cnt == 0) {
2078 		bond_set_carrier(bond);
2079 		eth_hw_addr_random(bond_dev);
2080 		bond->dev_addr_from_first = true;
2081 
2082 		if (bond_vlan_used(bond)) {
2083 			pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2084 				   bond_dev->name, bond_dev->name);
2085 			pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2086 				   bond_dev->name);
2087 		}
2088 	}
2089 
2090 	write_unlock_bh(&bond->lock);
2091 	unblock_netpoll_tx();
2092 
2093 	if (bond->slave_cnt == 0) {
2094 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2095 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2096 	}
2097 
2098 	bond_compute_features(bond);
2099 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2100 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2101 		pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2102 			bond_dev->name, slave_dev->name, bond_dev->name);
2103 
2104 	/* must do this from outside any spinlocks */
2105 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
2106 
2107 	bond_del_vlans_from_slave(bond, slave_dev);
2108 
2109 	/* If the mode USES_PRIMARY, then we should only remove its
2110 	 * promisc and mc settings if it was the curr_active_slave, but that was
2111 	 * already taken care of above when we detached the slave
2112 	 */
2113 	if (!USES_PRIMARY(bond->params.mode)) {
2114 		/* unset promiscuity level from slave */
2115 		if (bond_dev->flags & IFF_PROMISC)
2116 			dev_set_promiscuity(slave_dev, -1);
2117 
2118 		/* unset allmulti level from slave */
2119 		if (bond_dev->flags & IFF_ALLMULTI)
2120 			dev_set_allmulti(slave_dev, -1);
2121 
2122 		/* flush master's mc_list from slave */
2123 		netif_addr_lock_bh(bond_dev);
2124 		bond_mc_list_flush(bond_dev, slave_dev);
2125 		netif_addr_unlock_bh(bond_dev);
2126 	}
2127 
2128 	bond_upper_dev_unlink(bond_dev, slave_dev);
2129 
2130 	slave_disable_netpoll(slave);
2131 
2132 	/* close slave before restoring its mac address */
2133 	dev_close(slave_dev);
2134 
2135 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2136 		/* restore original ("permanent") mac address */
2137 		memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2138 		addr.sa_family = slave_dev->type;
2139 		dev_set_mac_address(slave_dev, &addr);
2140 	}
2141 
2142 	dev_set_mtu(slave_dev, slave->original_mtu);
2143 
2144 	slave_dev->priv_flags &= ~IFF_BONDING;
2145 
2146 	kfree(slave);
2147 
2148 	return 0;  /* deletion OK */
2149 }
2150 
2151 /* A wrapper used because of ndo_del_link */
2152 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2153 {
2154 	return __bond_release_one(bond_dev, slave_dev, false);
2155 }
2156 
2157 /*
2158 * First release a slave and then destroy the bond if no more slaves are left.
2159 * Must be under rtnl_lock when this function is called.
2160 */
2161 static int  bond_release_and_destroy(struct net_device *bond_dev,
2162 				     struct net_device *slave_dev)
2163 {
2164 	struct bonding *bond = netdev_priv(bond_dev);
2165 	int ret;
2166 
2167 	ret = bond_release(bond_dev, slave_dev);
2168 	if ((ret == 0) && (bond->slave_cnt == 0)) {
2169 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2170 		pr_info("%s: destroying bond %s.\n",
2171 			bond_dev->name, bond_dev->name);
2172 		unregister_netdevice(bond_dev);
2173 	}
2174 	return ret;
2175 }
2176 
2177 /*
2178  * This function changes the active slave to slave <slave_dev>.
2179  * It returns -EINVAL in the following cases.
2180  *  - <slave_dev> is not found in the list.
2181  *  - There is not active slave now.
2182  *  - <slave_dev> is already active.
2183  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2184  *  - <slave_dev> is not running.
2185  * In these cases, this function does nothing.
2186  * In the other cases, current_slave pointer is changed and 0 is returned.
2187  */
2188 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2189 {
2190 	struct bonding *bond = netdev_priv(bond_dev);
2191 	struct slave *old_active = NULL;
2192 	struct slave *new_active = NULL;
2193 	int res = 0;
2194 
2195 	if (!USES_PRIMARY(bond->params.mode))
2196 		return -EINVAL;
2197 
2198 	/* Verify that bond_dev is indeed the master of slave_dev */
2199 	if (!(slave_dev->flags & IFF_SLAVE) ||
2200 	    !netdev_has_upper_dev(slave_dev, bond_dev))
2201 		return -EINVAL;
2202 
2203 	read_lock(&bond->lock);
2204 
2205 	read_lock(&bond->curr_slave_lock);
2206 	old_active = bond->curr_active_slave;
2207 	read_unlock(&bond->curr_slave_lock);
2208 
2209 	new_active = bond_get_slave_by_dev(bond, slave_dev);
2210 
2211 	/*
2212 	 * Changing to the current active: do nothing; return success.
2213 	 */
2214 	if (new_active && (new_active == old_active)) {
2215 		read_unlock(&bond->lock);
2216 		return 0;
2217 	}
2218 
2219 	if ((new_active) &&
2220 	    (old_active) &&
2221 	    (new_active->link == BOND_LINK_UP) &&
2222 	    IS_UP(new_active->dev)) {
2223 		block_netpoll_tx();
2224 		write_lock_bh(&bond->curr_slave_lock);
2225 		bond_change_active_slave(bond, new_active);
2226 		write_unlock_bh(&bond->curr_slave_lock);
2227 		unblock_netpoll_tx();
2228 	} else
2229 		res = -EINVAL;
2230 
2231 	read_unlock(&bond->lock);
2232 
2233 	return res;
2234 }
2235 
2236 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2237 {
2238 	struct bonding *bond = netdev_priv(bond_dev);
2239 
2240 	info->bond_mode = bond->params.mode;
2241 	info->miimon = bond->params.miimon;
2242 
2243 	read_lock(&bond->lock);
2244 	info->num_slaves = bond->slave_cnt;
2245 	read_unlock(&bond->lock);
2246 
2247 	return 0;
2248 }
2249 
2250 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2251 {
2252 	struct bonding *bond = netdev_priv(bond_dev);
2253 	struct slave *slave;
2254 	int i, res = -ENODEV;
2255 
2256 	read_lock(&bond->lock);
2257 
2258 	bond_for_each_slave(bond, slave, i) {
2259 		if (i == (int)info->slave_id) {
2260 			res = 0;
2261 			strcpy(info->slave_name, slave->dev->name);
2262 			info->link = slave->link;
2263 			info->state = bond_slave_state(slave);
2264 			info->link_failure_count = slave->link_failure_count;
2265 			break;
2266 		}
2267 	}
2268 
2269 	read_unlock(&bond->lock);
2270 
2271 	return res;
2272 }
2273 
2274 /*-------------------------------- Monitoring -------------------------------*/
2275 
2276 
2277 static int bond_miimon_inspect(struct bonding *bond)
2278 {
2279 	struct slave *slave;
2280 	int i, link_state, commit = 0;
2281 	bool ignore_updelay;
2282 
2283 	ignore_updelay = !bond->curr_active_slave ? true : false;
2284 
2285 	bond_for_each_slave(bond, slave, i) {
2286 		slave->new_link = BOND_LINK_NOCHANGE;
2287 
2288 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2289 
2290 		switch (slave->link) {
2291 		case BOND_LINK_UP:
2292 			if (link_state)
2293 				continue;
2294 
2295 			slave->link = BOND_LINK_FAIL;
2296 			slave->delay = bond->params.downdelay;
2297 			if (slave->delay) {
2298 				pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2299 					bond->dev->name,
2300 					(bond->params.mode ==
2301 					 BOND_MODE_ACTIVEBACKUP) ?
2302 					(bond_is_active_slave(slave) ?
2303 					 "active " : "backup ") : "",
2304 					slave->dev->name,
2305 					bond->params.downdelay * bond->params.miimon);
2306 			}
2307 			/*FALLTHRU*/
2308 		case BOND_LINK_FAIL:
2309 			if (link_state) {
2310 				/*
2311 				 * recovered before downdelay expired
2312 				 */
2313 				slave->link = BOND_LINK_UP;
2314 				slave->jiffies = jiffies;
2315 				pr_info("%s: link status up again after %d ms for interface %s.\n",
2316 					bond->dev->name,
2317 					(bond->params.downdelay - slave->delay) *
2318 					bond->params.miimon,
2319 					slave->dev->name);
2320 				continue;
2321 			}
2322 
2323 			if (slave->delay <= 0) {
2324 				slave->new_link = BOND_LINK_DOWN;
2325 				commit++;
2326 				continue;
2327 			}
2328 
2329 			slave->delay--;
2330 			break;
2331 
2332 		case BOND_LINK_DOWN:
2333 			if (!link_state)
2334 				continue;
2335 
2336 			slave->link = BOND_LINK_BACK;
2337 			slave->delay = bond->params.updelay;
2338 
2339 			if (slave->delay) {
2340 				pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2341 					bond->dev->name, slave->dev->name,
2342 					ignore_updelay ? 0 :
2343 					bond->params.updelay *
2344 					bond->params.miimon);
2345 			}
2346 			/*FALLTHRU*/
2347 		case BOND_LINK_BACK:
2348 			if (!link_state) {
2349 				slave->link = BOND_LINK_DOWN;
2350 				pr_info("%s: link status down again after %d ms for interface %s.\n",
2351 					bond->dev->name,
2352 					(bond->params.updelay - slave->delay) *
2353 					bond->params.miimon,
2354 					slave->dev->name);
2355 
2356 				continue;
2357 			}
2358 
2359 			if (ignore_updelay)
2360 				slave->delay = 0;
2361 
2362 			if (slave->delay <= 0) {
2363 				slave->new_link = BOND_LINK_UP;
2364 				commit++;
2365 				ignore_updelay = false;
2366 				continue;
2367 			}
2368 
2369 			slave->delay--;
2370 			break;
2371 		}
2372 	}
2373 
2374 	return commit;
2375 }
2376 
2377 static void bond_miimon_commit(struct bonding *bond)
2378 {
2379 	struct slave *slave;
2380 	int i;
2381 
2382 	bond_for_each_slave(bond, slave, i) {
2383 		switch (slave->new_link) {
2384 		case BOND_LINK_NOCHANGE:
2385 			continue;
2386 
2387 		case BOND_LINK_UP:
2388 			slave->link = BOND_LINK_UP;
2389 			slave->jiffies = jiffies;
2390 
2391 			if (bond->params.mode == BOND_MODE_8023AD) {
2392 				/* prevent it from being the active one */
2393 				bond_set_backup_slave(slave);
2394 			} else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2395 				/* make it immediately active */
2396 				bond_set_active_slave(slave);
2397 			} else if (slave != bond->primary_slave) {
2398 				/* prevent it from being the active one */
2399 				bond_set_backup_slave(slave);
2400 			}
2401 
2402 			pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2403 				bond->dev->name, slave->dev->name,
2404 				slave->speed, slave->duplex ? "full" : "half");
2405 
2406 			/* notify ad that the link status has changed */
2407 			if (bond->params.mode == BOND_MODE_8023AD)
2408 				bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2409 
2410 			if (bond_is_lb(bond))
2411 				bond_alb_handle_link_change(bond, slave,
2412 							    BOND_LINK_UP);
2413 
2414 			if (!bond->curr_active_slave ||
2415 			    (slave == bond->primary_slave))
2416 				goto do_failover;
2417 
2418 			continue;
2419 
2420 		case BOND_LINK_DOWN:
2421 			if (slave->link_failure_count < UINT_MAX)
2422 				slave->link_failure_count++;
2423 
2424 			slave->link = BOND_LINK_DOWN;
2425 
2426 			if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2427 			    bond->params.mode == BOND_MODE_8023AD)
2428 				bond_set_slave_inactive_flags(slave);
2429 
2430 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
2431 				bond->dev->name, slave->dev->name);
2432 
2433 			if (bond->params.mode == BOND_MODE_8023AD)
2434 				bond_3ad_handle_link_change(slave,
2435 							    BOND_LINK_DOWN);
2436 
2437 			if (bond_is_lb(bond))
2438 				bond_alb_handle_link_change(bond, slave,
2439 							    BOND_LINK_DOWN);
2440 
2441 			if (slave == bond->curr_active_slave)
2442 				goto do_failover;
2443 
2444 			continue;
2445 
2446 		default:
2447 			pr_err("%s: invalid new link %d on slave %s\n",
2448 			       bond->dev->name, slave->new_link,
2449 			       slave->dev->name);
2450 			slave->new_link = BOND_LINK_NOCHANGE;
2451 
2452 			continue;
2453 		}
2454 
2455 do_failover:
2456 		ASSERT_RTNL();
2457 		block_netpoll_tx();
2458 		write_lock_bh(&bond->curr_slave_lock);
2459 		bond_select_active_slave(bond);
2460 		write_unlock_bh(&bond->curr_slave_lock);
2461 		unblock_netpoll_tx();
2462 	}
2463 
2464 	bond_set_carrier(bond);
2465 }
2466 
2467 /*
2468  * bond_mii_monitor
2469  *
2470  * Really a wrapper that splits the mii monitor into two phases: an
2471  * inspection, then (if inspection indicates something needs to be done)
2472  * an acquisition of appropriate locks followed by a commit phase to
2473  * implement whatever link state changes are indicated.
2474  */
2475 void bond_mii_monitor(struct work_struct *work)
2476 {
2477 	struct bonding *bond = container_of(work, struct bonding,
2478 					    mii_work.work);
2479 	bool should_notify_peers = false;
2480 	unsigned long delay;
2481 
2482 	read_lock(&bond->lock);
2483 
2484 	delay = msecs_to_jiffies(bond->params.miimon);
2485 
2486 	if (bond->slave_cnt == 0)
2487 		goto re_arm;
2488 
2489 	should_notify_peers = bond_should_notify_peers(bond);
2490 
2491 	if (bond_miimon_inspect(bond)) {
2492 		read_unlock(&bond->lock);
2493 
2494 		/* Race avoidance with bond_close cancel of workqueue */
2495 		if (!rtnl_trylock()) {
2496 			read_lock(&bond->lock);
2497 			delay = 1;
2498 			should_notify_peers = false;
2499 			goto re_arm;
2500 		}
2501 
2502 		read_lock(&bond->lock);
2503 
2504 		bond_miimon_commit(bond);
2505 
2506 		read_unlock(&bond->lock);
2507 		rtnl_unlock();	/* might sleep, hold no other locks */
2508 		read_lock(&bond->lock);
2509 	}
2510 
2511 re_arm:
2512 	if (bond->params.miimon)
2513 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2514 
2515 	read_unlock(&bond->lock);
2516 
2517 	if (should_notify_peers) {
2518 		if (!rtnl_trylock()) {
2519 			read_lock(&bond->lock);
2520 			bond->send_peer_notif++;
2521 			read_unlock(&bond->lock);
2522 			return;
2523 		}
2524 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2525 		rtnl_unlock();
2526 	}
2527 }
2528 
2529 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2530 {
2531 	struct vlan_entry *vlan;
2532 	struct net_device *vlan_dev;
2533 
2534 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2535 		return 1;
2536 
2537 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2538 		rcu_read_lock();
2539 		vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2540 						vlan->vlan_id);
2541 		rcu_read_unlock();
2542 		if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2543 			return 1;
2544 	}
2545 
2546 	return 0;
2547 }
2548 
2549 /*
2550  * We go to the (large) trouble of VLAN tagging ARP frames because
2551  * switches in VLAN mode (especially if ports are configured as
2552  * "native" to a VLAN) might not pass non-tagged frames.
2553  */
2554 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2555 {
2556 	struct sk_buff *skb;
2557 
2558 	pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2559 		 slave_dev->name, dest_ip, src_ip, vlan_id);
2560 
2561 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2562 			 NULL, slave_dev->dev_addr, NULL);
2563 
2564 	if (!skb) {
2565 		pr_err("ARP packet allocation failed\n");
2566 		return;
2567 	}
2568 	if (vlan_id) {
2569 		skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2570 		if (!skb) {
2571 			pr_err("failed to insert VLAN tag\n");
2572 			return;
2573 		}
2574 	}
2575 	arp_xmit(skb);
2576 }
2577 
2578 
2579 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2580 {
2581 	int i, vlan_id;
2582 	__be32 *targets = bond->params.arp_targets;
2583 	struct vlan_entry *vlan;
2584 	struct net_device *vlan_dev = NULL;
2585 	struct rtable *rt;
2586 
2587 	for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2588 		__be32 addr;
2589 		if (!targets[i])
2590 			break;
2591 		pr_debug("basa: target %x\n", targets[i]);
2592 		if (!bond_vlan_used(bond)) {
2593 			pr_debug("basa: empty vlan: arp_send\n");
2594 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2595 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2596 				      addr, 0);
2597 			continue;
2598 		}
2599 
2600 		/*
2601 		 * If VLANs are configured, we do a route lookup to
2602 		 * determine which VLAN interface would be used, so we
2603 		 * can tag the ARP with the proper VLAN tag.
2604 		 */
2605 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2606 				     RTO_ONLINK, 0);
2607 		if (IS_ERR(rt)) {
2608 			if (net_ratelimit()) {
2609 				pr_warning("%s: no route to arp_ip_target %pI4\n",
2610 					   bond->dev->name, &targets[i]);
2611 			}
2612 			continue;
2613 		}
2614 
2615 		/*
2616 		 * This target is not on a VLAN
2617 		 */
2618 		if (rt->dst.dev == bond->dev) {
2619 			ip_rt_put(rt);
2620 			pr_debug("basa: rtdev == bond->dev: arp_send\n");
2621 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2622 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2623 				      addr, 0);
2624 			continue;
2625 		}
2626 
2627 		vlan_id = 0;
2628 		list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2629 			rcu_read_lock();
2630 			vlan_dev = __vlan_find_dev_deep(bond->dev,
2631 							htons(ETH_P_8021Q),
2632 							vlan->vlan_id);
2633 			rcu_read_unlock();
2634 			if (vlan_dev == rt->dst.dev) {
2635 				vlan_id = vlan->vlan_id;
2636 				pr_debug("basa: vlan match on %s %d\n",
2637 				       vlan_dev->name, vlan_id);
2638 				break;
2639 			}
2640 		}
2641 
2642 		if (vlan_id && vlan_dev) {
2643 			ip_rt_put(rt);
2644 			addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2645 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2646 				      addr, vlan_id);
2647 			continue;
2648 		}
2649 
2650 		if (net_ratelimit()) {
2651 			pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2652 				   bond->dev->name, &targets[i],
2653 				   rt->dst.dev ? rt->dst.dev->name : "NULL");
2654 		}
2655 		ip_rt_put(rt);
2656 	}
2657 }
2658 
2659 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2660 {
2661 	int i;
2662 	__be32 *targets = bond->params.arp_targets;
2663 
2664 	for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2665 		pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2666 			 &sip, &tip, i, &targets[i],
2667 			 bond_has_this_ip(bond, tip));
2668 		if (sip == targets[i]) {
2669 			if (bond_has_this_ip(bond, tip))
2670 				slave->last_arp_rx = jiffies;
2671 			return;
2672 		}
2673 	}
2674 }
2675 
2676 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2677 			struct slave *slave)
2678 {
2679 	struct arphdr *arp = (struct arphdr *)skb->data;
2680 	unsigned char *arp_ptr;
2681 	__be32 sip, tip;
2682 	int alen;
2683 
2684 	if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2685 		return RX_HANDLER_ANOTHER;
2686 
2687 	read_lock(&bond->lock);
2688 	alen = arp_hdr_len(bond->dev);
2689 
2690 	pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2691 		 bond->dev->name, skb->dev->name);
2692 
2693 	if (alen > skb_headlen(skb)) {
2694 		arp = kmalloc(alen, GFP_ATOMIC);
2695 		if (!arp)
2696 			goto out_unlock;
2697 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
2698 			goto out_unlock;
2699 	}
2700 
2701 	if (arp->ar_hln != bond->dev->addr_len ||
2702 	    skb->pkt_type == PACKET_OTHERHOST ||
2703 	    skb->pkt_type == PACKET_LOOPBACK ||
2704 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
2705 	    arp->ar_pro != htons(ETH_P_IP) ||
2706 	    arp->ar_pln != 4)
2707 		goto out_unlock;
2708 
2709 	arp_ptr = (unsigned char *)(arp + 1);
2710 	arp_ptr += bond->dev->addr_len;
2711 	memcpy(&sip, arp_ptr, 4);
2712 	arp_ptr += 4 + bond->dev->addr_len;
2713 	memcpy(&tip, arp_ptr, 4);
2714 
2715 	pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2716 		 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2717 		 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2718 		 &sip, &tip);
2719 
2720 	/*
2721 	 * Backup slaves won't see the ARP reply, but do come through
2722 	 * here for each ARP probe (so we swap the sip/tip to validate
2723 	 * the probe).  In a "redundant switch, common router" type of
2724 	 * configuration, the ARP probe will (hopefully) travel from
2725 	 * the active, through one switch, the router, then the other
2726 	 * switch before reaching the backup.
2727 	 */
2728 	if (bond_is_active_slave(slave))
2729 		bond_validate_arp(bond, slave, sip, tip);
2730 	else
2731 		bond_validate_arp(bond, slave, tip, sip);
2732 
2733 out_unlock:
2734 	read_unlock(&bond->lock);
2735 	if (arp != (struct arphdr *)skb->data)
2736 		kfree(arp);
2737 	return RX_HANDLER_ANOTHER;
2738 }
2739 
2740 /*
2741  * this function is called regularly to monitor each slave's link
2742  * ensuring that traffic is being sent and received when arp monitoring
2743  * is used in load-balancing mode. if the adapter has been dormant, then an
2744  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2745  * arp monitoring in active backup mode.
2746  */
2747 void bond_loadbalance_arp_mon(struct work_struct *work)
2748 {
2749 	struct bonding *bond = container_of(work, struct bonding,
2750 					    arp_work.work);
2751 	struct slave *slave, *oldcurrent;
2752 	int do_failover = 0;
2753 	int delta_in_ticks, extra_ticks;
2754 	int i;
2755 
2756 	read_lock(&bond->lock);
2757 
2758 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2759 	extra_ticks = delta_in_ticks / 2;
2760 
2761 	if (bond->slave_cnt == 0)
2762 		goto re_arm;
2763 
2764 	read_lock(&bond->curr_slave_lock);
2765 	oldcurrent = bond->curr_active_slave;
2766 	read_unlock(&bond->curr_slave_lock);
2767 
2768 	/* see if any of the previous devices are up now (i.e. they have
2769 	 * xmt and rcv traffic). the curr_active_slave does not come into
2770 	 * the picture unless it is null. also, slave->jiffies is not needed
2771 	 * here because we send an arp on each slave and give a slave as
2772 	 * long as it needs to get the tx/rx within the delta.
2773 	 * TODO: what about up/down delay in arp mode? it wasn't here before
2774 	 *       so it can wait
2775 	 */
2776 	bond_for_each_slave(bond, slave, i) {
2777 		unsigned long trans_start = dev_trans_start(slave->dev);
2778 
2779 		if (slave->link != BOND_LINK_UP) {
2780 			if (time_in_range(jiffies,
2781 				trans_start - delta_in_ticks,
2782 				trans_start + delta_in_ticks + extra_ticks) &&
2783 			    time_in_range(jiffies,
2784 				slave->dev->last_rx - delta_in_ticks,
2785 				slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2786 
2787 				slave->link  = BOND_LINK_UP;
2788 				bond_set_active_slave(slave);
2789 
2790 				/* primary_slave has no meaning in round-robin
2791 				 * mode. the window of a slave being up and
2792 				 * curr_active_slave being null after enslaving
2793 				 * is closed.
2794 				 */
2795 				if (!oldcurrent) {
2796 					pr_info("%s: link status definitely up for interface %s, ",
2797 						bond->dev->name,
2798 						slave->dev->name);
2799 					do_failover = 1;
2800 				} else {
2801 					pr_info("%s: interface %s is now up\n",
2802 						bond->dev->name,
2803 						slave->dev->name);
2804 				}
2805 			}
2806 		} else {
2807 			/* slave->link == BOND_LINK_UP */
2808 
2809 			/* not all switches will respond to an arp request
2810 			 * when the source ip is 0, so don't take the link down
2811 			 * if we don't know our ip yet
2812 			 */
2813 			if (!time_in_range(jiffies,
2814 				trans_start - delta_in_ticks,
2815 				trans_start + 2 * delta_in_ticks + extra_ticks) ||
2816 			    !time_in_range(jiffies,
2817 				slave->dev->last_rx - delta_in_ticks,
2818 				slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2819 
2820 				slave->link  = BOND_LINK_DOWN;
2821 				bond_set_backup_slave(slave);
2822 
2823 				if (slave->link_failure_count < UINT_MAX)
2824 					slave->link_failure_count++;
2825 
2826 				pr_info("%s: interface %s is now down.\n",
2827 					bond->dev->name,
2828 					slave->dev->name);
2829 
2830 				if (slave == oldcurrent)
2831 					do_failover = 1;
2832 			}
2833 		}
2834 
2835 		/* note: if switch is in round-robin mode, all links
2836 		 * must tx arp to ensure all links rx an arp - otherwise
2837 		 * links may oscillate or not come up at all; if switch is
2838 		 * in something like xor mode, there is nothing we can
2839 		 * do - all replies will be rx'ed on same link causing slaves
2840 		 * to be unstable during low/no traffic periods
2841 		 */
2842 		if (IS_UP(slave->dev))
2843 			bond_arp_send_all(bond, slave);
2844 	}
2845 
2846 	if (do_failover) {
2847 		block_netpoll_tx();
2848 		write_lock_bh(&bond->curr_slave_lock);
2849 
2850 		bond_select_active_slave(bond);
2851 
2852 		write_unlock_bh(&bond->curr_slave_lock);
2853 		unblock_netpoll_tx();
2854 	}
2855 
2856 re_arm:
2857 	if (bond->params.arp_interval)
2858 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2859 
2860 	read_unlock(&bond->lock);
2861 }
2862 
2863 /*
2864  * Called to inspect slaves for active-backup mode ARP monitor link state
2865  * changes.  Sets new_link in slaves to specify what action should take
2866  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2867  * to link states must be committed.
2868  *
2869  * Called with bond->lock held for read.
2870  */
2871 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2872 {
2873 	struct slave *slave;
2874 	int i, commit = 0;
2875 	unsigned long trans_start;
2876 	int extra_ticks;
2877 
2878 	/* All the time comparisons below need some extra time. Otherwise, on
2879 	 * fast networks the ARP probe/reply may arrive within the same jiffy
2880 	 * as it was sent.  Then, the next time the ARP monitor is run, one
2881 	 * arp_interval will already have passed in the comparisons.
2882 	 */
2883 	extra_ticks = delta_in_ticks / 2;
2884 
2885 	bond_for_each_slave(bond, slave, i) {
2886 		slave->new_link = BOND_LINK_NOCHANGE;
2887 
2888 		if (slave->link != BOND_LINK_UP) {
2889 			if (time_in_range(jiffies,
2890 				slave_last_rx(bond, slave) - delta_in_ticks,
2891 				slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2892 
2893 				slave->new_link = BOND_LINK_UP;
2894 				commit++;
2895 			}
2896 
2897 			continue;
2898 		}
2899 
2900 		/*
2901 		 * Give slaves 2*delta after being enslaved or made
2902 		 * active.  This avoids bouncing, as the last receive
2903 		 * times need a full ARP monitor cycle to be updated.
2904 		 */
2905 		if (time_in_range(jiffies,
2906 				  slave->jiffies - delta_in_ticks,
2907 				  slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2908 			continue;
2909 
2910 		/*
2911 		 * Backup slave is down if:
2912 		 * - No current_arp_slave AND
2913 		 * - more than 3*delta since last receive AND
2914 		 * - the bond has an IP address
2915 		 *
2916 		 * Note: a non-null current_arp_slave indicates
2917 		 * the curr_active_slave went down and we are
2918 		 * searching for a new one; under this condition
2919 		 * we only take the curr_active_slave down - this
2920 		 * gives each slave a chance to tx/rx traffic
2921 		 * before being taken out
2922 		 */
2923 		if (!bond_is_active_slave(slave) &&
2924 		    !bond->current_arp_slave &&
2925 		    !time_in_range(jiffies,
2926 			slave_last_rx(bond, slave) - delta_in_ticks,
2927 			slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2928 
2929 			slave->new_link = BOND_LINK_DOWN;
2930 			commit++;
2931 		}
2932 
2933 		/*
2934 		 * Active slave is down if:
2935 		 * - more than 2*delta since transmitting OR
2936 		 * - (more than 2*delta since receive AND
2937 		 *    the bond has an IP address)
2938 		 */
2939 		trans_start = dev_trans_start(slave->dev);
2940 		if (bond_is_active_slave(slave) &&
2941 		    (!time_in_range(jiffies,
2942 			trans_start - delta_in_ticks,
2943 			trans_start + 2 * delta_in_ticks + extra_ticks) ||
2944 		     !time_in_range(jiffies,
2945 			slave_last_rx(bond, slave) - delta_in_ticks,
2946 			slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2947 
2948 			slave->new_link = BOND_LINK_DOWN;
2949 			commit++;
2950 		}
2951 	}
2952 
2953 	return commit;
2954 }
2955 
2956 /*
2957  * Called to commit link state changes noted by inspection step of
2958  * active-backup mode ARP monitor.
2959  *
2960  * Called with RTNL and bond->lock for read.
2961  */
2962 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2963 {
2964 	struct slave *slave;
2965 	int i;
2966 	unsigned long trans_start;
2967 
2968 	bond_for_each_slave(bond, slave, i) {
2969 		switch (slave->new_link) {
2970 		case BOND_LINK_NOCHANGE:
2971 			continue;
2972 
2973 		case BOND_LINK_UP:
2974 			trans_start = dev_trans_start(slave->dev);
2975 			if ((!bond->curr_active_slave &&
2976 			     time_in_range(jiffies,
2977 					   trans_start - delta_in_ticks,
2978 					   trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2979 			    bond->curr_active_slave != slave) {
2980 				slave->link = BOND_LINK_UP;
2981 				if (bond->current_arp_slave) {
2982 					bond_set_slave_inactive_flags(
2983 						bond->current_arp_slave);
2984 					bond->current_arp_slave = NULL;
2985 				}
2986 
2987 				pr_info("%s: link status definitely up for interface %s.\n",
2988 					bond->dev->name, slave->dev->name);
2989 
2990 				if (!bond->curr_active_slave ||
2991 				    (slave == bond->primary_slave))
2992 					goto do_failover;
2993 
2994 			}
2995 
2996 			continue;
2997 
2998 		case BOND_LINK_DOWN:
2999 			if (slave->link_failure_count < UINT_MAX)
3000 				slave->link_failure_count++;
3001 
3002 			slave->link = BOND_LINK_DOWN;
3003 			bond_set_slave_inactive_flags(slave);
3004 
3005 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
3006 				bond->dev->name, slave->dev->name);
3007 
3008 			if (slave == bond->curr_active_slave) {
3009 				bond->current_arp_slave = NULL;
3010 				goto do_failover;
3011 			}
3012 
3013 			continue;
3014 
3015 		default:
3016 			pr_err("%s: impossible: new_link %d on slave %s\n",
3017 			       bond->dev->name, slave->new_link,
3018 			       slave->dev->name);
3019 			continue;
3020 		}
3021 
3022 do_failover:
3023 		ASSERT_RTNL();
3024 		block_netpoll_tx();
3025 		write_lock_bh(&bond->curr_slave_lock);
3026 		bond_select_active_slave(bond);
3027 		write_unlock_bh(&bond->curr_slave_lock);
3028 		unblock_netpoll_tx();
3029 	}
3030 
3031 	bond_set_carrier(bond);
3032 }
3033 
3034 /*
3035  * Send ARP probes for active-backup mode ARP monitor.
3036  *
3037  * Called with bond->lock held for read.
3038  */
3039 static void bond_ab_arp_probe(struct bonding *bond)
3040 {
3041 	struct slave *slave;
3042 	int i;
3043 
3044 	read_lock(&bond->curr_slave_lock);
3045 
3046 	if (bond->current_arp_slave && bond->curr_active_slave)
3047 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
3048 			bond->current_arp_slave->dev->name,
3049 			bond->curr_active_slave->dev->name);
3050 
3051 	if (bond->curr_active_slave) {
3052 		bond_arp_send_all(bond, bond->curr_active_slave);
3053 		read_unlock(&bond->curr_slave_lock);
3054 		return;
3055 	}
3056 
3057 	read_unlock(&bond->curr_slave_lock);
3058 
3059 	/* if we don't have a curr_active_slave, search for the next available
3060 	 * backup slave from the current_arp_slave and make it the candidate
3061 	 * for becoming the curr_active_slave
3062 	 */
3063 
3064 	if (!bond->current_arp_slave) {
3065 		bond->current_arp_slave = bond->first_slave;
3066 		if (!bond->current_arp_slave)
3067 			return;
3068 	}
3069 
3070 	bond_set_slave_inactive_flags(bond->current_arp_slave);
3071 
3072 	/* search for next candidate */
3073 	bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3074 		if (IS_UP(slave->dev)) {
3075 			slave->link = BOND_LINK_BACK;
3076 			bond_set_slave_active_flags(slave);
3077 			bond_arp_send_all(bond, slave);
3078 			slave->jiffies = jiffies;
3079 			bond->current_arp_slave = slave;
3080 			break;
3081 		}
3082 
3083 		/* if the link state is up at this point, we
3084 		 * mark it down - this can happen if we have
3085 		 * simultaneous link failures and
3086 		 * reselect_active_interface doesn't make this
3087 		 * one the current slave so it is still marked
3088 		 * up when it is actually down
3089 		 */
3090 		if (slave->link == BOND_LINK_UP) {
3091 			slave->link = BOND_LINK_DOWN;
3092 			if (slave->link_failure_count < UINT_MAX)
3093 				slave->link_failure_count++;
3094 
3095 			bond_set_slave_inactive_flags(slave);
3096 
3097 			pr_info("%s: backup interface %s is now down.\n",
3098 				bond->dev->name, slave->dev->name);
3099 		}
3100 	}
3101 }
3102 
3103 void bond_activebackup_arp_mon(struct work_struct *work)
3104 {
3105 	struct bonding *bond = container_of(work, struct bonding,
3106 					    arp_work.work);
3107 	bool should_notify_peers = false;
3108 	int delta_in_ticks;
3109 
3110 	read_lock(&bond->lock);
3111 
3112 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3113 
3114 	if (bond->slave_cnt == 0)
3115 		goto re_arm;
3116 
3117 	should_notify_peers = bond_should_notify_peers(bond);
3118 
3119 	if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3120 		read_unlock(&bond->lock);
3121 
3122 		/* Race avoidance with bond_close flush of workqueue */
3123 		if (!rtnl_trylock()) {
3124 			read_lock(&bond->lock);
3125 			delta_in_ticks = 1;
3126 			should_notify_peers = false;
3127 			goto re_arm;
3128 		}
3129 
3130 		read_lock(&bond->lock);
3131 
3132 		bond_ab_arp_commit(bond, delta_in_ticks);
3133 
3134 		read_unlock(&bond->lock);
3135 		rtnl_unlock();
3136 		read_lock(&bond->lock);
3137 	}
3138 
3139 	bond_ab_arp_probe(bond);
3140 
3141 re_arm:
3142 	if (bond->params.arp_interval)
3143 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3144 
3145 	read_unlock(&bond->lock);
3146 
3147 	if (should_notify_peers) {
3148 		if (!rtnl_trylock()) {
3149 			read_lock(&bond->lock);
3150 			bond->send_peer_notif++;
3151 			read_unlock(&bond->lock);
3152 			return;
3153 		}
3154 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3155 		rtnl_unlock();
3156 	}
3157 }
3158 
3159 /*-------------------------- netdev event handling --------------------------*/
3160 
3161 /*
3162  * Change device name
3163  */
3164 static int bond_event_changename(struct bonding *bond)
3165 {
3166 	bond_remove_proc_entry(bond);
3167 	bond_create_proc_entry(bond);
3168 
3169 	bond_debug_reregister(bond);
3170 
3171 	return NOTIFY_DONE;
3172 }
3173 
3174 static int bond_master_netdev_event(unsigned long event,
3175 				    struct net_device *bond_dev)
3176 {
3177 	struct bonding *event_bond = netdev_priv(bond_dev);
3178 
3179 	switch (event) {
3180 	case NETDEV_CHANGENAME:
3181 		return bond_event_changename(event_bond);
3182 	case NETDEV_UNREGISTER:
3183 		bond_remove_proc_entry(event_bond);
3184 		break;
3185 	case NETDEV_REGISTER:
3186 		bond_create_proc_entry(event_bond);
3187 		break;
3188 	default:
3189 		break;
3190 	}
3191 
3192 	return NOTIFY_DONE;
3193 }
3194 
3195 static int bond_slave_netdev_event(unsigned long event,
3196 				   struct net_device *slave_dev)
3197 {
3198 	struct slave *slave = bond_slave_get_rtnl(slave_dev);
3199 	struct bonding *bond;
3200 	struct net_device *bond_dev;
3201 	u32 old_speed;
3202 	u8 old_duplex;
3203 
3204 	/* A netdev event can be generated while enslaving a device
3205 	 * before netdev_rx_handler_register is called in which case
3206 	 * slave will be NULL
3207 	 */
3208 	if (!slave)
3209 		return NOTIFY_DONE;
3210 	bond_dev = slave->bond->dev;
3211 	bond = slave->bond;
3212 
3213 	switch (event) {
3214 	case NETDEV_UNREGISTER:
3215 		if (bond->setup_by_slave)
3216 			bond_release_and_destroy(bond_dev, slave_dev);
3217 		else
3218 			bond_release(bond_dev, slave_dev);
3219 		break;
3220 	case NETDEV_UP:
3221 	case NETDEV_CHANGE:
3222 		old_speed = slave->speed;
3223 		old_duplex = slave->duplex;
3224 
3225 		bond_update_speed_duplex(slave);
3226 
3227 		if (bond->params.mode == BOND_MODE_8023AD) {
3228 			if (old_speed != slave->speed)
3229 				bond_3ad_adapter_speed_changed(slave);
3230 			if (old_duplex != slave->duplex)
3231 				bond_3ad_adapter_duplex_changed(slave);
3232 		}
3233 		break;
3234 	case NETDEV_DOWN:
3235 		/*
3236 		 * ... Or is it this?
3237 		 */
3238 		break;
3239 	case NETDEV_CHANGEMTU:
3240 		/*
3241 		 * TODO: Should slaves be allowed to
3242 		 * independently alter their MTU?  For
3243 		 * an active-backup bond, slaves need
3244 		 * not be the same type of device, so
3245 		 * MTUs may vary.  For other modes,
3246 		 * slaves arguably should have the
3247 		 * same MTUs. To do this, we'd need to
3248 		 * take over the slave's change_mtu
3249 		 * function for the duration of their
3250 		 * servitude.
3251 		 */
3252 		break;
3253 	case NETDEV_CHANGENAME:
3254 		/*
3255 		 * TODO: handle changing the primary's name
3256 		 */
3257 		break;
3258 	case NETDEV_FEAT_CHANGE:
3259 		bond_compute_features(bond);
3260 		break;
3261 	default:
3262 		break;
3263 	}
3264 
3265 	return NOTIFY_DONE;
3266 }
3267 
3268 /*
3269  * bond_netdev_event: handle netdev notifier chain events.
3270  *
3271  * This function receives events for the netdev chain.  The caller (an
3272  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3273  * locks for us to safely manipulate the slave devices (RTNL lock,
3274  * dev_probe_lock).
3275  */
3276 static int bond_netdev_event(struct notifier_block *this,
3277 			     unsigned long event, void *ptr)
3278 {
3279 	struct net_device *event_dev = (struct net_device *)ptr;
3280 
3281 	pr_debug("event_dev: %s, event: %lx\n",
3282 		 event_dev ? event_dev->name : "None",
3283 		 event);
3284 
3285 	if (!(event_dev->priv_flags & IFF_BONDING))
3286 		return NOTIFY_DONE;
3287 
3288 	if (event_dev->flags & IFF_MASTER) {
3289 		pr_debug("IFF_MASTER\n");
3290 		return bond_master_netdev_event(event, event_dev);
3291 	}
3292 
3293 	if (event_dev->flags & IFF_SLAVE) {
3294 		pr_debug("IFF_SLAVE\n");
3295 		return bond_slave_netdev_event(event, event_dev);
3296 	}
3297 
3298 	return NOTIFY_DONE;
3299 }
3300 
3301 static struct notifier_block bond_netdev_notifier = {
3302 	.notifier_call = bond_netdev_event,
3303 };
3304 
3305 /*---------------------------- Hashing Policies -----------------------------*/
3306 
3307 /*
3308  * Hash for the output device based upon layer 2 data
3309  */
3310 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3311 {
3312 	struct ethhdr *data = (struct ethhdr *)skb->data;
3313 
3314 	if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3315 		return (data->h_dest[5] ^ data->h_source[5]) % count;
3316 
3317 	return 0;
3318 }
3319 
3320 /*
3321  * Hash for the output device based upon layer 2 and layer 3 data. If
3322  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3323  */
3324 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3325 {
3326 	const struct ethhdr *data;
3327 	const struct iphdr *iph;
3328 	const struct ipv6hdr *ipv6h;
3329 	u32 v6hash;
3330 	const __be32 *s, *d;
3331 
3332 	if (skb->protocol == htons(ETH_P_IP) &&
3333 	    pskb_network_may_pull(skb, sizeof(*iph))) {
3334 		iph = ip_hdr(skb);
3335 		data = (struct ethhdr *)skb->data;
3336 		return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3337 			(data->h_dest[5] ^ data->h_source[5])) % count;
3338 	} else if (skb->protocol == htons(ETH_P_IPV6) &&
3339 		   pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3340 		ipv6h = ipv6_hdr(skb);
3341 		data = (struct ethhdr *)skb->data;
3342 		s = &ipv6h->saddr.s6_addr32[0];
3343 		d = &ipv6h->daddr.s6_addr32[0];
3344 		v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3345 		v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3346 		return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3347 	}
3348 
3349 	return bond_xmit_hash_policy_l2(skb, count);
3350 }
3351 
3352 /*
3353  * Hash for the output device based upon layer 3 and layer 4 data. If
3354  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3355  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3356  */
3357 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3358 {
3359 	u32 layer4_xor = 0;
3360 	const struct iphdr *iph;
3361 	const struct ipv6hdr *ipv6h;
3362 	const __be32 *s, *d;
3363 	const __be16 *l4 = NULL;
3364 	__be16 _l4[2];
3365 	int noff = skb_network_offset(skb);
3366 	int poff;
3367 
3368 	if (skb->protocol == htons(ETH_P_IP) &&
3369 	    pskb_may_pull(skb, noff + sizeof(*iph))) {
3370 		iph = ip_hdr(skb);
3371 		poff = proto_ports_offset(iph->protocol);
3372 
3373 		if (!ip_is_fragment(iph) && poff >= 0) {
3374 			l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3375 						sizeof(_l4), &_l4);
3376 			if (l4)
3377 				layer4_xor = ntohs(l4[0] ^ l4[1]);
3378 		}
3379 		return (layer4_xor ^
3380 			((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3381 	} else if (skb->protocol == htons(ETH_P_IPV6) &&
3382 		   pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3383 		ipv6h = ipv6_hdr(skb);
3384 		poff = proto_ports_offset(ipv6h->nexthdr);
3385 		if (poff >= 0) {
3386 			l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3387 						sizeof(_l4), &_l4);
3388 			if (l4)
3389 				layer4_xor = ntohs(l4[0] ^ l4[1]);
3390 		}
3391 		s = &ipv6h->saddr.s6_addr32[0];
3392 		d = &ipv6h->daddr.s6_addr32[0];
3393 		layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3394 		layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3395 			       (layer4_xor >> 8);
3396 		return layer4_xor % count;
3397 	}
3398 
3399 	return bond_xmit_hash_policy_l2(skb, count);
3400 }
3401 
3402 /*-------------------------- Device entry points ----------------------------*/
3403 
3404 static void bond_work_init_all(struct bonding *bond)
3405 {
3406 	INIT_DELAYED_WORK(&bond->mcast_work,
3407 			  bond_resend_igmp_join_requests_delayed);
3408 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3409 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3410 	if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3411 		INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3412 	else
3413 		INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3414 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3415 }
3416 
3417 static void bond_work_cancel_all(struct bonding *bond)
3418 {
3419 	cancel_delayed_work_sync(&bond->mii_work);
3420 	cancel_delayed_work_sync(&bond->arp_work);
3421 	cancel_delayed_work_sync(&bond->alb_work);
3422 	cancel_delayed_work_sync(&bond->ad_work);
3423 	cancel_delayed_work_sync(&bond->mcast_work);
3424 }
3425 
3426 static int bond_open(struct net_device *bond_dev)
3427 {
3428 	struct bonding *bond = netdev_priv(bond_dev);
3429 	struct slave *slave;
3430 	int i;
3431 
3432 	/* reset slave->backup and slave->inactive */
3433 	read_lock(&bond->lock);
3434 	if (bond->slave_cnt > 0) {
3435 		read_lock(&bond->curr_slave_lock);
3436 		bond_for_each_slave(bond, slave, i) {
3437 			if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3438 				&& (slave != bond->curr_active_slave)) {
3439 				bond_set_slave_inactive_flags(slave);
3440 			} else {
3441 				bond_set_slave_active_flags(slave);
3442 			}
3443 		}
3444 		read_unlock(&bond->curr_slave_lock);
3445 	}
3446 	read_unlock(&bond->lock);
3447 
3448 	bond_work_init_all(bond);
3449 
3450 	if (bond_is_lb(bond)) {
3451 		/* bond_alb_initialize must be called before the timer
3452 		 * is started.
3453 		 */
3454 		if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3455 			return -ENOMEM;
3456 		queue_delayed_work(bond->wq, &bond->alb_work, 0);
3457 	}
3458 
3459 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
3460 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
3461 
3462 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3463 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
3464 		if (bond->params.arp_validate)
3465 			bond->recv_probe = bond_arp_rcv;
3466 	}
3467 
3468 	if (bond->params.mode == BOND_MODE_8023AD) {
3469 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
3470 		/* register to receive LACPDUs */
3471 		bond->recv_probe = bond_3ad_lacpdu_recv;
3472 		bond_3ad_initiate_agg_selection(bond, 1);
3473 	}
3474 
3475 	return 0;
3476 }
3477 
3478 static int bond_close(struct net_device *bond_dev)
3479 {
3480 	struct bonding *bond = netdev_priv(bond_dev);
3481 
3482 	write_lock_bh(&bond->lock);
3483 	bond->send_peer_notif = 0;
3484 	write_unlock_bh(&bond->lock);
3485 
3486 	bond_work_cancel_all(bond);
3487 	if (bond_is_lb(bond)) {
3488 		/* Must be called only after all
3489 		 * slaves have been released
3490 		 */
3491 		bond_alb_deinitialize(bond);
3492 	}
3493 	bond->recv_probe = NULL;
3494 
3495 	return 0;
3496 }
3497 
3498 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3499 						struct rtnl_link_stats64 *stats)
3500 {
3501 	struct bonding *bond = netdev_priv(bond_dev);
3502 	struct rtnl_link_stats64 temp;
3503 	struct slave *slave;
3504 	int i;
3505 
3506 	memset(stats, 0, sizeof(*stats));
3507 
3508 	read_lock_bh(&bond->lock);
3509 
3510 	bond_for_each_slave(bond, slave, i) {
3511 		const struct rtnl_link_stats64 *sstats =
3512 			dev_get_stats(slave->dev, &temp);
3513 
3514 		stats->rx_packets += sstats->rx_packets;
3515 		stats->rx_bytes += sstats->rx_bytes;
3516 		stats->rx_errors += sstats->rx_errors;
3517 		stats->rx_dropped += sstats->rx_dropped;
3518 
3519 		stats->tx_packets += sstats->tx_packets;
3520 		stats->tx_bytes += sstats->tx_bytes;
3521 		stats->tx_errors += sstats->tx_errors;
3522 		stats->tx_dropped += sstats->tx_dropped;
3523 
3524 		stats->multicast += sstats->multicast;
3525 		stats->collisions += sstats->collisions;
3526 
3527 		stats->rx_length_errors += sstats->rx_length_errors;
3528 		stats->rx_over_errors += sstats->rx_over_errors;
3529 		stats->rx_crc_errors += sstats->rx_crc_errors;
3530 		stats->rx_frame_errors += sstats->rx_frame_errors;
3531 		stats->rx_fifo_errors += sstats->rx_fifo_errors;
3532 		stats->rx_missed_errors += sstats->rx_missed_errors;
3533 
3534 		stats->tx_aborted_errors += sstats->tx_aborted_errors;
3535 		stats->tx_carrier_errors += sstats->tx_carrier_errors;
3536 		stats->tx_fifo_errors += sstats->tx_fifo_errors;
3537 		stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3538 		stats->tx_window_errors += sstats->tx_window_errors;
3539 	}
3540 
3541 	read_unlock_bh(&bond->lock);
3542 
3543 	return stats;
3544 }
3545 
3546 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3547 {
3548 	struct net_device *slave_dev = NULL;
3549 	struct ifbond k_binfo;
3550 	struct ifbond __user *u_binfo = NULL;
3551 	struct ifslave k_sinfo;
3552 	struct ifslave __user *u_sinfo = NULL;
3553 	struct mii_ioctl_data *mii = NULL;
3554 	struct net *net;
3555 	int res = 0;
3556 
3557 	pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3558 
3559 	switch (cmd) {
3560 	case SIOCGMIIPHY:
3561 		mii = if_mii(ifr);
3562 		if (!mii)
3563 			return -EINVAL;
3564 
3565 		mii->phy_id = 0;
3566 		/* Fall Through */
3567 	case SIOCGMIIREG:
3568 		/*
3569 		 * We do this again just in case we were called by SIOCGMIIREG
3570 		 * instead of SIOCGMIIPHY.
3571 		 */
3572 		mii = if_mii(ifr);
3573 		if (!mii)
3574 			return -EINVAL;
3575 
3576 
3577 		if (mii->reg_num == 1) {
3578 			struct bonding *bond = netdev_priv(bond_dev);
3579 			mii->val_out = 0;
3580 			read_lock(&bond->lock);
3581 			read_lock(&bond->curr_slave_lock);
3582 			if (netif_carrier_ok(bond->dev))
3583 				mii->val_out = BMSR_LSTATUS;
3584 
3585 			read_unlock(&bond->curr_slave_lock);
3586 			read_unlock(&bond->lock);
3587 		}
3588 
3589 		return 0;
3590 	case BOND_INFO_QUERY_OLD:
3591 	case SIOCBONDINFOQUERY:
3592 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
3593 
3594 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3595 			return -EFAULT;
3596 
3597 		res = bond_info_query(bond_dev, &k_binfo);
3598 		if (res == 0 &&
3599 		    copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3600 			return -EFAULT;
3601 
3602 		return res;
3603 	case BOND_SLAVE_INFO_QUERY_OLD:
3604 	case SIOCBONDSLAVEINFOQUERY:
3605 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3606 
3607 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3608 			return -EFAULT;
3609 
3610 		res = bond_slave_info_query(bond_dev, &k_sinfo);
3611 		if (res == 0 &&
3612 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3613 			return -EFAULT;
3614 
3615 		return res;
3616 	default:
3617 		/* Go on */
3618 		break;
3619 	}
3620 
3621 	net = dev_net(bond_dev);
3622 
3623 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3624 		return -EPERM;
3625 
3626 	slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3627 
3628 	pr_debug("slave_dev=%p:\n", slave_dev);
3629 
3630 	if (!slave_dev)
3631 		res = -ENODEV;
3632 	else {
3633 		pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3634 		switch (cmd) {
3635 		case BOND_ENSLAVE_OLD:
3636 		case SIOCBONDENSLAVE:
3637 			res = bond_enslave(bond_dev, slave_dev);
3638 			break;
3639 		case BOND_RELEASE_OLD:
3640 		case SIOCBONDRELEASE:
3641 			res = bond_release(bond_dev, slave_dev);
3642 			break;
3643 		case BOND_SETHWADDR_OLD:
3644 		case SIOCBONDSETHWADDR:
3645 			bond_set_dev_addr(bond_dev, slave_dev);
3646 			res = 0;
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 		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 	read_unlock(&bond->curr_slave_lock);
4006 
4007 	if (res)
4008 		/* no suitable interface, frame not sent */
4009 		kfree_skb(skb);
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 		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 					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 		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 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = 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 		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 		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 int bond_ethtool_get_settings(struct net_device *bond_dev,
4268 				     struct ethtool_cmd *ecmd)
4269 {
4270 	struct bonding *bond = netdev_priv(bond_dev);
4271 	struct slave *slave;
4272 	int i;
4273 	unsigned long speed = 0;
4274 
4275 	ecmd->duplex = DUPLEX_UNKNOWN;
4276 	ecmd->port = PORT_OTHER;
4277 
4278 	/* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4279 	 * do not need to check mode.  Though link speed might not represent
4280 	 * the true receive or transmit bandwidth (not all modes are symmetric)
4281 	 * this is an accurate maximum.
4282 	 */
4283 	read_lock(&bond->lock);
4284 	bond_for_each_slave(bond, slave, i) {
4285 		if (SLAVE_IS_OK(slave)) {
4286 			if (slave->speed != SPEED_UNKNOWN)
4287 				speed += slave->speed;
4288 			if (ecmd->duplex == DUPLEX_UNKNOWN &&
4289 			    slave->duplex != DUPLEX_UNKNOWN)
4290 				ecmd->duplex = slave->duplex;
4291 		}
4292 	}
4293 	ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4294 	read_unlock(&bond->lock);
4295 	return 0;
4296 }
4297 
4298 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4299 				     struct ethtool_drvinfo *drvinfo)
4300 {
4301 	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4302 	strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4303 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4304 		 BOND_ABI_VERSION);
4305 }
4306 
4307 static const struct ethtool_ops bond_ethtool_ops = {
4308 	.get_drvinfo		= bond_ethtool_get_drvinfo,
4309 	.get_settings		= bond_ethtool_get_settings,
4310 	.get_link		= ethtool_op_get_link,
4311 };
4312 
4313 static const struct net_device_ops bond_netdev_ops = {
4314 	.ndo_init		= bond_init,
4315 	.ndo_uninit		= bond_uninit,
4316 	.ndo_open		= bond_open,
4317 	.ndo_stop		= bond_close,
4318 	.ndo_start_xmit		= bond_start_xmit,
4319 	.ndo_select_queue	= bond_select_queue,
4320 	.ndo_get_stats64	= bond_get_stats,
4321 	.ndo_do_ioctl		= bond_do_ioctl,
4322 	.ndo_change_rx_flags	= bond_change_rx_flags,
4323 	.ndo_set_rx_mode	= bond_set_multicast_list,
4324 	.ndo_change_mtu		= bond_change_mtu,
4325 	.ndo_set_mac_address	= bond_set_mac_address,
4326 	.ndo_neigh_setup	= bond_neigh_setup,
4327 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
4328 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
4329 #ifdef CONFIG_NET_POLL_CONTROLLER
4330 	.ndo_netpoll_setup	= bond_netpoll_setup,
4331 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
4332 	.ndo_poll_controller	= bond_poll_controller,
4333 #endif
4334 	.ndo_add_slave		= bond_enslave,
4335 	.ndo_del_slave		= bond_release,
4336 	.ndo_fix_features	= bond_fix_features,
4337 };
4338 
4339 static const struct device_type bond_type = {
4340 	.name = "bond",
4341 };
4342 
4343 static void bond_destructor(struct net_device *bond_dev)
4344 {
4345 	struct bonding *bond = netdev_priv(bond_dev);
4346 	if (bond->wq)
4347 		destroy_workqueue(bond->wq);
4348 	free_netdev(bond_dev);
4349 }
4350 
4351 static void bond_setup(struct net_device *bond_dev)
4352 {
4353 	struct bonding *bond = netdev_priv(bond_dev);
4354 
4355 	/* initialize rwlocks */
4356 	rwlock_init(&bond->lock);
4357 	rwlock_init(&bond->curr_slave_lock);
4358 
4359 	bond->params = bonding_defaults;
4360 
4361 	/* Initialize pointers */
4362 	bond->dev = bond_dev;
4363 	INIT_LIST_HEAD(&bond->vlan_list);
4364 
4365 	/* Initialize the device entry points */
4366 	ether_setup(bond_dev);
4367 	bond_dev->netdev_ops = &bond_netdev_ops;
4368 	bond_dev->ethtool_ops = &bond_ethtool_ops;
4369 	bond_set_mode_ops(bond, bond->params.mode);
4370 
4371 	bond_dev->destructor = bond_destructor;
4372 
4373 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4374 
4375 	/* Initialize the device options */
4376 	bond_dev->tx_queue_len = 0;
4377 	bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4378 	bond_dev->priv_flags |= IFF_BONDING;
4379 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4380 
4381 	/* At first, we block adding VLANs. That's the only way to
4382 	 * prevent problems that occur when adding VLANs over an
4383 	 * empty bond. The block will be removed once non-challenged
4384 	 * slaves are enslaved.
4385 	 */
4386 	bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4387 
4388 	/* don't acquire bond device's netif_tx_lock when
4389 	 * transmitting */
4390 	bond_dev->features |= NETIF_F_LLTX;
4391 
4392 	/* By default, we declare the bond to be fully
4393 	 * VLAN hardware accelerated capable. Special
4394 	 * care is taken in the various xmit functions
4395 	 * when there are slaves that are not hw accel
4396 	 * capable
4397 	 */
4398 
4399 	bond_dev->hw_features = BOND_VLAN_FEATURES |
4400 				NETIF_F_HW_VLAN_CTAG_TX |
4401 				NETIF_F_HW_VLAN_CTAG_RX |
4402 				NETIF_F_HW_VLAN_CTAG_FILTER;
4403 
4404 	bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4405 	bond_dev->features |= bond_dev->hw_features;
4406 }
4407 
4408 /*
4409 * Destroy a bonding device.
4410 * Must be under rtnl_lock when this function is called.
4411 */
4412 static void bond_uninit(struct net_device *bond_dev)
4413 {
4414 	struct bonding *bond = netdev_priv(bond_dev);
4415 	struct vlan_entry *vlan, *tmp;
4416 
4417 	bond_netpoll_cleanup(bond_dev);
4418 
4419 	/* Release the bonded slaves */
4420 	while (bond->first_slave != NULL)
4421 		__bond_release_one(bond_dev, bond->first_slave->dev, true);
4422 	pr_info("%s: released all slaves\n", bond_dev->name);
4423 
4424 	list_del(&bond->bond_list);
4425 
4426 	bond_debug_unregister(bond);
4427 
4428 	__hw_addr_flush(&bond->mc_list);
4429 
4430 	list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4431 		list_del(&vlan->vlan_list);
4432 		kfree(vlan);
4433 	}
4434 }
4435 
4436 /*------------------------- Module initialization ---------------------------*/
4437 
4438 /*
4439  * Convert string input module parms.  Accept either the
4440  * number of the mode or its string name.  A bit complicated because
4441  * some mode names are substrings of other names, and calls from sysfs
4442  * may have whitespace in the name (trailing newlines, for example).
4443  */
4444 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4445 {
4446 	int modeint = -1, i, rv;
4447 	char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4448 
4449 	for (p = (char *)buf; *p; p++)
4450 		if (!(isdigit(*p) || isspace(*p)))
4451 			break;
4452 
4453 	if (*p)
4454 		rv = sscanf(buf, "%20s", modestr);
4455 	else
4456 		rv = sscanf(buf, "%d", &modeint);
4457 
4458 	if (!rv)
4459 		return -1;
4460 
4461 	for (i = 0; tbl[i].modename; i++) {
4462 		if (modeint == tbl[i].mode)
4463 			return tbl[i].mode;
4464 		if (strcmp(modestr, tbl[i].modename) == 0)
4465 			return tbl[i].mode;
4466 	}
4467 
4468 	return -1;
4469 }
4470 
4471 static int bond_check_params(struct bond_params *params)
4472 {
4473 	int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4474 
4475 	/*
4476 	 * Convert string parameters.
4477 	 */
4478 	if (mode) {
4479 		bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4480 		if (bond_mode == -1) {
4481 			pr_err("Error: Invalid bonding mode \"%s\"\n",
4482 			       mode == NULL ? "NULL" : mode);
4483 			return -EINVAL;
4484 		}
4485 	}
4486 
4487 	if (xmit_hash_policy) {
4488 		if ((bond_mode != BOND_MODE_XOR) &&
4489 		    (bond_mode != BOND_MODE_8023AD)) {
4490 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4491 			       bond_mode_name(bond_mode));
4492 		} else {
4493 			xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4494 							xmit_hashtype_tbl);
4495 			if (xmit_hashtype == -1) {
4496 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4497 				       xmit_hash_policy == NULL ? "NULL" :
4498 				       xmit_hash_policy);
4499 				return -EINVAL;
4500 			}
4501 		}
4502 	}
4503 
4504 	if (lacp_rate) {
4505 		if (bond_mode != BOND_MODE_8023AD) {
4506 			pr_info("lacp_rate param is irrelevant in mode %s\n",
4507 				bond_mode_name(bond_mode));
4508 		} else {
4509 			lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4510 			if (lacp_fast == -1) {
4511 				pr_err("Error: Invalid lacp rate \"%s\"\n",
4512 				       lacp_rate == NULL ? "NULL" : lacp_rate);
4513 				return -EINVAL;
4514 			}
4515 		}
4516 	}
4517 
4518 	if (ad_select) {
4519 		params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4520 		if (params->ad_select == -1) {
4521 			pr_err("Error: Invalid ad_select \"%s\"\n",
4522 			       ad_select == NULL ? "NULL" : ad_select);
4523 			return -EINVAL;
4524 		}
4525 
4526 		if (bond_mode != BOND_MODE_8023AD) {
4527 			pr_warning("ad_select param only affects 802.3ad mode\n");
4528 		}
4529 	} else {
4530 		params->ad_select = BOND_AD_STABLE;
4531 	}
4532 
4533 	if (max_bonds < 0) {
4534 		pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4535 			   max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4536 		max_bonds = BOND_DEFAULT_MAX_BONDS;
4537 	}
4538 
4539 	if (miimon < 0) {
4540 		pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4541 			   miimon, INT_MAX, BOND_LINK_MON_INTERV);
4542 		miimon = BOND_LINK_MON_INTERV;
4543 	}
4544 
4545 	if (updelay < 0) {
4546 		pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547 			   updelay, INT_MAX);
4548 		updelay = 0;
4549 	}
4550 
4551 	if (downdelay < 0) {
4552 		pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4553 			   downdelay, INT_MAX);
4554 		downdelay = 0;
4555 	}
4556 
4557 	if ((use_carrier != 0) && (use_carrier != 1)) {
4558 		pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4559 			   use_carrier);
4560 		use_carrier = 1;
4561 	}
4562 
4563 	if (num_peer_notif < 0 || num_peer_notif > 255) {
4564 		pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4565 			   num_peer_notif);
4566 		num_peer_notif = 1;
4567 	}
4568 
4569 	/* reset values for 802.3ad */
4570 	if (bond_mode == BOND_MODE_8023AD) {
4571 		if (!miimon) {
4572 			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");
4573 			pr_warning("Forcing miimon to 100msec\n");
4574 			miimon = 100;
4575 		}
4576 	}
4577 
4578 	if (tx_queues < 1 || tx_queues > 255) {
4579 		pr_warning("Warning: tx_queues (%d) should be between "
4580 			   "1 and 255, resetting to %d\n",
4581 			   tx_queues, BOND_DEFAULT_TX_QUEUES);
4582 		tx_queues = BOND_DEFAULT_TX_QUEUES;
4583 	}
4584 
4585 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4586 		pr_warning("Warning: all_slaves_active module parameter (%d), "
4587 			   "not of valid value (0/1), so it was set to "
4588 			   "0\n", all_slaves_active);
4589 		all_slaves_active = 0;
4590 	}
4591 
4592 	if (resend_igmp < 0 || resend_igmp > 255) {
4593 		pr_warning("Warning: resend_igmp (%d) should be between "
4594 			   "0 and 255, resetting to %d\n",
4595 			   resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4596 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4597 	}
4598 
4599 	/* reset values for TLB/ALB */
4600 	if ((bond_mode == BOND_MODE_TLB) ||
4601 	    (bond_mode == BOND_MODE_ALB)) {
4602 		if (!miimon) {
4603 			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");
4604 			pr_warning("Forcing miimon to 100msec\n");
4605 			miimon = 100;
4606 		}
4607 	}
4608 
4609 	if (bond_mode == BOND_MODE_ALB) {
4610 		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",
4611 			  updelay);
4612 	}
4613 
4614 	if (!miimon) {
4615 		if (updelay || downdelay) {
4616 			/* just warn the user the up/down delay will have
4617 			 * no effect since miimon is zero...
4618 			 */
4619 			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",
4620 				   updelay, downdelay);
4621 		}
4622 	} else {
4623 		/* don't allow arp monitoring */
4624 		if (arp_interval) {
4625 			pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4626 				   miimon, arp_interval);
4627 			arp_interval = 0;
4628 		}
4629 
4630 		if ((updelay % miimon) != 0) {
4631 			pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4632 				   updelay, miimon,
4633 				   (updelay / miimon) * miimon);
4634 		}
4635 
4636 		updelay /= miimon;
4637 
4638 		if ((downdelay % miimon) != 0) {
4639 			pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4640 				   downdelay, miimon,
4641 				   (downdelay / miimon) * miimon);
4642 		}
4643 
4644 		downdelay /= miimon;
4645 	}
4646 
4647 	if (arp_interval < 0) {
4648 		pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4649 			   arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4650 		arp_interval = BOND_LINK_ARP_INTERV;
4651 	}
4652 
4653 	for (arp_ip_count = 0;
4654 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4655 	     arp_ip_count++) {
4656 		/* not complete check, but should be good enough to
4657 		   catch mistakes */
4658 		__be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4659 		if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4660 		    ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4661 			pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4662 				   arp_ip_target[arp_ip_count]);
4663 			arp_interval = 0;
4664 		} else {
4665 			arp_target[arp_ip_count] = ip;
4666 		}
4667 	}
4668 
4669 	if (arp_interval && !arp_ip_count) {
4670 		/* don't allow arping if no arp_ip_target given... */
4671 		pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4672 			   arp_interval);
4673 		arp_interval = 0;
4674 	}
4675 
4676 	if (arp_validate) {
4677 		if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4678 			pr_err("arp_validate only supported in active-backup mode\n");
4679 			return -EINVAL;
4680 		}
4681 		if (!arp_interval) {
4682 			pr_err("arp_validate requires arp_interval\n");
4683 			return -EINVAL;
4684 		}
4685 
4686 		arp_validate_value = bond_parse_parm(arp_validate,
4687 						     arp_validate_tbl);
4688 		if (arp_validate_value == -1) {
4689 			pr_err("Error: invalid arp_validate \"%s\"\n",
4690 			       arp_validate == NULL ? "NULL" : arp_validate);
4691 			return -EINVAL;
4692 		}
4693 	} else
4694 		arp_validate_value = 0;
4695 
4696 	if (miimon) {
4697 		pr_info("MII link monitoring set to %d ms\n", miimon);
4698 	} else if (arp_interval) {
4699 		int i;
4700 
4701 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4702 			arp_interval,
4703 			arp_validate_tbl[arp_validate_value].modename,
4704 			arp_ip_count);
4705 
4706 		for (i = 0; i < arp_ip_count; i++)
4707 			pr_info(" %s", arp_ip_target[i]);
4708 
4709 		pr_info("\n");
4710 
4711 	} else if (max_bonds) {
4712 		/* miimon and arp_interval not set, we need one so things
4713 		 * work as expected, see bonding.txt for details
4714 		 */
4715 		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");
4716 	}
4717 
4718 	if (primary && !USES_PRIMARY(bond_mode)) {
4719 		/* currently, using a primary only makes sense
4720 		 * in active backup, TLB or ALB modes
4721 		 */
4722 		pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4723 			   primary, bond_mode_name(bond_mode));
4724 		primary = NULL;
4725 	}
4726 
4727 	if (primary && primary_reselect) {
4728 		primary_reselect_value = bond_parse_parm(primary_reselect,
4729 							 pri_reselect_tbl);
4730 		if (primary_reselect_value == -1) {
4731 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
4732 			       primary_reselect ==
4733 					NULL ? "NULL" : primary_reselect);
4734 			return -EINVAL;
4735 		}
4736 	} else {
4737 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4738 	}
4739 
4740 	if (fail_over_mac) {
4741 		fail_over_mac_value = bond_parse_parm(fail_over_mac,
4742 						      fail_over_mac_tbl);
4743 		if (fail_over_mac_value == -1) {
4744 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
4745 			       arp_validate == NULL ? "NULL" : arp_validate);
4746 			return -EINVAL;
4747 		}
4748 
4749 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4750 			pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4751 	} else {
4752 		fail_over_mac_value = BOND_FOM_NONE;
4753 	}
4754 
4755 	/* fill params struct with the proper values */
4756 	params->mode = bond_mode;
4757 	params->xmit_policy = xmit_hashtype;
4758 	params->miimon = miimon;
4759 	params->num_peer_notif = num_peer_notif;
4760 	params->arp_interval = arp_interval;
4761 	params->arp_validate = arp_validate_value;
4762 	params->updelay = updelay;
4763 	params->downdelay = downdelay;
4764 	params->use_carrier = use_carrier;
4765 	params->lacp_fast = lacp_fast;
4766 	params->primary[0] = 0;
4767 	params->primary_reselect = primary_reselect_value;
4768 	params->fail_over_mac = fail_over_mac_value;
4769 	params->tx_queues = tx_queues;
4770 	params->all_slaves_active = all_slaves_active;
4771 	params->resend_igmp = resend_igmp;
4772 	params->min_links = min_links;
4773 
4774 	if (primary) {
4775 		strncpy(params->primary, primary, IFNAMSIZ);
4776 		params->primary[IFNAMSIZ - 1] = 0;
4777 	}
4778 
4779 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4780 
4781 	return 0;
4782 }
4783 
4784 static struct lock_class_key bonding_netdev_xmit_lock_key;
4785 static struct lock_class_key bonding_netdev_addr_lock_key;
4786 static struct lock_class_key bonding_tx_busylock_key;
4787 
4788 static void bond_set_lockdep_class_one(struct net_device *dev,
4789 				       struct netdev_queue *txq,
4790 				       void *_unused)
4791 {
4792 	lockdep_set_class(&txq->_xmit_lock,
4793 			  &bonding_netdev_xmit_lock_key);
4794 }
4795 
4796 static void bond_set_lockdep_class(struct net_device *dev)
4797 {
4798 	lockdep_set_class(&dev->addr_list_lock,
4799 			  &bonding_netdev_addr_lock_key);
4800 	netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4801 	dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4802 }
4803 
4804 /*
4805  * Called from registration process
4806  */
4807 static int bond_init(struct net_device *bond_dev)
4808 {
4809 	struct bonding *bond = netdev_priv(bond_dev);
4810 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4811 	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4812 
4813 	pr_debug("Begin bond_init for %s\n", bond_dev->name);
4814 
4815 	/*
4816 	 * Initialize locks that may be required during
4817 	 * en/deslave operations.  All of the bond_open work
4818 	 * (of which this is part) should really be moved to
4819 	 * a phase prior to dev_open
4820 	 */
4821 	spin_lock_init(&(bond_info->tx_hashtbl_lock));
4822 	spin_lock_init(&(bond_info->rx_hashtbl_lock));
4823 
4824 	bond->wq = create_singlethread_workqueue(bond_dev->name);
4825 	if (!bond->wq)
4826 		return -ENOMEM;
4827 
4828 	bond_set_lockdep_class(bond_dev);
4829 
4830 	list_add_tail(&bond->bond_list, &bn->dev_list);
4831 
4832 	bond_prepare_sysfs_group(bond);
4833 
4834 	bond_debug_register(bond);
4835 
4836 	/* Ensure valid dev_addr */
4837 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
4838 	    bond_dev->addr_assign_type == NET_ADDR_PERM) {
4839 		eth_hw_addr_random(bond_dev);
4840 		bond->dev_addr_from_first = true;
4841 	}
4842 
4843 	__hw_addr_init(&bond->mc_list);
4844 	return 0;
4845 }
4846 
4847 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4848 {
4849 	if (tb[IFLA_ADDRESS]) {
4850 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4851 			return -EINVAL;
4852 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4853 			return -EADDRNOTAVAIL;
4854 	}
4855 	return 0;
4856 }
4857 
4858 static unsigned int bond_get_num_tx_queues(void)
4859 {
4860 	return tx_queues;
4861 }
4862 
4863 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4864 	.kind			= "bond",
4865 	.priv_size		= sizeof(struct bonding),
4866 	.setup			= bond_setup,
4867 	.validate		= bond_validate,
4868 	.get_num_tx_queues	= bond_get_num_tx_queues,
4869 	.get_num_rx_queues	= bond_get_num_tx_queues, /* Use the same number
4870 							     as for TX queues */
4871 };
4872 
4873 /* Create a new bond based on the specified name and bonding parameters.
4874  * If name is NULL, obtain a suitable "bond%d" name for us.
4875  * Caller must NOT hold rtnl_lock; we need to release it here before we
4876  * set up our sysfs entries.
4877  */
4878 int bond_create(struct net *net, const char *name)
4879 {
4880 	struct net_device *bond_dev;
4881 	int res;
4882 
4883 	rtnl_lock();
4884 
4885 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4886 				   name ? name : "bond%d",
4887 				   bond_setup, tx_queues);
4888 	if (!bond_dev) {
4889 		pr_err("%s: eek! can't alloc netdev!\n", name);
4890 		rtnl_unlock();
4891 		return -ENOMEM;
4892 	}
4893 
4894 	dev_net_set(bond_dev, net);
4895 	bond_dev->rtnl_link_ops = &bond_link_ops;
4896 
4897 	res = register_netdevice(bond_dev);
4898 
4899 	netif_carrier_off(bond_dev);
4900 
4901 	rtnl_unlock();
4902 	if (res < 0)
4903 		bond_destructor(bond_dev);
4904 	return res;
4905 }
4906 
4907 static int __net_init bond_net_init(struct net *net)
4908 {
4909 	struct bond_net *bn = net_generic(net, bond_net_id);
4910 
4911 	bn->net = net;
4912 	INIT_LIST_HEAD(&bn->dev_list);
4913 
4914 	bond_create_proc_dir(bn);
4915 	bond_create_sysfs(bn);
4916 
4917 	return 0;
4918 }
4919 
4920 static void __net_exit bond_net_exit(struct net *net)
4921 {
4922 	struct bond_net *bn = net_generic(net, bond_net_id);
4923 	struct bonding *bond, *tmp_bond;
4924 	LIST_HEAD(list);
4925 
4926 	bond_destroy_sysfs(bn);
4927 	bond_destroy_proc_dir(bn);
4928 
4929 	/* Kill off any bonds created after unregistering bond rtnl ops */
4930 	rtnl_lock();
4931 	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4932 		unregister_netdevice_queue(bond->dev, &list);
4933 	unregister_netdevice_many(&list);
4934 	rtnl_unlock();
4935 }
4936 
4937 static struct pernet_operations bond_net_ops = {
4938 	.init = bond_net_init,
4939 	.exit = bond_net_exit,
4940 	.id   = &bond_net_id,
4941 	.size = sizeof(struct bond_net),
4942 };
4943 
4944 static int __init bonding_init(void)
4945 {
4946 	int i;
4947 	int res;
4948 
4949 	pr_info("%s", bond_version);
4950 
4951 	res = bond_check_params(&bonding_defaults);
4952 	if (res)
4953 		goto out;
4954 
4955 	res = register_pernet_subsys(&bond_net_ops);
4956 	if (res)
4957 		goto out;
4958 
4959 	res = rtnl_link_register(&bond_link_ops);
4960 	if (res)
4961 		goto err_link;
4962 
4963 	bond_create_debugfs();
4964 
4965 	for (i = 0; i < max_bonds; i++) {
4966 		res = bond_create(&init_net, NULL);
4967 		if (res)
4968 			goto err;
4969 	}
4970 
4971 	register_netdevice_notifier(&bond_netdev_notifier);
4972 out:
4973 	return res;
4974 err:
4975 	rtnl_link_unregister(&bond_link_ops);
4976 err_link:
4977 	unregister_pernet_subsys(&bond_net_ops);
4978 	goto out;
4979 
4980 }
4981 
4982 static void __exit bonding_exit(void)
4983 {
4984 	unregister_netdevice_notifier(&bond_netdev_notifier);
4985 
4986 	bond_destroy_debugfs();
4987 
4988 	rtnl_link_unregister(&bond_link_ops);
4989 	unregister_pernet_subsys(&bond_net_ops);
4990 
4991 #ifdef CONFIG_NET_POLL_CONTROLLER
4992 	/*
4993 	 * Make sure we don't have an imbalance on our netpoll blocking
4994 	 */
4995 	WARN_ON(atomic_read(&netpoll_block_tx));
4996 #endif
4997 }
4998 
4999 module_init(bonding_init);
5000 module_exit(bonding_exit);
5001 MODULE_LICENSE("GPL");
5002 MODULE_VERSION(DRV_VERSION);
5003 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5004 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5005 MODULE_ALIAS_RTNL_LINK("bond");
5006