xref: /openbmc/linux/drivers/net/bonding/bond_main.c (revision 39b6f3aa)
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 	features = netdev_add_tso_features(features, mask);
1366 
1367 out:
1368 	read_unlock(&bond->lock);
1369 	return features;
1370 }
1371 
1372 #define BOND_VLAN_FEATURES	(NETIF_F_ALL_CSUM | NETIF_F_SG | \
1373 				 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1374 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1375 
1376 static void bond_compute_features(struct bonding *bond)
1377 {
1378 	struct slave *slave;
1379 	struct net_device *bond_dev = bond->dev;
1380 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1381 	unsigned short max_hard_header_len = ETH_HLEN;
1382 	unsigned int gso_max_size = GSO_MAX_SIZE;
1383 	u16 gso_max_segs = GSO_MAX_SEGS;
1384 	int i;
1385 	unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1386 
1387 	read_lock(&bond->lock);
1388 
1389 	if (!bond->first_slave)
1390 		goto done;
1391 
1392 	bond_for_each_slave(bond, slave, i) {
1393 		vlan_features = netdev_increment_features(vlan_features,
1394 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1395 
1396 		dst_release_flag &= slave->dev->priv_flags;
1397 		if (slave->dev->hard_header_len > max_hard_header_len)
1398 			max_hard_header_len = slave->dev->hard_header_len;
1399 
1400 		gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1401 		gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1402 	}
1403 
1404 done:
1405 	bond_dev->vlan_features = vlan_features;
1406 	bond_dev->hard_header_len = max_hard_header_len;
1407 	bond_dev->gso_max_segs = gso_max_segs;
1408 	netif_set_gso_max_size(bond_dev, gso_max_size);
1409 
1410 	flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1411 	bond_dev->priv_flags = flags | dst_release_flag;
1412 
1413 	read_unlock(&bond->lock);
1414 
1415 	netdev_change_features(bond_dev);
1416 }
1417 
1418 static void bond_setup_by_slave(struct net_device *bond_dev,
1419 				struct net_device *slave_dev)
1420 {
1421 	struct bonding *bond = netdev_priv(bond_dev);
1422 
1423 	bond_dev->header_ops	    = slave_dev->header_ops;
1424 
1425 	bond_dev->type		    = slave_dev->type;
1426 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1427 	bond_dev->addr_len	    = slave_dev->addr_len;
1428 
1429 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1430 		slave_dev->addr_len);
1431 	bond->setup_by_slave = 1;
1432 }
1433 
1434 /* On bonding slaves other than the currently active slave, suppress
1435  * duplicates except for alb non-mcast/bcast.
1436  */
1437 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1438 					    struct slave *slave,
1439 					    struct bonding *bond)
1440 {
1441 	if (bond_is_slave_inactive(slave)) {
1442 		if (bond->params.mode == BOND_MODE_ALB &&
1443 		    skb->pkt_type != PACKET_BROADCAST &&
1444 		    skb->pkt_type != PACKET_MULTICAST)
1445 			return false;
1446 		return true;
1447 	}
1448 	return false;
1449 }
1450 
1451 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1452 {
1453 	struct sk_buff *skb = *pskb;
1454 	struct slave *slave;
1455 	struct bonding *bond;
1456 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1457 			  struct slave *);
1458 	int ret = RX_HANDLER_ANOTHER;
1459 
1460 	skb = skb_share_check(skb, GFP_ATOMIC);
1461 	if (unlikely(!skb))
1462 		return RX_HANDLER_CONSUMED;
1463 
1464 	*pskb = skb;
1465 
1466 	slave = bond_slave_get_rcu(skb->dev);
1467 	bond = slave->bond;
1468 
1469 	if (bond->params.arp_interval)
1470 		slave->dev->last_rx = jiffies;
1471 
1472 	recv_probe = ACCESS_ONCE(bond->recv_probe);
1473 	if (recv_probe) {
1474 		ret = recv_probe(skb, bond, slave);
1475 		if (ret == RX_HANDLER_CONSUMED) {
1476 			consume_skb(skb);
1477 			return ret;
1478 		}
1479 	}
1480 
1481 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1482 		return RX_HANDLER_EXACT;
1483 	}
1484 
1485 	skb->dev = bond->dev;
1486 
1487 	if (bond->params.mode == BOND_MODE_ALB &&
1488 	    bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1489 	    skb->pkt_type == PACKET_HOST) {
1490 
1491 		if (unlikely(skb_cow_head(skb,
1492 					  skb->data - skb_mac_header(skb)))) {
1493 			kfree_skb(skb);
1494 			return RX_HANDLER_CONSUMED;
1495 		}
1496 		memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1497 	}
1498 
1499 	return ret;
1500 }
1501 
1502 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1503 				      struct net_device *slave_dev)
1504 {
1505 	int err;
1506 
1507 	err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1508 	if (err)
1509 		return err;
1510 	slave_dev->flags |= IFF_SLAVE;
1511 	rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1512 	return 0;
1513 }
1514 
1515 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1516 				  struct net_device *slave_dev)
1517 {
1518 	netdev_upper_dev_unlink(slave_dev, bond_dev);
1519 	slave_dev->flags &= ~IFF_SLAVE;
1520 	rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1521 }
1522 
1523 /* enslave device <slave> to bond device <master> */
1524 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1525 {
1526 	struct bonding *bond = netdev_priv(bond_dev);
1527 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1528 	struct slave *new_slave = NULL;
1529 	struct netdev_hw_addr *ha;
1530 	struct sockaddr addr;
1531 	int link_reporting;
1532 	int res = 0;
1533 
1534 	if (!bond->params.use_carrier &&
1535 	    slave_dev->ethtool_ops->get_link == NULL &&
1536 	    slave_ops->ndo_do_ioctl == NULL) {
1537 		pr_warning("%s: Warning: no link monitoring support for %s\n",
1538 			   bond_dev->name, slave_dev->name);
1539 	}
1540 
1541 	/* already enslaved */
1542 	if (slave_dev->flags & IFF_SLAVE) {
1543 		pr_debug("Error, Device was already enslaved\n");
1544 		return -EBUSY;
1545 	}
1546 
1547 	/* vlan challenged mutual exclusion */
1548 	/* no need to lock since we're protected by rtnl_lock */
1549 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1550 		pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1551 		if (vlan_uses_dev(bond_dev)) {
1552 			pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1553 			       bond_dev->name, slave_dev->name, bond_dev->name);
1554 			return -EPERM;
1555 		} else {
1556 			pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1557 				   bond_dev->name, slave_dev->name,
1558 				   slave_dev->name, bond_dev->name);
1559 		}
1560 	} else {
1561 		pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1562 	}
1563 
1564 	/*
1565 	 * Old ifenslave binaries are no longer supported.  These can
1566 	 * be identified with moderate accuracy by the state of the slave:
1567 	 * the current ifenslave will set the interface down prior to
1568 	 * enslaving it; the old ifenslave will not.
1569 	 */
1570 	if ((slave_dev->flags & IFF_UP)) {
1571 		pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1572 		       slave_dev->name);
1573 		res = -EPERM;
1574 		goto err_undo_flags;
1575 	}
1576 
1577 	/* set bonding device ether type by slave - bonding netdevices are
1578 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1579 	 * there is a need to override some of the type dependent attribs/funcs.
1580 	 *
1581 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1582 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1583 	 */
1584 	if (bond->slave_cnt == 0) {
1585 		if (bond_dev->type != slave_dev->type) {
1586 			pr_debug("%s: change device type from %d to %d\n",
1587 				 bond_dev->name,
1588 				 bond_dev->type, slave_dev->type);
1589 
1590 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1591 						       bond_dev);
1592 			res = notifier_to_errno(res);
1593 			if (res) {
1594 				pr_err("%s: refused to change device type\n",
1595 				       bond_dev->name);
1596 				res = -EBUSY;
1597 				goto err_undo_flags;
1598 			}
1599 
1600 			/* Flush unicast and multicast addresses */
1601 			dev_uc_flush(bond_dev);
1602 			dev_mc_flush(bond_dev);
1603 
1604 			if (slave_dev->type != ARPHRD_ETHER)
1605 				bond_setup_by_slave(bond_dev, slave_dev);
1606 			else {
1607 				ether_setup(bond_dev);
1608 				bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1609 			}
1610 
1611 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1612 						 bond_dev);
1613 		}
1614 	} else if (bond_dev->type != slave_dev->type) {
1615 		pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1616 		       slave_dev->name,
1617 		       slave_dev->type, bond_dev->type);
1618 		res = -EINVAL;
1619 		goto err_undo_flags;
1620 	}
1621 
1622 	if (slave_ops->ndo_set_mac_address == NULL) {
1623 		if (bond->slave_cnt == 0) {
1624 			pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1625 				   bond_dev->name);
1626 			bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1627 		} else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1628 			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",
1629 			       bond_dev->name);
1630 			res = -EOPNOTSUPP;
1631 			goto err_undo_flags;
1632 		}
1633 	}
1634 
1635 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1636 
1637 	/* If this is the first slave, then we need to set the master's hardware
1638 	 * address to be the same as the slave's. */
1639 	if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1640 		bond_set_dev_addr(bond->dev, slave_dev);
1641 
1642 	new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1643 	if (!new_slave) {
1644 		res = -ENOMEM;
1645 		goto err_undo_flags;
1646 	}
1647 
1648 	/*
1649 	 * Set the new_slave's queue_id to be zero.  Queue ID mapping
1650 	 * is set via sysfs or module option if desired.
1651 	 */
1652 	new_slave->queue_id = 0;
1653 
1654 	/* Save slave's original mtu and then set it to match the bond */
1655 	new_slave->original_mtu = slave_dev->mtu;
1656 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1657 	if (res) {
1658 		pr_debug("Error %d calling dev_set_mtu\n", res);
1659 		goto err_free;
1660 	}
1661 
1662 	/*
1663 	 * Save slave's original ("permanent") mac address for modes
1664 	 * that need it, and for restoring it upon release, and then
1665 	 * set it to the master's address
1666 	 */
1667 	memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1668 
1669 	if (!bond->params.fail_over_mac) {
1670 		/*
1671 		 * Set slave to master's mac address.  The application already
1672 		 * set the master's mac address to that of the first slave
1673 		 */
1674 		memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1675 		addr.sa_family = slave_dev->type;
1676 		res = dev_set_mac_address(slave_dev, &addr);
1677 		if (res) {
1678 			pr_debug("Error %d calling set_mac_address\n", res);
1679 			goto err_restore_mtu;
1680 		}
1681 	}
1682 
1683 	res = bond_master_upper_dev_link(bond_dev, slave_dev);
1684 	if (res) {
1685 		pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1686 		goto err_restore_mac;
1687 	}
1688 
1689 	/* open the slave since the application closed it */
1690 	res = dev_open(slave_dev);
1691 	if (res) {
1692 		pr_debug("Opening slave %s failed\n", slave_dev->name);
1693 		goto err_unset_master;
1694 	}
1695 
1696 	new_slave->bond = bond;
1697 	new_slave->dev = slave_dev;
1698 	slave_dev->priv_flags |= IFF_BONDING;
1699 
1700 	if (bond_is_lb(bond)) {
1701 		/* bond_alb_init_slave() must be called before all other stages since
1702 		 * it might fail and we do not want to have to undo everything
1703 		 */
1704 		res = bond_alb_init_slave(bond, new_slave);
1705 		if (res)
1706 			goto err_close;
1707 	}
1708 
1709 	/* If the mode USES_PRIMARY, then the new slave gets the
1710 	 * master's promisc (and mc) settings only if it becomes the
1711 	 * curr_active_slave, and that is taken care of later when calling
1712 	 * bond_change_active()
1713 	 */
1714 	if (!USES_PRIMARY(bond->params.mode)) {
1715 		/* set promiscuity level to new slave */
1716 		if (bond_dev->flags & IFF_PROMISC) {
1717 			res = dev_set_promiscuity(slave_dev, 1);
1718 			if (res)
1719 				goto err_close;
1720 		}
1721 
1722 		/* set allmulti level to new slave */
1723 		if (bond_dev->flags & IFF_ALLMULTI) {
1724 			res = dev_set_allmulti(slave_dev, 1);
1725 			if (res)
1726 				goto err_close;
1727 		}
1728 
1729 		netif_addr_lock_bh(bond_dev);
1730 		/* upload master's mc_list to new slave */
1731 		netdev_for_each_mc_addr(ha, bond_dev)
1732 			dev_mc_add(slave_dev, ha->addr);
1733 		netif_addr_unlock_bh(bond_dev);
1734 	}
1735 
1736 	if (bond->params.mode == BOND_MODE_8023AD) {
1737 		/* add lacpdu mc addr to mc list */
1738 		u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1739 
1740 		dev_mc_add(slave_dev, lacpdu_multicast);
1741 	}
1742 
1743 	bond_add_vlans_on_slave(bond, slave_dev);
1744 
1745 	write_lock_bh(&bond->lock);
1746 
1747 	bond_attach_slave(bond, new_slave);
1748 
1749 	new_slave->delay = 0;
1750 	new_slave->link_failure_count = 0;
1751 
1752 	write_unlock_bh(&bond->lock);
1753 
1754 	bond_compute_features(bond);
1755 
1756 	bond_update_speed_duplex(new_slave);
1757 
1758 	read_lock(&bond->lock);
1759 
1760 	new_slave->last_arp_rx = jiffies -
1761 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
1762 
1763 	if (bond->params.miimon && !bond->params.use_carrier) {
1764 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1765 
1766 		if ((link_reporting == -1) && !bond->params.arp_interval) {
1767 			/*
1768 			 * miimon is set but a bonded network driver
1769 			 * does not support ETHTOOL/MII and
1770 			 * arp_interval is not set.  Note: if
1771 			 * use_carrier is enabled, we will never go
1772 			 * here (because netif_carrier is always
1773 			 * supported); thus, we don't need to change
1774 			 * the messages for netif_carrier.
1775 			 */
1776 			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",
1777 			       bond_dev->name, slave_dev->name);
1778 		} else if (link_reporting == -1) {
1779 			/* unable get link status using mii/ethtool */
1780 			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",
1781 				   bond_dev->name, slave_dev->name);
1782 		}
1783 	}
1784 
1785 	/* check for initial state */
1786 	if (bond->params.miimon) {
1787 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1788 			if (bond->params.updelay) {
1789 				new_slave->link = BOND_LINK_BACK;
1790 				new_slave->delay = bond->params.updelay;
1791 			} else {
1792 				new_slave->link = BOND_LINK_UP;
1793 			}
1794 		} else {
1795 			new_slave->link = BOND_LINK_DOWN;
1796 		}
1797 	} else if (bond->params.arp_interval) {
1798 		new_slave->link = (netif_carrier_ok(slave_dev) ?
1799 			BOND_LINK_UP : BOND_LINK_DOWN);
1800 	} else {
1801 		new_slave->link = BOND_LINK_UP;
1802 	}
1803 
1804 	if (new_slave->link != BOND_LINK_DOWN)
1805 		new_slave->jiffies = jiffies;
1806 	pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1807 		new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1808 			(new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1809 
1810 	if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1811 		/* if there is a primary slave, remember it */
1812 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1813 			bond->primary_slave = new_slave;
1814 			bond->force_primary = true;
1815 		}
1816 	}
1817 
1818 	write_lock_bh(&bond->curr_slave_lock);
1819 
1820 	switch (bond->params.mode) {
1821 	case BOND_MODE_ACTIVEBACKUP:
1822 		bond_set_slave_inactive_flags(new_slave);
1823 		bond_select_active_slave(bond);
1824 		break;
1825 	case BOND_MODE_8023AD:
1826 		/* in 802.3ad mode, the internal mechanism
1827 		 * will activate the slaves in the selected
1828 		 * aggregator
1829 		 */
1830 		bond_set_slave_inactive_flags(new_slave);
1831 		/* if this is the first slave */
1832 		if (bond->slave_cnt == 1) {
1833 			SLAVE_AD_INFO(new_slave).id = 1;
1834 			/* Initialize AD with the number of times that the AD timer is called in 1 second
1835 			 * can be called only after the mac address of the bond is set
1836 			 */
1837 			bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1838 		} else {
1839 			SLAVE_AD_INFO(new_slave).id =
1840 				SLAVE_AD_INFO(new_slave->prev).id + 1;
1841 		}
1842 
1843 		bond_3ad_bind_slave(new_slave);
1844 		break;
1845 	case BOND_MODE_TLB:
1846 	case BOND_MODE_ALB:
1847 		bond_set_active_slave(new_slave);
1848 		bond_set_slave_inactive_flags(new_slave);
1849 		bond_select_active_slave(bond);
1850 		break;
1851 	default:
1852 		pr_debug("This slave is always active in trunk mode\n");
1853 
1854 		/* always active in trunk mode */
1855 		bond_set_active_slave(new_slave);
1856 
1857 		/* In trunking mode there is little meaning to curr_active_slave
1858 		 * anyway (it holds no special properties of the bond device),
1859 		 * so we can change it without calling change_active_interface()
1860 		 */
1861 		if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1862 			bond->curr_active_slave = new_slave;
1863 
1864 		break;
1865 	} /* switch(bond_mode) */
1866 
1867 	write_unlock_bh(&bond->curr_slave_lock);
1868 
1869 	bond_set_carrier(bond);
1870 
1871 #ifdef CONFIG_NET_POLL_CONTROLLER
1872 	slave_dev->npinfo = bond_netpoll_info(bond);
1873 	if (slave_dev->npinfo) {
1874 		if (slave_enable_netpoll(new_slave)) {
1875 			read_unlock(&bond->lock);
1876 			pr_info("Error, %s: master_dev is using netpoll, "
1877 				 "but new slave device does not support netpoll.\n",
1878 				 bond_dev->name);
1879 			res = -EBUSY;
1880 			goto err_detach;
1881 		}
1882 	}
1883 #endif
1884 
1885 	read_unlock(&bond->lock);
1886 
1887 	res = bond_create_slave_symlinks(bond_dev, slave_dev);
1888 	if (res)
1889 		goto err_detach;
1890 
1891 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1892 					 new_slave);
1893 	if (res) {
1894 		pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1895 		goto err_dest_symlinks;
1896 	}
1897 
1898 	pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1899 		bond_dev->name, slave_dev->name,
1900 		bond_is_active_slave(new_slave) ? "n active" : " backup",
1901 		new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1902 
1903 	/* enslave is successful */
1904 	return 0;
1905 
1906 /* Undo stages on error */
1907 err_dest_symlinks:
1908 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
1909 
1910 err_detach:
1911 	if (!USES_PRIMARY(bond->params.mode)) {
1912 		netif_addr_lock_bh(bond_dev);
1913 		bond_mc_list_flush(bond_dev, slave_dev);
1914 		netif_addr_unlock_bh(bond_dev);
1915 	}
1916 	bond_del_vlans_from_slave(bond, slave_dev);
1917 	write_lock_bh(&bond->lock);
1918 	bond_detach_slave(bond, new_slave);
1919 	if (bond->primary_slave == new_slave)
1920 		bond->primary_slave = NULL;
1921 	if (bond->curr_active_slave == new_slave) {
1922 		bond_change_active_slave(bond, NULL);
1923 		write_unlock_bh(&bond->lock);
1924 		read_lock(&bond->lock);
1925 		write_lock_bh(&bond->curr_slave_lock);
1926 		bond_select_active_slave(bond);
1927 		write_unlock_bh(&bond->curr_slave_lock);
1928 		read_unlock(&bond->lock);
1929 	} else {
1930 		write_unlock_bh(&bond->lock);
1931 	}
1932 	slave_disable_netpoll(new_slave);
1933 
1934 err_close:
1935 	slave_dev->priv_flags &= ~IFF_BONDING;
1936 	dev_close(slave_dev);
1937 
1938 err_unset_master:
1939 	bond_upper_dev_unlink(bond_dev, slave_dev);
1940 
1941 err_restore_mac:
1942 	if (!bond->params.fail_over_mac) {
1943 		/* XXX TODO - fom follow mode needs to change master's
1944 		 * MAC if this slave's MAC is in use by the bond, or at
1945 		 * least print a warning.
1946 		 */
1947 		memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1948 		addr.sa_family = slave_dev->type;
1949 		dev_set_mac_address(slave_dev, &addr);
1950 	}
1951 
1952 err_restore_mtu:
1953 	dev_set_mtu(slave_dev, new_slave->original_mtu);
1954 
1955 err_free:
1956 	kfree(new_slave);
1957 
1958 err_undo_flags:
1959 	bond_compute_features(bond);
1960 
1961 	return res;
1962 }
1963 
1964 /*
1965  * Try to release the slave device <slave> from the bond device <master>
1966  * It is legal to access curr_active_slave without a lock because all the function
1967  * is write-locked. If "all" is true it means that the function is being called
1968  * while destroying a bond interface and all slaves are being released.
1969  *
1970  * The rules for slave state should be:
1971  *   for Active/Backup:
1972  *     Active stays on all backups go down
1973  *   for Bonded connections:
1974  *     The first up interface should be left on and all others downed.
1975  */
1976 static int __bond_release_one(struct net_device *bond_dev,
1977 			      struct net_device *slave_dev,
1978 			      bool all)
1979 {
1980 	struct bonding *bond = netdev_priv(bond_dev);
1981 	struct slave *slave, *oldcurrent;
1982 	struct sockaddr addr;
1983 	netdev_features_t old_features = bond_dev->features;
1984 
1985 	/* slave is not a slave or master is not master of this slave */
1986 	if (!(slave_dev->flags & IFF_SLAVE) ||
1987 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
1988 		pr_err("%s: Error: cannot release %s.\n",
1989 		       bond_dev->name, slave_dev->name);
1990 		return -EINVAL;
1991 	}
1992 
1993 	block_netpoll_tx();
1994 	write_lock_bh(&bond->lock);
1995 
1996 	slave = bond_get_slave_by_dev(bond, slave_dev);
1997 	if (!slave) {
1998 		/* not a slave of this bond */
1999 		pr_info("%s: %s not enslaved\n",
2000 			bond_dev->name, slave_dev->name);
2001 		write_unlock_bh(&bond->lock);
2002 		unblock_netpoll_tx();
2003 		return -EINVAL;
2004 	}
2005 
2006 	write_unlock_bh(&bond->lock);
2007 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2008 	 * for this slave anymore.
2009 	 */
2010 	netdev_rx_handler_unregister(slave_dev);
2011 	write_lock_bh(&bond->lock);
2012 
2013 	if (!all && !bond->params.fail_over_mac) {
2014 		if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2015 		    bond->slave_cnt > 1)
2016 			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",
2017 				   bond_dev->name, slave_dev->name,
2018 				   slave->perm_hwaddr,
2019 				   bond_dev->name, slave_dev->name);
2020 	}
2021 
2022 	/* Inform AD package of unbinding of slave. */
2023 	if (bond->params.mode == BOND_MODE_8023AD) {
2024 		/* must be called before the slave is
2025 		 * detached from the list
2026 		 */
2027 		bond_3ad_unbind_slave(slave);
2028 	}
2029 
2030 	pr_info("%s: releasing %s interface %s\n",
2031 		bond_dev->name,
2032 		bond_is_active_slave(slave) ? "active" : "backup",
2033 		slave_dev->name);
2034 
2035 	oldcurrent = bond->curr_active_slave;
2036 
2037 	bond->current_arp_slave = NULL;
2038 
2039 	/* release the slave from its bond */
2040 	bond_detach_slave(bond, slave);
2041 
2042 	if (bond->primary_slave == slave)
2043 		bond->primary_slave = NULL;
2044 
2045 	if (oldcurrent == slave)
2046 		bond_change_active_slave(bond, NULL);
2047 
2048 	if (bond_is_lb(bond)) {
2049 		/* Must be called only after the slave has been
2050 		 * detached from the list and the curr_active_slave
2051 		 * has been cleared (if our_slave == old_current),
2052 		 * but before a new active slave is selected.
2053 		 */
2054 		write_unlock_bh(&bond->lock);
2055 		bond_alb_deinit_slave(bond, slave);
2056 		write_lock_bh(&bond->lock);
2057 	}
2058 
2059 	if (all) {
2060 		bond->curr_active_slave = NULL;
2061 	} else if (oldcurrent == slave) {
2062 		/*
2063 		 * Note that we hold RTNL over this sequence, so there
2064 		 * is no concern that another slave add/remove event
2065 		 * will interfere.
2066 		 */
2067 		write_unlock_bh(&bond->lock);
2068 		read_lock(&bond->lock);
2069 		write_lock_bh(&bond->curr_slave_lock);
2070 
2071 		bond_select_active_slave(bond);
2072 
2073 		write_unlock_bh(&bond->curr_slave_lock);
2074 		read_unlock(&bond->lock);
2075 		write_lock_bh(&bond->lock);
2076 	}
2077 
2078 	if (bond->slave_cnt == 0) {
2079 		bond_set_carrier(bond);
2080 		eth_hw_addr_random(bond_dev);
2081 		bond->dev_addr_from_first = true;
2082 
2083 		if (bond_vlan_used(bond)) {
2084 			pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2085 				   bond_dev->name, bond_dev->name);
2086 			pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2087 				   bond_dev->name);
2088 		}
2089 	}
2090 
2091 	write_unlock_bh(&bond->lock);
2092 	unblock_netpoll_tx();
2093 
2094 	if (bond->slave_cnt == 0) {
2095 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2096 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2097 	}
2098 
2099 	bond_compute_features(bond);
2100 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2101 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2102 		pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2103 			bond_dev->name, slave_dev->name, bond_dev->name);
2104 
2105 	/* must do this from outside any spinlocks */
2106 	bond_destroy_slave_symlinks(bond_dev, slave_dev);
2107 
2108 	bond_del_vlans_from_slave(bond, slave_dev);
2109 
2110 	/* If the mode USES_PRIMARY, then we should only remove its
2111 	 * promisc and mc settings if it was the curr_active_slave, but that was
2112 	 * already taken care of above when we detached the slave
2113 	 */
2114 	if (!USES_PRIMARY(bond->params.mode)) {
2115 		/* unset promiscuity level from slave */
2116 		if (bond_dev->flags & IFF_PROMISC)
2117 			dev_set_promiscuity(slave_dev, -1);
2118 
2119 		/* unset allmulti level from slave */
2120 		if (bond_dev->flags & IFF_ALLMULTI)
2121 			dev_set_allmulti(slave_dev, -1);
2122 
2123 		/* flush master's mc_list from slave */
2124 		netif_addr_lock_bh(bond_dev);
2125 		bond_mc_list_flush(bond_dev, slave_dev);
2126 		netif_addr_unlock_bh(bond_dev);
2127 	}
2128 
2129 	bond_upper_dev_unlink(bond_dev, slave_dev);
2130 
2131 	slave_disable_netpoll(slave);
2132 
2133 	/* close slave before restoring its mac address */
2134 	dev_close(slave_dev);
2135 
2136 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2137 		/* restore original ("permanent") mac address */
2138 		memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2139 		addr.sa_family = slave_dev->type;
2140 		dev_set_mac_address(slave_dev, &addr);
2141 	}
2142 
2143 	dev_set_mtu(slave_dev, slave->original_mtu);
2144 
2145 	slave_dev->priv_flags &= ~IFF_BONDING;
2146 
2147 	kfree(slave);
2148 
2149 	return 0;  /* deletion OK */
2150 }
2151 
2152 /* A wrapper used because of ndo_del_link */
2153 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2154 {
2155 	return __bond_release_one(bond_dev, slave_dev, false);
2156 }
2157 
2158 /*
2159 * First release a slave and then destroy the bond if no more slaves are left.
2160 * Must be under rtnl_lock when this function is called.
2161 */
2162 static int  bond_release_and_destroy(struct net_device *bond_dev,
2163 				     struct net_device *slave_dev)
2164 {
2165 	struct bonding *bond = netdev_priv(bond_dev);
2166 	int ret;
2167 
2168 	ret = bond_release(bond_dev, slave_dev);
2169 	if ((ret == 0) && (bond->slave_cnt == 0)) {
2170 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2171 		pr_info("%s: destroying bond %s.\n",
2172 			bond_dev->name, bond_dev->name);
2173 		unregister_netdevice(bond_dev);
2174 	}
2175 	return ret;
2176 }
2177 
2178 /*
2179  * This function changes the active slave to slave <slave_dev>.
2180  * It returns -EINVAL in the following cases.
2181  *  - <slave_dev> is not found in the list.
2182  *  - There is not active slave now.
2183  *  - <slave_dev> is already active.
2184  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2185  *  - <slave_dev> is not running.
2186  * In these cases, this function does nothing.
2187  * In the other cases, current_slave pointer is changed and 0 is returned.
2188  */
2189 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2190 {
2191 	struct bonding *bond = netdev_priv(bond_dev);
2192 	struct slave *old_active = NULL;
2193 	struct slave *new_active = NULL;
2194 	int res = 0;
2195 
2196 	if (!USES_PRIMARY(bond->params.mode))
2197 		return -EINVAL;
2198 
2199 	/* Verify that bond_dev is indeed the master of slave_dev */
2200 	if (!(slave_dev->flags & IFF_SLAVE) ||
2201 	    !netdev_has_upper_dev(slave_dev, bond_dev))
2202 		return -EINVAL;
2203 
2204 	read_lock(&bond->lock);
2205 
2206 	read_lock(&bond->curr_slave_lock);
2207 	old_active = bond->curr_active_slave;
2208 	read_unlock(&bond->curr_slave_lock);
2209 
2210 	new_active = bond_get_slave_by_dev(bond, slave_dev);
2211 
2212 	/*
2213 	 * Changing to the current active: do nothing; return success.
2214 	 */
2215 	if (new_active && (new_active == old_active)) {
2216 		read_unlock(&bond->lock);
2217 		return 0;
2218 	}
2219 
2220 	if ((new_active) &&
2221 	    (old_active) &&
2222 	    (new_active->link == BOND_LINK_UP) &&
2223 	    IS_UP(new_active->dev)) {
2224 		block_netpoll_tx();
2225 		write_lock_bh(&bond->curr_slave_lock);
2226 		bond_change_active_slave(bond, new_active);
2227 		write_unlock_bh(&bond->curr_slave_lock);
2228 		unblock_netpoll_tx();
2229 	} else
2230 		res = -EINVAL;
2231 
2232 	read_unlock(&bond->lock);
2233 
2234 	return res;
2235 }
2236 
2237 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2238 {
2239 	struct bonding *bond = netdev_priv(bond_dev);
2240 
2241 	info->bond_mode = bond->params.mode;
2242 	info->miimon = bond->params.miimon;
2243 
2244 	read_lock(&bond->lock);
2245 	info->num_slaves = bond->slave_cnt;
2246 	read_unlock(&bond->lock);
2247 
2248 	return 0;
2249 }
2250 
2251 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2252 {
2253 	struct bonding *bond = netdev_priv(bond_dev);
2254 	struct slave *slave;
2255 	int i, res = -ENODEV;
2256 
2257 	read_lock(&bond->lock);
2258 
2259 	bond_for_each_slave(bond, slave, i) {
2260 		if (i == (int)info->slave_id) {
2261 			res = 0;
2262 			strcpy(info->slave_name, slave->dev->name);
2263 			info->link = slave->link;
2264 			info->state = bond_slave_state(slave);
2265 			info->link_failure_count = slave->link_failure_count;
2266 			break;
2267 		}
2268 	}
2269 
2270 	read_unlock(&bond->lock);
2271 
2272 	return res;
2273 }
2274 
2275 /*-------------------------------- Monitoring -------------------------------*/
2276 
2277 
2278 static int bond_miimon_inspect(struct bonding *bond)
2279 {
2280 	struct slave *slave;
2281 	int i, link_state, commit = 0;
2282 	bool ignore_updelay;
2283 
2284 	ignore_updelay = !bond->curr_active_slave ? true : false;
2285 
2286 	bond_for_each_slave(bond, slave, i) {
2287 		slave->new_link = BOND_LINK_NOCHANGE;
2288 
2289 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2290 
2291 		switch (slave->link) {
2292 		case BOND_LINK_UP:
2293 			if (link_state)
2294 				continue;
2295 
2296 			slave->link = BOND_LINK_FAIL;
2297 			slave->delay = bond->params.downdelay;
2298 			if (slave->delay) {
2299 				pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2300 					bond->dev->name,
2301 					(bond->params.mode ==
2302 					 BOND_MODE_ACTIVEBACKUP) ?
2303 					(bond_is_active_slave(slave) ?
2304 					 "active " : "backup ") : "",
2305 					slave->dev->name,
2306 					bond->params.downdelay * bond->params.miimon);
2307 			}
2308 			/*FALLTHRU*/
2309 		case BOND_LINK_FAIL:
2310 			if (link_state) {
2311 				/*
2312 				 * recovered before downdelay expired
2313 				 */
2314 				slave->link = BOND_LINK_UP;
2315 				slave->jiffies = jiffies;
2316 				pr_info("%s: link status up again after %d ms for interface %s.\n",
2317 					bond->dev->name,
2318 					(bond->params.downdelay - slave->delay) *
2319 					bond->params.miimon,
2320 					slave->dev->name);
2321 				continue;
2322 			}
2323 
2324 			if (slave->delay <= 0) {
2325 				slave->new_link = BOND_LINK_DOWN;
2326 				commit++;
2327 				continue;
2328 			}
2329 
2330 			slave->delay--;
2331 			break;
2332 
2333 		case BOND_LINK_DOWN:
2334 			if (!link_state)
2335 				continue;
2336 
2337 			slave->link = BOND_LINK_BACK;
2338 			slave->delay = bond->params.updelay;
2339 
2340 			if (slave->delay) {
2341 				pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2342 					bond->dev->name, slave->dev->name,
2343 					ignore_updelay ? 0 :
2344 					bond->params.updelay *
2345 					bond->params.miimon);
2346 			}
2347 			/*FALLTHRU*/
2348 		case BOND_LINK_BACK:
2349 			if (!link_state) {
2350 				slave->link = BOND_LINK_DOWN;
2351 				pr_info("%s: link status down again after %d ms for interface %s.\n",
2352 					bond->dev->name,
2353 					(bond->params.updelay - slave->delay) *
2354 					bond->params.miimon,
2355 					slave->dev->name);
2356 
2357 				continue;
2358 			}
2359 
2360 			if (ignore_updelay)
2361 				slave->delay = 0;
2362 
2363 			if (slave->delay <= 0) {
2364 				slave->new_link = BOND_LINK_UP;
2365 				commit++;
2366 				ignore_updelay = false;
2367 				continue;
2368 			}
2369 
2370 			slave->delay--;
2371 			break;
2372 		}
2373 	}
2374 
2375 	return commit;
2376 }
2377 
2378 static void bond_miimon_commit(struct bonding *bond)
2379 {
2380 	struct slave *slave;
2381 	int i;
2382 
2383 	bond_for_each_slave(bond, slave, i) {
2384 		switch (slave->new_link) {
2385 		case BOND_LINK_NOCHANGE:
2386 			continue;
2387 
2388 		case BOND_LINK_UP:
2389 			slave->link = BOND_LINK_UP;
2390 			slave->jiffies = jiffies;
2391 
2392 			if (bond->params.mode == BOND_MODE_8023AD) {
2393 				/* prevent it from being the active one */
2394 				bond_set_backup_slave(slave);
2395 			} else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2396 				/* make it immediately active */
2397 				bond_set_active_slave(slave);
2398 			} else if (slave != bond->primary_slave) {
2399 				/* prevent it from being the active one */
2400 				bond_set_backup_slave(slave);
2401 			}
2402 
2403 			pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2404 				bond->dev->name, slave->dev->name,
2405 				slave->speed, slave->duplex ? "full" : "half");
2406 
2407 			/* notify ad that the link status has changed */
2408 			if (bond->params.mode == BOND_MODE_8023AD)
2409 				bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2410 
2411 			if (bond_is_lb(bond))
2412 				bond_alb_handle_link_change(bond, slave,
2413 							    BOND_LINK_UP);
2414 
2415 			if (!bond->curr_active_slave ||
2416 			    (slave == bond->primary_slave))
2417 				goto do_failover;
2418 
2419 			continue;
2420 
2421 		case BOND_LINK_DOWN:
2422 			if (slave->link_failure_count < UINT_MAX)
2423 				slave->link_failure_count++;
2424 
2425 			slave->link = BOND_LINK_DOWN;
2426 
2427 			if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2428 			    bond->params.mode == BOND_MODE_8023AD)
2429 				bond_set_slave_inactive_flags(slave);
2430 
2431 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
2432 				bond->dev->name, slave->dev->name);
2433 
2434 			if (bond->params.mode == BOND_MODE_8023AD)
2435 				bond_3ad_handle_link_change(slave,
2436 							    BOND_LINK_DOWN);
2437 
2438 			if (bond_is_lb(bond))
2439 				bond_alb_handle_link_change(bond, slave,
2440 							    BOND_LINK_DOWN);
2441 
2442 			if (slave == bond->curr_active_slave)
2443 				goto do_failover;
2444 
2445 			continue;
2446 
2447 		default:
2448 			pr_err("%s: invalid new link %d on slave %s\n",
2449 			       bond->dev->name, slave->new_link,
2450 			       slave->dev->name);
2451 			slave->new_link = BOND_LINK_NOCHANGE;
2452 
2453 			continue;
2454 		}
2455 
2456 do_failover:
2457 		ASSERT_RTNL();
2458 		block_netpoll_tx();
2459 		write_lock_bh(&bond->curr_slave_lock);
2460 		bond_select_active_slave(bond);
2461 		write_unlock_bh(&bond->curr_slave_lock);
2462 		unblock_netpoll_tx();
2463 	}
2464 
2465 	bond_set_carrier(bond);
2466 }
2467 
2468 /*
2469  * bond_mii_monitor
2470  *
2471  * Really a wrapper that splits the mii monitor into two phases: an
2472  * inspection, then (if inspection indicates something needs to be done)
2473  * an acquisition of appropriate locks followed by a commit phase to
2474  * implement whatever link state changes are indicated.
2475  */
2476 void bond_mii_monitor(struct work_struct *work)
2477 {
2478 	struct bonding *bond = container_of(work, struct bonding,
2479 					    mii_work.work);
2480 	bool should_notify_peers = false;
2481 	unsigned long delay;
2482 
2483 	read_lock(&bond->lock);
2484 
2485 	delay = msecs_to_jiffies(bond->params.miimon);
2486 
2487 	if (bond->slave_cnt == 0)
2488 		goto re_arm;
2489 
2490 	should_notify_peers = bond_should_notify_peers(bond);
2491 
2492 	if (bond_miimon_inspect(bond)) {
2493 		read_unlock(&bond->lock);
2494 
2495 		/* Race avoidance with bond_close cancel of workqueue */
2496 		if (!rtnl_trylock()) {
2497 			read_lock(&bond->lock);
2498 			delay = 1;
2499 			should_notify_peers = false;
2500 			goto re_arm;
2501 		}
2502 
2503 		read_lock(&bond->lock);
2504 
2505 		bond_miimon_commit(bond);
2506 
2507 		read_unlock(&bond->lock);
2508 		rtnl_unlock();	/* might sleep, hold no other locks */
2509 		read_lock(&bond->lock);
2510 	}
2511 
2512 re_arm:
2513 	if (bond->params.miimon)
2514 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2515 
2516 	read_unlock(&bond->lock);
2517 
2518 	if (should_notify_peers) {
2519 		if (!rtnl_trylock()) {
2520 			read_lock(&bond->lock);
2521 			bond->send_peer_notif++;
2522 			read_unlock(&bond->lock);
2523 			return;
2524 		}
2525 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2526 		rtnl_unlock();
2527 	}
2528 }
2529 
2530 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2531 {
2532 	struct vlan_entry *vlan;
2533 	struct net_device *vlan_dev;
2534 
2535 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2536 		return 1;
2537 
2538 	list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2539 		rcu_read_lock();
2540 		vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2541 						vlan->vlan_id);
2542 		rcu_read_unlock();
2543 		if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2544 			return 1;
2545 	}
2546 
2547 	return 0;
2548 }
2549 
2550 /*
2551  * We go to the (large) trouble of VLAN tagging ARP frames because
2552  * switches in VLAN mode (especially if ports are configured as
2553  * "native" to a VLAN) might not pass non-tagged frames.
2554  */
2555 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2556 {
2557 	struct sk_buff *skb;
2558 
2559 	pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2560 		 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2561 
2562 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2563 			 NULL, slave_dev->dev_addr, NULL);
2564 
2565 	if (!skb) {
2566 		pr_err("ARP packet allocation failed\n");
2567 		return;
2568 	}
2569 	if (vlan_id) {
2570 		skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2571 		if (!skb) {
2572 			pr_err("failed to insert VLAN tag\n");
2573 			return;
2574 		}
2575 	}
2576 	arp_xmit(skb);
2577 }
2578 
2579 
2580 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2581 {
2582 	int i, vlan_id;
2583 	__be32 *targets = bond->params.arp_targets;
2584 	struct vlan_entry *vlan;
2585 	struct net_device *vlan_dev = NULL;
2586 	struct rtable *rt;
2587 
2588 	for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2589 		__be32 addr;
2590 		if (!targets[i])
2591 			break;
2592 		pr_debug("basa: target %pI4\n", &targets[i]);
2593 		if (!bond_vlan_used(bond)) {
2594 			pr_debug("basa: empty vlan: arp_send\n");
2595 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2596 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2597 				      addr, 0);
2598 			continue;
2599 		}
2600 
2601 		/*
2602 		 * If VLANs are configured, we do a route lookup to
2603 		 * determine which VLAN interface would be used, so we
2604 		 * can tag the ARP with the proper VLAN tag.
2605 		 */
2606 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2607 				     RTO_ONLINK, 0);
2608 		if (IS_ERR(rt)) {
2609 			if (net_ratelimit()) {
2610 				pr_warning("%s: no route to arp_ip_target %pI4\n",
2611 					   bond->dev->name, &targets[i]);
2612 			}
2613 			continue;
2614 		}
2615 
2616 		/*
2617 		 * This target is not on a VLAN
2618 		 */
2619 		if (rt->dst.dev == bond->dev) {
2620 			ip_rt_put(rt);
2621 			pr_debug("basa: rtdev == bond->dev: arp_send\n");
2622 			addr = bond_confirm_addr(bond->dev, targets[i], 0);
2623 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2624 				      addr, 0);
2625 			continue;
2626 		}
2627 
2628 		vlan_id = 0;
2629 		list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2630 			rcu_read_lock();
2631 			vlan_dev = __vlan_find_dev_deep(bond->dev,
2632 							htons(ETH_P_8021Q),
2633 							vlan->vlan_id);
2634 			rcu_read_unlock();
2635 			if (vlan_dev == rt->dst.dev) {
2636 				vlan_id = vlan->vlan_id;
2637 				pr_debug("basa: vlan match on %s %d\n",
2638 				       vlan_dev->name, vlan_id);
2639 				break;
2640 			}
2641 		}
2642 
2643 		if (vlan_id && vlan_dev) {
2644 			ip_rt_put(rt);
2645 			addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2646 			bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2647 				      addr, vlan_id);
2648 			continue;
2649 		}
2650 
2651 		if (net_ratelimit()) {
2652 			pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2653 				   bond->dev->name, &targets[i],
2654 				   rt->dst.dev ? rt->dst.dev->name : "NULL");
2655 		}
2656 		ip_rt_put(rt);
2657 	}
2658 }
2659 
2660 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2661 {
2662 	int i;
2663 	__be32 *targets = bond->params.arp_targets;
2664 
2665 	for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2666 		pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2667 			 &sip, &tip, i, &targets[i],
2668 			 bond_has_this_ip(bond, tip));
2669 		if (sip == targets[i]) {
2670 			if (bond_has_this_ip(bond, tip))
2671 				slave->last_arp_rx = jiffies;
2672 			return;
2673 		}
2674 	}
2675 }
2676 
2677 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2678 			struct slave *slave)
2679 {
2680 	struct arphdr *arp = (struct arphdr *)skb->data;
2681 	unsigned char *arp_ptr;
2682 	__be32 sip, tip;
2683 	int alen;
2684 
2685 	if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2686 		return RX_HANDLER_ANOTHER;
2687 
2688 	read_lock(&bond->lock);
2689 	alen = arp_hdr_len(bond->dev);
2690 
2691 	pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2692 		 bond->dev->name, skb->dev->name);
2693 
2694 	if (alen > skb_headlen(skb)) {
2695 		arp = kmalloc(alen, GFP_ATOMIC);
2696 		if (!arp)
2697 			goto out_unlock;
2698 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
2699 			goto out_unlock;
2700 	}
2701 
2702 	if (arp->ar_hln != bond->dev->addr_len ||
2703 	    skb->pkt_type == PACKET_OTHERHOST ||
2704 	    skb->pkt_type == PACKET_LOOPBACK ||
2705 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
2706 	    arp->ar_pro != htons(ETH_P_IP) ||
2707 	    arp->ar_pln != 4)
2708 		goto out_unlock;
2709 
2710 	arp_ptr = (unsigned char *)(arp + 1);
2711 	arp_ptr += bond->dev->addr_len;
2712 	memcpy(&sip, arp_ptr, 4);
2713 	arp_ptr += 4 + bond->dev->addr_len;
2714 	memcpy(&tip, arp_ptr, 4);
2715 
2716 	pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2717 		 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2718 		 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2719 		 &sip, &tip);
2720 
2721 	/*
2722 	 * Backup slaves won't see the ARP reply, but do come through
2723 	 * here for each ARP probe (so we swap the sip/tip to validate
2724 	 * the probe).  In a "redundant switch, common router" type of
2725 	 * configuration, the ARP probe will (hopefully) travel from
2726 	 * the active, through one switch, the router, then the other
2727 	 * switch before reaching the backup.
2728 	 */
2729 	if (bond_is_active_slave(slave))
2730 		bond_validate_arp(bond, slave, sip, tip);
2731 	else
2732 		bond_validate_arp(bond, slave, tip, sip);
2733 
2734 out_unlock:
2735 	read_unlock(&bond->lock);
2736 	if (arp != (struct arphdr *)skb->data)
2737 		kfree(arp);
2738 	return RX_HANDLER_ANOTHER;
2739 }
2740 
2741 /*
2742  * this function is called regularly to monitor each slave's link
2743  * ensuring that traffic is being sent and received when arp monitoring
2744  * is used in load-balancing mode. if the adapter has been dormant, then an
2745  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2746  * arp monitoring in active backup mode.
2747  */
2748 void bond_loadbalance_arp_mon(struct work_struct *work)
2749 {
2750 	struct bonding *bond = container_of(work, struct bonding,
2751 					    arp_work.work);
2752 	struct slave *slave, *oldcurrent;
2753 	int do_failover = 0;
2754 	int delta_in_ticks, extra_ticks;
2755 	int i;
2756 
2757 	read_lock(&bond->lock);
2758 
2759 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2760 	extra_ticks = delta_in_ticks / 2;
2761 
2762 	if (bond->slave_cnt == 0)
2763 		goto re_arm;
2764 
2765 	read_lock(&bond->curr_slave_lock);
2766 	oldcurrent = bond->curr_active_slave;
2767 	read_unlock(&bond->curr_slave_lock);
2768 
2769 	/* see if any of the previous devices are up now (i.e. they have
2770 	 * xmt and rcv traffic). the curr_active_slave does not come into
2771 	 * the picture unless it is null. also, slave->jiffies is not needed
2772 	 * here because we send an arp on each slave and give a slave as
2773 	 * long as it needs to get the tx/rx within the delta.
2774 	 * TODO: what about up/down delay in arp mode? it wasn't here before
2775 	 *       so it can wait
2776 	 */
2777 	bond_for_each_slave(bond, slave, i) {
2778 		unsigned long trans_start = dev_trans_start(slave->dev);
2779 
2780 		if (slave->link != BOND_LINK_UP) {
2781 			if (time_in_range(jiffies,
2782 				trans_start - delta_in_ticks,
2783 				trans_start + delta_in_ticks + extra_ticks) &&
2784 			    time_in_range(jiffies,
2785 				slave->dev->last_rx - delta_in_ticks,
2786 				slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2787 
2788 				slave->link  = BOND_LINK_UP;
2789 				bond_set_active_slave(slave);
2790 
2791 				/* primary_slave has no meaning in round-robin
2792 				 * mode. the window of a slave being up and
2793 				 * curr_active_slave being null after enslaving
2794 				 * is closed.
2795 				 */
2796 				if (!oldcurrent) {
2797 					pr_info("%s: link status definitely up for interface %s, ",
2798 						bond->dev->name,
2799 						slave->dev->name);
2800 					do_failover = 1;
2801 				} else {
2802 					pr_info("%s: interface %s is now up\n",
2803 						bond->dev->name,
2804 						slave->dev->name);
2805 				}
2806 			}
2807 		} else {
2808 			/* slave->link == BOND_LINK_UP */
2809 
2810 			/* not all switches will respond to an arp request
2811 			 * when the source ip is 0, so don't take the link down
2812 			 * if we don't know our ip yet
2813 			 */
2814 			if (!time_in_range(jiffies,
2815 				trans_start - delta_in_ticks,
2816 				trans_start + 2 * delta_in_ticks + extra_ticks) ||
2817 			    !time_in_range(jiffies,
2818 				slave->dev->last_rx - delta_in_ticks,
2819 				slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2820 
2821 				slave->link  = BOND_LINK_DOWN;
2822 				bond_set_backup_slave(slave);
2823 
2824 				if (slave->link_failure_count < UINT_MAX)
2825 					slave->link_failure_count++;
2826 
2827 				pr_info("%s: interface %s is now down.\n",
2828 					bond->dev->name,
2829 					slave->dev->name);
2830 
2831 				if (slave == oldcurrent)
2832 					do_failover = 1;
2833 			}
2834 		}
2835 
2836 		/* note: if switch is in round-robin mode, all links
2837 		 * must tx arp to ensure all links rx an arp - otherwise
2838 		 * links may oscillate or not come up at all; if switch is
2839 		 * in something like xor mode, there is nothing we can
2840 		 * do - all replies will be rx'ed on same link causing slaves
2841 		 * to be unstable during low/no traffic periods
2842 		 */
2843 		if (IS_UP(slave->dev))
2844 			bond_arp_send_all(bond, slave);
2845 	}
2846 
2847 	if (do_failover) {
2848 		block_netpoll_tx();
2849 		write_lock_bh(&bond->curr_slave_lock);
2850 
2851 		bond_select_active_slave(bond);
2852 
2853 		write_unlock_bh(&bond->curr_slave_lock);
2854 		unblock_netpoll_tx();
2855 	}
2856 
2857 re_arm:
2858 	if (bond->params.arp_interval)
2859 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2860 
2861 	read_unlock(&bond->lock);
2862 }
2863 
2864 /*
2865  * Called to inspect slaves for active-backup mode ARP monitor link state
2866  * changes.  Sets new_link in slaves to specify what action should take
2867  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2868  * to link states must be committed.
2869  *
2870  * Called with bond->lock held for read.
2871  */
2872 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2873 {
2874 	struct slave *slave;
2875 	int i, commit = 0;
2876 	unsigned long trans_start;
2877 	int extra_ticks;
2878 
2879 	/* All the time comparisons below need some extra time. Otherwise, on
2880 	 * fast networks the ARP probe/reply may arrive within the same jiffy
2881 	 * as it was sent.  Then, the next time the ARP monitor is run, one
2882 	 * arp_interval will already have passed in the comparisons.
2883 	 */
2884 	extra_ticks = delta_in_ticks / 2;
2885 
2886 	bond_for_each_slave(bond, slave, i) {
2887 		slave->new_link = BOND_LINK_NOCHANGE;
2888 
2889 		if (slave->link != BOND_LINK_UP) {
2890 			if (time_in_range(jiffies,
2891 				slave_last_rx(bond, slave) - delta_in_ticks,
2892 				slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2893 
2894 				slave->new_link = BOND_LINK_UP;
2895 				commit++;
2896 			}
2897 
2898 			continue;
2899 		}
2900 
2901 		/*
2902 		 * Give slaves 2*delta after being enslaved or made
2903 		 * active.  This avoids bouncing, as the last receive
2904 		 * times need a full ARP monitor cycle to be updated.
2905 		 */
2906 		if (time_in_range(jiffies,
2907 				  slave->jiffies - delta_in_ticks,
2908 				  slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2909 			continue;
2910 
2911 		/*
2912 		 * Backup slave is down if:
2913 		 * - No current_arp_slave AND
2914 		 * - more than 3*delta since last receive AND
2915 		 * - the bond has an IP address
2916 		 *
2917 		 * Note: a non-null current_arp_slave indicates
2918 		 * the curr_active_slave went down and we are
2919 		 * searching for a new one; under this condition
2920 		 * we only take the curr_active_slave down - this
2921 		 * gives each slave a chance to tx/rx traffic
2922 		 * before being taken out
2923 		 */
2924 		if (!bond_is_active_slave(slave) &&
2925 		    !bond->current_arp_slave &&
2926 		    !time_in_range(jiffies,
2927 			slave_last_rx(bond, slave) - delta_in_ticks,
2928 			slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2929 
2930 			slave->new_link = BOND_LINK_DOWN;
2931 			commit++;
2932 		}
2933 
2934 		/*
2935 		 * Active slave is down if:
2936 		 * - more than 2*delta since transmitting OR
2937 		 * - (more than 2*delta since receive AND
2938 		 *    the bond has an IP address)
2939 		 */
2940 		trans_start = dev_trans_start(slave->dev);
2941 		if (bond_is_active_slave(slave) &&
2942 		    (!time_in_range(jiffies,
2943 			trans_start - delta_in_ticks,
2944 			trans_start + 2 * delta_in_ticks + extra_ticks) ||
2945 		     !time_in_range(jiffies,
2946 			slave_last_rx(bond, slave) - delta_in_ticks,
2947 			slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2948 
2949 			slave->new_link = BOND_LINK_DOWN;
2950 			commit++;
2951 		}
2952 	}
2953 
2954 	return commit;
2955 }
2956 
2957 /*
2958  * Called to commit link state changes noted by inspection step of
2959  * active-backup mode ARP monitor.
2960  *
2961  * Called with RTNL and bond->lock for read.
2962  */
2963 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2964 {
2965 	struct slave *slave;
2966 	int i;
2967 	unsigned long trans_start;
2968 
2969 	bond_for_each_slave(bond, slave, i) {
2970 		switch (slave->new_link) {
2971 		case BOND_LINK_NOCHANGE:
2972 			continue;
2973 
2974 		case BOND_LINK_UP:
2975 			trans_start = dev_trans_start(slave->dev);
2976 			if ((!bond->curr_active_slave &&
2977 			     time_in_range(jiffies,
2978 					   trans_start - delta_in_ticks,
2979 					   trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2980 			    bond->curr_active_slave != slave) {
2981 				slave->link = BOND_LINK_UP;
2982 				if (bond->current_arp_slave) {
2983 					bond_set_slave_inactive_flags(
2984 						bond->current_arp_slave);
2985 					bond->current_arp_slave = NULL;
2986 				}
2987 
2988 				pr_info("%s: link status definitely up for interface %s.\n",
2989 					bond->dev->name, slave->dev->name);
2990 
2991 				if (!bond->curr_active_slave ||
2992 				    (slave == bond->primary_slave))
2993 					goto do_failover;
2994 
2995 			}
2996 
2997 			continue;
2998 
2999 		case BOND_LINK_DOWN:
3000 			if (slave->link_failure_count < UINT_MAX)
3001 				slave->link_failure_count++;
3002 
3003 			slave->link = BOND_LINK_DOWN;
3004 			bond_set_slave_inactive_flags(slave);
3005 
3006 			pr_info("%s: link status definitely down for interface %s, disabling it\n",
3007 				bond->dev->name, slave->dev->name);
3008 
3009 			if (slave == bond->curr_active_slave) {
3010 				bond->current_arp_slave = NULL;
3011 				goto do_failover;
3012 			}
3013 
3014 			continue;
3015 
3016 		default:
3017 			pr_err("%s: impossible: new_link %d on slave %s\n",
3018 			       bond->dev->name, slave->new_link,
3019 			       slave->dev->name);
3020 			continue;
3021 		}
3022 
3023 do_failover:
3024 		ASSERT_RTNL();
3025 		block_netpoll_tx();
3026 		write_lock_bh(&bond->curr_slave_lock);
3027 		bond_select_active_slave(bond);
3028 		write_unlock_bh(&bond->curr_slave_lock);
3029 		unblock_netpoll_tx();
3030 	}
3031 
3032 	bond_set_carrier(bond);
3033 }
3034 
3035 /*
3036  * Send ARP probes for active-backup mode ARP monitor.
3037  *
3038  * Called with bond->lock held for read.
3039  */
3040 static void bond_ab_arp_probe(struct bonding *bond)
3041 {
3042 	struct slave *slave;
3043 	int i;
3044 
3045 	read_lock(&bond->curr_slave_lock);
3046 
3047 	if (bond->current_arp_slave && bond->curr_active_slave)
3048 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
3049 			bond->current_arp_slave->dev->name,
3050 			bond->curr_active_slave->dev->name);
3051 
3052 	if (bond->curr_active_slave) {
3053 		bond_arp_send_all(bond, bond->curr_active_slave);
3054 		read_unlock(&bond->curr_slave_lock);
3055 		return;
3056 	}
3057 
3058 	read_unlock(&bond->curr_slave_lock);
3059 
3060 	/* if we don't have a curr_active_slave, search for the next available
3061 	 * backup slave from the current_arp_slave and make it the candidate
3062 	 * for becoming the curr_active_slave
3063 	 */
3064 
3065 	if (!bond->current_arp_slave) {
3066 		bond->current_arp_slave = bond->first_slave;
3067 		if (!bond->current_arp_slave)
3068 			return;
3069 	}
3070 
3071 	bond_set_slave_inactive_flags(bond->current_arp_slave);
3072 
3073 	/* search for next candidate */
3074 	bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3075 		if (IS_UP(slave->dev)) {
3076 			slave->link = BOND_LINK_BACK;
3077 			bond_set_slave_active_flags(slave);
3078 			bond_arp_send_all(bond, slave);
3079 			slave->jiffies = jiffies;
3080 			bond->current_arp_slave = slave;
3081 			break;
3082 		}
3083 
3084 		/* if the link state is up at this point, we
3085 		 * mark it down - this can happen if we have
3086 		 * simultaneous link failures and
3087 		 * reselect_active_interface doesn't make this
3088 		 * one the current slave so it is still marked
3089 		 * up when it is actually down
3090 		 */
3091 		if (slave->link == BOND_LINK_UP) {
3092 			slave->link = BOND_LINK_DOWN;
3093 			if (slave->link_failure_count < UINT_MAX)
3094 				slave->link_failure_count++;
3095 
3096 			bond_set_slave_inactive_flags(slave);
3097 
3098 			pr_info("%s: backup interface %s is now down.\n",
3099 				bond->dev->name, slave->dev->name);
3100 		}
3101 	}
3102 }
3103 
3104 void bond_activebackup_arp_mon(struct work_struct *work)
3105 {
3106 	struct bonding *bond = container_of(work, struct bonding,
3107 					    arp_work.work);
3108 	bool should_notify_peers = false;
3109 	int delta_in_ticks;
3110 
3111 	read_lock(&bond->lock);
3112 
3113 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3114 
3115 	if (bond->slave_cnt == 0)
3116 		goto re_arm;
3117 
3118 	should_notify_peers = bond_should_notify_peers(bond);
3119 
3120 	if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3121 		read_unlock(&bond->lock);
3122 
3123 		/* Race avoidance with bond_close flush of workqueue */
3124 		if (!rtnl_trylock()) {
3125 			read_lock(&bond->lock);
3126 			delta_in_ticks = 1;
3127 			should_notify_peers = false;
3128 			goto re_arm;
3129 		}
3130 
3131 		read_lock(&bond->lock);
3132 
3133 		bond_ab_arp_commit(bond, delta_in_ticks);
3134 
3135 		read_unlock(&bond->lock);
3136 		rtnl_unlock();
3137 		read_lock(&bond->lock);
3138 	}
3139 
3140 	bond_ab_arp_probe(bond);
3141 
3142 re_arm:
3143 	if (bond->params.arp_interval)
3144 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3145 
3146 	read_unlock(&bond->lock);
3147 
3148 	if (should_notify_peers) {
3149 		if (!rtnl_trylock()) {
3150 			read_lock(&bond->lock);
3151 			bond->send_peer_notif++;
3152 			read_unlock(&bond->lock);
3153 			return;
3154 		}
3155 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3156 		rtnl_unlock();
3157 	}
3158 }
3159 
3160 /*-------------------------- netdev event handling --------------------------*/
3161 
3162 /*
3163  * Change device name
3164  */
3165 static int bond_event_changename(struct bonding *bond)
3166 {
3167 	bond_remove_proc_entry(bond);
3168 	bond_create_proc_entry(bond);
3169 
3170 	bond_debug_reregister(bond);
3171 
3172 	return NOTIFY_DONE;
3173 }
3174 
3175 static int bond_master_netdev_event(unsigned long event,
3176 				    struct net_device *bond_dev)
3177 {
3178 	struct bonding *event_bond = netdev_priv(bond_dev);
3179 
3180 	switch (event) {
3181 	case NETDEV_CHANGENAME:
3182 		return bond_event_changename(event_bond);
3183 	case NETDEV_UNREGISTER:
3184 		bond_remove_proc_entry(event_bond);
3185 		break;
3186 	case NETDEV_REGISTER:
3187 		bond_create_proc_entry(event_bond);
3188 		break;
3189 	default:
3190 		break;
3191 	}
3192 
3193 	return NOTIFY_DONE;
3194 }
3195 
3196 static int bond_slave_netdev_event(unsigned long event,
3197 				   struct net_device *slave_dev)
3198 {
3199 	struct slave *slave = bond_slave_get_rtnl(slave_dev);
3200 	struct bonding *bond;
3201 	struct net_device *bond_dev;
3202 	u32 old_speed;
3203 	u8 old_duplex;
3204 
3205 	/* A netdev event can be generated while enslaving a device
3206 	 * before netdev_rx_handler_register is called in which case
3207 	 * slave will be NULL
3208 	 */
3209 	if (!slave)
3210 		return NOTIFY_DONE;
3211 	bond_dev = slave->bond->dev;
3212 	bond = slave->bond;
3213 
3214 	switch (event) {
3215 	case NETDEV_UNREGISTER:
3216 		if (bond->setup_by_slave)
3217 			bond_release_and_destroy(bond_dev, slave_dev);
3218 		else
3219 			bond_release(bond_dev, slave_dev);
3220 		break;
3221 	case NETDEV_UP:
3222 	case NETDEV_CHANGE:
3223 		old_speed = slave->speed;
3224 		old_duplex = slave->duplex;
3225 
3226 		bond_update_speed_duplex(slave);
3227 
3228 		if (bond->params.mode == BOND_MODE_8023AD) {
3229 			if (old_speed != slave->speed)
3230 				bond_3ad_adapter_speed_changed(slave);
3231 			if (old_duplex != slave->duplex)
3232 				bond_3ad_adapter_duplex_changed(slave);
3233 		}
3234 		break;
3235 	case NETDEV_DOWN:
3236 		/*
3237 		 * ... Or is it this?
3238 		 */
3239 		break;
3240 	case NETDEV_CHANGEMTU:
3241 		/*
3242 		 * TODO: Should slaves be allowed to
3243 		 * independently alter their MTU?  For
3244 		 * an active-backup bond, slaves need
3245 		 * not be the same type of device, so
3246 		 * MTUs may vary.  For other modes,
3247 		 * slaves arguably should have the
3248 		 * same MTUs. To do this, we'd need to
3249 		 * take over the slave's change_mtu
3250 		 * function for the duration of their
3251 		 * servitude.
3252 		 */
3253 		break;
3254 	case NETDEV_CHANGENAME:
3255 		/*
3256 		 * TODO: handle changing the primary's name
3257 		 */
3258 		break;
3259 	case NETDEV_FEAT_CHANGE:
3260 		bond_compute_features(bond);
3261 		break;
3262 	default:
3263 		break;
3264 	}
3265 
3266 	return NOTIFY_DONE;
3267 }
3268 
3269 /*
3270  * bond_netdev_event: handle netdev notifier chain events.
3271  *
3272  * This function receives events for the netdev chain.  The caller (an
3273  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3274  * locks for us to safely manipulate the slave devices (RTNL lock,
3275  * dev_probe_lock).
3276  */
3277 static int bond_netdev_event(struct notifier_block *this,
3278 			     unsigned long event, void *ptr)
3279 {
3280 	struct net_device *event_dev = (struct net_device *)ptr;
3281 
3282 	pr_debug("event_dev: %s, event: %lx\n",
3283 		 event_dev ? event_dev->name : "None",
3284 		 event);
3285 
3286 	if (!(event_dev->priv_flags & IFF_BONDING))
3287 		return NOTIFY_DONE;
3288 
3289 	if (event_dev->flags & IFF_MASTER) {
3290 		pr_debug("IFF_MASTER\n");
3291 		return bond_master_netdev_event(event, event_dev);
3292 	}
3293 
3294 	if (event_dev->flags & IFF_SLAVE) {
3295 		pr_debug("IFF_SLAVE\n");
3296 		return bond_slave_netdev_event(event, event_dev);
3297 	}
3298 
3299 	return NOTIFY_DONE;
3300 }
3301 
3302 static struct notifier_block bond_netdev_notifier = {
3303 	.notifier_call = bond_netdev_event,
3304 };
3305 
3306 /*---------------------------- Hashing Policies -----------------------------*/
3307 
3308 /*
3309  * Hash for the output device based upon layer 2 data
3310  */
3311 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3312 {
3313 	struct ethhdr *data = (struct ethhdr *)skb->data;
3314 
3315 	if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3316 		return (data->h_dest[5] ^ data->h_source[5]) % count;
3317 
3318 	return 0;
3319 }
3320 
3321 /*
3322  * Hash for the output device based upon layer 2 and layer 3 data. If
3323  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3324  */
3325 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3326 {
3327 	const struct ethhdr *data;
3328 	const struct iphdr *iph;
3329 	const struct ipv6hdr *ipv6h;
3330 	u32 v6hash;
3331 	const __be32 *s, *d;
3332 
3333 	if (skb->protocol == htons(ETH_P_IP) &&
3334 	    pskb_network_may_pull(skb, sizeof(*iph))) {
3335 		iph = ip_hdr(skb);
3336 		data = (struct ethhdr *)skb->data;
3337 		return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3338 			(data->h_dest[5] ^ data->h_source[5])) % count;
3339 	} else if (skb->protocol == htons(ETH_P_IPV6) &&
3340 		   pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3341 		ipv6h = ipv6_hdr(skb);
3342 		data = (struct ethhdr *)skb->data;
3343 		s = &ipv6h->saddr.s6_addr32[0];
3344 		d = &ipv6h->daddr.s6_addr32[0];
3345 		v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3346 		v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3347 		return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3348 	}
3349 
3350 	return bond_xmit_hash_policy_l2(skb, count);
3351 }
3352 
3353 /*
3354  * Hash for the output device based upon layer 3 and layer 4 data. If
3355  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3356  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3357  */
3358 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3359 {
3360 	u32 layer4_xor = 0;
3361 	const struct iphdr *iph;
3362 	const struct ipv6hdr *ipv6h;
3363 	const __be32 *s, *d;
3364 	const __be16 *l4 = NULL;
3365 	__be16 _l4[2];
3366 	int noff = skb_network_offset(skb);
3367 	int poff;
3368 
3369 	if (skb->protocol == htons(ETH_P_IP) &&
3370 	    pskb_may_pull(skb, noff + sizeof(*iph))) {
3371 		iph = ip_hdr(skb);
3372 		poff = proto_ports_offset(iph->protocol);
3373 
3374 		if (!ip_is_fragment(iph) && poff >= 0) {
3375 			l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3376 						sizeof(_l4), &_l4);
3377 			if (l4)
3378 				layer4_xor = ntohs(l4[0] ^ l4[1]);
3379 		}
3380 		return (layer4_xor ^
3381 			((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3382 	} else if (skb->protocol == htons(ETH_P_IPV6) &&
3383 		   pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3384 		ipv6h = ipv6_hdr(skb);
3385 		poff = proto_ports_offset(ipv6h->nexthdr);
3386 		if (poff >= 0) {
3387 			l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3388 						sizeof(_l4), &_l4);
3389 			if (l4)
3390 				layer4_xor = ntohs(l4[0] ^ l4[1]);
3391 		}
3392 		s = &ipv6h->saddr.s6_addr32[0];
3393 		d = &ipv6h->daddr.s6_addr32[0];
3394 		layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3395 		layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3396 			       (layer4_xor >> 8);
3397 		return layer4_xor % count;
3398 	}
3399 
3400 	return bond_xmit_hash_policy_l2(skb, count);
3401 }
3402 
3403 /*-------------------------- Device entry points ----------------------------*/
3404 
3405 static void bond_work_init_all(struct bonding *bond)
3406 {
3407 	INIT_DELAYED_WORK(&bond->mcast_work,
3408 			  bond_resend_igmp_join_requests_delayed);
3409 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3410 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3411 	if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3412 		INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3413 	else
3414 		INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3415 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3416 }
3417 
3418 static void bond_work_cancel_all(struct bonding *bond)
3419 {
3420 	cancel_delayed_work_sync(&bond->mii_work);
3421 	cancel_delayed_work_sync(&bond->arp_work);
3422 	cancel_delayed_work_sync(&bond->alb_work);
3423 	cancel_delayed_work_sync(&bond->ad_work);
3424 	cancel_delayed_work_sync(&bond->mcast_work);
3425 }
3426 
3427 static int bond_open(struct net_device *bond_dev)
3428 {
3429 	struct bonding *bond = netdev_priv(bond_dev);
3430 	struct slave *slave;
3431 	int i;
3432 
3433 	/* reset slave->backup and slave->inactive */
3434 	read_lock(&bond->lock);
3435 	if (bond->slave_cnt > 0) {
3436 		read_lock(&bond->curr_slave_lock);
3437 		bond_for_each_slave(bond, slave, i) {
3438 			if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3439 				&& (slave != bond->curr_active_slave)) {
3440 				bond_set_slave_inactive_flags(slave);
3441 			} else {
3442 				bond_set_slave_active_flags(slave);
3443 			}
3444 		}
3445 		read_unlock(&bond->curr_slave_lock);
3446 	}
3447 	read_unlock(&bond->lock);
3448 
3449 	bond_work_init_all(bond);
3450 
3451 	if (bond_is_lb(bond)) {
3452 		/* bond_alb_initialize must be called before the timer
3453 		 * is started.
3454 		 */
3455 		if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3456 			return -ENOMEM;
3457 		queue_delayed_work(bond->wq, &bond->alb_work, 0);
3458 	}
3459 
3460 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
3461 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
3462 
3463 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3464 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
3465 		if (bond->params.arp_validate)
3466 			bond->recv_probe = bond_arp_rcv;
3467 	}
3468 
3469 	if (bond->params.mode == BOND_MODE_8023AD) {
3470 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
3471 		/* register to receive LACPDUs */
3472 		bond->recv_probe = bond_3ad_lacpdu_recv;
3473 		bond_3ad_initiate_agg_selection(bond, 1);
3474 	}
3475 
3476 	return 0;
3477 }
3478 
3479 static int bond_close(struct net_device *bond_dev)
3480 {
3481 	struct bonding *bond = netdev_priv(bond_dev);
3482 
3483 	write_lock_bh(&bond->lock);
3484 	bond->send_peer_notif = 0;
3485 	write_unlock_bh(&bond->lock);
3486 
3487 	bond_work_cancel_all(bond);
3488 	if (bond_is_lb(bond)) {
3489 		/* Must be called only after all
3490 		 * slaves have been released
3491 		 */
3492 		bond_alb_deinitialize(bond);
3493 	}
3494 	bond->recv_probe = NULL;
3495 
3496 	return 0;
3497 }
3498 
3499 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3500 						struct rtnl_link_stats64 *stats)
3501 {
3502 	struct bonding *bond = netdev_priv(bond_dev);
3503 	struct rtnl_link_stats64 temp;
3504 	struct slave *slave;
3505 	int i;
3506 
3507 	memset(stats, 0, sizeof(*stats));
3508 
3509 	read_lock_bh(&bond->lock);
3510 
3511 	bond_for_each_slave(bond, slave, i) {
3512 		const struct rtnl_link_stats64 *sstats =
3513 			dev_get_stats(slave->dev, &temp);
3514 
3515 		stats->rx_packets += sstats->rx_packets;
3516 		stats->rx_bytes += sstats->rx_bytes;
3517 		stats->rx_errors += sstats->rx_errors;
3518 		stats->rx_dropped += sstats->rx_dropped;
3519 
3520 		stats->tx_packets += sstats->tx_packets;
3521 		stats->tx_bytes += sstats->tx_bytes;
3522 		stats->tx_errors += sstats->tx_errors;
3523 		stats->tx_dropped += sstats->tx_dropped;
3524 
3525 		stats->multicast += sstats->multicast;
3526 		stats->collisions += sstats->collisions;
3527 
3528 		stats->rx_length_errors += sstats->rx_length_errors;
3529 		stats->rx_over_errors += sstats->rx_over_errors;
3530 		stats->rx_crc_errors += sstats->rx_crc_errors;
3531 		stats->rx_frame_errors += sstats->rx_frame_errors;
3532 		stats->rx_fifo_errors += sstats->rx_fifo_errors;
3533 		stats->rx_missed_errors += sstats->rx_missed_errors;
3534 
3535 		stats->tx_aborted_errors += sstats->tx_aborted_errors;
3536 		stats->tx_carrier_errors += sstats->tx_carrier_errors;
3537 		stats->tx_fifo_errors += sstats->tx_fifo_errors;
3538 		stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3539 		stats->tx_window_errors += sstats->tx_window_errors;
3540 	}
3541 
3542 	read_unlock_bh(&bond->lock);
3543 
3544 	return stats;
3545 }
3546 
3547 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3548 {
3549 	struct net_device *slave_dev = NULL;
3550 	struct ifbond k_binfo;
3551 	struct ifbond __user *u_binfo = NULL;
3552 	struct ifslave k_sinfo;
3553 	struct ifslave __user *u_sinfo = NULL;
3554 	struct mii_ioctl_data *mii = NULL;
3555 	struct net *net;
3556 	int res = 0;
3557 
3558 	pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3559 
3560 	switch (cmd) {
3561 	case SIOCGMIIPHY:
3562 		mii = if_mii(ifr);
3563 		if (!mii)
3564 			return -EINVAL;
3565 
3566 		mii->phy_id = 0;
3567 		/* Fall Through */
3568 	case SIOCGMIIREG:
3569 		/*
3570 		 * We do this again just in case we were called by SIOCGMIIREG
3571 		 * instead of SIOCGMIIPHY.
3572 		 */
3573 		mii = if_mii(ifr);
3574 		if (!mii)
3575 			return -EINVAL;
3576 
3577 
3578 		if (mii->reg_num == 1) {
3579 			struct bonding *bond = netdev_priv(bond_dev);
3580 			mii->val_out = 0;
3581 			read_lock(&bond->lock);
3582 			read_lock(&bond->curr_slave_lock);
3583 			if (netif_carrier_ok(bond->dev))
3584 				mii->val_out = BMSR_LSTATUS;
3585 
3586 			read_unlock(&bond->curr_slave_lock);
3587 			read_unlock(&bond->lock);
3588 		}
3589 
3590 		return 0;
3591 	case BOND_INFO_QUERY_OLD:
3592 	case SIOCBONDINFOQUERY:
3593 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
3594 
3595 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3596 			return -EFAULT;
3597 
3598 		res = bond_info_query(bond_dev, &k_binfo);
3599 		if (res == 0 &&
3600 		    copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3601 			return -EFAULT;
3602 
3603 		return res;
3604 	case BOND_SLAVE_INFO_QUERY_OLD:
3605 	case SIOCBONDSLAVEINFOQUERY:
3606 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3607 
3608 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3609 			return -EFAULT;
3610 
3611 		res = bond_slave_info_query(bond_dev, &k_sinfo);
3612 		if (res == 0 &&
3613 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3614 			return -EFAULT;
3615 
3616 		return res;
3617 	default:
3618 		/* Go on */
3619 		break;
3620 	}
3621 
3622 	net = dev_net(bond_dev);
3623 
3624 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3625 		return -EPERM;
3626 
3627 	slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3628 
3629 	pr_debug("slave_dev=%p:\n", slave_dev);
3630 
3631 	if (!slave_dev)
3632 		res = -ENODEV;
3633 	else {
3634 		pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3635 		switch (cmd) {
3636 		case BOND_ENSLAVE_OLD:
3637 		case SIOCBONDENSLAVE:
3638 			res = bond_enslave(bond_dev, slave_dev);
3639 			break;
3640 		case BOND_RELEASE_OLD:
3641 		case SIOCBONDRELEASE:
3642 			res = bond_release(bond_dev, slave_dev);
3643 			break;
3644 		case BOND_SETHWADDR_OLD:
3645 		case SIOCBONDSETHWADDR:
3646 			bond_set_dev_addr(bond_dev, slave_dev);
3647 			res = 0;
3648 			break;
3649 		case BOND_CHANGE_ACTIVE_OLD:
3650 		case SIOCBONDCHANGEACTIVE:
3651 			res = bond_ioctl_change_active(bond_dev, slave_dev);
3652 			break;
3653 		default:
3654 			res = -EOPNOTSUPP;
3655 		}
3656 
3657 		dev_put(slave_dev);
3658 	}
3659 
3660 	return res;
3661 }
3662 
3663 static bool bond_addr_in_mc_list(unsigned char *addr,
3664 				 struct netdev_hw_addr_list *list,
3665 				 int addrlen)
3666 {
3667 	struct netdev_hw_addr *ha;
3668 
3669 	netdev_hw_addr_list_for_each(ha, list)
3670 		if (!memcmp(ha->addr, addr, addrlen))
3671 			return true;
3672 
3673 	return false;
3674 }
3675 
3676 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3677 {
3678 	struct bonding *bond = netdev_priv(bond_dev);
3679 
3680 	if (change & IFF_PROMISC)
3681 		bond_set_promiscuity(bond,
3682 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
3683 
3684 	if (change & IFF_ALLMULTI)
3685 		bond_set_allmulti(bond,
3686 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3687 }
3688 
3689 static void bond_set_multicast_list(struct net_device *bond_dev)
3690 {
3691 	struct bonding *bond = netdev_priv(bond_dev);
3692 	struct netdev_hw_addr *ha;
3693 	bool found;
3694 
3695 	read_lock(&bond->lock);
3696 
3697 	/* looking for addresses to add to slaves' mc list */
3698 	netdev_for_each_mc_addr(ha, bond_dev) {
3699 		found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3700 					     bond_dev->addr_len);
3701 		if (!found)
3702 			bond_mc_add(bond, ha->addr);
3703 	}
3704 
3705 	/* looking for addresses to delete from slaves' list */
3706 	netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3707 		found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3708 					     bond_dev->addr_len);
3709 		if (!found)
3710 			bond_mc_del(bond, ha->addr);
3711 	}
3712 
3713 	/* save master's multicast list */
3714 	__hw_addr_flush(&bond->mc_list);
3715 	__hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3716 			       bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3717 
3718 	read_unlock(&bond->lock);
3719 }
3720 
3721 static int bond_neigh_init(struct neighbour *n)
3722 {
3723 	struct bonding *bond = netdev_priv(n->dev);
3724 	struct slave *slave = bond->first_slave;
3725 	const struct net_device_ops *slave_ops;
3726 	struct neigh_parms parms;
3727 	int ret;
3728 
3729 	if (!slave)
3730 		return 0;
3731 
3732 	slave_ops = slave->dev->netdev_ops;
3733 
3734 	if (!slave_ops->ndo_neigh_setup)
3735 		return 0;
3736 
3737 	parms.neigh_setup = NULL;
3738 	parms.neigh_cleanup = NULL;
3739 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3740 	if (ret)
3741 		return ret;
3742 
3743 	/*
3744 	 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3745 	 * after the last slave has been detached.  Assumes that all slaves
3746 	 * utilize the same neigh_cleanup (true at this writing as only user
3747 	 * is ipoib).
3748 	 */
3749 	n->parms->neigh_cleanup = parms.neigh_cleanup;
3750 
3751 	if (!parms.neigh_setup)
3752 		return 0;
3753 
3754 	return parms.neigh_setup(n);
3755 }
3756 
3757 /*
3758  * The bonding ndo_neigh_setup is called at init time beofre any
3759  * slave exists. So we must declare proxy setup function which will
3760  * be used at run time to resolve the actual slave neigh param setup.
3761  */
3762 static int bond_neigh_setup(struct net_device *dev,
3763 			    struct neigh_parms *parms)
3764 {
3765 	parms->neigh_setup   = bond_neigh_init;
3766 
3767 	return 0;
3768 }
3769 
3770 /*
3771  * Change the MTU of all of a master's slaves to match the master
3772  */
3773 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3774 {
3775 	struct bonding *bond = netdev_priv(bond_dev);
3776 	struct slave *slave, *stop_at;
3777 	int res = 0;
3778 	int i;
3779 
3780 	pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3781 		 (bond_dev ? bond_dev->name : "None"), new_mtu);
3782 
3783 	/* Can't hold bond->lock with bh disabled here since
3784 	 * some base drivers panic. On the other hand we can't
3785 	 * hold bond->lock without bh disabled because we'll
3786 	 * deadlock. The only solution is to rely on the fact
3787 	 * that we're under rtnl_lock here, and the slaves
3788 	 * list won't change. This doesn't solve the problem
3789 	 * of setting the slave's MTU while it is
3790 	 * transmitting, but the assumption is that the base
3791 	 * driver can handle that.
3792 	 *
3793 	 * TODO: figure out a way to safely iterate the slaves
3794 	 * list, but without holding a lock around the actual
3795 	 * call to the base driver.
3796 	 */
3797 
3798 	bond_for_each_slave(bond, slave, i) {
3799 		pr_debug("s %p s->p %p c_m %p\n",
3800 			 slave,
3801 			 slave->prev,
3802 			 slave->dev->netdev_ops->ndo_change_mtu);
3803 
3804 		res = dev_set_mtu(slave->dev, new_mtu);
3805 
3806 		if (res) {
3807 			/* If we failed to set the slave's mtu to the new value
3808 			 * we must abort the operation even in ACTIVE_BACKUP
3809 			 * mode, because if we allow the backup slaves to have
3810 			 * different mtu values than the active slave we'll
3811 			 * need to change their mtu when doing a failover. That
3812 			 * means changing their mtu from timer context, which
3813 			 * is probably not a good idea.
3814 			 */
3815 			pr_debug("err %d %s\n", res, slave->dev->name);
3816 			goto unwind;
3817 		}
3818 	}
3819 
3820 	bond_dev->mtu = new_mtu;
3821 
3822 	return 0;
3823 
3824 unwind:
3825 	/* unwind from head to the slave that failed */
3826 	stop_at = slave;
3827 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3828 		int tmp_res;
3829 
3830 		tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3831 		if (tmp_res) {
3832 			pr_debug("unwind err %d dev %s\n",
3833 				 tmp_res, slave->dev->name);
3834 		}
3835 	}
3836 
3837 	return res;
3838 }
3839 
3840 /*
3841  * Change HW address
3842  *
3843  * Note that many devices must be down to change the HW address, and
3844  * downing the master releases all slaves.  We can make bonds full of
3845  * bonding devices to test this, however.
3846  */
3847 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3848 {
3849 	struct bonding *bond = netdev_priv(bond_dev);
3850 	struct sockaddr *sa = addr, tmp_sa;
3851 	struct slave *slave, *stop_at;
3852 	int res = 0;
3853 	int i;
3854 
3855 	if (bond->params.mode == BOND_MODE_ALB)
3856 		return bond_alb_set_mac_address(bond_dev, addr);
3857 
3858 
3859 	pr_debug("bond=%p, name=%s\n",
3860 		 bond, bond_dev ? bond_dev->name : "None");
3861 
3862 	/*
3863 	 * If fail_over_mac is set to active, do nothing and return
3864 	 * success.  Returning an error causes ifenslave to fail.
3865 	 */
3866 	if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3867 		return 0;
3868 
3869 	if (!is_valid_ether_addr(sa->sa_data))
3870 		return -EADDRNOTAVAIL;
3871 
3872 	/* Can't hold bond->lock with bh disabled here since
3873 	 * some base drivers panic. On the other hand we can't
3874 	 * hold bond->lock without bh disabled because we'll
3875 	 * deadlock. The only solution is to rely on the fact
3876 	 * that we're under rtnl_lock here, and the slaves
3877 	 * list won't change. This doesn't solve the problem
3878 	 * of setting the slave's hw address while it is
3879 	 * transmitting, but the assumption is that the base
3880 	 * driver can handle that.
3881 	 *
3882 	 * TODO: figure out a way to safely iterate the slaves
3883 	 * list, but without holding a lock around the actual
3884 	 * call to the base driver.
3885 	 */
3886 
3887 	bond_for_each_slave(bond, slave, i) {
3888 		const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3889 		pr_debug("slave %p %s\n", slave, slave->dev->name);
3890 
3891 		if (slave_ops->ndo_set_mac_address == NULL) {
3892 			res = -EOPNOTSUPP;
3893 			pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3894 			goto unwind;
3895 		}
3896 
3897 		res = dev_set_mac_address(slave->dev, addr);
3898 		if (res) {
3899 			/* TODO: consider downing the slave
3900 			 * and retry ?
3901 			 * User should expect communications
3902 			 * breakage anyway until ARP finish
3903 			 * updating, so...
3904 			 */
3905 			pr_debug("err %d %s\n", res, slave->dev->name);
3906 			goto unwind;
3907 		}
3908 	}
3909 
3910 	/* success */
3911 	memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3912 	return 0;
3913 
3914 unwind:
3915 	memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3916 	tmp_sa.sa_family = bond_dev->type;
3917 
3918 	/* unwind from head to the slave that failed */
3919 	stop_at = slave;
3920 	bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3921 		int tmp_res;
3922 
3923 		tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3924 		if (tmp_res) {
3925 			pr_debug("unwind err %d dev %s\n",
3926 				 tmp_res, slave->dev->name);
3927 		}
3928 	}
3929 
3930 	return res;
3931 }
3932 
3933 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3934 {
3935 	struct bonding *bond = netdev_priv(bond_dev);
3936 	struct slave *slave, *start_at;
3937 	int i, slave_no, res = 1;
3938 	struct iphdr *iph = ip_hdr(skb);
3939 
3940 	/*
3941 	 * Start with the curr_active_slave that joined the bond as the
3942 	 * default for sending IGMP traffic.  For failover purposes one
3943 	 * needs to maintain some consistency for the interface that will
3944 	 * send the join/membership reports.  The curr_active_slave found
3945 	 * will send all of this type of traffic.
3946 	 */
3947 	if ((iph->protocol == IPPROTO_IGMP) &&
3948 	    (skb->protocol == htons(ETH_P_IP))) {
3949 
3950 		read_lock(&bond->curr_slave_lock);
3951 		slave = bond->curr_active_slave;
3952 		read_unlock(&bond->curr_slave_lock);
3953 
3954 		if (!slave)
3955 			goto out;
3956 	} else {
3957 		/*
3958 		 * Concurrent TX may collide on rr_tx_counter; we accept
3959 		 * that as being rare enough not to justify using an
3960 		 * atomic op here.
3961 		 */
3962 		slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3963 
3964 		bond_for_each_slave(bond, slave, i) {
3965 			slave_no--;
3966 			if (slave_no < 0)
3967 				break;
3968 		}
3969 	}
3970 
3971 	start_at = slave;
3972 	bond_for_each_slave_from(bond, slave, i, start_at) {
3973 		if (IS_UP(slave->dev) &&
3974 		    (slave->link == BOND_LINK_UP) &&
3975 		    bond_is_active_slave(slave)) {
3976 			res = bond_dev_queue_xmit(bond, skb, slave->dev);
3977 			break;
3978 		}
3979 	}
3980 
3981 out:
3982 	if (res) {
3983 		/* no suitable interface, frame not sent */
3984 		kfree_skb(skb);
3985 	}
3986 
3987 	return NETDEV_TX_OK;
3988 }
3989 
3990 
3991 /*
3992  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3993  * the bond has a usable interface.
3994  */
3995 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3996 {
3997 	struct bonding *bond = netdev_priv(bond_dev);
3998 	int res = 1;
3999 
4000 	read_lock(&bond->curr_slave_lock);
4001 
4002 	if (bond->curr_active_slave)
4003 		res = bond_dev_queue_xmit(bond, skb,
4004 			bond->curr_active_slave->dev);
4005 
4006 	read_unlock(&bond->curr_slave_lock);
4007 
4008 	if (res)
4009 		/* no suitable interface, frame not sent */
4010 		kfree_skb(skb);
4011 
4012 	return NETDEV_TX_OK;
4013 }
4014 
4015 /*
4016  * In bond_xmit_xor() , we determine the output device by using a pre-
4017  * determined xmit_hash_policy(), If the selected device is not enabled,
4018  * find the next active slave.
4019  */
4020 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4021 {
4022 	struct bonding *bond = netdev_priv(bond_dev);
4023 	struct slave *slave, *start_at;
4024 	int slave_no;
4025 	int i;
4026 	int res = 1;
4027 
4028 	slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4029 
4030 	bond_for_each_slave(bond, slave, i) {
4031 		slave_no--;
4032 		if (slave_no < 0)
4033 			break;
4034 	}
4035 
4036 	start_at = slave;
4037 
4038 	bond_for_each_slave_from(bond, slave, i, start_at) {
4039 		if (IS_UP(slave->dev) &&
4040 		    (slave->link == BOND_LINK_UP) &&
4041 		    bond_is_active_slave(slave)) {
4042 			res = bond_dev_queue_xmit(bond, skb, slave->dev);
4043 			break;
4044 		}
4045 	}
4046 
4047 	if (res) {
4048 		/* no suitable interface, frame not sent */
4049 		kfree_skb(skb);
4050 	}
4051 
4052 	return NETDEV_TX_OK;
4053 }
4054 
4055 /*
4056  * in broadcast mode, we send everything to all usable interfaces.
4057  */
4058 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4059 {
4060 	struct bonding *bond = netdev_priv(bond_dev);
4061 	struct slave *slave, *start_at;
4062 	struct net_device *tx_dev = NULL;
4063 	int i;
4064 	int res = 1;
4065 
4066 	read_lock(&bond->curr_slave_lock);
4067 	start_at = bond->curr_active_slave;
4068 	read_unlock(&bond->curr_slave_lock);
4069 
4070 	if (!start_at)
4071 		goto out;
4072 
4073 	bond_for_each_slave_from(bond, slave, i, start_at) {
4074 		if (IS_UP(slave->dev) &&
4075 		    (slave->link == BOND_LINK_UP) &&
4076 		    bond_is_active_slave(slave)) {
4077 			if (tx_dev) {
4078 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4079 				if (!skb2) {
4080 					pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4081 					       bond_dev->name);
4082 					continue;
4083 				}
4084 
4085 				res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4086 				if (res) {
4087 					kfree_skb(skb2);
4088 					continue;
4089 				}
4090 			}
4091 			tx_dev = slave->dev;
4092 		}
4093 	}
4094 
4095 	if (tx_dev)
4096 		res = bond_dev_queue_xmit(bond, skb, tx_dev);
4097 
4098 out:
4099 	if (res)
4100 		/* no suitable interface, frame not sent */
4101 		kfree_skb(skb);
4102 
4103 	/* frame sent to all suitable interfaces */
4104 	return NETDEV_TX_OK;
4105 }
4106 
4107 /*------------------------- Device initialization ---------------------------*/
4108 
4109 static void bond_set_xmit_hash_policy(struct bonding *bond)
4110 {
4111 	switch (bond->params.xmit_policy) {
4112 	case BOND_XMIT_POLICY_LAYER23:
4113 		bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4114 		break;
4115 	case BOND_XMIT_POLICY_LAYER34:
4116 		bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4117 		break;
4118 	case BOND_XMIT_POLICY_LAYER2:
4119 	default:
4120 		bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4121 		break;
4122 	}
4123 }
4124 
4125 /*
4126  * Lookup the slave that corresponds to a qid
4127  */
4128 static inline int bond_slave_override(struct bonding *bond,
4129 				      struct sk_buff *skb)
4130 {
4131 	int i, res = 1;
4132 	struct slave *slave = NULL;
4133 	struct slave *check_slave;
4134 
4135 	if (!skb->queue_mapping)
4136 		return 1;
4137 
4138 	/* Find out if any slaves have the same mapping as this skb. */
4139 	bond_for_each_slave(bond, check_slave, i) {
4140 		if (check_slave->queue_id == skb->queue_mapping) {
4141 			slave = check_slave;
4142 			break;
4143 		}
4144 	}
4145 
4146 	/* If the slave isn't UP, use default transmit policy. */
4147 	if (slave && slave->queue_id && IS_UP(slave->dev) &&
4148 	    (slave->link == BOND_LINK_UP)) {
4149 		res = bond_dev_queue_xmit(bond, skb, slave->dev);
4150 	}
4151 
4152 	return res;
4153 }
4154 
4155 
4156 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4157 {
4158 	/*
4159 	 * This helper function exists to help dev_pick_tx get the correct
4160 	 * destination queue.  Using a helper function skips a call to
4161 	 * skb_tx_hash and will put the skbs in the queue we expect on their
4162 	 * way down to the bonding driver.
4163 	 */
4164 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4165 
4166 	/*
4167 	 * Save the original txq to restore before passing to the driver
4168 	 */
4169 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4170 
4171 	if (unlikely(txq >= dev->real_num_tx_queues)) {
4172 		do {
4173 			txq -= dev->real_num_tx_queues;
4174 		} while (txq >= dev->real_num_tx_queues);
4175 	}
4176 	return txq;
4177 }
4178 
4179 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4180 {
4181 	struct bonding *bond = netdev_priv(dev);
4182 
4183 	if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4184 		if (!bond_slave_override(bond, skb))
4185 			return NETDEV_TX_OK;
4186 	}
4187 
4188 	switch (bond->params.mode) {
4189 	case BOND_MODE_ROUNDROBIN:
4190 		return bond_xmit_roundrobin(skb, dev);
4191 	case BOND_MODE_ACTIVEBACKUP:
4192 		return bond_xmit_activebackup(skb, dev);
4193 	case BOND_MODE_XOR:
4194 		return bond_xmit_xor(skb, dev);
4195 	case BOND_MODE_BROADCAST:
4196 		return bond_xmit_broadcast(skb, dev);
4197 	case BOND_MODE_8023AD:
4198 		return bond_3ad_xmit_xor(skb, dev);
4199 	case BOND_MODE_ALB:
4200 	case BOND_MODE_TLB:
4201 		return bond_alb_xmit(skb, dev);
4202 	default:
4203 		/* Should never happen, mode already checked */
4204 		pr_err("%s: Error: Unknown bonding mode %d\n",
4205 		       dev->name, bond->params.mode);
4206 		WARN_ON_ONCE(1);
4207 		kfree_skb(skb);
4208 		return NETDEV_TX_OK;
4209 	}
4210 }
4211 
4212 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4213 {
4214 	struct bonding *bond = netdev_priv(dev);
4215 	netdev_tx_t ret = NETDEV_TX_OK;
4216 
4217 	/*
4218 	 * If we risk deadlock from transmitting this in the
4219 	 * netpoll path, tell netpoll to queue the frame for later tx
4220 	 */
4221 	if (is_netpoll_tx_blocked(dev))
4222 		return NETDEV_TX_BUSY;
4223 
4224 	read_lock(&bond->lock);
4225 
4226 	if (bond->slave_cnt)
4227 		ret = __bond_start_xmit(skb, dev);
4228 	else
4229 		kfree_skb(skb);
4230 
4231 	read_unlock(&bond->lock);
4232 
4233 	return ret;
4234 }
4235 
4236 /*
4237  * set bond mode specific net device operations
4238  */
4239 void bond_set_mode_ops(struct bonding *bond, int mode)
4240 {
4241 	struct net_device *bond_dev = bond->dev;
4242 
4243 	switch (mode) {
4244 	case BOND_MODE_ROUNDROBIN:
4245 		break;
4246 	case BOND_MODE_ACTIVEBACKUP:
4247 		break;
4248 	case BOND_MODE_XOR:
4249 		bond_set_xmit_hash_policy(bond);
4250 		break;
4251 	case BOND_MODE_BROADCAST:
4252 		break;
4253 	case BOND_MODE_8023AD:
4254 		bond_set_xmit_hash_policy(bond);
4255 		break;
4256 	case BOND_MODE_ALB:
4257 		/* FALLTHRU */
4258 	case BOND_MODE_TLB:
4259 		break;
4260 	default:
4261 		/* Should never happen, mode already checked */
4262 		pr_err("%s: Error: Unknown bonding mode %d\n",
4263 		       bond_dev->name, mode);
4264 		break;
4265 	}
4266 }
4267 
4268 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4269 				     struct ethtool_cmd *ecmd)
4270 {
4271 	struct bonding *bond = netdev_priv(bond_dev);
4272 	struct slave *slave;
4273 	int i;
4274 	unsigned long speed = 0;
4275 
4276 	ecmd->duplex = DUPLEX_UNKNOWN;
4277 	ecmd->port = PORT_OTHER;
4278 
4279 	/* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4280 	 * do not need to check mode.  Though link speed might not represent
4281 	 * the true receive or transmit bandwidth (not all modes are symmetric)
4282 	 * this is an accurate maximum.
4283 	 */
4284 	read_lock(&bond->lock);
4285 	bond_for_each_slave(bond, slave, i) {
4286 		if (SLAVE_IS_OK(slave)) {
4287 			if (slave->speed != SPEED_UNKNOWN)
4288 				speed += slave->speed;
4289 			if (ecmd->duplex == DUPLEX_UNKNOWN &&
4290 			    slave->duplex != DUPLEX_UNKNOWN)
4291 				ecmd->duplex = slave->duplex;
4292 		}
4293 	}
4294 	ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4295 	read_unlock(&bond->lock);
4296 	return 0;
4297 }
4298 
4299 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4300 				     struct ethtool_drvinfo *drvinfo)
4301 {
4302 	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4303 	strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4304 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4305 		 BOND_ABI_VERSION);
4306 }
4307 
4308 static const struct ethtool_ops bond_ethtool_ops = {
4309 	.get_drvinfo		= bond_ethtool_get_drvinfo,
4310 	.get_settings		= bond_ethtool_get_settings,
4311 	.get_link		= ethtool_op_get_link,
4312 };
4313 
4314 static const struct net_device_ops bond_netdev_ops = {
4315 	.ndo_init		= bond_init,
4316 	.ndo_uninit		= bond_uninit,
4317 	.ndo_open		= bond_open,
4318 	.ndo_stop		= bond_close,
4319 	.ndo_start_xmit		= bond_start_xmit,
4320 	.ndo_select_queue	= bond_select_queue,
4321 	.ndo_get_stats64	= bond_get_stats,
4322 	.ndo_do_ioctl		= bond_do_ioctl,
4323 	.ndo_change_rx_flags	= bond_change_rx_flags,
4324 	.ndo_set_rx_mode	= bond_set_multicast_list,
4325 	.ndo_change_mtu		= bond_change_mtu,
4326 	.ndo_set_mac_address	= bond_set_mac_address,
4327 	.ndo_neigh_setup	= bond_neigh_setup,
4328 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
4329 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
4330 #ifdef CONFIG_NET_POLL_CONTROLLER
4331 	.ndo_netpoll_setup	= bond_netpoll_setup,
4332 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
4333 	.ndo_poll_controller	= bond_poll_controller,
4334 #endif
4335 	.ndo_add_slave		= bond_enslave,
4336 	.ndo_del_slave		= bond_release,
4337 	.ndo_fix_features	= bond_fix_features,
4338 };
4339 
4340 static const struct device_type bond_type = {
4341 	.name = "bond",
4342 };
4343 
4344 static void bond_destructor(struct net_device *bond_dev)
4345 {
4346 	struct bonding *bond = netdev_priv(bond_dev);
4347 	if (bond->wq)
4348 		destroy_workqueue(bond->wq);
4349 	free_netdev(bond_dev);
4350 }
4351 
4352 static void bond_setup(struct net_device *bond_dev)
4353 {
4354 	struct bonding *bond = netdev_priv(bond_dev);
4355 
4356 	/* initialize rwlocks */
4357 	rwlock_init(&bond->lock);
4358 	rwlock_init(&bond->curr_slave_lock);
4359 
4360 	bond->params = bonding_defaults;
4361 
4362 	/* Initialize pointers */
4363 	bond->dev = bond_dev;
4364 	INIT_LIST_HEAD(&bond->vlan_list);
4365 
4366 	/* Initialize the device entry points */
4367 	ether_setup(bond_dev);
4368 	bond_dev->netdev_ops = &bond_netdev_ops;
4369 	bond_dev->ethtool_ops = &bond_ethtool_ops;
4370 	bond_set_mode_ops(bond, bond->params.mode);
4371 
4372 	bond_dev->destructor = bond_destructor;
4373 
4374 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4375 
4376 	/* Initialize the device options */
4377 	bond_dev->tx_queue_len = 0;
4378 	bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4379 	bond_dev->priv_flags |= IFF_BONDING;
4380 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4381 
4382 	/* At first, we block adding VLANs. That's the only way to
4383 	 * prevent problems that occur when adding VLANs over an
4384 	 * empty bond. The block will be removed once non-challenged
4385 	 * slaves are enslaved.
4386 	 */
4387 	bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4388 
4389 	/* don't acquire bond device's netif_tx_lock when
4390 	 * transmitting */
4391 	bond_dev->features |= NETIF_F_LLTX;
4392 
4393 	/* By default, we declare the bond to be fully
4394 	 * VLAN hardware accelerated capable. Special
4395 	 * care is taken in the various xmit functions
4396 	 * when there are slaves that are not hw accel
4397 	 * capable
4398 	 */
4399 
4400 	bond_dev->hw_features = BOND_VLAN_FEATURES |
4401 				NETIF_F_HW_VLAN_CTAG_TX |
4402 				NETIF_F_HW_VLAN_CTAG_RX |
4403 				NETIF_F_HW_VLAN_CTAG_FILTER;
4404 
4405 	bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4406 	bond_dev->features |= bond_dev->hw_features;
4407 }
4408 
4409 /*
4410 * Destroy a bonding device.
4411 * Must be under rtnl_lock when this function is called.
4412 */
4413 static void bond_uninit(struct net_device *bond_dev)
4414 {
4415 	struct bonding *bond = netdev_priv(bond_dev);
4416 	struct vlan_entry *vlan, *tmp;
4417 
4418 	bond_netpoll_cleanup(bond_dev);
4419 
4420 	/* Release the bonded slaves */
4421 	while (bond->first_slave != NULL)
4422 		__bond_release_one(bond_dev, bond->first_slave->dev, true);
4423 	pr_info("%s: released all slaves\n", bond_dev->name);
4424 
4425 	list_del(&bond->bond_list);
4426 
4427 	bond_debug_unregister(bond);
4428 
4429 	__hw_addr_flush(&bond->mc_list);
4430 
4431 	list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4432 		list_del(&vlan->vlan_list);
4433 		kfree(vlan);
4434 	}
4435 }
4436 
4437 /*------------------------- Module initialization ---------------------------*/
4438 
4439 /*
4440  * Convert string input module parms.  Accept either the
4441  * number of the mode or its string name.  A bit complicated because
4442  * some mode names are substrings of other names, and calls from sysfs
4443  * may have whitespace in the name (trailing newlines, for example).
4444  */
4445 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4446 {
4447 	int modeint = -1, i, rv;
4448 	char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4449 
4450 	for (p = (char *)buf; *p; p++)
4451 		if (!(isdigit(*p) || isspace(*p)))
4452 			break;
4453 
4454 	if (*p)
4455 		rv = sscanf(buf, "%20s", modestr);
4456 	else
4457 		rv = sscanf(buf, "%d", &modeint);
4458 
4459 	if (!rv)
4460 		return -1;
4461 
4462 	for (i = 0; tbl[i].modename; i++) {
4463 		if (modeint == tbl[i].mode)
4464 			return tbl[i].mode;
4465 		if (strcmp(modestr, tbl[i].modename) == 0)
4466 			return tbl[i].mode;
4467 	}
4468 
4469 	return -1;
4470 }
4471 
4472 static int bond_check_params(struct bond_params *params)
4473 {
4474 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4475 
4476 	/*
4477 	 * Convert string parameters.
4478 	 */
4479 	if (mode) {
4480 		bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4481 		if (bond_mode == -1) {
4482 			pr_err("Error: Invalid bonding mode \"%s\"\n",
4483 			       mode == NULL ? "NULL" : mode);
4484 			return -EINVAL;
4485 		}
4486 	}
4487 
4488 	if (xmit_hash_policy) {
4489 		if ((bond_mode != BOND_MODE_XOR) &&
4490 		    (bond_mode != BOND_MODE_8023AD)) {
4491 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4492 			       bond_mode_name(bond_mode));
4493 		} else {
4494 			xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4495 							xmit_hashtype_tbl);
4496 			if (xmit_hashtype == -1) {
4497 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4498 				       xmit_hash_policy == NULL ? "NULL" :
4499 				       xmit_hash_policy);
4500 				return -EINVAL;
4501 			}
4502 		}
4503 	}
4504 
4505 	if (lacp_rate) {
4506 		if (bond_mode != BOND_MODE_8023AD) {
4507 			pr_info("lacp_rate param is irrelevant in mode %s\n",
4508 				bond_mode_name(bond_mode));
4509 		} else {
4510 			lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4511 			if (lacp_fast == -1) {
4512 				pr_err("Error: Invalid lacp rate \"%s\"\n",
4513 				       lacp_rate == NULL ? "NULL" : lacp_rate);
4514 				return -EINVAL;
4515 			}
4516 		}
4517 	}
4518 
4519 	if (ad_select) {
4520 		params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4521 		if (params->ad_select == -1) {
4522 			pr_err("Error: Invalid ad_select \"%s\"\n",
4523 			       ad_select == NULL ? "NULL" : ad_select);
4524 			return -EINVAL;
4525 		}
4526 
4527 		if (bond_mode != BOND_MODE_8023AD) {
4528 			pr_warning("ad_select param only affects 802.3ad mode\n");
4529 		}
4530 	} else {
4531 		params->ad_select = BOND_AD_STABLE;
4532 	}
4533 
4534 	if (max_bonds < 0) {
4535 		pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4536 			   max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4537 		max_bonds = BOND_DEFAULT_MAX_BONDS;
4538 	}
4539 
4540 	if (miimon < 0) {
4541 		pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4542 			   miimon, INT_MAX, BOND_LINK_MON_INTERV);
4543 		miimon = BOND_LINK_MON_INTERV;
4544 	}
4545 
4546 	if (updelay < 0) {
4547 		pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4548 			   updelay, INT_MAX);
4549 		updelay = 0;
4550 	}
4551 
4552 	if (downdelay < 0) {
4553 		pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4554 			   downdelay, INT_MAX);
4555 		downdelay = 0;
4556 	}
4557 
4558 	if ((use_carrier != 0) && (use_carrier != 1)) {
4559 		pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4560 			   use_carrier);
4561 		use_carrier = 1;
4562 	}
4563 
4564 	if (num_peer_notif < 0 || num_peer_notif > 255) {
4565 		pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4566 			   num_peer_notif);
4567 		num_peer_notif = 1;
4568 	}
4569 
4570 	/* reset values for 802.3ad */
4571 	if (bond_mode == BOND_MODE_8023AD) {
4572 		if (!miimon) {
4573 			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");
4574 			pr_warning("Forcing miimon to 100msec\n");
4575 			miimon = 100;
4576 		}
4577 	}
4578 
4579 	if (tx_queues < 1 || tx_queues > 255) {
4580 		pr_warning("Warning: tx_queues (%d) should be between "
4581 			   "1 and 255, resetting to %d\n",
4582 			   tx_queues, BOND_DEFAULT_TX_QUEUES);
4583 		tx_queues = BOND_DEFAULT_TX_QUEUES;
4584 	}
4585 
4586 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4587 		pr_warning("Warning: all_slaves_active module parameter (%d), "
4588 			   "not of valid value (0/1), so it was set to "
4589 			   "0\n", all_slaves_active);
4590 		all_slaves_active = 0;
4591 	}
4592 
4593 	if (resend_igmp < 0 || resend_igmp > 255) {
4594 		pr_warning("Warning: resend_igmp (%d) should be between "
4595 			   "0 and 255, resetting to %d\n",
4596 			   resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4597 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4598 	}
4599 
4600 	/* reset values for TLB/ALB */
4601 	if ((bond_mode == BOND_MODE_TLB) ||
4602 	    (bond_mode == BOND_MODE_ALB)) {
4603 		if (!miimon) {
4604 			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");
4605 			pr_warning("Forcing miimon to 100msec\n");
4606 			miimon = 100;
4607 		}
4608 	}
4609 
4610 	if (bond_mode == BOND_MODE_ALB) {
4611 		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",
4612 			  updelay);
4613 	}
4614 
4615 	if (!miimon) {
4616 		if (updelay || downdelay) {
4617 			/* just warn the user the up/down delay will have
4618 			 * no effect since miimon is zero...
4619 			 */
4620 			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",
4621 				   updelay, downdelay);
4622 		}
4623 	} else {
4624 		/* don't allow arp monitoring */
4625 		if (arp_interval) {
4626 			pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4627 				   miimon, arp_interval);
4628 			arp_interval = 0;
4629 		}
4630 
4631 		if ((updelay % miimon) != 0) {
4632 			pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4633 				   updelay, miimon,
4634 				   (updelay / miimon) * miimon);
4635 		}
4636 
4637 		updelay /= miimon;
4638 
4639 		if ((downdelay % miimon) != 0) {
4640 			pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4641 				   downdelay, miimon,
4642 				   (downdelay / miimon) * miimon);
4643 		}
4644 
4645 		downdelay /= miimon;
4646 	}
4647 
4648 	if (arp_interval < 0) {
4649 		pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4650 			   arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4651 		arp_interval = BOND_LINK_ARP_INTERV;
4652 	}
4653 
4654 	for (arp_ip_count = 0, i = 0;
4655 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4656 		/* not complete check, but should be good enough to
4657 		   catch mistakes */
4658 		__be32 ip = in_aton(arp_ip_target[i]);
4659 		if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4660 		    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[i]);
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 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4700 			arp_interval,
4701 			arp_validate_tbl[arp_validate_value].modename,
4702 			arp_ip_count);
4703 
4704 		for (i = 0; i < arp_ip_count; i++)
4705 			pr_info(" %s", arp_ip_target[i]);
4706 
4707 		pr_info("\n");
4708 
4709 	} else if (max_bonds) {
4710 		/* miimon and arp_interval not set, we need one so things
4711 		 * work as expected, see bonding.txt for details
4712 		 */
4713 		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");
4714 	}
4715 
4716 	if (primary && !USES_PRIMARY(bond_mode)) {
4717 		/* currently, using a primary only makes sense
4718 		 * in active backup, TLB or ALB modes
4719 		 */
4720 		pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4721 			   primary, bond_mode_name(bond_mode));
4722 		primary = NULL;
4723 	}
4724 
4725 	if (primary && primary_reselect) {
4726 		primary_reselect_value = bond_parse_parm(primary_reselect,
4727 							 pri_reselect_tbl);
4728 		if (primary_reselect_value == -1) {
4729 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
4730 			       primary_reselect ==
4731 					NULL ? "NULL" : primary_reselect);
4732 			return -EINVAL;
4733 		}
4734 	} else {
4735 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4736 	}
4737 
4738 	if (fail_over_mac) {
4739 		fail_over_mac_value = bond_parse_parm(fail_over_mac,
4740 						      fail_over_mac_tbl);
4741 		if (fail_over_mac_value == -1) {
4742 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
4743 			       arp_validate == NULL ? "NULL" : arp_validate);
4744 			return -EINVAL;
4745 		}
4746 
4747 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4748 			pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4749 	} else {
4750 		fail_over_mac_value = BOND_FOM_NONE;
4751 	}
4752 
4753 	/* fill params struct with the proper values */
4754 	params->mode = bond_mode;
4755 	params->xmit_policy = xmit_hashtype;
4756 	params->miimon = miimon;
4757 	params->num_peer_notif = num_peer_notif;
4758 	params->arp_interval = arp_interval;
4759 	params->arp_validate = arp_validate_value;
4760 	params->updelay = updelay;
4761 	params->downdelay = downdelay;
4762 	params->use_carrier = use_carrier;
4763 	params->lacp_fast = lacp_fast;
4764 	params->primary[0] = 0;
4765 	params->primary_reselect = primary_reselect_value;
4766 	params->fail_over_mac = fail_over_mac_value;
4767 	params->tx_queues = tx_queues;
4768 	params->all_slaves_active = all_slaves_active;
4769 	params->resend_igmp = resend_igmp;
4770 	params->min_links = min_links;
4771 
4772 	if (primary) {
4773 		strncpy(params->primary, primary, IFNAMSIZ);
4774 		params->primary[IFNAMSIZ - 1] = 0;
4775 	}
4776 
4777 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4778 
4779 	return 0;
4780 }
4781 
4782 static struct lock_class_key bonding_netdev_xmit_lock_key;
4783 static struct lock_class_key bonding_netdev_addr_lock_key;
4784 static struct lock_class_key bonding_tx_busylock_key;
4785 
4786 static void bond_set_lockdep_class_one(struct net_device *dev,
4787 				       struct netdev_queue *txq,
4788 				       void *_unused)
4789 {
4790 	lockdep_set_class(&txq->_xmit_lock,
4791 			  &bonding_netdev_xmit_lock_key);
4792 }
4793 
4794 static void bond_set_lockdep_class(struct net_device *dev)
4795 {
4796 	lockdep_set_class(&dev->addr_list_lock,
4797 			  &bonding_netdev_addr_lock_key);
4798 	netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4799 	dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4800 }
4801 
4802 /*
4803  * Called from registration process
4804  */
4805 static int bond_init(struct net_device *bond_dev)
4806 {
4807 	struct bonding *bond = netdev_priv(bond_dev);
4808 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4809 	struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4810 
4811 	pr_debug("Begin bond_init for %s\n", bond_dev->name);
4812 
4813 	/*
4814 	 * Initialize locks that may be required during
4815 	 * en/deslave operations.  All of the bond_open work
4816 	 * (of which this is part) should really be moved to
4817 	 * a phase prior to dev_open
4818 	 */
4819 	spin_lock_init(&(bond_info->tx_hashtbl_lock));
4820 	spin_lock_init(&(bond_info->rx_hashtbl_lock));
4821 
4822 	bond->wq = create_singlethread_workqueue(bond_dev->name);
4823 	if (!bond->wq)
4824 		return -ENOMEM;
4825 
4826 	bond_set_lockdep_class(bond_dev);
4827 
4828 	list_add_tail(&bond->bond_list, &bn->dev_list);
4829 
4830 	bond_prepare_sysfs_group(bond);
4831 
4832 	bond_debug_register(bond);
4833 
4834 	/* Ensure valid dev_addr */
4835 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
4836 	    bond_dev->addr_assign_type == NET_ADDR_PERM) {
4837 		eth_hw_addr_random(bond_dev);
4838 		bond->dev_addr_from_first = true;
4839 	}
4840 
4841 	__hw_addr_init(&bond->mc_list);
4842 	return 0;
4843 }
4844 
4845 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4846 {
4847 	if (tb[IFLA_ADDRESS]) {
4848 		if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4849 			return -EINVAL;
4850 		if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4851 			return -EADDRNOTAVAIL;
4852 	}
4853 	return 0;
4854 }
4855 
4856 static unsigned int bond_get_num_tx_queues(void)
4857 {
4858 	return tx_queues;
4859 }
4860 
4861 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4862 	.kind			= "bond",
4863 	.priv_size		= sizeof(struct bonding),
4864 	.setup			= bond_setup,
4865 	.validate		= bond_validate,
4866 	.get_num_tx_queues	= bond_get_num_tx_queues,
4867 	.get_num_rx_queues	= bond_get_num_tx_queues, /* Use the same number
4868 							     as for TX queues */
4869 };
4870 
4871 /* Create a new bond based on the specified name and bonding parameters.
4872  * If name is NULL, obtain a suitable "bond%d" name for us.
4873  * Caller must NOT hold rtnl_lock; we need to release it here before we
4874  * set up our sysfs entries.
4875  */
4876 int bond_create(struct net *net, const char *name)
4877 {
4878 	struct net_device *bond_dev;
4879 	int res;
4880 
4881 	rtnl_lock();
4882 
4883 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4884 				   name ? name : "bond%d",
4885 				   bond_setup, tx_queues);
4886 	if (!bond_dev) {
4887 		pr_err("%s: eek! can't alloc netdev!\n", name);
4888 		rtnl_unlock();
4889 		return -ENOMEM;
4890 	}
4891 
4892 	dev_net_set(bond_dev, net);
4893 	bond_dev->rtnl_link_ops = &bond_link_ops;
4894 
4895 	res = register_netdevice(bond_dev);
4896 
4897 	netif_carrier_off(bond_dev);
4898 
4899 	rtnl_unlock();
4900 	if (res < 0)
4901 		bond_destructor(bond_dev);
4902 	return res;
4903 }
4904 
4905 static int __net_init bond_net_init(struct net *net)
4906 {
4907 	struct bond_net *bn = net_generic(net, bond_net_id);
4908 
4909 	bn->net = net;
4910 	INIT_LIST_HEAD(&bn->dev_list);
4911 
4912 	bond_create_proc_dir(bn);
4913 	bond_create_sysfs(bn);
4914 
4915 	return 0;
4916 }
4917 
4918 static void __net_exit bond_net_exit(struct net *net)
4919 {
4920 	struct bond_net *bn = net_generic(net, bond_net_id);
4921 	struct bonding *bond, *tmp_bond;
4922 	LIST_HEAD(list);
4923 
4924 	bond_destroy_sysfs(bn);
4925 	bond_destroy_proc_dir(bn);
4926 
4927 	/* Kill off any bonds created after unregistering bond rtnl ops */
4928 	rtnl_lock();
4929 	list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4930 		unregister_netdevice_queue(bond->dev, &list);
4931 	unregister_netdevice_many(&list);
4932 	rtnl_unlock();
4933 }
4934 
4935 static struct pernet_operations bond_net_ops = {
4936 	.init = bond_net_init,
4937 	.exit = bond_net_exit,
4938 	.id   = &bond_net_id,
4939 	.size = sizeof(struct bond_net),
4940 };
4941 
4942 static int __init bonding_init(void)
4943 {
4944 	int i;
4945 	int res;
4946 
4947 	pr_info("%s", bond_version);
4948 
4949 	res = bond_check_params(&bonding_defaults);
4950 	if (res)
4951 		goto out;
4952 
4953 	res = register_pernet_subsys(&bond_net_ops);
4954 	if (res)
4955 		goto out;
4956 
4957 	res = rtnl_link_register(&bond_link_ops);
4958 	if (res)
4959 		goto err_link;
4960 
4961 	bond_create_debugfs();
4962 
4963 	for (i = 0; i < max_bonds; i++) {
4964 		res = bond_create(&init_net, NULL);
4965 		if (res)
4966 			goto err;
4967 	}
4968 
4969 	register_netdevice_notifier(&bond_netdev_notifier);
4970 out:
4971 	return res;
4972 err:
4973 	rtnl_link_unregister(&bond_link_ops);
4974 err_link:
4975 	unregister_pernet_subsys(&bond_net_ops);
4976 	goto out;
4977 
4978 }
4979 
4980 static void __exit bonding_exit(void)
4981 {
4982 	unregister_netdevice_notifier(&bond_netdev_notifier);
4983 
4984 	bond_destroy_debugfs();
4985 
4986 	rtnl_link_unregister(&bond_link_ops);
4987 	unregister_pernet_subsys(&bond_net_ops);
4988 
4989 #ifdef CONFIG_NET_POLL_CONTROLLER
4990 	/*
4991 	 * Make sure we don't have an imbalance on our netpoll blocking
4992 	 */
4993 	WARN_ON(atomic_read(&netpoll_block_tx));
4994 #endif
4995 }
4996 
4997 module_init(bonding_init);
4998 module_exit(bonding_exit);
4999 MODULE_LICENSE("GPL");
5000 MODULE_VERSION(DRV_VERSION);
5001 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5002 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5003 MODULE_ALIAS_RTNL_LINK("bond");
5004