1 // SPDX-License-Identifier: GPL-1.0+
2 /*
3  * originally based on the dummy device.
4  *
5  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
6  * Based on dummy.c, and eql.c devices.
7  *
8  * bonding.c: an Ethernet Bonding driver
9  *
10  * This is useful to talk to a Cisco EtherChannel compatible equipment:
11  *	Cisco 5500
12  *	Sun Trunking (Solaris)
13  *	Alteon AceDirector Trunks
14  *	Linux Bonding
15  *	and probably many L2 switches ...
16  *
17  * How it works:
18  *    ifconfig bond0 ipaddress netmask up
19  *      will setup a network device, with an ip address.  No mac address
20  *	will be assigned at this time.  The hw mac address will come from
21  *	the first slave bonded to the channel.  All slaves will then use
22  *	this hw mac address.
23  *
24  *    ifconfig bond0 down
25  *         will release all slaves, marking them as down.
26  *
27  *    ifenslave bond0 eth0
28  *	will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
29  *	a: be used as initial mac address
30  *	b: if a hw mac address already is there, eth0's hw mac address
31  *	   will then be set from bond0.
32  *
33  */
34 
35 #include <linux/kernel.h>
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/fcntl.h>
39 #include <linux/filter.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/icmp.h>
47 #include <linux/icmpv6.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <linux/io.h>
59 #include <asm/dma.h>
60 #include <linux/uaccess.h>
61 #include <linux/errno.h>
62 #include <linux/netdevice.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/smp.h>
70 #include <linux/if_ether.h>
71 #include <net/arp.h>
72 #include <linux/mii.h>
73 #include <linux/ethtool.h>
74 #include <linux/if_vlan.h>
75 #include <linux/if_bonding.h>
76 #include <linux/phy.h>
77 #include <linux/jiffies.h>
78 #include <linux/preempt.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include <net/pkt_sched.h>
83 #include <linux/rculist.h>
84 #include <net/flow_dissector.h>
85 #include <net/xfrm.h>
86 #include <net/bonding.h>
87 #include <net/bond_3ad.h>
88 #include <net/bond_alb.h>
89 #if IS_ENABLED(CONFIG_TLS_DEVICE)
90 #include <net/tls.h>
91 #endif
92 #include <net/ip6_route.h>
93 #include <net/xdp.h>
94 
95 #include "bonding_priv.h"
96 
97 /*---------------------------- Module parameters ----------------------------*/
98 
99 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
100 
101 static int max_bonds	= BOND_DEFAULT_MAX_BONDS;
102 static int tx_queues	= BOND_DEFAULT_TX_QUEUES;
103 static int num_peer_notif = 1;
104 static int miimon;
105 static int updelay;
106 static int downdelay;
107 static int use_carrier	= 1;
108 static char *mode;
109 static char *primary;
110 static char *primary_reselect;
111 static char *lacp_rate;
112 static int min_links;
113 static char *ad_select;
114 static char *xmit_hash_policy;
115 static int arp_interval;
116 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
117 static char *arp_validate;
118 static char *arp_all_targets;
119 static char *fail_over_mac;
120 static int all_slaves_active;
121 static struct bond_params bonding_defaults;
122 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
123 static int packets_per_slave = 1;
124 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
125 
126 module_param(max_bonds, int, 0);
127 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
128 module_param(tx_queues, int, 0);
129 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
130 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
131 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
132 			       "failover event (alias of num_unsol_na)");
133 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
134 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
135 			       "failover event (alias of num_grat_arp)");
136 module_param(miimon, int, 0);
137 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
138 module_param(updelay, int, 0);
139 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
140 module_param(downdelay, int, 0);
141 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
142 			    "in milliseconds");
143 module_param(use_carrier, int, 0);
144 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
145 			      "0 for off, 1 for on (default)");
146 module_param(mode, charp, 0);
147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
148 		       "1 for active-backup, 2 for balance-xor, "
149 		       "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
150 		       "6 for balance-alb");
151 module_param(primary, charp, 0);
152 MODULE_PARM_DESC(primary, "Primary network device to use");
153 module_param(primary_reselect, charp, 0);
154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
155 				   "once it comes up; "
156 				   "0 for always (default), "
157 				   "1 for only if speed of primary is "
158 				   "better, "
159 				   "2 for only on active slave "
160 				   "failure");
161 module_param(lacp_rate, charp, 0);
162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163 			    "0 for slow, 1 for fast");
164 module_param(ad_select, charp, 0);
165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166 			    "0 for stable (default), 1 for bandwidth, "
167 			    "2 for count");
168 module_param(min_links, int, 0);
169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
170 
171 module_param(xmit_hash_policy, charp, 0);
172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
173 				   "0 for layer 2 (default), 1 for layer 3+4, "
174 				   "2 for layer 2+3, 3 for encap layer 2+3, "
175 				   "4 for encap layer 3+4, 5 for vlan+srcmac");
176 module_param(arp_interval, int, 0);
177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178 module_param_array(arp_ip_target, charp, NULL, 0);
179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180 module_param(arp_validate, charp, 0);
181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
182 			       "0 for none (default), 1 for active, "
183 			       "2 for backup, 3 for all");
184 module_param(arp_all_targets, charp, 0);
185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186 module_param(fail_over_mac, charp, 0);
187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
188 				"the same MAC; 0 for none (default), "
189 				"1 for active, 2 for follow");
190 module_param(all_slaves_active, int, 0);
191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
192 				     "by setting active flag for all slaves; "
193 				     "0 for never (default), 1 for always.");
194 module_param(resend_igmp, int, 0);
195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
196 			      "link failure");
197 module_param(packets_per_slave, int, 0);
198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
199 				    "mode; 0 for a random slave, 1 packet per "
200 				    "slave (default), >1 packets per slave.");
201 module_param(lp_interval, uint, 0);
202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
203 			      "the bonding driver sends learning packets to "
204 			      "each slaves peer switch. The default is 1.");
205 
206 /*----------------------------- Global variables ----------------------------*/
207 
208 #ifdef CONFIG_NET_POLL_CONTROLLER
209 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210 #endif
211 
212 unsigned int bond_net_id __read_mostly;
213 
214 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
215 	{
216 		.key_id = FLOW_DISSECTOR_KEY_CONTROL,
217 		.offset = offsetof(struct flow_keys, control),
218 	},
219 	{
220 		.key_id = FLOW_DISSECTOR_KEY_BASIC,
221 		.offset = offsetof(struct flow_keys, basic),
222 	},
223 	{
224 		.key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
225 		.offset = offsetof(struct flow_keys, addrs.v4addrs),
226 	},
227 	{
228 		.key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
229 		.offset = offsetof(struct flow_keys, addrs.v6addrs),
230 	},
231 	{
232 		.key_id = FLOW_DISSECTOR_KEY_TIPC,
233 		.offset = offsetof(struct flow_keys, addrs.tipckey),
234 	},
235 	{
236 		.key_id = FLOW_DISSECTOR_KEY_PORTS,
237 		.offset = offsetof(struct flow_keys, ports),
238 	},
239 	{
240 		.key_id = FLOW_DISSECTOR_KEY_ICMP,
241 		.offset = offsetof(struct flow_keys, icmp),
242 	},
243 	{
244 		.key_id = FLOW_DISSECTOR_KEY_VLAN,
245 		.offset = offsetof(struct flow_keys, vlan),
246 	},
247 	{
248 		.key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
249 		.offset = offsetof(struct flow_keys, tags),
250 	},
251 	{
252 		.key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
253 		.offset = offsetof(struct flow_keys, keyid),
254 	},
255 };
256 
257 static struct flow_dissector flow_keys_bonding __read_mostly;
258 
259 /*-------------------------- Forward declarations ---------------------------*/
260 
261 static int bond_init(struct net_device *bond_dev);
262 static void bond_uninit(struct net_device *bond_dev);
263 static void bond_get_stats(struct net_device *bond_dev,
264 			   struct rtnl_link_stats64 *stats);
265 static void bond_slave_arr_handler(struct work_struct *work);
266 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
267 				  int mod);
268 static void bond_netdev_notify_work(struct work_struct *work);
269 
270 /*---------------------------- General routines -----------------------------*/
271 
bond_mode_name(int mode)272 const char *bond_mode_name(int mode)
273 {
274 	static const char *names[] = {
275 		[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
276 		[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
277 		[BOND_MODE_XOR] = "load balancing (xor)",
278 		[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
279 		[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
280 		[BOND_MODE_TLB] = "transmit load balancing",
281 		[BOND_MODE_ALB] = "adaptive load balancing",
282 	};
283 
284 	if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
285 		return "unknown";
286 
287 	return names[mode];
288 }
289 
290 /**
291  * bond_dev_queue_xmit - Prepare skb for xmit.
292  *
293  * @bond: bond device that got this skb for tx.
294  * @skb: hw accel VLAN tagged skb to transmit
295  * @slave_dev: slave that is supposed to xmit this skbuff
296  */
bond_dev_queue_xmit(struct bonding * bond,struct sk_buff * skb,struct net_device * slave_dev)297 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
298 			struct net_device *slave_dev)
299 {
300 	skb->dev = slave_dev;
301 
302 	BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
303 		     sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
304 	skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
305 
306 	if (unlikely(netpoll_tx_running(bond->dev)))
307 		return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
308 
309 	return dev_queue_xmit(skb);
310 }
311 
bond_sk_check(struct bonding * bond)312 static bool bond_sk_check(struct bonding *bond)
313 {
314 	switch (BOND_MODE(bond)) {
315 	case BOND_MODE_8023AD:
316 	case BOND_MODE_XOR:
317 		if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
318 			return true;
319 		fallthrough;
320 	default:
321 		return false;
322 	}
323 }
324 
bond_xdp_check(struct bonding * bond)325 static bool bond_xdp_check(struct bonding *bond)
326 {
327 	switch (BOND_MODE(bond)) {
328 	case BOND_MODE_ROUNDROBIN:
329 	case BOND_MODE_ACTIVEBACKUP:
330 		return true;
331 	case BOND_MODE_8023AD:
332 	case BOND_MODE_XOR:
333 		/* vlan+srcmac is not supported with XDP as in most cases the 802.1q
334 		 * payload is not in the packet due to hardware offload.
335 		 */
336 		if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
337 			return true;
338 		fallthrough;
339 	default:
340 		return false;
341 	}
342 }
343 
344 /*---------------------------------- VLAN -----------------------------------*/
345 
346 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
347  * We don't protect the slave list iteration with a lock because:
348  * a. This operation is performed in IOCTL context,
349  * b. The operation is protected by the RTNL semaphore in the 8021q code,
350  * c. Holding a lock with BH disabled while directly calling a base driver
351  *    entry point is generally a BAD idea.
352  *
353  * The design of synchronization/protection for this operation in the 8021q
354  * module is good for one or more VLAN devices over a single physical device
355  * and cannot be extended for a teaming solution like bonding, so there is a
356  * potential race condition here where a net device from the vlan group might
357  * be referenced (either by a base driver or the 8021q code) while it is being
358  * removed from the system. However, it turns out we're not making matters
359  * worse, and if it works for regular VLAN usage it will work here too.
360 */
361 
362 /**
363  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
364  * @bond_dev: bonding net device that got called
365  * @proto: network protocol ID
366  * @vid: vlan id being added
367  */
bond_vlan_rx_add_vid(struct net_device * bond_dev,__be16 proto,u16 vid)368 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
369 				__be16 proto, u16 vid)
370 {
371 	struct bonding *bond = netdev_priv(bond_dev);
372 	struct slave *slave, *rollback_slave;
373 	struct list_head *iter;
374 	int res;
375 
376 	bond_for_each_slave(bond, slave, iter) {
377 		res = vlan_vid_add(slave->dev, proto, vid);
378 		if (res)
379 			goto unwind;
380 	}
381 
382 	return 0;
383 
384 unwind:
385 	/* unwind to the slave that failed */
386 	bond_for_each_slave(bond, rollback_slave, iter) {
387 		if (rollback_slave == slave)
388 			break;
389 
390 		vlan_vid_del(rollback_slave->dev, proto, vid);
391 	}
392 
393 	return res;
394 }
395 
396 /**
397  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
398  * @bond_dev: bonding net device that got called
399  * @proto: network protocol ID
400  * @vid: vlan id being removed
401  */
bond_vlan_rx_kill_vid(struct net_device * bond_dev,__be16 proto,u16 vid)402 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
403 				 __be16 proto, u16 vid)
404 {
405 	struct bonding *bond = netdev_priv(bond_dev);
406 	struct list_head *iter;
407 	struct slave *slave;
408 
409 	bond_for_each_slave(bond, slave, iter)
410 		vlan_vid_del(slave->dev, proto, vid);
411 
412 	if (bond_is_lb(bond))
413 		bond_alb_clear_vlan(bond, vid);
414 
415 	return 0;
416 }
417 
418 /*---------------------------------- XFRM -----------------------------------*/
419 
420 #ifdef CONFIG_XFRM_OFFLOAD
421 /**
422  * bond_ipsec_add_sa - program device with a security association
423  * @xs: pointer to transformer state struct
424  * @extack: extack point to fill failure reason
425  **/
bond_ipsec_add_sa(struct xfrm_state * xs,struct netlink_ext_ack * extack)426 static int bond_ipsec_add_sa(struct xfrm_state *xs,
427 			     struct netlink_ext_ack *extack)
428 {
429 	struct net_device *bond_dev = xs->xso.dev;
430 	struct bond_ipsec *ipsec;
431 	struct bonding *bond;
432 	struct slave *slave;
433 	int err;
434 
435 	if (!bond_dev)
436 		return -EINVAL;
437 
438 	rcu_read_lock();
439 	bond = netdev_priv(bond_dev);
440 	slave = rcu_dereference(bond->curr_active_slave);
441 	if (!slave) {
442 		rcu_read_unlock();
443 		return -ENODEV;
444 	}
445 
446 	if (!slave->dev->xfrmdev_ops ||
447 	    !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
448 	    netif_is_bond_master(slave->dev)) {
449 		NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
450 		rcu_read_unlock();
451 		return -EINVAL;
452 	}
453 
454 	ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
455 	if (!ipsec) {
456 		rcu_read_unlock();
457 		return -ENOMEM;
458 	}
459 	xs->xso.real_dev = slave->dev;
460 
461 	err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
462 	if (!err) {
463 		ipsec->xs = xs;
464 		INIT_LIST_HEAD(&ipsec->list);
465 		spin_lock_bh(&bond->ipsec_lock);
466 		list_add(&ipsec->list, &bond->ipsec_list);
467 		spin_unlock_bh(&bond->ipsec_lock);
468 	} else {
469 		kfree(ipsec);
470 	}
471 	rcu_read_unlock();
472 	return err;
473 }
474 
bond_ipsec_add_sa_all(struct bonding * bond)475 static void bond_ipsec_add_sa_all(struct bonding *bond)
476 {
477 	struct net_device *bond_dev = bond->dev;
478 	struct bond_ipsec *ipsec;
479 	struct slave *slave;
480 
481 	rcu_read_lock();
482 	slave = rcu_dereference(bond->curr_active_slave);
483 	if (!slave)
484 		goto out;
485 
486 	if (!slave->dev->xfrmdev_ops ||
487 	    !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
488 	    netif_is_bond_master(slave->dev)) {
489 		spin_lock_bh(&bond->ipsec_lock);
490 		if (!list_empty(&bond->ipsec_list))
491 			slave_warn(bond_dev, slave->dev,
492 				   "%s: no slave xdo_dev_state_add\n",
493 				   __func__);
494 		spin_unlock_bh(&bond->ipsec_lock);
495 		goto out;
496 	}
497 
498 	spin_lock_bh(&bond->ipsec_lock);
499 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
500 		ipsec->xs->xso.real_dev = slave->dev;
501 		if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
502 			slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
503 			ipsec->xs->xso.real_dev = NULL;
504 		}
505 	}
506 	spin_unlock_bh(&bond->ipsec_lock);
507 out:
508 	rcu_read_unlock();
509 }
510 
511 /**
512  * bond_ipsec_del_sa - clear out this specific SA
513  * @xs: pointer to transformer state struct
514  **/
bond_ipsec_del_sa(struct xfrm_state * xs)515 static void bond_ipsec_del_sa(struct xfrm_state *xs)
516 {
517 	struct net_device *bond_dev = xs->xso.dev;
518 	struct bond_ipsec *ipsec;
519 	struct bonding *bond;
520 	struct slave *slave;
521 
522 	if (!bond_dev)
523 		return;
524 
525 	rcu_read_lock();
526 	bond = netdev_priv(bond_dev);
527 	slave = rcu_dereference(bond->curr_active_slave);
528 
529 	if (!slave)
530 		goto out;
531 
532 	if (!xs->xso.real_dev)
533 		goto out;
534 
535 	WARN_ON(xs->xso.real_dev != slave->dev);
536 
537 	if (!slave->dev->xfrmdev_ops ||
538 	    !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
539 	    netif_is_bond_master(slave->dev)) {
540 		slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
541 		goto out;
542 	}
543 
544 	slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
545 out:
546 	spin_lock_bh(&bond->ipsec_lock);
547 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
548 		if (ipsec->xs == xs) {
549 			list_del(&ipsec->list);
550 			kfree(ipsec);
551 			break;
552 		}
553 	}
554 	spin_unlock_bh(&bond->ipsec_lock);
555 	rcu_read_unlock();
556 }
557 
bond_ipsec_del_sa_all(struct bonding * bond)558 static void bond_ipsec_del_sa_all(struct bonding *bond)
559 {
560 	struct net_device *bond_dev = bond->dev;
561 	struct bond_ipsec *ipsec;
562 	struct slave *slave;
563 
564 	rcu_read_lock();
565 	slave = rcu_dereference(bond->curr_active_slave);
566 	if (!slave) {
567 		rcu_read_unlock();
568 		return;
569 	}
570 
571 	spin_lock_bh(&bond->ipsec_lock);
572 	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
573 		if (!ipsec->xs->xso.real_dev)
574 			continue;
575 
576 		if (!slave->dev->xfrmdev_ops ||
577 		    !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
578 		    netif_is_bond_master(slave->dev)) {
579 			slave_warn(bond_dev, slave->dev,
580 				   "%s: no slave xdo_dev_state_delete\n",
581 				   __func__);
582 		} else {
583 			slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
584 		}
585 		ipsec->xs->xso.real_dev = NULL;
586 	}
587 	spin_unlock_bh(&bond->ipsec_lock);
588 	rcu_read_unlock();
589 }
590 
591 /**
592  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
593  * @skb: current data packet
594  * @xs: pointer to transformer state struct
595  **/
bond_ipsec_offload_ok(struct sk_buff * skb,struct xfrm_state * xs)596 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
597 {
598 	struct net_device *bond_dev = xs->xso.dev;
599 	struct net_device *real_dev;
600 	struct slave *curr_active;
601 	struct bonding *bond;
602 	int err;
603 
604 	bond = netdev_priv(bond_dev);
605 	rcu_read_lock();
606 	curr_active = rcu_dereference(bond->curr_active_slave);
607 	real_dev = curr_active->dev;
608 
609 	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
610 		err = false;
611 		goto out;
612 	}
613 
614 	if (!xs->xso.real_dev) {
615 		err = false;
616 		goto out;
617 	}
618 
619 	if (!real_dev->xfrmdev_ops ||
620 	    !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
621 	    netif_is_bond_master(real_dev)) {
622 		err = false;
623 		goto out;
624 	}
625 
626 	err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
627 out:
628 	rcu_read_unlock();
629 	return err;
630 }
631 
632 static const struct xfrmdev_ops bond_xfrmdev_ops = {
633 	.xdo_dev_state_add = bond_ipsec_add_sa,
634 	.xdo_dev_state_delete = bond_ipsec_del_sa,
635 	.xdo_dev_offload_ok = bond_ipsec_offload_ok,
636 };
637 #endif /* CONFIG_XFRM_OFFLOAD */
638 
639 /*------------------------------- Link status -------------------------------*/
640 
641 /* Set the carrier state for the master according to the state of its
642  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
643  * do special 802.3ad magic.
644  *
645  * Returns zero if carrier state does not change, nonzero if it does.
646  */
bond_set_carrier(struct bonding * bond)647 int bond_set_carrier(struct bonding *bond)
648 {
649 	struct list_head *iter;
650 	struct slave *slave;
651 
652 	if (!bond_has_slaves(bond))
653 		goto down;
654 
655 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
656 		return bond_3ad_set_carrier(bond);
657 
658 	bond_for_each_slave(bond, slave, iter) {
659 		if (slave->link == BOND_LINK_UP) {
660 			if (!netif_carrier_ok(bond->dev)) {
661 				netif_carrier_on(bond->dev);
662 				return 1;
663 			}
664 			return 0;
665 		}
666 	}
667 
668 down:
669 	if (netif_carrier_ok(bond->dev)) {
670 		netif_carrier_off(bond->dev);
671 		return 1;
672 	}
673 	return 0;
674 }
675 
676 /* Get link speed and duplex from the slave's base driver
677  * using ethtool. If for some reason the call fails or the
678  * values are invalid, set speed and duplex to -1,
679  * and return. Return 1 if speed or duplex settings are
680  * UNKNOWN; 0 otherwise.
681  */
bond_update_speed_duplex(struct slave * slave)682 static int bond_update_speed_duplex(struct slave *slave)
683 {
684 	struct net_device *slave_dev = slave->dev;
685 	struct ethtool_link_ksettings ecmd;
686 	int res;
687 
688 	slave->speed = SPEED_UNKNOWN;
689 	slave->duplex = DUPLEX_UNKNOWN;
690 
691 	res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
692 	if (res < 0)
693 		return 1;
694 	if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
695 		return 1;
696 	switch (ecmd.base.duplex) {
697 	case DUPLEX_FULL:
698 	case DUPLEX_HALF:
699 		break;
700 	default:
701 		return 1;
702 	}
703 
704 	slave->speed = ecmd.base.speed;
705 	slave->duplex = ecmd.base.duplex;
706 
707 	return 0;
708 }
709 
bond_slave_link_status(s8 link)710 const char *bond_slave_link_status(s8 link)
711 {
712 	switch (link) {
713 	case BOND_LINK_UP:
714 		return "up";
715 	case BOND_LINK_FAIL:
716 		return "going down";
717 	case BOND_LINK_DOWN:
718 		return "down";
719 	case BOND_LINK_BACK:
720 		return "going back";
721 	default:
722 		return "unknown";
723 	}
724 }
725 
726 /* if <dev> supports MII link status reporting, check its link status.
727  *
728  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
729  * depending upon the setting of the use_carrier parameter.
730  *
731  * Return either BMSR_LSTATUS, meaning that the link is up (or we
732  * can't tell and just pretend it is), or 0, meaning that the link is
733  * down.
734  *
735  * If reporting is non-zero, instead of faking link up, return -1 if
736  * both ETHTOOL and MII ioctls fail (meaning the device does not
737  * support them).  If use_carrier is set, return whatever it says.
738  * It'd be nice if there was a good way to tell if a driver supports
739  * netif_carrier, but there really isn't.
740  */
bond_check_dev_link(struct bonding * bond,struct net_device * slave_dev,int reporting)741 static int bond_check_dev_link(struct bonding *bond,
742 			       struct net_device *slave_dev, int reporting)
743 {
744 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
745 	int (*ioctl)(struct net_device *, struct ifreq *, int);
746 	struct ifreq ifr;
747 	struct mii_ioctl_data *mii;
748 
749 	if (!reporting && !netif_running(slave_dev))
750 		return 0;
751 
752 	if (bond->params.use_carrier)
753 		return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
754 
755 	/* Try to get link status using Ethtool first. */
756 	if (slave_dev->ethtool_ops->get_link)
757 		return slave_dev->ethtool_ops->get_link(slave_dev) ?
758 			BMSR_LSTATUS : 0;
759 
760 	/* Ethtool can't be used, fallback to MII ioctls. */
761 	ioctl = slave_ops->ndo_eth_ioctl;
762 	if (ioctl) {
763 		/* TODO: set pointer to correct ioctl on a per team member
764 		 *       bases to make this more efficient. that is, once
765 		 *       we determine the correct ioctl, we will always
766 		 *       call it and not the others for that team
767 		 *       member.
768 		 */
769 
770 		/* We cannot assume that SIOCGMIIPHY will also read a
771 		 * register; not all network drivers (e.g., e100)
772 		 * support that.
773 		 */
774 
775 		/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
776 		strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
777 		mii = if_mii(&ifr);
778 		if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
779 			mii->reg_num = MII_BMSR;
780 			if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
781 				return mii->val_out & BMSR_LSTATUS;
782 		}
783 	}
784 
785 	/* If reporting, report that either there's no ndo_eth_ioctl,
786 	 * or both SIOCGMIIREG and get_link failed (meaning that we
787 	 * cannot report link status).  If not reporting, pretend
788 	 * we're ok.
789 	 */
790 	return reporting ? -1 : BMSR_LSTATUS;
791 }
792 
793 /*----------------------------- Multicast list ------------------------------*/
794 
795 /* Push the promiscuity flag down to appropriate slaves */
bond_set_promiscuity(struct bonding * bond,int inc)796 static int bond_set_promiscuity(struct bonding *bond, int inc)
797 {
798 	struct list_head *iter;
799 	int err = 0;
800 
801 	if (bond_uses_primary(bond)) {
802 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
803 
804 		if (curr_active)
805 			err = dev_set_promiscuity(curr_active->dev, inc);
806 	} else {
807 		struct slave *slave;
808 
809 		bond_for_each_slave(bond, slave, iter) {
810 			err = dev_set_promiscuity(slave->dev, inc);
811 			if (err)
812 				return err;
813 		}
814 	}
815 	return err;
816 }
817 
818 /* Push the allmulti flag down to all slaves */
bond_set_allmulti(struct bonding * bond,int inc)819 static int bond_set_allmulti(struct bonding *bond, int inc)
820 {
821 	struct list_head *iter;
822 	int err = 0;
823 
824 	if (bond_uses_primary(bond)) {
825 		struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
826 
827 		if (curr_active)
828 			err = dev_set_allmulti(curr_active->dev, inc);
829 	} else {
830 		struct slave *slave;
831 
832 		bond_for_each_slave(bond, slave, iter) {
833 			err = dev_set_allmulti(slave->dev, inc);
834 			if (err)
835 				return err;
836 		}
837 	}
838 	return err;
839 }
840 
841 /* Retrieve the list of registered multicast addresses for the bonding
842  * device and retransmit an IGMP JOIN request to the current active
843  * slave.
844  */
bond_resend_igmp_join_requests_delayed(struct work_struct * work)845 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
846 {
847 	struct bonding *bond = container_of(work, struct bonding,
848 					    mcast_work.work);
849 
850 	if (!rtnl_trylock()) {
851 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
852 		return;
853 	}
854 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
855 
856 	if (bond->igmp_retrans > 1) {
857 		bond->igmp_retrans--;
858 		queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
859 	}
860 	rtnl_unlock();
861 }
862 
863 /* Flush bond's hardware addresses from slave */
bond_hw_addr_flush(struct net_device * bond_dev,struct net_device * slave_dev)864 static void bond_hw_addr_flush(struct net_device *bond_dev,
865 			       struct net_device *slave_dev)
866 {
867 	struct bonding *bond = netdev_priv(bond_dev);
868 
869 	dev_uc_unsync(slave_dev, bond_dev);
870 	dev_mc_unsync(slave_dev, bond_dev);
871 
872 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
873 		dev_mc_del(slave_dev, lacpdu_mcast_addr);
874 }
875 
876 /*--------------------------- Active slave change ---------------------------*/
877 
878 /* Update the hardware address list and promisc/allmulti for the new and
879  * old active slaves (if any).  Modes that are not using primary keep all
880  * slaves up date at all times; only the modes that use primary need to call
881  * this function to swap these settings during a failover.
882  */
bond_hw_addr_swap(struct bonding * bond,struct slave * new_active,struct slave * old_active)883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
884 			      struct slave *old_active)
885 {
886 	if (old_active) {
887 		if (bond->dev->flags & IFF_PROMISC)
888 			dev_set_promiscuity(old_active->dev, -1);
889 
890 		if (bond->dev->flags & IFF_ALLMULTI)
891 			dev_set_allmulti(old_active->dev, -1);
892 
893 		if (bond->dev->flags & IFF_UP)
894 			bond_hw_addr_flush(bond->dev, old_active->dev);
895 	}
896 
897 	if (new_active) {
898 		/* FIXME: Signal errors upstream. */
899 		if (bond->dev->flags & IFF_PROMISC)
900 			dev_set_promiscuity(new_active->dev, 1);
901 
902 		if (bond->dev->flags & IFF_ALLMULTI)
903 			dev_set_allmulti(new_active->dev, 1);
904 
905 		if (bond->dev->flags & IFF_UP) {
906 			netif_addr_lock_bh(bond->dev);
907 			dev_uc_sync(new_active->dev, bond->dev);
908 			dev_mc_sync(new_active->dev, bond->dev);
909 			netif_addr_unlock_bh(bond->dev);
910 		}
911 	}
912 }
913 
914 /**
915  * bond_set_dev_addr - clone slave's address to bond
916  * @bond_dev: bond net device
917  * @slave_dev: slave net device
918  *
919  * Should be called with RTNL held.
920  */
bond_set_dev_addr(struct net_device * bond_dev,struct net_device * slave_dev)921 static int bond_set_dev_addr(struct net_device *bond_dev,
922 			     struct net_device *slave_dev)
923 {
924 	int err;
925 
926 	slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
927 		  bond_dev, slave_dev, slave_dev->addr_len);
928 	err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
929 	if (err)
930 		return err;
931 
932 	__dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
933 	bond_dev->addr_assign_type = NET_ADDR_STOLEN;
934 	call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
935 	return 0;
936 }
937 
bond_get_old_active(struct bonding * bond,struct slave * new_active)938 static struct slave *bond_get_old_active(struct bonding *bond,
939 					 struct slave *new_active)
940 {
941 	struct slave *slave;
942 	struct list_head *iter;
943 
944 	bond_for_each_slave(bond, slave, iter) {
945 		if (slave == new_active)
946 			continue;
947 
948 		if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
949 			return slave;
950 	}
951 
952 	return NULL;
953 }
954 
955 /* bond_do_fail_over_mac
956  *
957  * Perform special MAC address swapping for fail_over_mac settings
958  *
959  * Called with RTNL
960  */
bond_do_fail_over_mac(struct bonding * bond,struct slave * new_active,struct slave * old_active)961 static void bond_do_fail_over_mac(struct bonding *bond,
962 				  struct slave *new_active,
963 				  struct slave *old_active)
964 {
965 	u8 tmp_mac[MAX_ADDR_LEN];
966 	struct sockaddr_storage ss;
967 	int rv;
968 
969 	switch (bond->params.fail_over_mac) {
970 	case BOND_FOM_ACTIVE:
971 		if (new_active) {
972 			rv = bond_set_dev_addr(bond->dev, new_active->dev);
973 			if (rv)
974 				slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
975 					  -rv);
976 		}
977 		break;
978 	case BOND_FOM_FOLLOW:
979 		/* if new_active && old_active, swap them
980 		 * if just old_active, do nothing (going to no active slave)
981 		 * if just new_active, set new_active to bond's MAC
982 		 */
983 		if (!new_active)
984 			return;
985 
986 		if (!old_active)
987 			old_active = bond_get_old_active(bond, new_active);
988 
989 		if (old_active) {
990 			bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
991 					  new_active->dev->addr_len);
992 			bond_hw_addr_copy(ss.__data,
993 					  old_active->dev->dev_addr,
994 					  old_active->dev->addr_len);
995 			ss.ss_family = new_active->dev->type;
996 		} else {
997 			bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
998 					  bond->dev->addr_len);
999 			ss.ss_family = bond->dev->type;
1000 		}
1001 
1002 		rv = dev_set_mac_address(new_active->dev,
1003 					 (struct sockaddr *)&ss, NULL);
1004 		if (rv) {
1005 			slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1006 				  -rv);
1007 			goto out;
1008 		}
1009 
1010 		if (!old_active)
1011 			goto out;
1012 
1013 		bond_hw_addr_copy(ss.__data, tmp_mac,
1014 				  new_active->dev->addr_len);
1015 		ss.ss_family = old_active->dev->type;
1016 
1017 		rv = dev_set_mac_address(old_active->dev,
1018 					 (struct sockaddr *)&ss, NULL);
1019 		if (rv)
1020 			slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1021 				  -rv);
1022 out:
1023 		break;
1024 	default:
1025 		netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1026 			   bond->params.fail_over_mac);
1027 		break;
1028 	}
1029 
1030 }
1031 
1032 /**
1033  * bond_choose_primary_or_current - select the primary or high priority slave
1034  * @bond: our bonding struct
1035  *
1036  * - Check if there is a primary link. If the primary link was set and is up,
1037  *   go on and do link reselection.
1038  *
1039  * - If primary link is not set or down, find the highest priority link.
1040  *   If the highest priority link is not current slave, set it as primary
1041  *   link and do link reselection.
1042  */
bond_choose_primary_or_current(struct bonding * bond)1043 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1044 {
1045 	struct slave *prim = rtnl_dereference(bond->primary_slave);
1046 	struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1047 	struct slave *slave, *hprio = NULL;
1048 	struct list_head *iter;
1049 
1050 	if (!prim || prim->link != BOND_LINK_UP) {
1051 		bond_for_each_slave(bond, slave, iter) {
1052 			if (slave->link == BOND_LINK_UP) {
1053 				hprio = hprio ?: slave;
1054 				if (slave->prio > hprio->prio)
1055 					hprio = slave;
1056 			}
1057 		}
1058 
1059 		if (hprio && hprio != curr) {
1060 			prim = hprio;
1061 			goto link_reselect;
1062 		}
1063 
1064 		if (!curr || curr->link != BOND_LINK_UP)
1065 			return NULL;
1066 		return curr;
1067 	}
1068 
1069 	if (bond->force_primary) {
1070 		bond->force_primary = false;
1071 		return prim;
1072 	}
1073 
1074 link_reselect:
1075 	if (!curr || curr->link != BOND_LINK_UP)
1076 		return prim;
1077 
1078 	/* At this point, prim and curr are both up */
1079 	switch (bond->params.primary_reselect) {
1080 	case BOND_PRI_RESELECT_ALWAYS:
1081 		return prim;
1082 	case BOND_PRI_RESELECT_BETTER:
1083 		if (prim->speed < curr->speed)
1084 			return curr;
1085 		if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1086 			return curr;
1087 		return prim;
1088 	case BOND_PRI_RESELECT_FAILURE:
1089 		return curr;
1090 	default:
1091 		netdev_err(bond->dev, "impossible primary_reselect %d\n",
1092 			   bond->params.primary_reselect);
1093 		return curr;
1094 	}
1095 }
1096 
1097 /**
1098  * bond_find_best_slave - select the best available slave to be the active one
1099  * @bond: our bonding struct
1100  */
bond_find_best_slave(struct bonding * bond)1101 static struct slave *bond_find_best_slave(struct bonding *bond)
1102 {
1103 	struct slave *slave, *bestslave = NULL;
1104 	struct list_head *iter;
1105 	int mintime = bond->params.updelay;
1106 
1107 	slave = bond_choose_primary_or_current(bond);
1108 	if (slave)
1109 		return slave;
1110 
1111 	bond_for_each_slave(bond, slave, iter) {
1112 		if (slave->link == BOND_LINK_UP)
1113 			return slave;
1114 		if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1115 		    slave->delay < mintime) {
1116 			mintime = slave->delay;
1117 			bestslave = slave;
1118 		}
1119 	}
1120 
1121 	return bestslave;
1122 }
1123 
bond_should_notify_peers(struct bonding * bond)1124 static bool bond_should_notify_peers(struct bonding *bond)
1125 {
1126 	struct slave *slave;
1127 
1128 	rcu_read_lock();
1129 	slave = rcu_dereference(bond->curr_active_slave);
1130 	rcu_read_unlock();
1131 
1132 	if (!slave || !bond->send_peer_notif ||
1133 	    bond->send_peer_notif %
1134 	    max(1, bond->params.peer_notif_delay) != 0 ||
1135 	    !netif_carrier_ok(bond->dev) ||
1136 	    test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1137 		return false;
1138 
1139 	netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1140 		   slave ? slave->dev->name : "NULL");
1141 
1142 	return true;
1143 }
1144 
1145 /**
1146  * bond_change_active_slave - change the active slave into the specified one
1147  * @bond: our bonding struct
1148  * @new_active: the new slave to make the active one
1149  *
1150  * Set the new slave to the bond's settings and unset them on the old
1151  * curr_active_slave.
1152  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1153  *
1154  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1155  * because it is apparently the best available slave we have, even though its
1156  * updelay hasn't timed out yet.
1157  *
1158  * Caller must hold RTNL.
1159  */
bond_change_active_slave(struct bonding * bond,struct slave * new_active)1160 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1161 {
1162 	struct slave *old_active;
1163 
1164 	ASSERT_RTNL();
1165 
1166 	old_active = rtnl_dereference(bond->curr_active_slave);
1167 
1168 	if (old_active == new_active)
1169 		return;
1170 
1171 #ifdef CONFIG_XFRM_OFFLOAD
1172 	bond_ipsec_del_sa_all(bond);
1173 #endif /* CONFIG_XFRM_OFFLOAD */
1174 
1175 	if (new_active) {
1176 		new_active->last_link_up = jiffies;
1177 
1178 		if (new_active->link == BOND_LINK_BACK) {
1179 			if (bond_uses_primary(bond)) {
1180 				slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1181 					   (bond->params.updelay - new_active->delay) * bond->params.miimon);
1182 			}
1183 
1184 			new_active->delay = 0;
1185 			bond_set_slave_link_state(new_active, BOND_LINK_UP,
1186 						  BOND_SLAVE_NOTIFY_NOW);
1187 
1188 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
1189 				bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1190 
1191 			if (bond_is_lb(bond))
1192 				bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1193 		} else {
1194 			if (bond_uses_primary(bond))
1195 				slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1196 		}
1197 	}
1198 
1199 	if (bond_uses_primary(bond))
1200 		bond_hw_addr_swap(bond, new_active, old_active);
1201 
1202 	if (bond_is_lb(bond)) {
1203 		bond_alb_handle_active_change(bond, new_active);
1204 		if (old_active)
1205 			bond_set_slave_inactive_flags(old_active,
1206 						      BOND_SLAVE_NOTIFY_NOW);
1207 		if (new_active)
1208 			bond_set_slave_active_flags(new_active,
1209 						    BOND_SLAVE_NOTIFY_NOW);
1210 	} else {
1211 		rcu_assign_pointer(bond->curr_active_slave, new_active);
1212 	}
1213 
1214 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1215 		if (old_active)
1216 			bond_set_slave_inactive_flags(old_active,
1217 						      BOND_SLAVE_NOTIFY_NOW);
1218 
1219 		if (new_active) {
1220 			bool should_notify_peers = false;
1221 
1222 			bond_set_slave_active_flags(new_active,
1223 						    BOND_SLAVE_NOTIFY_NOW);
1224 
1225 			if (bond->params.fail_over_mac)
1226 				bond_do_fail_over_mac(bond, new_active,
1227 						      old_active);
1228 
1229 			if (netif_running(bond->dev)) {
1230 				bond->send_peer_notif =
1231 					bond->params.num_peer_notif *
1232 					max(1, bond->params.peer_notif_delay);
1233 				should_notify_peers =
1234 					bond_should_notify_peers(bond);
1235 			}
1236 
1237 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1238 			if (should_notify_peers) {
1239 				bond->send_peer_notif--;
1240 				call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1241 							 bond->dev);
1242 			}
1243 		}
1244 	}
1245 
1246 #ifdef CONFIG_XFRM_OFFLOAD
1247 	bond_ipsec_add_sa_all(bond);
1248 #endif /* CONFIG_XFRM_OFFLOAD */
1249 
1250 	/* resend IGMP joins since active slave has changed or
1251 	 * all were sent on curr_active_slave.
1252 	 * resend only if bond is brought up with the affected
1253 	 * bonding modes and the retransmission is enabled
1254 	 */
1255 	if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1256 	    ((bond_uses_primary(bond) && new_active) ||
1257 	     BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1258 		bond->igmp_retrans = bond->params.resend_igmp;
1259 		queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1260 	}
1261 }
1262 
1263 /**
1264  * bond_select_active_slave - select a new active slave, if needed
1265  * @bond: our bonding struct
1266  *
1267  * This functions should be called when one of the following occurs:
1268  * - The old curr_active_slave has been released or lost its link.
1269  * - The primary_slave has got its link back.
1270  * - A slave has got its link back and there's no old curr_active_slave.
1271  *
1272  * Caller must hold RTNL.
1273  */
bond_select_active_slave(struct bonding * bond)1274 void bond_select_active_slave(struct bonding *bond)
1275 {
1276 	struct slave *best_slave;
1277 	int rv;
1278 
1279 	ASSERT_RTNL();
1280 
1281 	best_slave = bond_find_best_slave(bond);
1282 	if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1283 		bond_change_active_slave(bond, best_slave);
1284 		rv = bond_set_carrier(bond);
1285 		if (!rv)
1286 			return;
1287 
1288 		if (netif_carrier_ok(bond->dev))
1289 			netdev_info(bond->dev, "active interface up!\n");
1290 		else
1291 			netdev_info(bond->dev, "now running without any active interface!\n");
1292 	}
1293 }
1294 
1295 #ifdef CONFIG_NET_POLL_CONTROLLER
slave_enable_netpoll(struct slave * slave)1296 static inline int slave_enable_netpoll(struct slave *slave)
1297 {
1298 	struct netpoll *np;
1299 	int err = 0;
1300 
1301 	np = kzalloc(sizeof(*np), GFP_KERNEL);
1302 	err = -ENOMEM;
1303 	if (!np)
1304 		goto out;
1305 
1306 	err = __netpoll_setup(np, slave->dev);
1307 	if (err) {
1308 		kfree(np);
1309 		goto out;
1310 	}
1311 	slave->np = np;
1312 out:
1313 	return err;
1314 }
slave_disable_netpoll(struct slave * slave)1315 static inline void slave_disable_netpoll(struct slave *slave)
1316 {
1317 	struct netpoll *np = slave->np;
1318 
1319 	if (!np)
1320 		return;
1321 
1322 	slave->np = NULL;
1323 
1324 	__netpoll_free(np);
1325 }
1326 
bond_poll_controller(struct net_device * bond_dev)1327 static void bond_poll_controller(struct net_device *bond_dev)
1328 {
1329 	struct bonding *bond = netdev_priv(bond_dev);
1330 	struct slave *slave = NULL;
1331 	struct list_head *iter;
1332 	struct ad_info ad_info;
1333 
1334 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1335 		if (bond_3ad_get_active_agg_info(bond, &ad_info))
1336 			return;
1337 
1338 	bond_for_each_slave_rcu(bond, slave, iter) {
1339 		if (!bond_slave_is_up(slave))
1340 			continue;
1341 
1342 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1343 			struct aggregator *agg =
1344 			    SLAVE_AD_INFO(slave)->port.aggregator;
1345 
1346 			if (agg &&
1347 			    agg->aggregator_identifier != ad_info.aggregator_id)
1348 				continue;
1349 		}
1350 
1351 		netpoll_poll_dev(slave->dev);
1352 	}
1353 }
1354 
bond_netpoll_cleanup(struct net_device * bond_dev)1355 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1356 {
1357 	struct bonding *bond = netdev_priv(bond_dev);
1358 	struct list_head *iter;
1359 	struct slave *slave;
1360 
1361 	bond_for_each_slave(bond, slave, iter)
1362 		if (bond_slave_is_up(slave))
1363 			slave_disable_netpoll(slave);
1364 }
1365 
bond_netpoll_setup(struct net_device * dev,struct netpoll_info * ni)1366 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1367 {
1368 	struct bonding *bond = netdev_priv(dev);
1369 	struct list_head *iter;
1370 	struct slave *slave;
1371 	int err = 0;
1372 
1373 	bond_for_each_slave(bond, slave, iter) {
1374 		err = slave_enable_netpoll(slave);
1375 		if (err) {
1376 			bond_netpoll_cleanup(dev);
1377 			break;
1378 		}
1379 	}
1380 	return err;
1381 }
1382 #else
slave_enable_netpoll(struct slave * slave)1383 static inline int slave_enable_netpoll(struct slave *slave)
1384 {
1385 	return 0;
1386 }
slave_disable_netpoll(struct slave * slave)1387 static inline void slave_disable_netpoll(struct slave *slave)
1388 {
1389 }
bond_netpoll_cleanup(struct net_device * bond_dev)1390 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1391 {
1392 }
1393 #endif
1394 
1395 /*---------------------------------- IOCTL ----------------------------------*/
1396 
bond_fix_features(struct net_device * dev,netdev_features_t features)1397 static netdev_features_t bond_fix_features(struct net_device *dev,
1398 					   netdev_features_t features)
1399 {
1400 	struct bonding *bond = netdev_priv(dev);
1401 	struct list_head *iter;
1402 	netdev_features_t mask;
1403 	struct slave *slave;
1404 
1405 	mask = features;
1406 
1407 	features &= ~NETIF_F_ONE_FOR_ALL;
1408 	features |= NETIF_F_ALL_FOR_ALL;
1409 
1410 	bond_for_each_slave(bond, slave, iter) {
1411 		features = netdev_increment_features(features,
1412 						     slave->dev->features,
1413 						     mask);
1414 	}
1415 	features = netdev_add_tso_features(features, mask);
1416 
1417 	return features;
1418 }
1419 
1420 #define BOND_VLAN_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1421 				 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1422 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
1423 
1424 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1425 				 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1426 
1427 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
1428 				 NETIF_F_GSO_SOFTWARE)
1429 
1430 
bond_compute_features(struct bonding * bond)1431 static void bond_compute_features(struct bonding *bond)
1432 {
1433 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1434 					IFF_XMIT_DST_RELEASE_PERM;
1435 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1436 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
1437 #ifdef CONFIG_XFRM_OFFLOAD
1438 	netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1439 #endif /* CONFIG_XFRM_OFFLOAD */
1440 	netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1441 	struct net_device *bond_dev = bond->dev;
1442 	struct list_head *iter;
1443 	struct slave *slave;
1444 	unsigned short max_hard_header_len = ETH_HLEN;
1445 	unsigned int tso_max_size = TSO_MAX_SIZE;
1446 	u16 tso_max_segs = TSO_MAX_SEGS;
1447 
1448 	if (!bond_has_slaves(bond))
1449 		goto done;
1450 	vlan_features &= NETIF_F_ALL_FOR_ALL;
1451 	mpls_features &= NETIF_F_ALL_FOR_ALL;
1452 
1453 	bond_for_each_slave(bond, slave, iter) {
1454 		vlan_features = netdev_increment_features(vlan_features,
1455 			slave->dev->vlan_features, BOND_VLAN_FEATURES);
1456 
1457 		enc_features = netdev_increment_features(enc_features,
1458 							 slave->dev->hw_enc_features,
1459 							 BOND_ENC_FEATURES);
1460 
1461 #ifdef CONFIG_XFRM_OFFLOAD
1462 		xfrm_features = netdev_increment_features(xfrm_features,
1463 							  slave->dev->hw_enc_features,
1464 							  BOND_XFRM_FEATURES);
1465 #endif /* CONFIG_XFRM_OFFLOAD */
1466 
1467 		mpls_features = netdev_increment_features(mpls_features,
1468 							  slave->dev->mpls_features,
1469 							  BOND_MPLS_FEATURES);
1470 
1471 		dst_release_flag &= slave->dev->priv_flags;
1472 		if (slave->dev->hard_header_len > max_hard_header_len)
1473 			max_hard_header_len = slave->dev->hard_header_len;
1474 
1475 		tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1476 		tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1477 	}
1478 	bond_dev->hard_header_len = max_hard_header_len;
1479 
1480 done:
1481 	bond_dev->vlan_features = vlan_features;
1482 	bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1483 				    NETIF_F_HW_VLAN_CTAG_TX |
1484 				    NETIF_F_HW_VLAN_STAG_TX;
1485 #ifdef CONFIG_XFRM_OFFLOAD
1486 	bond_dev->hw_enc_features |= xfrm_features;
1487 #endif /* CONFIG_XFRM_OFFLOAD */
1488 	bond_dev->mpls_features = mpls_features;
1489 	netif_set_tso_max_segs(bond_dev, tso_max_segs);
1490 	netif_set_tso_max_size(bond_dev, tso_max_size);
1491 
1492 	bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1493 	if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1494 	    dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1495 		bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1496 
1497 	netdev_change_features(bond_dev);
1498 }
1499 
bond_setup_by_slave(struct net_device * bond_dev,struct net_device * slave_dev)1500 static void bond_setup_by_slave(struct net_device *bond_dev,
1501 				struct net_device *slave_dev)
1502 {
1503 	bool was_up = !!(bond_dev->flags & IFF_UP);
1504 
1505 	dev_close(bond_dev);
1506 
1507 	bond_dev->header_ops	    = slave_dev->header_ops;
1508 
1509 	bond_dev->type		    = slave_dev->type;
1510 	bond_dev->hard_header_len   = slave_dev->hard_header_len;
1511 	bond_dev->needed_headroom   = slave_dev->needed_headroom;
1512 	bond_dev->addr_len	    = slave_dev->addr_len;
1513 
1514 	memcpy(bond_dev->broadcast, slave_dev->broadcast,
1515 		slave_dev->addr_len);
1516 
1517 	if (slave_dev->flags & IFF_POINTOPOINT) {
1518 		bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1519 		bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1520 	}
1521 	if (was_up)
1522 		dev_open(bond_dev, NULL);
1523 }
1524 
1525 /* On bonding slaves other than the currently active slave, suppress
1526  * duplicates except for alb non-mcast/bcast.
1527  */
bond_should_deliver_exact_match(struct sk_buff * skb,struct slave * slave,struct bonding * bond)1528 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1529 					    struct slave *slave,
1530 					    struct bonding *bond)
1531 {
1532 	if (bond_is_slave_inactive(slave)) {
1533 		if (BOND_MODE(bond) == BOND_MODE_ALB &&
1534 		    skb->pkt_type != PACKET_BROADCAST &&
1535 		    skb->pkt_type != PACKET_MULTICAST)
1536 			return false;
1537 		return true;
1538 	}
1539 	return false;
1540 }
1541 
bond_handle_frame(struct sk_buff ** pskb)1542 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1543 {
1544 	struct sk_buff *skb = *pskb;
1545 	struct slave *slave;
1546 	struct bonding *bond;
1547 	int (*recv_probe)(const struct sk_buff *, struct bonding *,
1548 			  struct slave *);
1549 	int ret = RX_HANDLER_ANOTHER;
1550 
1551 	skb = skb_share_check(skb, GFP_ATOMIC);
1552 	if (unlikely(!skb))
1553 		return RX_HANDLER_CONSUMED;
1554 
1555 	*pskb = skb;
1556 
1557 	slave = bond_slave_get_rcu(skb->dev);
1558 	bond = slave->bond;
1559 
1560 	recv_probe = READ_ONCE(bond->recv_probe);
1561 	if (recv_probe) {
1562 		ret = recv_probe(skb, bond, slave);
1563 		if (ret == RX_HANDLER_CONSUMED) {
1564 			consume_skb(skb);
1565 			return ret;
1566 		}
1567 	}
1568 
1569 	/*
1570 	 * For packets determined by bond_should_deliver_exact_match() call to
1571 	 * be suppressed we want to make an exception for link-local packets.
1572 	 * This is necessary for e.g. LLDP daemons to be able to monitor
1573 	 * inactive slave links without being forced to bind to them
1574 	 * explicitly.
1575 	 *
1576 	 * At the same time, packets that are passed to the bonding master
1577 	 * (including link-local ones) can have their originating interface
1578 	 * determined via PACKET_ORIGDEV socket option.
1579 	 */
1580 	if (bond_should_deliver_exact_match(skb, slave, bond)) {
1581 		if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1582 			return RX_HANDLER_PASS;
1583 		return RX_HANDLER_EXACT;
1584 	}
1585 
1586 	skb->dev = bond->dev;
1587 
1588 	if (BOND_MODE(bond) == BOND_MODE_ALB &&
1589 	    netif_is_bridge_port(bond->dev) &&
1590 	    skb->pkt_type == PACKET_HOST) {
1591 
1592 		if (unlikely(skb_cow_head(skb,
1593 					  skb->data - skb_mac_header(skb)))) {
1594 			kfree_skb(skb);
1595 			return RX_HANDLER_CONSUMED;
1596 		}
1597 		bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1598 				  bond->dev->addr_len);
1599 	}
1600 
1601 	return ret;
1602 }
1603 
bond_lag_tx_type(struct bonding * bond)1604 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1605 {
1606 	switch (BOND_MODE(bond)) {
1607 	case BOND_MODE_ROUNDROBIN:
1608 		return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1609 	case BOND_MODE_ACTIVEBACKUP:
1610 		return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1611 	case BOND_MODE_BROADCAST:
1612 		return NETDEV_LAG_TX_TYPE_BROADCAST;
1613 	case BOND_MODE_XOR:
1614 	case BOND_MODE_8023AD:
1615 		return NETDEV_LAG_TX_TYPE_HASH;
1616 	default:
1617 		return NETDEV_LAG_TX_TYPE_UNKNOWN;
1618 	}
1619 }
1620 
bond_lag_hash_type(struct bonding * bond,enum netdev_lag_tx_type type)1621 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1622 					       enum netdev_lag_tx_type type)
1623 {
1624 	if (type != NETDEV_LAG_TX_TYPE_HASH)
1625 		return NETDEV_LAG_HASH_NONE;
1626 
1627 	switch (bond->params.xmit_policy) {
1628 	case BOND_XMIT_POLICY_LAYER2:
1629 		return NETDEV_LAG_HASH_L2;
1630 	case BOND_XMIT_POLICY_LAYER34:
1631 		return NETDEV_LAG_HASH_L34;
1632 	case BOND_XMIT_POLICY_LAYER23:
1633 		return NETDEV_LAG_HASH_L23;
1634 	case BOND_XMIT_POLICY_ENCAP23:
1635 		return NETDEV_LAG_HASH_E23;
1636 	case BOND_XMIT_POLICY_ENCAP34:
1637 		return NETDEV_LAG_HASH_E34;
1638 	case BOND_XMIT_POLICY_VLAN_SRCMAC:
1639 		return NETDEV_LAG_HASH_VLAN_SRCMAC;
1640 	default:
1641 		return NETDEV_LAG_HASH_UNKNOWN;
1642 	}
1643 }
1644 
bond_master_upper_dev_link(struct bonding * bond,struct slave * slave,struct netlink_ext_ack * extack)1645 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1646 				      struct netlink_ext_ack *extack)
1647 {
1648 	struct netdev_lag_upper_info lag_upper_info;
1649 	enum netdev_lag_tx_type type;
1650 	int err;
1651 
1652 	type = bond_lag_tx_type(bond);
1653 	lag_upper_info.tx_type = type;
1654 	lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1655 
1656 	err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1657 					   &lag_upper_info, extack);
1658 	if (err)
1659 		return err;
1660 
1661 	slave->dev->flags |= IFF_SLAVE;
1662 	return 0;
1663 }
1664 
bond_upper_dev_unlink(struct bonding * bond,struct slave * slave)1665 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1666 {
1667 	netdev_upper_dev_unlink(slave->dev, bond->dev);
1668 	slave->dev->flags &= ~IFF_SLAVE;
1669 }
1670 
slave_kobj_release(struct kobject * kobj)1671 static void slave_kobj_release(struct kobject *kobj)
1672 {
1673 	struct slave *slave = to_slave(kobj);
1674 	struct bonding *bond = bond_get_bond_by_slave(slave);
1675 
1676 	cancel_delayed_work_sync(&slave->notify_work);
1677 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
1678 		kfree(SLAVE_AD_INFO(slave));
1679 
1680 	kfree(slave);
1681 }
1682 
1683 static struct kobj_type slave_ktype = {
1684 	.release = slave_kobj_release,
1685 #ifdef CONFIG_SYSFS
1686 	.sysfs_ops = &slave_sysfs_ops,
1687 #endif
1688 };
1689 
bond_kobj_init(struct slave * slave)1690 static int bond_kobj_init(struct slave *slave)
1691 {
1692 	int err;
1693 
1694 	err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1695 				   &(slave->dev->dev.kobj), "bonding_slave");
1696 	if (err)
1697 		kobject_put(&slave->kobj);
1698 
1699 	return err;
1700 }
1701 
bond_alloc_slave(struct bonding * bond,struct net_device * slave_dev)1702 static struct slave *bond_alloc_slave(struct bonding *bond,
1703 				      struct net_device *slave_dev)
1704 {
1705 	struct slave *slave = NULL;
1706 
1707 	slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1708 	if (!slave)
1709 		return NULL;
1710 
1711 	slave->bond = bond;
1712 	slave->dev = slave_dev;
1713 	INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1714 
1715 	if (bond_kobj_init(slave))
1716 		return NULL;
1717 
1718 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1719 		SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1720 					       GFP_KERNEL);
1721 		if (!SLAVE_AD_INFO(slave)) {
1722 			kobject_put(&slave->kobj);
1723 			return NULL;
1724 		}
1725 	}
1726 
1727 	return slave;
1728 }
1729 
bond_fill_ifbond(struct bonding * bond,struct ifbond * info)1730 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1731 {
1732 	info->bond_mode = BOND_MODE(bond);
1733 	info->miimon = bond->params.miimon;
1734 	info->num_slaves = bond->slave_cnt;
1735 }
1736 
bond_fill_ifslave(struct slave * slave,struct ifslave * info)1737 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1738 {
1739 	strcpy(info->slave_name, slave->dev->name);
1740 	info->link = slave->link;
1741 	info->state = bond_slave_state(slave);
1742 	info->link_failure_count = slave->link_failure_count;
1743 }
1744 
bond_netdev_notify_work(struct work_struct * _work)1745 static void bond_netdev_notify_work(struct work_struct *_work)
1746 {
1747 	struct slave *slave = container_of(_work, struct slave,
1748 					   notify_work.work);
1749 
1750 	if (rtnl_trylock()) {
1751 		struct netdev_bonding_info binfo;
1752 
1753 		bond_fill_ifslave(slave, &binfo.slave);
1754 		bond_fill_ifbond(slave->bond, &binfo.master);
1755 		netdev_bonding_info_change(slave->dev, &binfo);
1756 		rtnl_unlock();
1757 	} else {
1758 		queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1759 	}
1760 }
1761 
bond_queue_slave_event(struct slave * slave)1762 void bond_queue_slave_event(struct slave *slave)
1763 {
1764 	queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1765 }
1766 
bond_lower_state_changed(struct slave * slave)1767 void bond_lower_state_changed(struct slave *slave)
1768 {
1769 	struct netdev_lag_lower_state_info info;
1770 
1771 	info.link_up = slave->link == BOND_LINK_UP ||
1772 		       slave->link == BOND_LINK_FAIL;
1773 	info.tx_enabled = bond_is_active_slave(slave);
1774 	netdev_lower_state_changed(slave->dev, &info);
1775 }
1776 
1777 #define BOND_NL_ERR(bond_dev, extack, errmsg) do {		\
1778 	if (extack)						\
1779 		NL_SET_ERR_MSG(extack, errmsg);			\
1780 	else							\
1781 		netdev_err(bond_dev, "Error: %s\n", errmsg);	\
1782 } while (0)
1783 
1784 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {		\
1785 	if (extack)							\
1786 		NL_SET_ERR_MSG(extack, errmsg);				\
1787 	else								\
1788 		slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);	\
1789 } while (0)
1790 
1791 /* The bonding driver uses ether_setup() to convert a master bond device
1792  * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1793  * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1794  * if they were set
1795  */
bond_ether_setup(struct net_device * bond_dev)1796 static void bond_ether_setup(struct net_device *bond_dev)
1797 {
1798 	unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1799 
1800 	ether_setup(bond_dev);
1801 	bond_dev->flags |= IFF_MASTER | flags;
1802 	bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1803 }
1804 
bond_xdp_set_features(struct net_device * bond_dev)1805 void bond_xdp_set_features(struct net_device *bond_dev)
1806 {
1807 	struct bonding *bond = netdev_priv(bond_dev);
1808 	xdp_features_t val = NETDEV_XDP_ACT_MASK;
1809 	struct list_head *iter;
1810 	struct slave *slave;
1811 
1812 	ASSERT_RTNL();
1813 
1814 	if (!bond_xdp_check(bond)) {
1815 		xdp_clear_features_flag(bond_dev);
1816 		return;
1817 	}
1818 
1819 	bond_for_each_slave(bond, slave, iter)
1820 		val &= slave->dev->xdp_features;
1821 
1822 	val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1823 
1824 	xdp_set_features_flag(bond_dev, val);
1825 }
1826 
1827 /* enslave device <slave> to bond device <master> */
bond_enslave(struct net_device * bond_dev,struct net_device * slave_dev,struct netlink_ext_ack * extack)1828 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1829 		 struct netlink_ext_ack *extack)
1830 {
1831 	struct bonding *bond = netdev_priv(bond_dev);
1832 	const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1833 	struct slave *new_slave = NULL, *prev_slave;
1834 	struct sockaddr_storage ss;
1835 	int link_reporting;
1836 	int res = 0, i;
1837 
1838 	if (slave_dev->flags & IFF_MASTER &&
1839 	    !netif_is_bond_master(slave_dev)) {
1840 		BOND_NL_ERR(bond_dev, extack,
1841 			    "Device type (master device) cannot be enslaved");
1842 		return -EPERM;
1843 	}
1844 
1845 	if (!bond->params.use_carrier &&
1846 	    slave_dev->ethtool_ops->get_link == NULL &&
1847 	    slave_ops->ndo_eth_ioctl == NULL) {
1848 		slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1849 	}
1850 
1851 	/* already in-use? */
1852 	if (netdev_is_rx_handler_busy(slave_dev)) {
1853 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1854 			     "Device is in use and cannot be enslaved");
1855 		return -EBUSY;
1856 	}
1857 
1858 	if (bond_dev == slave_dev) {
1859 		BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1860 		return -EPERM;
1861 	}
1862 
1863 	/* vlan challenged mutual exclusion */
1864 	/* no need to lock since we're protected by rtnl_lock */
1865 	if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1866 		slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1867 		if (vlan_uses_dev(bond_dev)) {
1868 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1869 				     "Can not enslave VLAN challenged device to VLAN enabled bond");
1870 			return -EPERM;
1871 		} else {
1872 			slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1873 		}
1874 	} else {
1875 		slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1876 	}
1877 
1878 	if (slave_dev->features & NETIF_F_HW_ESP)
1879 		slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1880 
1881 	/* Old ifenslave binaries are no longer supported.  These can
1882 	 * be identified with moderate accuracy by the state of the slave:
1883 	 * the current ifenslave will set the interface down prior to
1884 	 * enslaving it; the old ifenslave will not.
1885 	 */
1886 	if (slave_dev->flags & IFF_UP) {
1887 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1888 			     "Device can not be enslaved while up");
1889 		return -EPERM;
1890 	}
1891 
1892 	/* set bonding device ether type by slave - bonding netdevices are
1893 	 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1894 	 * there is a need to override some of the type dependent attribs/funcs.
1895 	 *
1896 	 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1897 	 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1898 	 */
1899 	if (!bond_has_slaves(bond)) {
1900 		if (bond_dev->type != slave_dev->type) {
1901 			slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1902 				  bond_dev->type, slave_dev->type);
1903 
1904 			res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1905 						       bond_dev);
1906 			res = notifier_to_errno(res);
1907 			if (res) {
1908 				slave_err(bond_dev, slave_dev, "refused to change device type\n");
1909 				return -EBUSY;
1910 			}
1911 
1912 			/* Flush unicast and multicast addresses */
1913 			dev_uc_flush(bond_dev);
1914 			dev_mc_flush(bond_dev);
1915 
1916 			if (slave_dev->type != ARPHRD_ETHER)
1917 				bond_setup_by_slave(bond_dev, slave_dev);
1918 			else
1919 				bond_ether_setup(bond_dev);
1920 
1921 			call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1922 						 bond_dev);
1923 		}
1924 	} else if (bond_dev->type != slave_dev->type) {
1925 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1926 			     "Device type is different from other slaves");
1927 		return -EINVAL;
1928 	}
1929 
1930 	if (slave_dev->type == ARPHRD_INFINIBAND &&
1931 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1932 		SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1933 			     "Only active-backup mode is supported for infiniband slaves");
1934 		res = -EOPNOTSUPP;
1935 		goto err_undo_flags;
1936 	}
1937 
1938 	if (!slave_ops->ndo_set_mac_address ||
1939 	    slave_dev->type == ARPHRD_INFINIBAND) {
1940 		slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1941 		if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1942 		    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1943 			if (!bond_has_slaves(bond)) {
1944 				bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1945 				slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1946 			} else {
1947 				SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1948 					     "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1949 				res = -EOPNOTSUPP;
1950 				goto err_undo_flags;
1951 			}
1952 		}
1953 	}
1954 
1955 	call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1956 
1957 	/* If this is the first slave, then we need to set the master's hardware
1958 	 * address to be the same as the slave's.
1959 	 */
1960 	if (!bond_has_slaves(bond) &&
1961 	    bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1962 		res = bond_set_dev_addr(bond->dev, slave_dev);
1963 		if (res)
1964 			goto err_undo_flags;
1965 	}
1966 
1967 	new_slave = bond_alloc_slave(bond, slave_dev);
1968 	if (!new_slave) {
1969 		res = -ENOMEM;
1970 		goto err_undo_flags;
1971 	}
1972 
1973 	/* Set the new_slave's queue_id to be zero.  Queue ID mapping
1974 	 * is set via sysfs or module option if desired.
1975 	 */
1976 	new_slave->queue_id = 0;
1977 
1978 	/* Save slave's original mtu and then set it to match the bond */
1979 	new_slave->original_mtu = slave_dev->mtu;
1980 	res = dev_set_mtu(slave_dev, bond->dev->mtu);
1981 	if (res) {
1982 		slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1983 		goto err_free;
1984 	}
1985 
1986 	/* Save slave's original ("permanent") mac address for modes
1987 	 * that need it, and for restoring it upon release, and then
1988 	 * set it to the master's address
1989 	 */
1990 	bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1991 			  slave_dev->addr_len);
1992 
1993 	if (!bond->params.fail_over_mac ||
1994 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1995 		/* Set slave to master's mac address.  The application already
1996 		 * set the master's mac address to that of the first slave
1997 		 */
1998 		memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1999 		ss.ss_family = slave_dev->type;
2000 		res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
2001 					  extack);
2002 		if (res) {
2003 			slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
2004 			goto err_restore_mtu;
2005 		}
2006 	}
2007 
2008 	/* set no_addrconf flag before open to prevent IPv6 addrconf */
2009 	slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2010 
2011 	/* open the slave since the application closed it */
2012 	res = dev_open(slave_dev, extack);
2013 	if (res) {
2014 		slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2015 		goto err_restore_mac;
2016 	}
2017 
2018 	slave_dev->priv_flags |= IFF_BONDING;
2019 	/* initialize slave stats */
2020 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2021 
2022 	if (bond_is_lb(bond)) {
2023 		/* bond_alb_init_slave() must be called before all other stages since
2024 		 * it might fail and we do not want to have to undo everything
2025 		 */
2026 		res = bond_alb_init_slave(bond, new_slave);
2027 		if (res)
2028 			goto err_close;
2029 	}
2030 
2031 	res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2032 	if (res) {
2033 		slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2034 		goto err_close;
2035 	}
2036 
2037 	prev_slave = bond_last_slave(bond);
2038 
2039 	new_slave->delay = 0;
2040 	new_slave->link_failure_count = 0;
2041 
2042 	if (bond_update_speed_duplex(new_slave) &&
2043 	    bond_needs_speed_duplex(bond))
2044 		new_slave->link = BOND_LINK_DOWN;
2045 
2046 	new_slave->last_rx = jiffies -
2047 		(msecs_to_jiffies(bond->params.arp_interval) + 1);
2048 	for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2049 		new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2050 
2051 	new_slave->last_tx = new_slave->last_rx;
2052 
2053 	if (bond->params.miimon && !bond->params.use_carrier) {
2054 		link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2055 
2056 		if ((link_reporting == -1) && !bond->params.arp_interval) {
2057 			/* miimon is set but a bonded network driver
2058 			 * does not support ETHTOOL/MII and
2059 			 * arp_interval is not set.  Note: if
2060 			 * use_carrier is enabled, we will never go
2061 			 * here (because netif_carrier is always
2062 			 * supported); thus, we don't need to change
2063 			 * the messages for netif_carrier.
2064 			 */
2065 			slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
2066 		} else if (link_reporting == -1) {
2067 			/* unable get link status using mii/ethtool */
2068 			slave_warn(bond_dev, slave_dev, "can't get link status from slave; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n");
2069 		}
2070 	}
2071 
2072 	/* check for initial state */
2073 	new_slave->link = BOND_LINK_NOCHANGE;
2074 	if (bond->params.miimon) {
2075 		if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2076 			if (bond->params.updelay) {
2077 				bond_set_slave_link_state(new_slave,
2078 							  BOND_LINK_BACK,
2079 							  BOND_SLAVE_NOTIFY_NOW);
2080 				new_slave->delay = bond->params.updelay;
2081 			} else {
2082 				bond_set_slave_link_state(new_slave,
2083 							  BOND_LINK_UP,
2084 							  BOND_SLAVE_NOTIFY_NOW);
2085 			}
2086 		} else {
2087 			bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2088 						  BOND_SLAVE_NOTIFY_NOW);
2089 		}
2090 	} else if (bond->params.arp_interval) {
2091 		bond_set_slave_link_state(new_slave,
2092 					  (netif_carrier_ok(slave_dev) ?
2093 					  BOND_LINK_UP : BOND_LINK_DOWN),
2094 					  BOND_SLAVE_NOTIFY_NOW);
2095 	} else {
2096 		bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2097 					  BOND_SLAVE_NOTIFY_NOW);
2098 	}
2099 
2100 	if (new_slave->link != BOND_LINK_DOWN)
2101 		new_slave->last_link_up = jiffies;
2102 	slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2103 		  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2104 		  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2105 
2106 	if (bond_uses_primary(bond) && bond->params.primary[0]) {
2107 		/* if there is a primary slave, remember it */
2108 		if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2109 			rcu_assign_pointer(bond->primary_slave, new_slave);
2110 			bond->force_primary = true;
2111 		}
2112 	}
2113 
2114 	switch (BOND_MODE(bond)) {
2115 	case BOND_MODE_ACTIVEBACKUP:
2116 		bond_set_slave_inactive_flags(new_slave,
2117 					      BOND_SLAVE_NOTIFY_NOW);
2118 		break;
2119 	case BOND_MODE_8023AD:
2120 		/* in 802.3ad mode, the internal mechanism
2121 		 * will activate the slaves in the selected
2122 		 * aggregator
2123 		 */
2124 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2125 		/* if this is the first slave */
2126 		if (!prev_slave) {
2127 			SLAVE_AD_INFO(new_slave)->id = 1;
2128 			/* Initialize AD with the number of times that the AD timer is called in 1 second
2129 			 * can be called only after the mac address of the bond is set
2130 			 */
2131 			bond_3ad_initialize(bond);
2132 		} else {
2133 			SLAVE_AD_INFO(new_slave)->id =
2134 				SLAVE_AD_INFO(prev_slave)->id + 1;
2135 		}
2136 
2137 		bond_3ad_bind_slave(new_slave);
2138 		break;
2139 	case BOND_MODE_TLB:
2140 	case BOND_MODE_ALB:
2141 		bond_set_active_slave(new_slave);
2142 		bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2143 		break;
2144 	default:
2145 		slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2146 
2147 		/* always active in trunk mode */
2148 		bond_set_active_slave(new_slave);
2149 
2150 		/* In trunking mode there is little meaning to curr_active_slave
2151 		 * anyway (it holds no special properties of the bond device),
2152 		 * so we can change it without calling change_active_interface()
2153 		 */
2154 		if (!rcu_access_pointer(bond->curr_active_slave) &&
2155 		    new_slave->link == BOND_LINK_UP)
2156 			rcu_assign_pointer(bond->curr_active_slave, new_slave);
2157 
2158 		break;
2159 	} /* switch(bond_mode) */
2160 
2161 #ifdef CONFIG_NET_POLL_CONTROLLER
2162 	if (bond->dev->npinfo) {
2163 		if (slave_enable_netpoll(new_slave)) {
2164 			slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2165 			res = -EBUSY;
2166 			goto err_detach;
2167 		}
2168 	}
2169 #endif
2170 
2171 	if (!(bond_dev->features & NETIF_F_LRO))
2172 		dev_disable_lro(slave_dev);
2173 
2174 	res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2175 					 new_slave);
2176 	if (res) {
2177 		slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2178 		goto err_detach;
2179 	}
2180 
2181 	res = bond_master_upper_dev_link(bond, new_slave, extack);
2182 	if (res) {
2183 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2184 		goto err_unregister;
2185 	}
2186 
2187 	bond_lower_state_changed(new_slave);
2188 
2189 	res = bond_sysfs_slave_add(new_slave);
2190 	if (res) {
2191 		slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2192 		goto err_upper_unlink;
2193 	}
2194 
2195 	/* If the mode uses primary, then the following is handled by
2196 	 * bond_change_active_slave().
2197 	 */
2198 	if (!bond_uses_primary(bond)) {
2199 		/* set promiscuity level to new slave */
2200 		if (bond_dev->flags & IFF_PROMISC) {
2201 			res = dev_set_promiscuity(slave_dev, 1);
2202 			if (res)
2203 				goto err_sysfs_del;
2204 		}
2205 
2206 		/* set allmulti level to new slave */
2207 		if (bond_dev->flags & IFF_ALLMULTI) {
2208 			res = dev_set_allmulti(slave_dev, 1);
2209 			if (res) {
2210 				if (bond_dev->flags & IFF_PROMISC)
2211 					dev_set_promiscuity(slave_dev, -1);
2212 				goto err_sysfs_del;
2213 			}
2214 		}
2215 
2216 		if (bond_dev->flags & IFF_UP) {
2217 			netif_addr_lock_bh(bond_dev);
2218 			dev_mc_sync_multiple(slave_dev, bond_dev);
2219 			dev_uc_sync_multiple(slave_dev, bond_dev);
2220 			netif_addr_unlock_bh(bond_dev);
2221 
2222 			if (BOND_MODE(bond) == BOND_MODE_8023AD)
2223 				dev_mc_add(slave_dev, lacpdu_mcast_addr);
2224 		}
2225 	}
2226 
2227 	bond->slave_cnt++;
2228 	bond_compute_features(bond);
2229 	bond_set_carrier(bond);
2230 
2231 	if (bond_uses_primary(bond)) {
2232 		block_netpoll_tx();
2233 		bond_select_active_slave(bond);
2234 		unblock_netpoll_tx();
2235 	}
2236 
2237 	if (bond_mode_can_use_xmit_hash(bond))
2238 		bond_update_slave_arr(bond, NULL);
2239 
2240 
2241 	if (!slave_dev->netdev_ops->ndo_bpf ||
2242 	    !slave_dev->netdev_ops->ndo_xdp_xmit) {
2243 		if (bond->xdp_prog) {
2244 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2245 				     "Slave does not support XDP");
2246 			res = -EOPNOTSUPP;
2247 			goto err_sysfs_del;
2248 		}
2249 	} else if (bond->xdp_prog) {
2250 		struct netdev_bpf xdp = {
2251 			.command = XDP_SETUP_PROG,
2252 			.flags   = 0,
2253 			.prog    = bond->xdp_prog,
2254 			.extack  = extack,
2255 		};
2256 
2257 		if (dev_xdp_prog_count(slave_dev) > 0) {
2258 			SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2259 				     "Slave has XDP program loaded, please unload before enslaving");
2260 			res = -EOPNOTSUPP;
2261 			goto err_sysfs_del;
2262 		}
2263 
2264 		res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2265 		if (res < 0) {
2266 			/* ndo_bpf() sets extack error message */
2267 			slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2268 			goto err_sysfs_del;
2269 		}
2270 		if (bond->xdp_prog)
2271 			bpf_prog_inc(bond->xdp_prog);
2272 	}
2273 
2274 	bond_xdp_set_features(bond_dev);
2275 
2276 	slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2277 		   bond_is_active_slave(new_slave) ? "an active" : "a backup",
2278 		   new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2279 
2280 	/* enslave is successful */
2281 	bond_queue_slave_event(new_slave);
2282 	return 0;
2283 
2284 /* Undo stages on error */
2285 err_sysfs_del:
2286 	bond_sysfs_slave_del(new_slave);
2287 
2288 err_upper_unlink:
2289 	bond_upper_dev_unlink(bond, new_slave);
2290 
2291 err_unregister:
2292 	netdev_rx_handler_unregister(slave_dev);
2293 
2294 err_detach:
2295 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2296 	if (rcu_access_pointer(bond->primary_slave) == new_slave)
2297 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2298 	if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2299 		block_netpoll_tx();
2300 		bond_change_active_slave(bond, NULL);
2301 		bond_select_active_slave(bond);
2302 		unblock_netpoll_tx();
2303 	}
2304 	/* either primary_slave or curr_active_slave might've changed */
2305 	synchronize_rcu();
2306 	slave_disable_netpoll(new_slave);
2307 
2308 err_close:
2309 	if (!netif_is_bond_master(slave_dev))
2310 		slave_dev->priv_flags &= ~IFF_BONDING;
2311 	dev_close(slave_dev);
2312 
2313 err_restore_mac:
2314 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2315 	if (!bond->params.fail_over_mac ||
2316 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2317 		/* XXX TODO - fom follow mode needs to change master's
2318 		 * MAC if this slave's MAC is in use by the bond, or at
2319 		 * least print a warning.
2320 		 */
2321 		bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2322 				  new_slave->dev->addr_len);
2323 		ss.ss_family = slave_dev->type;
2324 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2325 	}
2326 
2327 err_restore_mtu:
2328 	dev_set_mtu(slave_dev, new_slave->original_mtu);
2329 
2330 err_free:
2331 	kobject_put(&new_slave->kobj);
2332 
2333 err_undo_flags:
2334 	/* Enslave of first slave has failed and we need to fix master's mac */
2335 	if (!bond_has_slaves(bond)) {
2336 		if (ether_addr_equal_64bits(bond_dev->dev_addr,
2337 					    slave_dev->dev_addr))
2338 			eth_hw_addr_random(bond_dev);
2339 		if (bond_dev->type != ARPHRD_ETHER) {
2340 			dev_close(bond_dev);
2341 			bond_ether_setup(bond_dev);
2342 		}
2343 	}
2344 
2345 	return res;
2346 }
2347 
2348 /* Try to release the slave device <slave> from the bond device <master>
2349  * It is legal to access curr_active_slave without a lock because all the function
2350  * is RTNL-locked. If "all" is true it means that the function is being called
2351  * while destroying a bond interface and all slaves are being released.
2352  *
2353  * The rules for slave state should be:
2354  *   for Active/Backup:
2355  *     Active stays on all backups go down
2356  *   for Bonded connections:
2357  *     The first up interface should be left on and all others downed.
2358  */
__bond_release_one(struct net_device * bond_dev,struct net_device * slave_dev,bool all,bool unregister)2359 static int __bond_release_one(struct net_device *bond_dev,
2360 			      struct net_device *slave_dev,
2361 			      bool all, bool unregister)
2362 {
2363 	struct bonding *bond = netdev_priv(bond_dev);
2364 	struct slave *slave, *oldcurrent;
2365 	struct sockaddr_storage ss;
2366 	int old_flags = bond_dev->flags;
2367 	netdev_features_t old_features = bond_dev->features;
2368 
2369 	/* slave is not a slave or master is not master of this slave */
2370 	if (!(slave_dev->flags & IFF_SLAVE) ||
2371 	    !netdev_has_upper_dev(slave_dev, bond_dev)) {
2372 		slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2373 		return -EINVAL;
2374 	}
2375 
2376 	block_netpoll_tx();
2377 
2378 	slave = bond_get_slave_by_dev(bond, slave_dev);
2379 	if (!slave) {
2380 		/* not a slave of this bond */
2381 		slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2382 		unblock_netpoll_tx();
2383 		return -EINVAL;
2384 	}
2385 
2386 	bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2387 
2388 	bond_sysfs_slave_del(slave);
2389 
2390 	/* recompute stats just before removing the slave */
2391 	bond_get_stats(bond->dev, &bond->bond_stats);
2392 
2393 	if (bond->xdp_prog) {
2394 		struct netdev_bpf xdp = {
2395 			.command = XDP_SETUP_PROG,
2396 			.flags   = 0,
2397 			.prog	 = NULL,
2398 			.extack  = NULL,
2399 		};
2400 		if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2401 			slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2402 	}
2403 
2404 	/* unregister rx_handler early so bond_handle_frame wouldn't be called
2405 	 * for this slave anymore.
2406 	 */
2407 	netdev_rx_handler_unregister(slave_dev);
2408 
2409 	if (BOND_MODE(bond) == BOND_MODE_8023AD)
2410 		bond_3ad_unbind_slave(slave);
2411 
2412 	bond_upper_dev_unlink(bond, slave);
2413 
2414 	if (bond_mode_can_use_xmit_hash(bond))
2415 		bond_update_slave_arr(bond, slave);
2416 
2417 	slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2418 		    bond_is_active_slave(slave) ? "active" : "backup");
2419 
2420 	oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2421 
2422 	RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2423 
2424 	if (!all && (!bond->params.fail_over_mac ||
2425 		     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2426 		if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2427 		    bond_has_slaves(bond))
2428 			slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
2429 				   slave->perm_hwaddr);
2430 	}
2431 
2432 	if (rtnl_dereference(bond->primary_slave) == slave)
2433 		RCU_INIT_POINTER(bond->primary_slave, NULL);
2434 
2435 	if (oldcurrent == slave)
2436 		bond_change_active_slave(bond, NULL);
2437 
2438 	if (bond_is_lb(bond)) {
2439 		/* Must be called only after the slave has been
2440 		 * detached from the list and the curr_active_slave
2441 		 * has been cleared (if our_slave == old_current),
2442 		 * but before a new active slave is selected.
2443 		 */
2444 		bond_alb_deinit_slave(bond, slave);
2445 	}
2446 
2447 	if (all) {
2448 		RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2449 	} else if (oldcurrent == slave) {
2450 		/* Note that we hold RTNL over this sequence, so there
2451 		 * is no concern that another slave add/remove event
2452 		 * will interfere.
2453 		 */
2454 		bond_select_active_slave(bond);
2455 	}
2456 
2457 	bond_set_carrier(bond);
2458 	if (!bond_has_slaves(bond))
2459 		eth_hw_addr_random(bond_dev);
2460 
2461 	unblock_netpoll_tx();
2462 	synchronize_rcu();
2463 	bond->slave_cnt--;
2464 
2465 	if (!bond_has_slaves(bond)) {
2466 		call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2467 		call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2468 	}
2469 
2470 	bond_compute_features(bond);
2471 	if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2472 	    (old_features & NETIF_F_VLAN_CHALLENGED))
2473 		slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2474 
2475 	vlan_vids_del_by_dev(slave_dev, bond_dev);
2476 
2477 	/* If the mode uses primary, then this case was handled above by
2478 	 * bond_change_active_slave(..., NULL)
2479 	 */
2480 	if (!bond_uses_primary(bond)) {
2481 		/* unset promiscuity level from slave
2482 		 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2483 		 * of the IFF_PROMISC flag in the bond_dev, but we need the
2484 		 * value of that flag before that change, as that was the value
2485 		 * when this slave was attached, so we cache at the start of the
2486 		 * function and use it here. Same goes for ALLMULTI below
2487 		 */
2488 		if (old_flags & IFF_PROMISC)
2489 			dev_set_promiscuity(slave_dev, -1);
2490 
2491 		/* unset allmulti level from slave */
2492 		if (old_flags & IFF_ALLMULTI)
2493 			dev_set_allmulti(slave_dev, -1);
2494 
2495 		if (old_flags & IFF_UP)
2496 			bond_hw_addr_flush(bond_dev, slave_dev);
2497 	}
2498 
2499 	slave_disable_netpoll(slave);
2500 
2501 	/* close slave before restoring its mac address */
2502 	dev_close(slave_dev);
2503 
2504 	slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2505 
2506 	if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2507 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2508 		/* restore original ("permanent") mac address */
2509 		bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2510 				  slave->dev->addr_len);
2511 		ss.ss_family = slave_dev->type;
2512 		dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2513 	}
2514 
2515 	if (unregister)
2516 		__dev_set_mtu(slave_dev, slave->original_mtu);
2517 	else
2518 		dev_set_mtu(slave_dev, slave->original_mtu);
2519 
2520 	if (!netif_is_bond_master(slave_dev))
2521 		slave_dev->priv_flags &= ~IFF_BONDING;
2522 
2523 	bond_xdp_set_features(bond_dev);
2524 	kobject_put(&slave->kobj);
2525 
2526 	return 0;
2527 }
2528 
2529 /* A wrapper used because of ndo_del_link */
bond_release(struct net_device * bond_dev,struct net_device * slave_dev)2530 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2531 {
2532 	return __bond_release_one(bond_dev, slave_dev, false, false);
2533 }
2534 
2535 /* First release a slave and then destroy the bond if no more slaves are left.
2536  * Must be under rtnl_lock when this function is called.
2537  */
bond_release_and_destroy(struct net_device * bond_dev,struct net_device * slave_dev)2538 static int bond_release_and_destroy(struct net_device *bond_dev,
2539 				    struct net_device *slave_dev)
2540 {
2541 	struct bonding *bond = netdev_priv(bond_dev);
2542 	int ret;
2543 
2544 	ret = __bond_release_one(bond_dev, slave_dev, false, true);
2545 	if (ret == 0 && !bond_has_slaves(bond) &&
2546 	    bond_dev->reg_state != NETREG_UNREGISTERING) {
2547 		bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2548 		netdev_info(bond_dev, "Destroying bond\n");
2549 		bond_remove_proc_entry(bond);
2550 		unregister_netdevice(bond_dev);
2551 	}
2552 	return ret;
2553 }
2554 
bond_info_query(struct net_device * bond_dev,struct ifbond * info)2555 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2556 {
2557 	struct bonding *bond = netdev_priv(bond_dev);
2558 
2559 	bond_fill_ifbond(bond, info);
2560 }
2561 
bond_slave_info_query(struct net_device * bond_dev,struct ifslave * info)2562 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2563 {
2564 	struct bonding *bond = netdev_priv(bond_dev);
2565 	struct list_head *iter;
2566 	int i = 0, res = -ENODEV;
2567 	struct slave *slave;
2568 
2569 	bond_for_each_slave(bond, slave, iter) {
2570 		if (i++ == (int)info->slave_id) {
2571 			res = 0;
2572 			bond_fill_ifslave(slave, info);
2573 			break;
2574 		}
2575 	}
2576 
2577 	return res;
2578 }
2579 
2580 /*-------------------------------- Monitoring -------------------------------*/
2581 
2582 /* called with rcu_read_lock() */
bond_miimon_inspect(struct bonding * bond)2583 static int bond_miimon_inspect(struct bonding *bond)
2584 {
2585 	bool ignore_updelay = false;
2586 	int link_state, commit = 0;
2587 	struct list_head *iter;
2588 	struct slave *slave;
2589 
2590 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2591 		ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2592 	} else {
2593 		struct bond_up_slave *usable_slaves;
2594 
2595 		usable_slaves = rcu_dereference(bond->usable_slaves);
2596 
2597 		if (usable_slaves && usable_slaves->count == 0)
2598 			ignore_updelay = true;
2599 	}
2600 
2601 	bond_for_each_slave_rcu(bond, slave, iter) {
2602 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2603 
2604 		link_state = bond_check_dev_link(bond, slave->dev, 0);
2605 
2606 		switch (slave->link) {
2607 		case BOND_LINK_UP:
2608 			if (link_state)
2609 				continue;
2610 
2611 			bond_propose_link_state(slave, BOND_LINK_FAIL);
2612 			commit++;
2613 			slave->delay = bond->params.downdelay;
2614 			if (slave->delay) {
2615 				slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2616 					   (BOND_MODE(bond) ==
2617 					    BOND_MODE_ACTIVEBACKUP) ?
2618 					    (bond_is_active_slave(slave) ?
2619 					     "active " : "backup ") : "",
2620 					   bond->params.downdelay * bond->params.miimon);
2621 			}
2622 			fallthrough;
2623 		case BOND_LINK_FAIL:
2624 			if (link_state) {
2625 				/* recovered before downdelay expired */
2626 				bond_propose_link_state(slave, BOND_LINK_UP);
2627 				slave->last_link_up = jiffies;
2628 				slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2629 					   (bond->params.downdelay - slave->delay) *
2630 					   bond->params.miimon);
2631 				commit++;
2632 				continue;
2633 			}
2634 
2635 			if (slave->delay <= 0) {
2636 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2637 				commit++;
2638 				continue;
2639 			}
2640 
2641 			slave->delay--;
2642 			break;
2643 
2644 		case BOND_LINK_DOWN:
2645 			if (!link_state)
2646 				continue;
2647 
2648 			bond_propose_link_state(slave, BOND_LINK_BACK);
2649 			commit++;
2650 			slave->delay = bond->params.updelay;
2651 
2652 			if (slave->delay) {
2653 				slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2654 					   ignore_updelay ? 0 :
2655 					   bond->params.updelay *
2656 					   bond->params.miimon);
2657 			}
2658 			fallthrough;
2659 		case BOND_LINK_BACK:
2660 			if (!link_state) {
2661 				bond_propose_link_state(slave, BOND_LINK_DOWN);
2662 				slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2663 					   (bond->params.updelay - slave->delay) *
2664 					   bond->params.miimon);
2665 				commit++;
2666 				continue;
2667 			}
2668 
2669 			if (ignore_updelay)
2670 				slave->delay = 0;
2671 
2672 			if (slave->delay <= 0) {
2673 				bond_propose_link_state(slave, BOND_LINK_UP);
2674 				commit++;
2675 				ignore_updelay = false;
2676 				continue;
2677 			}
2678 
2679 			slave->delay--;
2680 			break;
2681 		}
2682 	}
2683 
2684 	return commit;
2685 }
2686 
bond_miimon_link_change(struct bonding * bond,struct slave * slave,char link)2687 static void bond_miimon_link_change(struct bonding *bond,
2688 				    struct slave *slave,
2689 				    char link)
2690 {
2691 	switch (BOND_MODE(bond)) {
2692 	case BOND_MODE_8023AD:
2693 		bond_3ad_handle_link_change(slave, link);
2694 		break;
2695 	case BOND_MODE_TLB:
2696 	case BOND_MODE_ALB:
2697 		bond_alb_handle_link_change(bond, slave, link);
2698 		break;
2699 	case BOND_MODE_XOR:
2700 		bond_update_slave_arr(bond, NULL);
2701 		break;
2702 	}
2703 }
2704 
bond_miimon_commit(struct bonding * bond)2705 static void bond_miimon_commit(struct bonding *bond)
2706 {
2707 	struct slave *slave, *primary, *active;
2708 	bool do_failover = false;
2709 	struct list_head *iter;
2710 
2711 	ASSERT_RTNL();
2712 
2713 	bond_for_each_slave(bond, slave, iter) {
2714 		switch (slave->link_new_state) {
2715 		case BOND_LINK_NOCHANGE:
2716 			/* For 802.3ad mode, check current slave speed and
2717 			 * duplex again in case its port was disabled after
2718 			 * invalid speed/duplex reporting but recovered before
2719 			 * link monitoring could make a decision on the actual
2720 			 * link status
2721 			 */
2722 			if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2723 			    slave->link == BOND_LINK_UP)
2724 				bond_3ad_adapter_speed_duplex_changed(slave);
2725 			continue;
2726 
2727 		case BOND_LINK_UP:
2728 			if (bond_update_speed_duplex(slave) &&
2729 			    bond_needs_speed_duplex(bond)) {
2730 				slave->link = BOND_LINK_DOWN;
2731 				if (net_ratelimit())
2732 					slave_warn(bond->dev, slave->dev,
2733 						   "failed to get link speed/duplex\n");
2734 				continue;
2735 			}
2736 			bond_set_slave_link_state(slave, BOND_LINK_UP,
2737 						  BOND_SLAVE_NOTIFY_NOW);
2738 			slave->last_link_up = jiffies;
2739 
2740 			primary = rtnl_dereference(bond->primary_slave);
2741 			if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2742 				/* prevent it from being the active one */
2743 				bond_set_backup_slave(slave);
2744 			} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2745 				/* make it immediately active */
2746 				bond_set_active_slave(slave);
2747 			}
2748 
2749 			slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2750 				   slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2751 				   slave->duplex ? "full" : "half");
2752 
2753 			bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2754 
2755 			active = rtnl_dereference(bond->curr_active_slave);
2756 			if (!active || slave == primary || slave->prio > active->prio)
2757 				do_failover = true;
2758 
2759 			continue;
2760 
2761 		case BOND_LINK_DOWN:
2762 			if (slave->link_failure_count < UINT_MAX)
2763 				slave->link_failure_count++;
2764 
2765 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2766 						  BOND_SLAVE_NOTIFY_NOW);
2767 
2768 			if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2769 			    BOND_MODE(bond) == BOND_MODE_8023AD)
2770 				bond_set_slave_inactive_flags(slave,
2771 							      BOND_SLAVE_NOTIFY_NOW);
2772 
2773 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2774 
2775 			bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2776 
2777 			if (slave == rcu_access_pointer(bond->curr_active_slave))
2778 				do_failover = true;
2779 
2780 			continue;
2781 
2782 		default:
2783 			slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2784 				  slave->link_new_state);
2785 			bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2786 
2787 			continue;
2788 		}
2789 	}
2790 
2791 	if (do_failover) {
2792 		block_netpoll_tx();
2793 		bond_select_active_slave(bond);
2794 		unblock_netpoll_tx();
2795 	}
2796 
2797 	bond_set_carrier(bond);
2798 }
2799 
2800 /* bond_mii_monitor
2801  *
2802  * Really a wrapper that splits the mii monitor into two phases: an
2803  * inspection, then (if inspection indicates something needs to be done)
2804  * an acquisition of appropriate locks followed by a commit phase to
2805  * implement whatever link state changes are indicated.
2806  */
bond_mii_monitor(struct work_struct * work)2807 static void bond_mii_monitor(struct work_struct *work)
2808 {
2809 	struct bonding *bond = container_of(work, struct bonding,
2810 					    mii_work.work);
2811 	bool should_notify_peers = false;
2812 	bool commit;
2813 	unsigned long delay;
2814 	struct slave *slave;
2815 	struct list_head *iter;
2816 
2817 	delay = msecs_to_jiffies(bond->params.miimon);
2818 
2819 	if (!bond_has_slaves(bond))
2820 		goto re_arm;
2821 
2822 	rcu_read_lock();
2823 	should_notify_peers = bond_should_notify_peers(bond);
2824 	commit = !!bond_miimon_inspect(bond);
2825 	if (bond->send_peer_notif) {
2826 		rcu_read_unlock();
2827 		if (rtnl_trylock()) {
2828 			bond->send_peer_notif--;
2829 			rtnl_unlock();
2830 		}
2831 	} else {
2832 		rcu_read_unlock();
2833 	}
2834 
2835 	if (commit) {
2836 		/* Race avoidance with bond_close cancel of workqueue */
2837 		if (!rtnl_trylock()) {
2838 			delay = 1;
2839 			should_notify_peers = false;
2840 			goto re_arm;
2841 		}
2842 
2843 		bond_for_each_slave(bond, slave, iter) {
2844 			bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2845 		}
2846 		bond_miimon_commit(bond);
2847 
2848 		rtnl_unlock();	/* might sleep, hold no other locks */
2849 	}
2850 
2851 re_arm:
2852 	if (bond->params.miimon)
2853 		queue_delayed_work(bond->wq, &bond->mii_work, delay);
2854 
2855 	if (should_notify_peers) {
2856 		if (!rtnl_trylock())
2857 			return;
2858 		call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2859 		rtnl_unlock();
2860 	}
2861 }
2862 
bond_upper_dev_walk(struct net_device * upper,struct netdev_nested_priv * priv)2863 static int bond_upper_dev_walk(struct net_device *upper,
2864 			       struct netdev_nested_priv *priv)
2865 {
2866 	__be32 ip = *(__be32 *)priv->data;
2867 
2868 	return ip == bond_confirm_addr(upper, 0, ip);
2869 }
2870 
bond_has_this_ip(struct bonding * bond,__be32 ip)2871 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2872 {
2873 	struct netdev_nested_priv priv = {
2874 		.data = (void *)&ip,
2875 	};
2876 	bool ret = false;
2877 
2878 	if (ip == bond_confirm_addr(bond->dev, 0, ip))
2879 		return true;
2880 
2881 	rcu_read_lock();
2882 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2883 		ret = true;
2884 	rcu_read_unlock();
2885 
2886 	return ret;
2887 }
2888 
2889 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
2890 
bond_handle_vlan(struct slave * slave,struct bond_vlan_tag * tags,struct sk_buff * skb)2891 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2892 			     struct sk_buff *skb)
2893 {
2894 	struct net_device *bond_dev = slave->bond->dev;
2895 	struct net_device *slave_dev = slave->dev;
2896 	struct bond_vlan_tag *outer_tag = tags;
2897 
2898 	if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
2899 		return true;
2900 
2901 	tags++;
2902 
2903 	/* Go through all the tags backwards and add them to the packet */
2904 	while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
2905 		if (!tags->vlan_id) {
2906 			tags++;
2907 			continue;
2908 		}
2909 
2910 		slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2911 			  ntohs(outer_tag->vlan_proto), tags->vlan_id);
2912 		skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2913 						tags->vlan_id);
2914 		if (!skb) {
2915 			net_err_ratelimited("failed to insert inner VLAN tag\n");
2916 			return false;
2917 		}
2918 
2919 		tags++;
2920 	}
2921 	/* Set the outer tag */
2922 	if (outer_tag->vlan_id) {
2923 		slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2924 			  ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2925 		__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2926 				       outer_tag->vlan_id);
2927 	}
2928 
2929 	return true;
2930 }
2931 
2932 /* We go to the (large) trouble of VLAN tagging ARP frames because
2933  * switches in VLAN mode (especially if ports are configured as
2934  * "native" to a VLAN) might not pass non-tagged frames.
2935  */
bond_arp_send(struct slave * slave,int arp_op,__be32 dest_ip,__be32 src_ip,struct bond_vlan_tag * tags)2936 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2937 			  __be32 src_ip, struct bond_vlan_tag *tags)
2938 {
2939 	struct net_device *bond_dev = slave->bond->dev;
2940 	struct net_device *slave_dev = slave->dev;
2941 	struct sk_buff *skb;
2942 
2943 	slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2944 		  arp_op, &dest_ip, &src_ip);
2945 
2946 	skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2947 			 NULL, slave_dev->dev_addr, NULL);
2948 
2949 	if (!skb) {
2950 		net_err_ratelimited("ARP packet allocation failed\n");
2951 		return;
2952 	}
2953 
2954 	if (bond_handle_vlan(slave, tags, skb)) {
2955 		slave_update_last_tx(slave);
2956 		arp_xmit(skb);
2957 	}
2958 
2959 	return;
2960 }
2961 
2962 /* Validate the device path between the @start_dev and the @end_dev.
2963  * The path is valid if the @end_dev is reachable through device
2964  * stacking.
2965  * When the path is validated, collect any vlan information in the
2966  * path.
2967  */
bond_verify_device_path(struct net_device * start_dev,struct net_device * end_dev,int level)2968 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2969 					      struct net_device *end_dev,
2970 					      int level)
2971 {
2972 	struct bond_vlan_tag *tags;
2973 	struct net_device *upper;
2974 	struct list_head  *iter;
2975 
2976 	if (start_dev == end_dev) {
2977 		tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2978 		if (!tags)
2979 			return ERR_PTR(-ENOMEM);
2980 		tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
2981 		return tags;
2982 	}
2983 
2984 	netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2985 		tags = bond_verify_device_path(upper, end_dev, level + 1);
2986 		if (IS_ERR_OR_NULL(tags)) {
2987 			if (IS_ERR(tags))
2988 				return tags;
2989 			continue;
2990 		}
2991 		if (is_vlan_dev(upper)) {
2992 			tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2993 			tags[level].vlan_id = vlan_dev_vlan_id(upper);
2994 		}
2995 
2996 		return tags;
2997 	}
2998 
2999 	return NULL;
3000 }
3001 
bond_arp_send_all(struct bonding * bond,struct slave * slave)3002 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
3003 {
3004 	struct rtable *rt;
3005 	struct bond_vlan_tag *tags;
3006 	__be32 *targets = bond->params.arp_targets, addr;
3007 	int i;
3008 
3009 	for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3010 		slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3011 			  __func__, &targets[i]);
3012 		tags = NULL;
3013 
3014 		/* Find out through which dev should the packet go */
3015 		rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
3016 				     RTO_ONLINK, 0);
3017 		if (IS_ERR(rt)) {
3018 			/* there's no route to target - try to send arp
3019 			 * probe to generate any traffic (arp_validate=0)
3020 			 */
3021 			if (bond->params.arp_validate)
3022 				pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3023 					     bond->dev->name,
3024 					     &targets[i]);
3025 			bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3026 				      0, tags);
3027 			continue;
3028 		}
3029 
3030 		/* bond device itself */
3031 		if (rt->dst.dev == bond->dev)
3032 			goto found;
3033 
3034 		rcu_read_lock();
3035 		tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3036 		rcu_read_unlock();
3037 
3038 		if (!IS_ERR_OR_NULL(tags))
3039 			goto found;
3040 
3041 		/* Not our device - skip */
3042 		slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3043 			   &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3044 
3045 		ip_rt_put(rt);
3046 		continue;
3047 
3048 found:
3049 		addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3050 		ip_rt_put(rt);
3051 		bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3052 		kfree(tags);
3053 	}
3054 }
3055 
bond_validate_arp(struct bonding * bond,struct slave * slave,__be32 sip,__be32 tip)3056 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3057 {
3058 	int i;
3059 
3060 	if (!sip || !bond_has_this_ip(bond, tip)) {
3061 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3062 			   __func__, &sip, &tip);
3063 		return;
3064 	}
3065 
3066 	i = bond_get_targets_ip(bond->params.arp_targets, sip);
3067 	if (i == -1) {
3068 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3069 			   __func__, &sip);
3070 		return;
3071 	}
3072 	slave->last_rx = jiffies;
3073 	slave->target_last_arp_rx[i] = jiffies;
3074 }
3075 
bond_arp_rcv(const struct sk_buff * skb,struct bonding * bond,struct slave * slave)3076 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3077 			struct slave *slave)
3078 {
3079 	struct arphdr *arp = (struct arphdr *)skb->data;
3080 	struct slave *curr_active_slave, *curr_arp_slave;
3081 	unsigned char *arp_ptr;
3082 	__be32 sip, tip;
3083 	unsigned int alen;
3084 
3085 	alen = arp_hdr_len(bond->dev);
3086 
3087 	if (alen > skb_headlen(skb)) {
3088 		arp = kmalloc(alen, GFP_ATOMIC);
3089 		if (!arp)
3090 			goto out_unlock;
3091 		if (skb_copy_bits(skb, 0, arp, alen) < 0)
3092 			goto out_unlock;
3093 	}
3094 
3095 	if (arp->ar_hln != bond->dev->addr_len ||
3096 	    skb->pkt_type == PACKET_OTHERHOST ||
3097 	    skb->pkt_type == PACKET_LOOPBACK ||
3098 	    arp->ar_hrd != htons(ARPHRD_ETHER) ||
3099 	    arp->ar_pro != htons(ETH_P_IP) ||
3100 	    arp->ar_pln != 4)
3101 		goto out_unlock;
3102 
3103 	arp_ptr = (unsigned char *)(arp + 1);
3104 	arp_ptr += bond->dev->addr_len;
3105 	memcpy(&sip, arp_ptr, 4);
3106 	arp_ptr += 4 + bond->dev->addr_len;
3107 	memcpy(&tip, arp_ptr, 4);
3108 
3109 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3110 		  __func__, slave->dev->name, bond_slave_state(slave),
3111 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3112 		  &sip, &tip);
3113 
3114 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3115 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3116 
3117 	/* We 'trust' the received ARP enough to validate it if:
3118 	 *
3119 	 * (a) the slave receiving the ARP is active (which includes the
3120 	 * current ARP slave, if any), or
3121 	 *
3122 	 * (b) the receiving slave isn't active, but there is a currently
3123 	 * active slave and it received valid arp reply(s) after it became
3124 	 * the currently active slave, or
3125 	 *
3126 	 * (c) there is an ARP slave that sent an ARP during the prior ARP
3127 	 * interval, and we receive an ARP reply on any slave.  We accept
3128 	 * these because switch FDB update delays may deliver the ARP
3129 	 * reply to a slave other than the sender of the ARP request.
3130 	 *
3131 	 * Note: for (b), backup slaves are receiving the broadcast ARP
3132 	 * request, not a reply.  This request passes from the sending
3133 	 * slave through the L2 switch(es) to the receiving slave.  Since
3134 	 * this is checking the request, sip/tip are swapped for
3135 	 * validation.
3136 	 *
3137 	 * This is done to avoid endless looping when we can't reach the
3138 	 * arp_ip_target and fool ourselves with our own arp requests.
3139 	 */
3140 	if (bond_is_active_slave(slave))
3141 		bond_validate_arp(bond, slave, sip, tip);
3142 	else if (curr_active_slave &&
3143 		 time_after(slave_last_rx(bond, curr_active_slave),
3144 			    curr_active_slave->last_link_up))
3145 		bond_validate_arp(bond, slave, tip, sip);
3146 	else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3147 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3148 		bond_validate_arp(bond, slave, sip, tip);
3149 
3150 out_unlock:
3151 	if (arp != (struct arphdr *)skb->data)
3152 		kfree(arp);
3153 	return RX_HANDLER_ANOTHER;
3154 }
3155 
3156 #if IS_ENABLED(CONFIG_IPV6)
bond_ns_send(struct slave * slave,const struct in6_addr * daddr,const struct in6_addr * saddr,struct bond_vlan_tag * tags)3157 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3158 			 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3159 {
3160 	struct net_device *bond_dev = slave->bond->dev;
3161 	struct net_device *slave_dev = slave->dev;
3162 	struct in6_addr mcaddr;
3163 	struct sk_buff *skb;
3164 
3165 	slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3166 		  daddr, saddr);
3167 
3168 	skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3169 	if (!skb) {
3170 		net_err_ratelimited("NS packet allocation failed\n");
3171 		return;
3172 	}
3173 
3174 	addrconf_addr_solict_mult(daddr, &mcaddr);
3175 	if (bond_handle_vlan(slave, tags, skb)) {
3176 		slave_update_last_tx(slave);
3177 		ndisc_send_skb(skb, &mcaddr, saddr);
3178 	}
3179 }
3180 
bond_ns_send_all(struct bonding * bond,struct slave * slave)3181 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3182 {
3183 	struct in6_addr *targets = bond->params.ns_targets;
3184 	struct bond_vlan_tag *tags;
3185 	struct dst_entry *dst;
3186 	struct in6_addr saddr;
3187 	struct flowi6 fl6;
3188 	int i;
3189 
3190 	for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3191 		slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3192 			  __func__, &targets[i]);
3193 		tags = NULL;
3194 
3195 		/* Find out through which dev should the packet go */
3196 		memset(&fl6, 0, sizeof(struct flowi6));
3197 		fl6.daddr = targets[i];
3198 		fl6.flowi6_oif = bond->dev->ifindex;
3199 
3200 		dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3201 		if (dst->error) {
3202 			dst_release(dst);
3203 			/* there's no route to target - try to send arp
3204 			 * probe to generate any traffic (arp_validate=0)
3205 			 */
3206 			if (bond->params.arp_validate)
3207 				pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3208 					     bond->dev->name,
3209 					     &targets[i]);
3210 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3211 			continue;
3212 		}
3213 
3214 		/* bond device itself */
3215 		if (dst->dev == bond->dev)
3216 			goto found;
3217 
3218 		rcu_read_lock();
3219 		tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3220 		rcu_read_unlock();
3221 
3222 		if (!IS_ERR_OR_NULL(tags))
3223 			goto found;
3224 
3225 		/* Not our device - skip */
3226 		slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3227 			  &targets[i], dst->dev ? dst->dev->name : "NULL");
3228 
3229 		dst_release(dst);
3230 		continue;
3231 
3232 found:
3233 		if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3234 			bond_ns_send(slave, &targets[i], &saddr, tags);
3235 		else
3236 			bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3237 
3238 		dst_release(dst);
3239 		kfree(tags);
3240 	}
3241 }
3242 
bond_confirm_addr6(struct net_device * dev,struct netdev_nested_priv * priv)3243 static int bond_confirm_addr6(struct net_device *dev,
3244 			      struct netdev_nested_priv *priv)
3245 {
3246 	struct in6_addr *addr = (struct in6_addr *)priv->data;
3247 
3248 	return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3249 }
3250 
bond_has_this_ip6(struct bonding * bond,struct in6_addr * addr)3251 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3252 {
3253 	struct netdev_nested_priv priv = {
3254 		.data = addr,
3255 	};
3256 	int ret = false;
3257 
3258 	if (bond_confirm_addr6(bond->dev, &priv))
3259 		return true;
3260 
3261 	rcu_read_lock();
3262 	if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3263 		ret = true;
3264 	rcu_read_unlock();
3265 
3266 	return ret;
3267 }
3268 
bond_validate_na(struct bonding * bond,struct slave * slave,struct in6_addr * saddr,struct in6_addr * daddr)3269 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3270 			     struct in6_addr *saddr, struct in6_addr *daddr)
3271 {
3272 	int i;
3273 
3274 	/* Ignore NAs that:
3275 	 * 1. Source address is unspecified address.
3276 	 * 2. Dest address is neither all-nodes multicast address nor
3277 	 *    exist on bond interface.
3278 	 */
3279 	if (ipv6_addr_any(saddr) ||
3280 	    (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3281 	     !bond_has_this_ip6(bond, daddr))) {
3282 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3283 			  __func__, saddr, daddr);
3284 		return;
3285 	}
3286 
3287 	i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3288 	if (i == -1) {
3289 		slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3290 			  __func__, saddr);
3291 		return;
3292 	}
3293 	slave->last_rx = jiffies;
3294 	slave->target_last_arp_rx[i] = jiffies;
3295 }
3296 
bond_na_rcv(const struct sk_buff * skb,struct bonding * bond,struct slave * slave)3297 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3298 		       struct slave *slave)
3299 {
3300 	struct slave *curr_active_slave, *curr_arp_slave;
3301 	struct in6_addr *saddr, *daddr;
3302 	struct {
3303 		struct ipv6hdr ip6;
3304 		struct icmp6hdr icmp6;
3305 	} *combined, _combined;
3306 
3307 	if (skb->pkt_type == PACKET_OTHERHOST ||
3308 	    skb->pkt_type == PACKET_LOOPBACK)
3309 		goto out;
3310 
3311 	combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3312 	if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3313 	    (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3314 	     combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3315 		goto out;
3316 
3317 	saddr = &combined->ip6.saddr;
3318 	daddr = &combined->ip6.daddr;
3319 
3320 	slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3321 		  __func__, slave->dev->name, bond_slave_state(slave),
3322 		  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3323 		  saddr, daddr);
3324 
3325 	curr_active_slave = rcu_dereference(bond->curr_active_slave);
3326 	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3327 
3328 	/* We 'trust' the received ARP enough to validate it if:
3329 	 * see bond_arp_rcv().
3330 	 */
3331 	if (bond_is_active_slave(slave))
3332 		bond_validate_na(bond, slave, saddr, daddr);
3333 	else if (curr_active_slave &&
3334 		 time_after(slave_last_rx(bond, curr_active_slave),
3335 			    curr_active_slave->last_link_up))
3336 		bond_validate_na(bond, slave, daddr, saddr);
3337 	else if (curr_arp_slave &&
3338 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3339 		bond_validate_na(bond, slave, saddr, daddr);
3340 
3341 out:
3342 	return RX_HANDLER_ANOTHER;
3343 }
3344 #endif
3345 
bond_rcv_validate(const struct sk_buff * skb,struct bonding * bond,struct slave * slave)3346 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3347 		      struct slave *slave)
3348 {
3349 #if IS_ENABLED(CONFIG_IPV6)
3350 	bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3351 #endif
3352 	bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3353 
3354 	slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3355 		  __func__, skb->dev->name);
3356 
3357 	/* Use arp validate logic for both ARP and NS */
3358 	if (!slave_do_arp_validate(bond, slave)) {
3359 		if ((slave_do_arp_validate_only(bond) && is_arp) ||
3360 #if IS_ENABLED(CONFIG_IPV6)
3361 		    (slave_do_arp_validate_only(bond) && is_ipv6) ||
3362 #endif
3363 		    !slave_do_arp_validate_only(bond))
3364 			slave->last_rx = jiffies;
3365 		return RX_HANDLER_ANOTHER;
3366 	} else if (is_arp) {
3367 		return bond_arp_rcv(skb, bond, slave);
3368 #if IS_ENABLED(CONFIG_IPV6)
3369 	} else if (is_ipv6) {
3370 		return bond_na_rcv(skb, bond, slave);
3371 #endif
3372 	} else {
3373 		return RX_HANDLER_ANOTHER;
3374 	}
3375 }
3376 
bond_send_validate(struct bonding * bond,struct slave * slave)3377 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3378 {
3379 	bond_arp_send_all(bond, slave);
3380 #if IS_ENABLED(CONFIG_IPV6)
3381 	bond_ns_send_all(bond, slave);
3382 #endif
3383 }
3384 
3385 /* function to verify if we're in the arp_interval timeslice, returns true if
3386  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3387  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3388  */
bond_time_in_interval(struct bonding * bond,unsigned long last_act,int mod)3389 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3390 				  int mod)
3391 {
3392 	int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3393 
3394 	return time_in_range(jiffies,
3395 			     last_act - delta_in_ticks,
3396 			     last_act + mod * delta_in_ticks + delta_in_ticks/2);
3397 }
3398 
3399 /* This function is called regularly to monitor each slave's link
3400  * ensuring that traffic is being sent and received when arp monitoring
3401  * is used in load-balancing mode. if the adapter has been dormant, then an
3402  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3403  * arp monitoring in active backup mode.
3404  */
bond_loadbalance_arp_mon(struct bonding * bond)3405 static void bond_loadbalance_arp_mon(struct bonding *bond)
3406 {
3407 	struct slave *slave, *oldcurrent;
3408 	struct list_head *iter;
3409 	int do_failover = 0, slave_state_changed = 0;
3410 
3411 	if (!bond_has_slaves(bond))
3412 		goto re_arm;
3413 
3414 	rcu_read_lock();
3415 
3416 	oldcurrent = rcu_dereference(bond->curr_active_slave);
3417 	/* see if any of the previous devices are up now (i.e. they have
3418 	 * xmt and rcv traffic). the curr_active_slave does not come into
3419 	 * the picture unless it is null. also, slave->last_link_up is not
3420 	 * needed here because we send an arp on each slave and give a slave
3421 	 * as long as it needs to get the tx/rx within the delta.
3422 	 * TODO: what about up/down delay in arp mode? it wasn't here before
3423 	 *       so it can wait
3424 	 */
3425 	bond_for_each_slave_rcu(bond, slave, iter) {
3426 		unsigned long last_tx = slave_last_tx(slave);
3427 
3428 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3429 
3430 		if (slave->link != BOND_LINK_UP) {
3431 			if (bond_time_in_interval(bond, last_tx, 1) &&
3432 			    bond_time_in_interval(bond, slave->last_rx, 1)) {
3433 
3434 				bond_propose_link_state(slave, BOND_LINK_UP);
3435 				slave_state_changed = 1;
3436 
3437 				/* primary_slave has no meaning in round-robin
3438 				 * mode. the window of a slave being up and
3439 				 * curr_active_slave being null after enslaving
3440 				 * is closed.
3441 				 */
3442 				if (!oldcurrent) {
3443 					slave_info(bond->dev, slave->dev, "link status definitely up\n");
3444 					do_failover = 1;
3445 				} else {
3446 					slave_info(bond->dev, slave->dev, "interface is now up\n");
3447 				}
3448 			}
3449 		} else {
3450 			/* slave->link == BOND_LINK_UP */
3451 
3452 			/* not all switches will respond to an arp request
3453 			 * when the source ip is 0, so don't take the link down
3454 			 * if we don't know our ip yet
3455 			 */
3456 			if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3457 			    !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3458 
3459 				bond_propose_link_state(slave, BOND_LINK_DOWN);
3460 				slave_state_changed = 1;
3461 
3462 				if (slave->link_failure_count < UINT_MAX)
3463 					slave->link_failure_count++;
3464 
3465 				slave_info(bond->dev, slave->dev, "interface is now down\n");
3466 
3467 				if (slave == oldcurrent)
3468 					do_failover = 1;
3469 			}
3470 		}
3471 
3472 		/* note: if switch is in round-robin mode, all links
3473 		 * must tx arp to ensure all links rx an arp - otherwise
3474 		 * links may oscillate or not come up at all; if switch is
3475 		 * in something like xor mode, there is nothing we can
3476 		 * do - all replies will be rx'ed on same link causing slaves
3477 		 * to be unstable during low/no traffic periods
3478 		 */
3479 		if (bond_slave_is_up(slave))
3480 			bond_send_validate(bond, slave);
3481 	}
3482 
3483 	rcu_read_unlock();
3484 
3485 	if (do_failover || slave_state_changed) {
3486 		if (!rtnl_trylock())
3487 			goto re_arm;
3488 
3489 		bond_for_each_slave(bond, slave, iter) {
3490 			if (slave->link_new_state != BOND_LINK_NOCHANGE)
3491 				slave->link = slave->link_new_state;
3492 		}
3493 
3494 		if (slave_state_changed) {
3495 			bond_slave_state_change(bond);
3496 			if (BOND_MODE(bond) == BOND_MODE_XOR)
3497 				bond_update_slave_arr(bond, NULL);
3498 		}
3499 		if (do_failover) {
3500 			block_netpoll_tx();
3501 			bond_select_active_slave(bond);
3502 			unblock_netpoll_tx();
3503 		}
3504 		rtnl_unlock();
3505 	}
3506 
3507 re_arm:
3508 	if (bond->params.arp_interval)
3509 		queue_delayed_work(bond->wq, &bond->arp_work,
3510 				   msecs_to_jiffies(bond->params.arp_interval));
3511 }
3512 
3513 /* Called to inspect slaves for active-backup mode ARP monitor link state
3514  * changes.  Sets proposed link state in slaves to specify what action
3515  * should take place for the slave.  Returns 0 if no changes are found, >0
3516  * if changes to link states must be committed.
3517  *
3518  * Called with rcu_read_lock held.
3519  */
bond_ab_arp_inspect(struct bonding * bond)3520 static int bond_ab_arp_inspect(struct bonding *bond)
3521 {
3522 	unsigned long last_tx, last_rx;
3523 	struct list_head *iter;
3524 	struct slave *slave;
3525 	int commit = 0;
3526 
3527 	bond_for_each_slave_rcu(bond, slave, iter) {
3528 		bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3529 		last_rx = slave_last_rx(bond, slave);
3530 
3531 		if (slave->link != BOND_LINK_UP) {
3532 			if (bond_time_in_interval(bond, last_rx, 1)) {
3533 				bond_propose_link_state(slave, BOND_LINK_UP);
3534 				commit++;
3535 			} else if (slave->link == BOND_LINK_BACK) {
3536 				bond_propose_link_state(slave, BOND_LINK_FAIL);
3537 				commit++;
3538 			}
3539 			continue;
3540 		}
3541 
3542 		/* Give slaves 2*delta after being enslaved or made
3543 		 * active.  This avoids bouncing, as the last receive
3544 		 * times need a full ARP monitor cycle to be updated.
3545 		 */
3546 		if (bond_time_in_interval(bond, slave->last_link_up, 2))
3547 			continue;
3548 
3549 		/* Backup slave is down if:
3550 		 * - No current_arp_slave AND
3551 		 * - more than (missed_max+1)*delta since last receive AND
3552 		 * - the bond has an IP address
3553 		 *
3554 		 * Note: a non-null current_arp_slave indicates
3555 		 * the curr_active_slave went down and we are
3556 		 * searching for a new one; under this condition
3557 		 * we only take the curr_active_slave down - this
3558 		 * gives each slave a chance to tx/rx traffic
3559 		 * before being taken out
3560 		 */
3561 		if (!bond_is_active_slave(slave) &&
3562 		    !rcu_access_pointer(bond->current_arp_slave) &&
3563 		    !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3564 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3565 			commit++;
3566 		}
3567 
3568 		/* Active slave is down if:
3569 		 * - more than missed_max*delta since transmitting OR
3570 		 * - (more than missed_max*delta since receive AND
3571 		 *    the bond has an IP address)
3572 		 */
3573 		last_tx = slave_last_tx(slave);
3574 		if (bond_is_active_slave(slave) &&
3575 		    (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3576 		     !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3577 			bond_propose_link_state(slave, BOND_LINK_DOWN);
3578 			commit++;
3579 		}
3580 	}
3581 
3582 	return commit;
3583 }
3584 
3585 /* Called to commit link state changes noted by inspection step of
3586  * active-backup mode ARP monitor.
3587  *
3588  * Called with RTNL hold.
3589  */
bond_ab_arp_commit(struct bonding * bond)3590 static void bond_ab_arp_commit(struct bonding *bond)
3591 {
3592 	bool do_failover = false;
3593 	struct list_head *iter;
3594 	unsigned long last_tx;
3595 	struct slave *slave;
3596 
3597 	bond_for_each_slave(bond, slave, iter) {
3598 		switch (slave->link_new_state) {
3599 		case BOND_LINK_NOCHANGE:
3600 			continue;
3601 
3602 		case BOND_LINK_UP:
3603 			last_tx = slave_last_tx(slave);
3604 			if (rtnl_dereference(bond->curr_active_slave) != slave ||
3605 			    (!rtnl_dereference(bond->curr_active_slave) &&
3606 			     bond_time_in_interval(bond, last_tx, 1))) {
3607 				struct slave *current_arp_slave;
3608 
3609 				current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3610 				bond_set_slave_link_state(slave, BOND_LINK_UP,
3611 							  BOND_SLAVE_NOTIFY_NOW);
3612 				if (current_arp_slave) {
3613 					bond_set_slave_inactive_flags(
3614 						current_arp_slave,
3615 						BOND_SLAVE_NOTIFY_NOW);
3616 					RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3617 				}
3618 
3619 				slave_info(bond->dev, slave->dev, "link status definitely up\n");
3620 
3621 				if (!rtnl_dereference(bond->curr_active_slave) ||
3622 				    slave == rtnl_dereference(bond->primary_slave) ||
3623 				    slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3624 					do_failover = true;
3625 
3626 			}
3627 
3628 			continue;
3629 
3630 		case BOND_LINK_DOWN:
3631 			if (slave->link_failure_count < UINT_MAX)
3632 				slave->link_failure_count++;
3633 
3634 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3635 						  BOND_SLAVE_NOTIFY_NOW);
3636 			bond_set_slave_inactive_flags(slave,
3637 						      BOND_SLAVE_NOTIFY_NOW);
3638 
3639 			slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3640 
3641 			if (slave == rtnl_dereference(bond->curr_active_slave)) {
3642 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3643 				do_failover = true;
3644 			}
3645 
3646 			continue;
3647 
3648 		case BOND_LINK_FAIL:
3649 			bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3650 						  BOND_SLAVE_NOTIFY_NOW);
3651 			bond_set_slave_inactive_flags(slave,
3652 						      BOND_SLAVE_NOTIFY_NOW);
3653 
3654 			/* A slave has just been enslaved and has become
3655 			 * the current active slave.
3656 			 */
3657 			if (rtnl_dereference(bond->curr_active_slave))
3658 				RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3659 			continue;
3660 
3661 		default:
3662 			slave_err(bond->dev, slave->dev,
3663 				  "impossible: link_new_state %d on slave\n",
3664 				  slave->link_new_state);
3665 			continue;
3666 		}
3667 	}
3668 
3669 	if (do_failover) {
3670 		block_netpoll_tx();
3671 		bond_select_active_slave(bond);
3672 		unblock_netpoll_tx();
3673 	}
3674 
3675 	bond_set_carrier(bond);
3676 }
3677 
3678 /* Send ARP probes for active-backup mode ARP monitor.
3679  *
3680  * Called with rcu_read_lock held.
3681  */
bond_ab_arp_probe(struct bonding * bond)3682 static bool bond_ab_arp_probe(struct bonding *bond)
3683 {
3684 	struct slave *slave, *before = NULL, *new_slave = NULL,
3685 		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3686 		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3687 	struct list_head *iter;
3688 	bool found = false;
3689 	bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3690 
3691 	if (curr_arp_slave && curr_active_slave)
3692 		netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3693 			    curr_arp_slave->dev->name,
3694 			    curr_active_slave->dev->name);
3695 
3696 	if (curr_active_slave) {
3697 		bond_send_validate(bond, curr_active_slave);
3698 		return should_notify_rtnl;
3699 	}
3700 
3701 	/* if we don't have a curr_active_slave, search for the next available
3702 	 * backup slave from the current_arp_slave and make it the candidate
3703 	 * for becoming the curr_active_slave
3704 	 */
3705 
3706 	if (!curr_arp_slave) {
3707 		curr_arp_slave = bond_first_slave_rcu(bond);
3708 		if (!curr_arp_slave)
3709 			return should_notify_rtnl;
3710 	}
3711 
3712 	bond_for_each_slave_rcu(bond, slave, iter) {
3713 		if (!found && !before && bond_slave_is_up(slave))
3714 			before = slave;
3715 
3716 		if (found && !new_slave && bond_slave_is_up(slave))
3717 			new_slave = slave;
3718 		/* if the link state is up at this point, we
3719 		 * mark it down - this can happen if we have
3720 		 * simultaneous link failures and
3721 		 * reselect_active_interface doesn't make this
3722 		 * one the current slave so it is still marked
3723 		 * up when it is actually down
3724 		 */
3725 		if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3726 			bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3727 						  BOND_SLAVE_NOTIFY_LATER);
3728 			if (slave->link_failure_count < UINT_MAX)
3729 				slave->link_failure_count++;
3730 
3731 			bond_set_slave_inactive_flags(slave,
3732 						      BOND_SLAVE_NOTIFY_LATER);
3733 
3734 			slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3735 		}
3736 		if (slave == curr_arp_slave)
3737 			found = true;
3738 	}
3739 
3740 	if (!new_slave && before)
3741 		new_slave = before;
3742 
3743 	if (!new_slave)
3744 		goto check_state;
3745 
3746 	bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3747 				  BOND_SLAVE_NOTIFY_LATER);
3748 	bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3749 	bond_send_validate(bond, new_slave);
3750 	new_slave->last_link_up = jiffies;
3751 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
3752 
3753 check_state:
3754 	bond_for_each_slave_rcu(bond, slave, iter) {
3755 		if (slave->should_notify || slave->should_notify_link) {
3756 			should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3757 			break;
3758 		}
3759 	}
3760 	return should_notify_rtnl;
3761 }
3762 
bond_activebackup_arp_mon(struct bonding * bond)3763 static void bond_activebackup_arp_mon(struct bonding *bond)
3764 {
3765 	bool should_notify_peers = false;
3766 	bool should_notify_rtnl = false;
3767 	int delta_in_ticks;
3768 
3769 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3770 
3771 	if (!bond_has_slaves(bond))
3772 		goto re_arm;
3773 
3774 	rcu_read_lock();
3775 
3776 	should_notify_peers = bond_should_notify_peers(bond);
3777 
3778 	if (bond_ab_arp_inspect(bond)) {
3779 		rcu_read_unlock();
3780 
3781 		/* Race avoidance with bond_close flush of workqueue */
3782 		if (!rtnl_trylock()) {
3783 			delta_in_ticks = 1;
3784 			should_notify_peers = false;
3785 			goto re_arm;
3786 		}
3787 
3788 		bond_ab_arp_commit(bond);
3789 
3790 		rtnl_unlock();
3791 		rcu_read_lock();
3792 	}
3793 
3794 	should_notify_rtnl = bond_ab_arp_probe(bond);
3795 	rcu_read_unlock();
3796 
3797 re_arm:
3798 	if (bond->params.arp_interval)
3799 		queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3800 
3801 	if (should_notify_peers || should_notify_rtnl) {
3802 		if (!rtnl_trylock())
3803 			return;
3804 
3805 		if (should_notify_peers) {
3806 			bond->send_peer_notif--;
3807 			call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3808 						 bond->dev);
3809 		}
3810 		if (should_notify_rtnl) {
3811 			bond_slave_state_notify(bond);
3812 			bond_slave_link_notify(bond);
3813 		}
3814 
3815 		rtnl_unlock();
3816 	}
3817 }
3818 
bond_arp_monitor(struct work_struct * work)3819 static void bond_arp_monitor(struct work_struct *work)
3820 {
3821 	struct bonding *bond = container_of(work, struct bonding,
3822 					    arp_work.work);
3823 
3824 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3825 		bond_activebackup_arp_mon(bond);
3826 	else
3827 		bond_loadbalance_arp_mon(bond);
3828 }
3829 
3830 /*-------------------------- netdev event handling --------------------------*/
3831 
3832 /* Change device name */
bond_event_changename(struct bonding * bond)3833 static int bond_event_changename(struct bonding *bond)
3834 {
3835 	bond_remove_proc_entry(bond);
3836 	bond_create_proc_entry(bond);
3837 
3838 	bond_debug_reregister(bond);
3839 
3840 	return NOTIFY_DONE;
3841 }
3842 
bond_master_netdev_event(unsigned long event,struct net_device * bond_dev)3843 static int bond_master_netdev_event(unsigned long event,
3844 				    struct net_device *bond_dev)
3845 {
3846 	struct bonding *event_bond = netdev_priv(bond_dev);
3847 
3848 	netdev_dbg(bond_dev, "%s called\n", __func__);
3849 
3850 	switch (event) {
3851 	case NETDEV_CHANGENAME:
3852 		return bond_event_changename(event_bond);
3853 	case NETDEV_UNREGISTER:
3854 		bond_remove_proc_entry(event_bond);
3855 #ifdef CONFIG_XFRM_OFFLOAD
3856 		xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3857 #endif /* CONFIG_XFRM_OFFLOAD */
3858 		break;
3859 	case NETDEV_REGISTER:
3860 		bond_create_proc_entry(event_bond);
3861 		break;
3862 	default:
3863 		break;
3864 	}
3865 
3866 	return NOTIFY_DONE;
3867 }
3868 
bond_slave_netdev_event(unsigned long event,struct net_device * slave_dev)3869 static int bond_slave_netdev_event(unsigned long event,
3870 				   struct net_device *slave_dev)
3871 {
3872 	struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3873 	struct bonding *bond;
3874 	struct net_device *bond_dev;
3875 
3876 	/* A netdev event can be generated while enslaving a device
3877 	 * before netdev_rx_handler_register is called in which case
3878 	 * slave will be NULL
3879 	 */
3880 	if (!slave) {
3881 		netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3882 		return NOTIFY_DONE;
3883 	}
3884 
3885 	bond_dev = slave->bond->dev;
3886 	bond = slave->bond;
3887 	primary = rtnl_dereference(bond->primary_slave);
3888 
3889 	slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3890 
3891 	switch (event) {
3892 	case NETDEV_UNREGISTER:
3893 		if (bond_dev->type != ARPHRD_ETHER)
3894 			bond_release_and_destroy(bond_dev, slave_dev);
3895 		else
3896 			__bond_release_one(bond_dev, slave_dev, false, true);
3897 		break;
3898 	case NETDEV_UP:
3899 	case NETDEV_CHANGE:
3900 		/* For 802.3ad mode only:
3901 		 * Getting invalid Speed/Duplex values here will put slave
3902 		 * in weird state. Mark it as link-fail if the link was
3903 		 * previously up or link-down if it hasn't yet come up, and
3904 		 * let link-monitoring (miimon) set it right when correct
3905 		 * speeds/duplex are available.
3906 		 */
3907 		if (bond_update_speed_duplex(slave) &&
3908 		    BOND_MODE(bond) == BOND_MODE_8023AD) {
3909 			if (slave->last_link_up)
3910 				slave->link = BOND_LINK_FAIL;
3911 			else
3912 				slave->link = BOND_LINK_DOWN;
3913 		}
3914 
3915 		if (BOND_MODE(bond) == BOND_MODE_8023AD)
3916 			bond_3ad_adapter_speed_duplex_changed(slave);
3917 		fallthrough;
3918 	case NETDEV_DOWN:
3919 		/* Refresh slave-array if applicable!
3920 		 * If the setup does not use miimon or arpmon (mode-specific!),
3921 		 * then these events will not cause the slave-array to be
3922 		 * refreshed. This will cause xmit to use a slave that is not
3923 		 * usable. Avoid such situation by refeshing the array at these
3924 		 * events. If these (miimon/arpmon) parameters are configured
3925 		 * then array gets refreshed twice and that should be fine!
3926 		 */
3927 		if (bond_mode_can_use_xmit_hash(bond))
3928 			bond_update_slave_arr(bond, NULL);
3929 		break;
3930 	case NETDEV_CHANGEMTU:
3931 		/* TODO: Should slaves be allowed to
3932 		 * independently alter their MTU?  For
3933 		 * an active-backup bond, slaves need
3934 		 * not be the same type of device, so
3935 		 * MTUs may vary.  For other modes,
3936 		 * slaves arguably should have the
3937 		 * same MTUs. To do this, we'd need to
3938 		 * take over the slave's change_mtu
3939 		 * function for the duration of their
3940 		 * servitude.
3941 		 */
3942 		break;
3943 	case NETDEV_CHANGENAME:
3944 		/* we don't care if we don't have primary set */
3945 		if (!bond_uses_primary(bond) ||
3946 		    !bond->params.primary[0])
3947 			break;
3948 
3949 		if (slave == primary) {
3950 			/* slave's name changed - he's no longer primary */
3951 			RCU_INIT_POINTER(bond->primary_slave, NULL);
3952 		} else if (!strcmp(slave_dev->name, bond->params.primary)) {
3953 			/* we have a new primary slave */
3954 			rcu_assign_pointer(bond->primary_slave, slave);
3955 		} else { /* we didn't change primary - exit */
3956 			break;
3957 		}
3958 
3959 		netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3960 			    primary ? slave_dev->name : "none");
3961 
3962 		block_netpoll_tx();
3963 		bond_select_active_slave(bond);
3964 		unblock_netpoll_tx();
3965 		break;
3966 	case NETDEV_FEAT_CHANGE:
3967 		if (!bond->notifier_ctx) {
3968 			bond->notifier_ctx = true;
3969 			bond_compute_features(bond);
3970 			bond->notifier_ctx = false;
3971 		}
3972 		break;
3973 	case NETDEV_RESEND_IGMP:
3974 		/* Propagate to master device */
3975 		call_netdevice_notifiers(event, slave->bond->dev);
3976 		break;
3977 	case NETDEV_XDP_FEAT_CHANGE:
3978 		bond_xdp_set_features(bond_dev);
3979 		break;
3980 	default:
3981 		break;
3982 	}
3983 
3984 	return NOTIFY_DONE;
3985 }
3986 
3987 /* bond_netdev_event: handle netdev notifier chain events.
3988  *
3989  * This function receives events for the netdev chain.  The caller (an
3990  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3991  * locks for us to safely manipulate the slave devices (RTNL lock,
3992  * dev_probe_lock).
3993  */
bond_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)3994 static int bond_netdev_event(struct notifier_block *this,
3995 			     unsigned long event, void *ptr)
3996 {
3997 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3998 
3999 	netdev_dbg(event_dev, "%s received %s\n",
4000 		   __func__, netdev_cmd_to_name(event));
4001 
4002 	if (!(event_dev->priv_flags & IFF_BONDING))
4003 		return NOTIFY_DONE;
4004 
4005 	if (event_dev->flags & IFF_MASTER) {
4006 		int ret;
4007 
4008 		ret = bond_master_netdev_event(event, event_dev);
4009 		if (ret != NOTIFY_DONE)
4010 			return ret;
4011 	}
4012 
4013 	if (event_dev->flags & IFF_SLAVE)
4014 		return bond_slave_netdev_event(event, event_dev);
4015 
4016 	return NOTIFY_DONE;
4017 }
4018 
4019 static struct notifier_block bond_netdev_notifier = {
4020 	.notifier_call = bond_netdev_event,
4021 };
4022 
4023 /*---------------------------- Hashing Policies -----------------------------*/
4024 
4025 /* Helper to access data in a packet, with or without a backing skb.
4026  * If skb is given the data is linearized if necessary via pskb_may_pull.
4027  */
bond_pull_data(struct sk_buff * skb,const void * data,int hlen,int n)4028 static inline const void *bond_pull_data(struct sk_buff *skb,
4029 					 const void *data, int hlen, int n)
4030 {
4031 	if (likely(n <= hlen))
4032 		return data;
4033 	else if (skb && likely(pskb_may_pull(skb, n)))
4034 		return skb->data;
4035 
4036 	return NULL;
4037 }
4038 
4039 /* L2 hash helper */
bond_eth_hash(struct sk_buff * skb,const void * data,int mhoff,int hlen)4040 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4041 {
4042 	struct ethhdr *ep;
4043 
4044 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4045 	if (!data)
4046 		return 0;
4047 
4048 	ep = (struct ethhdr *)(data + mhoff);
4049 	return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4050 }
4051 
bond_flow_ip(struct sk_buff * skb,struct flow_keys * fk,const void * data,int hlen,__be16 l2_proto,int * nhoff,int * ip_proto,bool l34)4052 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4053 			 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4054 {
4055 	const struct ipv6hdr *iph6;
4056 	const struct iphdr *iph;
4057 
4058 	if (l2_proto == htons(ETH_P_IP)) {
4059 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4060 		if (!data)
4061 			return false;
4062 
4063 		iph = (const struct iphdr *)(data + *nhoff);
4064 		iph_to_flow_copy_v4addrs(fk, iph);
4065 		*nhoff += iph->ihl << 2;
4066 		if (!ip_is_fragment(iph))
4067 			*ip_proto = iph->protocol;
4068 	} else if (l2_proto == htons(ETH_P_IPV6)) {
4069 		data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4070 		if (!data)
4071 			return false;
4072 
4073 		iph6 = (const struct ipv6hdr *)(data + *nhoff);
4074 		iph_to_flow_copy_v6addrs(fk, iph6);
4075 		*nhoff += sizeof(*iph6);
4076 		*ip_proto = iph6->nexthdr;
4077 	} else {
4078 		return false;
4079 	}
4080 
4081 	if (l34 && *ip_proto >= 0)
4082 		fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4083 
4084 	return true;
4085 }
4086 
bond_vlan_srcmac_hash(struct sk_buff * skb,const void * data,int mhoff,int hlen)4087 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4088 {
4089 	u32 srcmac_vendor = 0, srcmac_dev = 0;
4090 	struct ethhdr *mac_hdr;
4091 	u16 vlan = 0;
4092 	int i;
4093 
4094 	data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4095 	if (!data)
4096 		return 0;
4097 	mac_hdr = (struct ethhdr *)(data + mhoff);
4098 
4099 	for (i = 0; i < 3; i++)
4100 		srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4101 
4102 	for (i = 3; i < ETH_ALEN; i++)
4103 		srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4104 
4105 	if (skb && skb_vlan_tag_present(skb))
4106 		vlan = skb_vlan_tag_get(skb);
4107 
4108 	return vlan ^ srcmac_vendor ^ srcmac_dev;
4109 }
4110 
4111 /* Extract the appropriate headers based on bond's xmit policy */
bond_flow_dissect(struct bonding * bond,struct sk_buff * skb,const void * data,__be16 l2_proto,int nhoff,int hlen,struct flow_keys * fk)4112 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4113 			      __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4114 {
4115 	bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4116 	int ip_proto = -1;
4117 
4118 	switch (bond->params.xmit_policy) {
4119 	case BOND_XMIT_POLICY_ENCAP23:
4120 	case BOND_XMIT_POLICY_ENCAP34:
4121 		memset(fk, 0, sizeof(*fk));
4122 		return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4123 					  fk, data, l2_proto, nhoff, hlen, 0);
4124 	default:
4125 		break;
4126 	}
4127 
4128 	fk->ports.ports = 0;
4129 	memset(&fk->icmp, 0, sizeof(fk->icmp));
4130 	if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4131 		return false;
4132 
4133 	/* ICMP error packets contains at least 8 bytes of the header
4134 	 * of the packet which generated the error. Use this information
4135 	 * to correlate ICMP error packets within the same flow which
4136 	 * generated the error.
4137 	 */
4138 	if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4139 		skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4140 		if (ip_proto == IPPROTO_ICMP) {
4141 			if (!icmp_is_err(fk->icmp.type))
4142 				return true;
4143 
4144 			nhoff += sizeof(struct icmphdr);
4145 		} else if (ip_proto == IPPROTO_ICMPV6) {
4146 			if (!icmpv6_is_err(fk->icmp.type))
4147 				return true;
4148 
4149 			nhoff += sizeof(struct icmp6hdr);
4150 		}
4151 		return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4152 	}
4153 
4154 	return true;
4155 }
4156 
bond_ip_hash(u32 hash,struct flow_keys * flow,int xmit_policy)4157 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4158 {
4159 	hash ^= (__force u32)flow_get_u32_dst(flow) ^
4160 		(__force u32)flow_get_u32_src(flow);
4161 	hash ^= (hash >> 16);
4162 	hash ^= (hash >> 8);
4163 
4164 	/* discard lowest hash bit to deal with the common even ports pattern */
4165 	if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4166 		xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4167 		return hash >> 1;
4168 
4169 	return hash;
4170 }
4171 
4172 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4173  * the data as required, but this function can be used without it if the data is
4174  * known to be linear (e.g. with xdp_buff).
4175  */
__bond_xmit_hash(struct bonding * bond,struct sk_buff * skb,const void * data,__be16 l2_proto,int mhoff,int nhoff,int hlen)4176 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4177 			    __be16 l2_proto, int mhoff, int nhoff, int hlen)
4178 {
4179 	struct flow_keys flow;
4180 	u32 hash;
4181 
4182 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4183 		return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4184 
4185 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4186 	    !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4187 		return bond_eth_hash(skb, data, mhoff, hlen);
4188 
4189 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4190 	    bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4191 		hash = bond_eth_hash(skb, data, mhoff, hlen);
4192 	} else {
4193 		if (flow.icmp.id)
4194 			memcpy(&hash, &flow.icmp, sizeof(hash));
4195 		else
4196 			memcpy(&hash, &flow.ports.ports, sizeof(hash));
4197 	}
4198 
4199 	return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4200 }
4201 
4202 /**
4203  * bond_xmit_hash - generate a hash value based on the xmit policy
4204  * @bond: bonding device
4205  * @skb: buffer to use for headers
4206  *
4207  * This function will extract the necessary headers from the skb buffer and use
4208  * them to generate a hash based on the xmit_policy set in the bonding device
4209  */
bond_xmit_hash(struct bonding * bond,struct sk_buff * skb)4210 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4211 {
4212 	if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4213 	    skb->l4_hash)
4214 		return skb->hash;
4215 
4216 	return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4217 				0, skb_network_offset(skb),
4218 				skb_headlen(skb));
4219 }
4220 
4221 /**
4222  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4223  * @bond: bonding device
4224  * @xdp: buffer to use for headers
4225  *
4226  * The XDP variant of bond_xmit_hash.
4227  */
bond_xmit_hash_xdp(struct bonding * bond,struct xdp_buff * xdp)4228 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4229 {
4230 	struct ethhdr *eth;
4231 
4232 	if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4233 		return 0;
4234 
4235 	eth = (struct ethhdr *)xdp->data;
4236 
4237 	return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4238 				sizeof(struct ethhdr), xdp->data_end - xdp->data);
4239 }
4240 
4241 /*-------------------------- Device entry points ----------------------------*/
4242 
bond_work_init_all(struct bonding * bond)4243 void bond_work_init_all(struct bonding *bond)
4244 {
4245 	INIT_DELAYED_WORK(&bond->mcast_work,
4246 			  bond_resend_igmp_join_requests_delayed);
4247 	INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4248 	INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4249 	INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4250 	INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4251 	INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4252 }
4253 
bond_work_cancel_all(struct bonding * bond)4254 static void bond_work_cancel_all(struct bonding *bond)
4255 {
4256 	cancel_delayed_work_sync(&bond->mii_work);
4257 	cancel_delayed_work_sync(&bond->arp_work);
4258 	cancel_delayed_work_sync(&bond->alb_work);
4259 	cancel_delayed_work_sync(&bond->ad_work);
4260 	cancel_delayed_work_sync(&bond->mcast_work);
4261 	cancel_delayed_work_sync(&bond->slave_arr_work);
4262 }
4263 
bond_open(struct net_device * bond_dev)4264 static int bond_open(struct net_device *bond_dev)
4265 {
4266 	struct bonding *bond = netdev_priv(bond_dev);
4267 	struct list_head *iter;
4268 	struct slave *slave;
4269 
4270 	if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4271 		bond->rr_tx_counter = alloc_percpu(u32);
4272 		if (!bond->rr_tx_counter)
4273 			return -ENOMEM;
4274 	}
4275 
4276 	/* reset slave->backup and slave->inactive */
4277 	if (bond_has_slaves(bond)) {
4278 		bond_for_each_slave(bond, slave, iter) {
4279 			if (bond_uses_primary(bond) &&
4280 			    slave != rcu_access_pointer(bond->curr_active_slave)) {
4281 				bond_set_slave_inactive_flags(slave,
4282 							      BOND_SLAVE_NOTIFY_NOW);
4283 			} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4284 				bond_set_slave_active_flags(slave,
4285 							    BOND_SLAVE_NOTIFY_NOW);
4286 			}
4287 		}
4288 	}
4289 
4290 	if (bond_is_lb(bond)) {
4291 		/* bond_alb_initialize must be called before the timer
4292 		 * is started.
4293 		 */
4294 		if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4295 			return -ENOMEM;
4296 		if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4297 			queue_delayed_work(bond->wq, &bond->alb_work, 0);
4298 	}
4299 
4300 	if (bond->params.miimon)  /* link check interval, in milliseconds. */
4301 		queue_delayed_work(bond->wq, &bond->mii_work, 0);
4302 
4303 	if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4304 		queue_delayed_work(bond->wq, &bond->arp_work, 0);
4305 		bond->recv_probe = bond_rcv_validate;
4306 	}
4307 
4308 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4309 		queue_delayed_work(bond->wq, &bond->ad_work, 0);
4310 		/* register to receive LACPDUs */
4311 		bond->recv_probe = bond_3ad_lacpdu_recv;
4312 		bond_3ad_initiate_agg_selection(bond, 1);
4313 
4314 		bond_for_each_slave(bond, slave, iter)
4315 			dev_mc_add(slave->dev, lacpdu_mcast_addr);
4316 	}
4317 
4318 	if (bond_mode_can_use_xmit_hash(bond))
4319 		bond_update_slave_arr(bond, NULL);
4320 
4321 	return 0;
4322 }
4323 
bond_close(struct net_device * bond_dev)4324 static int bond_close(struct net_device *bond_dev)
4325 {
4326 	struct bonding *bond = netdev_priv(bond_dev);
4327 	struct slave *slave;
4328 
4329 	bond_work_cancel_all(bond);
4330 	bond->send_peer_notif = 0;
4331 	if (bond_is_lb(bond))
4332 		bond_alb_deinitialize(bond);
4333 	bond->recv_probe = NULL;
4334 
4335 	if (bond_uses_primary(bond)) {
4336 		rcu_read_lock();
4337 		slave = rcu_dereference(bond->curr_active_slave);
4338 		if (slave)
4339 			bond_hw_addr_flush(bond_dev, slave->dev);
4340 		rcu_read_unlock();
4341 	} else {
4342 		struct list_head *iter;
4343 
4344 		bond_for_each_slave(bond, slave, iter)
4345 			bond_hw_addr_flush(bond_dev, slave->dev);
4346 	}
4347 
4348 	return 0;
4349 }
4350 
4351 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4352  * that some drivers can provide 32bit values only.
4353  */
bond_fold_stats(struct rtnl_link_stats64 * _res,const struct rtnl_link_stats64 * _new,const struct rtnl_link_stats64 * _old)4354 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4355 			    const struct rtnl_link_stats64 *_new,
4356 			    const struct rtnl_link_stats64 *_old)
4357 {
4358 	const u64 *new = (const u64 *)_new;
4359 	const u64 *old = (const u64 *)_old;
4360 	u64 *res = (u64 *)_res;
4361 	int i;
4362 
4363 	for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4364 		u64 nv = new[i];
4365 		u64 ov = old[i];
4366 		s64 delta = nv - ov;
4367 
4368 		/* detects if this particular field is 32bit only */
4369 		if (((nv | ov) >> 32) == 0)
4370 			delta = (s64)(s32)((u32)nv - (u32)ov);
4371 
4372 		/* filter anomalies, some drivers reset their stats
4373 		 * at down/up events.
4374 		 */
4375 		if (delta > 0)
4376 			res[i] += delta;
4377 	}
4378 }
4379 
4380 #ifdef CONFIG_LOCKDEP
bond_get_lowest_level_rcu(struct net_device * dev)4381 static int bond_get_lowest_level_rcu(struct net_device *dev)
4382 {
4383 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4384 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4385 	int cur = 0, max = 0;
4386 
4387 	now = dev;
4388 	iter = &dev->adj_list.lower;
4389 
4390 	while (1) {
4391 		next = NULL;
4392 		while (1) {
4393 			ldev = netdev_next_lower_dev_rcu(now, &iter);
4394 			if (!ldev)
4395 				break;
4396 
4397 			next = ldev;
4398 			niter = &ldev->adj_list.lower;
4399 			dev_stack[cur] = now;
4400 			iter_stack[cur++] = iter;
4401 			if (max <= cur)
4402 				max = cur;
4403 			break;
4404 		}
4405 
4406 		if (!next) {
4407 			if (!cur)
4408 				return max;
4409 			next = dev_stack[--cur];
4410 			niter = iter_stack[cur];
4411 		}
4412 
4413 		now = next;
4414 		iter = niter;
4415 	}
4416 
4417 	return max;
4418 }
4419 #endif
4420 
bond_get_stats(struct net_device * bond_dev,struct rtnl_link_stats64 * stats)4421 static void bond_get_stats(struct net_device *bond_dev,
4422 			   struct rtnl_link_stats64 *stats)
4423 {
4424 	struct bonding *bond = netdev_priv(bond_dev);
4425 	struct rtnl_link_stats64 temp;
4426 	struct list_head *iter;
4427 	struct slave *slave;
4428 	int nest_level = 0;
4429 
4430 
4431 	rcu_read_lock();
4432 #ifdef CONFIG_LOCKDEP
4433 	nest_level = bond_get_lowest_level_rcu(bond_dev);
4434 #endif
4435 
4436 	spin_lock_nested(&bond->stats_lock, nest_level);
4437 	memcpy(stats, &bond->bond_stats, sizeof(*stats));
4438 
4439 	bond_for_each_slave_rcu(bond, slave, iter) {
4440 		const struct rtnl_link_stats64 *new =
4441 			dev_get_stats(slave->dev, &temp);
4442 
4443 		bond_fold_stats(stats, new, &slave->slave_stats);
4444 
4445 		/* save off the slave stats for the next run */
4446 		memcpy(&slave->slave_stats, new, sizeof(*new));
4447 	}
4448 
4449 	memcpy(&bond->bond_stats, stats, sizeof(*stats));
4450 	spin_unlock(&bond->stats_lock);
4451 	rcu_read_unlock();
4452 }
4453 
bond_eth_ioctl(struct net_device * bond_dev,struct ifreq * ifr,int cmd)4454 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4455 {
4456 	struct bonding *bond = netdev_priv(bond_dev);
4457 	struct mii_ioctl_data *mii = NULL;
4458 
4459 	netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4460 
4461 	switch (cmd) {
4462 	case SIOCGMIIPHY:
4463 		mii = if_mii(ifr);
4464 		if (!mii)
4465 			return -EINVAL;
4466 
4467 		mii->phy_id = 0;
4468 		fallthrough;
4469 	case SIOCGMIIREG:
4470 		/* We do this again just in case we were called by SIOCGMIIREG
4471 		 * instead of SIOCGMIIPHY.
4472 		 */
4473 		mii = if_mii(ifr);
4474 		if (!mii)
4475 			return -EINVAL;
4476 
4477 		if (mii->reg_num == 1) {
4478 			mii->val_out = 0;
4479 			if (netif_carrier_ok(bond->dev))
4480 				mii->val_out = BMSR_LSTATUS;
4481 		}
4482 
4483 		break;
4484 	default:
4485 		return -EOPNOTSUPP;
4486 	}
4487 
4488 	return 0;
4489 }
4490 
bond_do_ioctl(struct net_device * bond_dev,struct ifreq * ifr,int cmd)4491 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4492 {
4493 	struct bonding *bond = netdev_priv(bond_dev);
4494 	struct net_device *slave_dev = NULL;
4495 	struct ifbond k_binfo;
4496 	struct ifbond __user *u_binfo = NULL;
4497 	struct ifslave k_sinfo;
4498 	struct ifslave __user *u_sinfo = NULL;
4499 	struct bond_opt_value newval;
4500 	struct net *net;
4501 	int res = 0;
4502 
4503 	netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4504 
4505 	switch (cmd) {
4506 	case SIOCBONDINFOQUERY:
4507 		u_binfo = (struct ifbond __user *)ifr->ifr_data;
4508 
4509 		if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4510 			return -EFAULT;
4511 
4512 		bond_info_query(bond_dev, &k_binfo);
4513 		if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4514 			return -EFAULT;
4515 
4516 		return 0;
4517 	case SIOCBONDSLAVEINFOQUERY:
4518 		u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4519 
4520 		if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4521 			return -EFAULT;
4522 
4523 		res = bond_slave_info_query(bond_dev, &k_sinfo);
4524 		if (res == 0 &&
4525 		    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4526 			return -EFAULT;
4527 
4528 		return res;
4529 	default:
4530 		break;
4531 	}
4532 
4533 	net = dev_net(bond_dev);
4534 
4535 	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4536 		return -EPERM;
4537 
4538 	slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4539 
4540 	slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4541 
4542 	if (!slave_dev)
4543 		return -ENODEV;
4544 
4545 	switch (cmd) {
4546 	case SIOCBONDENSLAVE:
4547 		res = bond_enslave(bond_dev, slave_dev, NULL);
4548 		break;
4549 	case SIOCBONDRELEASE:
4550 		res = bond_release(bond_dev, slave_dev);
4551 		break;
4552 	case SIOCBONDSETHWADDR:
4553 		res = bond_set_dev_addr(bond_dev, slave_dev);
4554 		break;
4555 	case SIOCBONDCHANGEACTIVE:
4556 		bond_opt_initstr(&newval, slave_dev->name);
4557 		res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4558 					    &newval);
4559 		break;
4560 	default:
4561 		res = -EOPNOTSUPP;
4562 	}
4563 
4564 	return res;
4565 }
4566 
bond_siocdevprivate(struct net_device * bond_dev,struct ifreq * ifr,void __user * data,int cmd)4567 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4568 			       void __user *data, int cmd)
4569 {
4570 	struct ifreq ifrdata = { .ifr_data = data };
4571 
4572 	switch (cmd) {
4573 	case BOND_INFO_QUERY_OLD:
4574 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4575 	case BOND_SLAVE_INFO_QUERY_OLD:
4576 		return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4577 	case BOND_ENSLAVE_OLD:
4578 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4579 	case BOND_RELEASE_OLD:
4580 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4581 	case BOND_SETHWADDR_OLD:
4582 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4583 	case BOND_CHANGE_ACTIVE_OLD:
4584 		return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4585 	}
4586 
4587 	return -EOPNOTSUPP;
4588 }
4589 
bond_change_rx_flags(struct net_device * bond_dev,int change)4590 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4591 {
4592 	struct bonding *bond = netdev_priv(bond_dev);
4593 
4594 	if (change & IFF_PROMISC)
4595 		bond_set_promiscuity(bond,
4596 				     bond_dev->flags & IFF_PROMISC ? 1 : -1);
4597 
4598 	if (change & IFF_ALLMULTI)
4599 		bond_set_allmulti(bond,
4600 				  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4601 }
4602 
bond_set_rx_mode(struct net_device * bond_dev)4603 static void bond_set_rx_mode(struct net_device *bond_dev)
4604 {
4605 	struct bonding *bond = netdev_priv(bond_dev);
4606 	struct list_head *iter;
4607 	struct slave *slave;
4608 
4609 	rcu_read_lock();
4610 	if (bond_uses_primary(bond)) {
4611 		slave = rcu_dereference(bond->curr_active_slave);
4612 		if (slave) {
4613 			dev_uc_sync(slave->dev, bond_dev);
4614 			dev_mc_sync(slave->dev, bond_dev);
4615 		}
4616 	} else {
4617 		bond_for_each_slave_rcu(bond, slave, iter) {
4618 			dev_uc_sync_multiple(slave->dev, bond_dev);
4619 			dev_mc_sync_multiple(slave->dev, bond_dev);
4620 		}
4621 	}
4622 	rcu_read_unlock();
4623 }
4624 
bond_neigh_init(struct neighbour * n)4625 static int bond_neigh_init(struct neighbour *n)
4626 {
4627 	struct bonding *bond = netdev_priv(n->dev);
4628 	const struct net_device_ops *slave_ops;
4629 	struct neigh_parms parms;
4630 	struct slave *slave;
4631 	int ret = 0;
4632 
4633 	rcu_read_lock();
4634 	slave = bond_first_slave_rcu(bond);
4635 	if (!slave)
4636 		goto out;
4637 	slave_ops = slave->dev->netdev_ops;
4638 	if (!slave_ops->ndo_neigh_setup)
4639 		goto out;
4640 
4641 	/* TODO: find another way [1] to implement this.
4642 	 * Passing a zeroed structure is fragile,
4643 	 * but at least we do not pass garbage.
4644 	 *
4645 	 * [1] One way would be that ndo_neigh_setup() never touch
4646 	 *     struct neigh_parms, but propagate the new neigh_setup()
4647 	 *     back to ___neigh_create() / neigh_parms_alloc()
4648 	 */
4649 	memset(&parms, 0, sizeof(parms));
4650 	ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4651 
4652 	if (ret)
4653 		goto out;
4654 
4655 	if (parms.neigh_setup)
4656 		ret = parms.neigh_setup(n);
4657 out:
4658 	rcu_read_unlock();
4659 	return ret;
4660 }
4661 
4662 /* The bonding ndo_neigh_setup is called at init time beofre any
4663  * slave exists. So we must declare proxy setup function which will
4664  * be used at run time to resolve the actual slave neigh param setup.
4665  *
4666  * It's also called by master devices (such as vlans) to setup their
4667  * underlying devices. In that case - do nothing, we're already set up from
4668  * our init.
4669  */
bond_neigh_setup(struct net_device * dev,struct neigh_parms * parms)4670 static int bond_neigh_setup(struct net_device *dev,
4671 			    struct neigh_parms *parms)
4672 {
4673 	/* modify only our neigh_parms */
4674 	if (parms->dev == dev)
4675 		parms->neigh_setup = bond_neigh_init;
4676 
4677 	return 0;
4678 }
4679 
4680 /* Change the MTU of all of a master's slaves to match the master */
bond_change_mtu(struct net_device * bond_dev,int new_mtu)4681 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4682 {
4683 	struct bonding *bond = netdev_priv(bond_dev);
4684 	struct slave *slave, *rollback_slave;
4685 	struct list_head *iter;
4686 	int res = 0;
4687 
4688 	netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4689 
4690 	bond_for_each_slave(bond, slave, iter) {
4691 		slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4692 			   slave, slave->dev->netdev_ops->ndo_change_mtu);
4693 
4694 		res = dev_set_mtu(slave->dev, new_mtu);
4695 
4696 		if (res) {
4697 			/* If we failed to set the slave's mtu to the new value
4698 			 * we must abort the operation even in ACTIVE_BACKUP
4699 			 * mode, because if we allow the backup slaves to have
4700 			 * different mtu values than the active slave we'll
4701 			 * need to change their mtu when doing a failover. That
4702 			 * means changing their mtu from timer context, which
4703 			 * is probably not a good idea.
4704 			 */
4705 			slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4706 				  res, new_mtu);
4707 			goto unwind;
4708 		}
4709 	}
4710 
4711 	bond_dev->mtu = new_mtu;
4712 
4713 	return 0;
4714 
4715 unwind:
4716 	/* unwind from head to the slave that failed */
4717 	bond_for_each_slave(bond, rollback_slave, iter) {
4718 		int tmp_res;
4719 
4720 		if (rollback_slave == slave)
4721 			break;
4722 
4723 		tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4724 		if (tmp_res)
4725 			slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4726 				  tmp_res);
4727 	}
4728 
4729 	return res;
4730 }
4731 
4732 /* Change HW address
4733  *
4734  * Note that many devices must be down to change the HW address, and
4735  * downing the master releases all slaves.  We can make bonds full of
4736  * bonding devices to test this, however.
4737  */
bond_set_mac_address(struct net_device * bond_dev,void * addr)4738 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4739 {
4740 	struct bonding *bond = netdev_priv(bond_dev);
4741 	struct slave *slave, *rollback_slave;
4742 	struct sockaddr_storage *ss = addr, tmp_ss;
4743 	struct list_head *iter;
4744 	int res = 0;
4745 
4746 	if (BOND_MODE(bond) == BOND_MODE_ALB)
4747 		return bond_alb_set_mac_address(bond_dev, addr);
4748 
4749 
4750 	netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4751 
4752 	/* If fail_over_mac is enabled, do nothing and return success.
4753 	 * Returning an error causes ifenslave to fail.
4754 	 */
4755 	if (bond->params.fail_over_mac &&
4756 	    BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4757 		return 0;
4758 
4759 	if (!is_valid_ether_addr(ss->__data))
4760 		return -EADDRNOTAVAIL;
4761 
4762 	bond_for_each_slave(bond, slave, iter) {
4763 		slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4764 			  __func__, slave);
4765 		res = dev_set_mac_address(slave->dev, addr, NULL);
4766 		if (res) {
4767 			/* TODO: consider downing the slave
4768 			 * and retry ?
4769 			 * User should expect communications
4770 			 * breakage anyway until ARP finish
4771 			 * updating, so...
4772 			 */
4773 			slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4774 				  __func__, res);
4775 			goto unwind;
4776 		}
4777 	}
4778 
4779 	/* success */
4780 	dev_addr_set(bond_dev, ss->__data);
4781 	return 0;
4782 
4783 unwind:
4784 	memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4785 	tmp_ss.ss_family = bond_dev->type;
4786 
4787 	/* unwind from head to the slave that failed */
4788 	bond_for_each_slave(bond, rollback_slave, iter) {
4789 		int tmp_res;
4790 
4791 		if (rollback_slave == slave)
4792 			break;
4793 
4794 		tmp_res = dev_set_mac_address(rollback_slave->dev,
4795 					      (struct sockaddr *)&tmp_ss, NULL);
4796 		if (tmp_res) {
4797 			slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4798 				   __func__, tmp_res);
4799 		}
4800 	}
4801 
4802 	return res;
4803 }
4804 
4805 /**
4806  * bond_get_slave_by_id - get xmit slave with slave_id
4807  * @bond: bonding device that is transmitting
4808  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4809  *
4810  * This function tries to get slave with slave_id but in case
4811  * it fails, it tries to find the first available slave for transmission.
4812  */
bond_get_slave_by_id(struct bonding * bond,int slave_id)4813 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4814 					  int slave_id)
4815 {
4816 	struct list_head *iter;
4817 	struct slave *slave;
4818 	int i = slave_id;
4819 
4820 	/* Here we start from the slave with slave_id */
4821 	bond_for_each_slave_rcu(bond, slave, iter) {
4822 		if (--i < 0) {
4823 			if (bond_slave_can_tx(slave))
4824 				return slave;
4825 		}
4826 	}
4827 
4828 	/* Here we start from the first slave up to slave_id */
4829 	i = slave_id;
4830 	bond_for_each_slave_rcu(bond, slave, iter) {
4831 		if (--i < 0)
4832 			break;
4833 		if (bond_slave_can_tx(slave))
4834 			return slave;
4835 	}
4836 	/* no slave that can tx has been found */
4837 	return NULL;
4838 }
4839 
4840 /**
4841  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4842  * @bond: bonding device to use
4843  *
4844  * Based on the value of the bonding device's packets_per_slave parameter
4845  * this function generates a slave id, which is usually used as the next
4846  * slave to transmit through.
4847  */
bond_rr_gen_slave_id(struct bonding * bond)4848 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4849 {
4850 	u32 slave_id;
4851 	struct reciprocal_value reciprocal_packets_per_slave;
4852 	int packets_per_slave = bond->params.packets_per_slave;
4853 
4854 	switch (packets_per_slave) {
4855 	case 0:
4856 		slave_id = get_random_u32();
4857 		break;
4858 	case 1:
4859 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4860 		break;
4861 	default:
4862 		reciprocal_packets_per_slave =
4863 			bond->params.reciprocal_packets_per_slave;
4864 		slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4865 		slave_id = reciprocal_divide(slave_id,
4866 					     reciprocal_packets_per_slave);
4867 		break;
4868 	}
4869 
4870 	return slave_id;
4871 }
4872 
bond_xmit_roundrobin_slave_get(struct bonding * bond,struct sk_buff * skb)4873 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4874 						    struct sk_buff *skb)
4875 {
4876 	struct slave *slave;
4877 	int slave_cnt;
4878 	u32 slave_id;
4879 
4880 	/* Start with the curr_active_slave that joined the bond as the
4881 	 * default for sending IGMP traffic.  For failover purposes one
4882 	 * needs to maintain some consistency for the interface that will
4883 	 * send the join/membership reports.  The curr_active_slave found
4884 	 * will send all of this type of traffic.
4885 	 */
4886 	if (skb->protocol == htons(ETH_P_IP)) {
4887 		int noff = skb_network_offset(skb);
4888 		struct iphdr *iph;
4889 
4890 		if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4891 			goto non_igmp;
4892 
4893 		iph = ip_hdr(skb);
4894 		if (iph->protocol == IPPROTO_IGMP) {
4895 			slave = rcu_dereference(bond->curr_active_slave);
4896 			if (slave)
4897 				return slave;
4898 			return bond_get_slave_by_id(bond, 0);
4899 		}
4900 	}
4901 
4902 non_igmp:
4903 	slave_cnt = READ_ONCE(bond->slave_cnt);
4904 	if (likely(slave_cnt)) {
4905 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4906 		return bond_get_slave_by_id(bond, slave_id);
4907 	}
4908 	return NULL;
4909 }
4910 
bond_xdp_xmit_roundrobin_slave_get(struct bonding * bond,struct xdp_buff * xdp)4911 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4912 							struct xdp_buff *xdp)
4913 {
4914 	struct slave *slave;
4915 	int slave_cnt;
4916 	u32 slave_id;
4917 	const struct ethhdr *eth;
4918 	void *data = xdp->data;
4919 
4920 	if (data + sizeof(struct ethhdr) > xdp->data_end)
4921 		goto non_igmp;
4922 
4923 	eth = (struct ethhdr *)data;
4924 	data += sizeof(struct ethhdr);
4925 
4926 	/* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4927 	if (eth->h_proto == htons(ETH_P_IP)) {
4928 		const struct iphdr *iph;
4929 
4930 		if (data + sizeof(struct iphdr) > xdp->data_end)
4931 			goto non_igmp;
4932 
4933 		iph = (struct iphdr *)data;
4934 
4935 		if (iph->protocol == IPPROTO_IGMP) {
4936 			slave = rcu_dereference(bond->curr_active_slave);
4937 			if (slave)
4938 				return slave;
4939 			return bond_get_slave_by_id(bond, 0);
4940 		}
4941 	}
4942 
4943 non_igmp:
4944 	slave_cnt = READ_ONCE(bond->slave_cnt);
4945 	if (likely(slave_cnt)) {
4946 		slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4947 		return bond_get_slave_by_id(bond, slave_id);
4948 	}
4949 	return NULL;
4950 }
4951 
bond_xmit_roundrobin(struct sk_buff * skb,struct net_device * bond_dev)4952 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4953 					struct net_device *bond_dev)
4954 {
4955 	struct bonding *bond = netdev_priv(bond_dev);
4956 	struct slave *slave;
4957 
4958 	slave = bond_xmit_roundrobin_slave_get(bond, skb);
4959 	if (likely(slave))
4960 		return bond_dev_queue_xmit(bond, skb, slave->dev);
4961 
4962 	return bond_tx_drop(bond_dev, skb);
4963 }
4964 
bond_xmit_activebackup_slave_get(struct bonding * bond)4965 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4966 {
4967 	return rcu_dereference(bond->curr_active_slave);
4968 }
4969 
4970 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4971  * the bond has a usable interface.
4972  */
bond_xmit_activebackup(struct sk_buff * skb,struct net_device * bond_dev)4973 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4974 					  struct net_device *bond_dev)
4975 {
4976 	struct bonding *bond = netdev_priv(bond_dev);
4977 	struct slave *slave;
4978 
4979 	slave = bond_xmit_activebackup_slave_get(bond);
4980 	if (slave)
4981 		return bond_dev_queue_xmit(bond, skb, slave->dev);
4982 
4983 	return bond_tx_drop(bond_dev, skb);
4984 }
4985 
4986 /* Use this to update slave_array when (a) it's not appropriate to update
4987  * slave_array right away (note that update_slave_array() may sleep)
4988  * and / or (b) RTNL is not held.
4989  */
bond_slave_arr_work_rearm(struct bonding * bond,unsigned long delay)4990 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4991 {
4992 	queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4993 }
4994 
4995 /* Slave array work handler. Holds only RTNL */
bond_slave_arr_handler(struct work_struct * work)4996 static void bond_slave_arr_handler(struct work_struct *work)
4997 {
4998 	struct bonding *bond = container_of(work, struct bonding,
4999 					    slave_arr_work.work);
5000 	int ret;
5001 
5002 	if (!rtnl_trylock())
5003 		goto err;
5004 
5005 	ret = bond_update_slave_arr(bond, NULL);
5006 	rtnl_unlock();
5007 	if (ret) {
5008 		pr_warn_ratelimited("Failed to update slave array from WT\n");
5009 		goto err;
5010 	}
5011 	return;
5012 
5013 err:
5014 	bond_slave_arr_work_rearm(bond, 1);
5015 }
5016 
bond_skip_slave(struct bond_up_slave * slaves,struct slave * skipslave)5017 static void bond_skip_slave(struct bond_up_slave *slaves,
5018 			    struct slave *skipslave)
5019 {
5020 	int idx;
5021 
5022 	/* Rare situation where caller has asked to skip a specific
5023 	 * slave but allocation failed (most likely!). BTW this is
5024 	 * only possible when the call is initiated from
5025 	 * __bond_release_one(). In this situation; overwrite the
5026 	 * skipslave entry in the array with the last entry from the
5027 	 * array to avoid a situation where the xmit path may choose
5028 	 * this to-be-skipped slave to send a packet out.
5029 	 */
5030 	for (idx = 0; slaves && idx < slaves->count; idx++) {
5031 		if (skipslave == slaves->arr[idx]) {
5032 			slaves->arr[idx] =
5033 				slaves->arr[slaves->count - 1];
5034 			slaves->count--;
5035 			break;
5036 		}
5037 	}
5038 }
5039 
bond_set_slave_arr(struct bonding * bond,struct bond_up_slave * usable_slaves,struct bond_up_slave * all_slaves)5040 static void bond_set_slave_arr(struct bonding *bond,
5041 			       struct bond_up_slave *usable_slaves,
5042 			       struct bond_up_slave *all_slaves)
5043 {
5044 	struct bond_up_slave *usable, *all;
5045 
5046 	usable = rtnl_dereference(bond->usable_slaves);
5047 	rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5048 	kfree_rcu(usable, rcu);
5049 
5050 	all = rtnl_dereference(bond->all_slaves);
5051 	rcu_assign_pointer(bond->all_slaves, all_slaves);
5052 	kfree_rcu(all, rcu);
5053 }
5054 
bond_reset_slave_arr(struct bonding * bond)5055 static void bond_reset_slave_arr(struct bonding *bond)
5056 {
5057 	bond_set_slave_arr(bond, NULL, NULL);
5058 }
5059 
5060 /* Build the usable slaves array in control path for modes that use xmit-hash
5061  * to determine the slave interface -
5062  * (a) BOND_MODE_8023AD
5063  * (b) BOND_MODE_XOR
5064  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5065  *
5066  * The caller is expected to hold RTNL only and NO other lock!
5067  */
bond_update_slave_arr(struct bonding * bond,struct slave * skipslave)5068 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5069 {
5070 	struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5071 	struct slave *slave;
5072 	struct list_head *iter;
5073 	int agg_id = 0;
5074 	int ret = 0;
5075 
5076 	might_sleep();
5077 
5078 	usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5079 					    bond->slave_cnt), GFP_KERNEL);
5080 	all_slaves = kzalloc(struct_size(all_slaves, arr,
5081 					 bond->slave_cnt), GFP_KERNEL);
5082 	if (!usable_slaves || !all_slaves) {
5083 		ret = -ENOMEM;
5084 		goto out;
5085 	}
5086 	if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5087 		struct ad_info ad_info;
5088 
5089 		spin_lock_bh(&bond->mode_lock);
5090 		if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5091 			spin_unlock_bh(&bond->mode_lock);
5092 			pr_debug("bond_3ad_get_active_agg_info failed\n");
5093 			/* No active aggragator means it's not safe to use
5094 			 * the previous array.
5095 			 */
5096 			bond_reset_slave_arr(bond);
5097 			goto out;
5098 		}
5099 		spin_unlock_bh(&bond->mode_lock);
5100 		agg_id = ad_info.aggregator_id;
5101 	}
5102 	bond_for_each_slave(bond, slave, iter) {
5103 		if (skipslave == slave)
5104 			continue;
5105 
5106 		all_slaves->arr[all_slaves->count++] = slave;
5107 		if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5108 			struct aggregator *agg;
5109 
5110 			agg = SLAVE_AD_INFO(slave)->port.aggregator;
5111 			if (!agg || agg->aggregator_identifier != agg_id)
5112 				continue;
5113 		}
5114 		if (!bond_slave_can_tx(slave))
5115 			continue;
5116 
5117 		slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5118 			  usable_slaves->count);
5119 
5120 		usable_slaves->arr[usable_slaves->count++] = slave;
5121 	}
5122 
5123 	bond_set_slave_arr(bond, usable_slaves, all_slaves);
5124 	return ret;
5125 out:
5126 	if (ret != 0 && skipslave) {
5127 		bond_skip_slave(rtnl_dereference(bond->all_slaves),
5128 				skipslave);
5129 		bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5130 				skipslave);
5131 	}
5132 	kfree_rcu(all_slaves, rcu);
5133 	kfree_rcu(usable_slaves, rcu);
5134 
5135 	return ret;
5136 }
5137 
bond_xmit_3ad_xor_slave_get(struct bonding * bond,struct sk_buff * skb,struct bond_up_slave * slaves)5138 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5139 						 struct sk_buff *skb,
5140 						 struct bond_up_slave *slaves)
5141 {
5142 	struct slave *slave;
5143 	unsigned int count;
5144 	u32 hash;
5145 
5146 	hash = bond_xmit_hash(bond, skb);
5147 	count = slaves ? READ_ONCE(slaves->count) : 0;
5148 	if (unlikely(!count))
5149 		return NULL;
5150 
5151 	slave = slaves->arr[hash % count];
5152 	return slave;
5153 }
5154 
bond_xdp_xmit_3ad_xor_slave_get(struct bonding * bond,struct xdp_buff * xdp)5155 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5156 						     struct xdp_buff *xdp)
5157 {
5158 	struct bond_up_slave *slaves;
5159 	unsigned int count;
5160 	u32 hash;
5161 
5162 	hash = bond_xmit_hash_xdp(bond, xdp);
5163 	slaves = rcu_dereference(bond->usable_slaves);
5164 	count = slaves ? READ_ONCE(slaves->count) : 0;
5165 	if (unlikely(!count))
5166 		return NULL;
5167 
5168 	return slaves->arr[hash % count];
5169 }
5170 
5171 /* Use this Xmit function for 3AD as well as XOR modes. The current
5172  * usable slave array is formed in the control path. The xmit function
5173  * just calculates hash and sends the packet out.
5174  */
bond_3ad_xor_xmit(struct sk_buff * skb,struct net_device * dev)5175 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5176 				     struct net_device *dev)
5177 {
5178 	struct bonding *bond = netdev_priv(dev);
5179 	struct bond_up_slave *slaves;
5180 	struct slave *slave;
5181 
5182 	slaves = rcu_dereference(bond->usable_slaves);
5183 	slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5184 	if (likely(slave))
5185 		return bond_dev_queue_xmit(bond, skb, slave->dev);
5186 
5187 	return bond_tx_drop(dev, skb);
5188 }
5189 
5190 /* in broadcast mode, we send everything to all usable interfaces. */
bond_xmit_broadcast(struct sk_buff * skb,struct net_device * bond_dev)5191 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5192 				       struct net_device *bond_dev)
5193 {
5194 	struct bonding *bond = netdev_priv(bond_dev);
5195 	struct slave *slave = NULL;
5196 	struct list_head *iter;
5197 	bool xmit_suc = false;
5198 	bool skb_used = false;
5199 
5200 	bond_for_each_slave_rcu(bond, slave, iter) {
5201 		struct sk_buff *skb2;
5202 
5203 		if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5204 			continue;
5205 
5206 		if (bond_is_last_slave(bond, slave)) {
5207 			skb2 = skb;
5208 			skb_used = true;
5209 		} else {
5210 			skb2 = skb_clone(skb, GFP_ATOMIC);
5211 			if (!skb2) {
5212 				net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5213 						    bond_dev->name, __func__);
5214 				continue;
5215 			}
5216 		}
5217 
5218 		if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5219 			xmit_suc = true;
5220 	}
5221 
5222 	if (!skb_used)
5223 		dev_kfree_skb_any(skb);
5224 
5225 	if (xmit_suc)
5226 		return NETDEV_TX_OK;
5227 
5228 	dev_core_stats_tx_dropped_inc(bond_dev);
5229 	return NET_XMIT_DROP;
5230 }
5231 
5232 /*------------------------- Device initialization ---------------------------*/
5233 
5234 /* Lookup the slave that corresponds to a qid */
bond_slave_override(struct bonding * bond,struct sk_buff * skb)5235 static inline int bond_slave_override(struct bonding *bond,
5236 				      struct sk_buff *skb)
5237 {
5238 	struct slave *slave = NULL;
5239 	struct list_head *iter;
5240 
5241 	if (!skb_rx_queue_recorded(skb))
5242 		return 1;
5243 
5244 	/* Find out if any slaves have the same mapping as this skb. */
5245 	bond_for_each_slave_rcu(bond, slave, iter) {
5246 		if (slave->queue_id == skb_get_queue_mapping(skb)) {
5247 			if (bond_slave_is_up(slave) &&
5248 			    slave->link == BOND_LINK_UP) {
5249 				bond_dev_queue_xmit(bond, skb, slave->dev);
5250 				return 0;
5251 			}
5252 			/* If the slave isn't UP, use default transmit policy. */
5253 			break;
5254 		}
5255 	}
5256 
5257 	return 1;
5258 }
5259 
5260 
bond_select_queue(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)5261 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5262 			     struct net_device *sb_dev)
5263 {
5264 	/* This helper function exists to help dev_pick_tx get the correct
5265 	 * destination queue.  Using a helper function skips a call to
5266 	 * skb_tx_hash and will put the skbs in the queue we expect on their
5267 	 * way down to the bonding driver.
5268 	 */
5269 	u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5270 
5271 	/* Save the original txq to restore before passing to the driver */
5272 	qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5273 
5274 	if (unlikely(txq >= dev->real_num_tx_queues)) {
5275 		do {
5276 			txq -= dev->real_num_tx_queues;
5277 		} while (txq >= dev->real_num_tx_queues);
5278 	}
5279 	return txq;
5280 }
5281 
bond_xmit_get_slave(struct net_device * master_dev,struct sk_buff * skb,bool all_slaves)5282 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5283 					      struct sk_buff *skb,
5284 					      bool all_slaves)
5285 {
5286 	struct bonding *bond = netdev_priv(master_dev);
5287 	struct bond_up_slave *slaves;
5288 	struct slave *slave = NULL;
5289 
5290 	switch (BOND_MODE(bond)) {
5291 	case BOND_MODE_ROUNDROBIN:
5292 		slave = bond_xmit_roundrobin_slave_get(bond, skb);
5293 		break;
5294 	case BOND_MODE_ACTIVEBACKUP:
5295 		slave = bond_xmit_activebackup_slave_get(bond);
5296 		break;
5297 	case BOND_MODE_8023AD:
5298 	case BOND_MODE_XOR:
5299 		if (all_slaves)
5300 			slaves = rcu_dereference(bond->all_slaves);
5301 		else
5302 			slaves = rcu_dereference(bond->usable_slaves);
5303 		slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5304 		break;
5305 	case BOND_MODE_BROADCAST:
5306 		break;
5307 	case BOND_MODE_ALB:
5308 		slave = bond_xmit_alb_slave_get(bond, skb);
5309 		break;
5310 	case BOND_MODE_TLB:
5311 		slave = bond_xmit_tlb_slave_get(bond, skb);
5312 		break;
5313 	default:
5314 		/* Should never happen, mode already checked */
5315 		WARN_ONCE(true, "Unknown bonding mode");
5316 		break;
5317 	}
5318 
5319 	if (slave)
5320 		return slave->dev;
5321 	return NULL;
5322 }
5323 
bond_sk_to_flow(struct sock * sk,struct flow_keys * flow)5324 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5325 {
5326 	switch (sk->sk_family) {
5327 #if IS_ENABLED(CONFIG_IPV6)
5328 	case AF_INET6:
5329 		if (ipv6_only_sock(sk) ||
5330 		    ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5331 			flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5332 			flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5333 			flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5334 			break;
5335 		}
5336 		fallthrough;
5337 #endif
5338 	default: /* AF_INET */
5339 		flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5340 		flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5341 		flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5342 		break;
5343 	}
5344 
5345 	flow->ports.src = inet_sk(sk)->inet_sport;
5346 	flow->ports.dst = inet_sk(sk)->inet_dport;
5347 }
5348 
5349 /**
5350  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5351  * @sk: socket to use for headers
5352  *
5353  * This function will extract the necessary field from the socket and use
5354  * them to generate a hash based on the LAYER34 xmit_policy.
5355  * Assumes that sk is a TCP or UDP socket.
5356  */
bond_sk_hash_l34(struct sock * sk)5357 static u32 bond_sk_hash_l34(struct sock *sk)
5358 {
5359 	struct flow_keys flow;
5360 	u32 hash;
5361 
5362 	bond_sk_to_flow(sk, &flow);
5363 
5364 	/* L4 */
5365 	memcpy(&hash, &flow.ports.ports, sizeof(hash));
5366 	/* L3 */
5367 	return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5368 }
5369 
__bond_sk_get_lower_dev(struct bonding * bond,struct sock * sk)5370 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5371 						  struct sock *sk)
5372 {
5373 	struct bond_up_slave *slaves;
5374 	struct slave *slave;
5375 	unsigned int count;
5376 	u32 hash;
5377 
5378 	slaves = rcu_dereference(bond->usable_slaves);
5379 	count = slaves ? READ_ONCE(slaves->count) : 0;
5380 	if (unlikely(!count))
5381 		return NULL;
5382 
5383 	hash = bond_sk_hash_l34(sk);
5384 	slave = slaves->arr[hash % count];
5385 
5386 	return slave->dev;
5387 }
5388 
bond_sk_get_lower_dev(struct net_device * dev,struct sock * sk)5389 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5390 						struct sock *sk)
5391 {
5392 	struct bonding *bond = netdev_priv(dev);
5393 	struct net_device *lower = NULL;
5394 
5395 	rcu_read_lock();
5396 	if (bond_sk_check(bond))
5397 		lower = __bond_sk_get_lower_dev(bond, sk);
5398 	rcu_read_unlock();
5399 
5400 	return lower;
5401 }
5402 
5403 #if IS_ENABLED(CONFIG_TLS_DEVICE)
bond_tls_device_xmit(struct bonding * bond,struct sk_buff * skb,struct net_device * dev)5404 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5405 					struct net_device *dev)
5406 {
5407 	struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5408 
5409 	/* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5410 	 * was true, if tls_device_down is running in parallel, but it's OK,
5411 	 * because bond_get_slave_by_dev has a NULL check.
5412 	 */
5413 	if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5414 		return bond_dev_queue_xmit(bond, skb, tls_netdev);
5415 	return bond_tx_drop(dev, skb);
5416 }
5417 #endif
5418 
__bond_start_xmit(struct sk_buff * skb,struct net_device * dev)5419 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5420 {
5421 	struct bonding *bond = netdev_priv(dev);
5422 
5423 	if (bond_should_override_tx_queue(bond) &&
5424 	    !bond_slave_override(bond, skb))
5425 		return NETDEV_TX_OK;
5426 
5427 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5428 	if (tls_is_skb_tx_device_offloaded(skb))
5429 		return bond_tls_device_xmit(bond, skb, dev);
5430 #endif
5431 
5432 	switch (BOND_MODE(bond)) {
5433 	case BOND_MODE_ROUNDROBIN:
5434 		return bond_xmit_roundrobin(skb, dev);
5435 	case BOND_MODE_ACTIVEBACKUP:
5436 		return bond_xmit_activebackup(skb, dev);
5437 	case BOND_MODE_8023AD:
5438 	case BOND_MODE_XOR:
5439 		return bond_3ad_xor_xmit(skb, dev);
5440 	case BOND_MODE_BROADCAST:
5441 		return bond_xmit_broadcast(skb, dev);
5442 	case BOND_MODE_ALB:
5443 		return bond_alb_xmit(skb, dev);
5444 	case BOND_MODE_TLB:
5445 		return bond_tlb_xmit(skb, dev);
5446 	default:
5447 		/* Should never happen, mode already checked */
5448 		netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5449 		WARN_ON_ONCE(1);
5450 		return bond_tx_drop(dev, skb);
5451 	}
5452 }
5453 
bond_start_xmit(struct sk_buff * skb,struct net_device * dev)5454 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5455 {
5456 	struct bonding *bond = netdev_priv(dev);
5457 	netdev_tx_t ret = NETDEV_TX_OK;
5458 
5459 	/* If we risk deadlock from transmitting this in the
5460 	 * netpoll path, tell netpoll to queue the frame for later tx
5461 	 */
5462 	if (unlikely(is_netpoll_tx_blocked(dev)))
5463 		return NETDEV_TX_BUSY;
5464 
5465 	rcu_read_lock();
5466 	if (bond_has_slaves(bond))
5467 		ret = __bond_start_xmit(skb, dev);
5468 	else
5469 		ret = bond_tx_drop(dev, skb);
5470 	rcu_read_unlock();
5471 
5472 	return ret;
5473 }
5474 
5475 static struct net_device *
bond_xdp_get_xmit_slave(struct net_device * bond_dev,struct xdp_buff * xdp)5476 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5477 {
5478 	struct bonding *bond = netdev_priv(bond_dev);
5479 	struct slave *slave;
5480 
5481 	/* Caller needs to hold rcu_read_lock() */
5482 
5483 	switch (BOND_MODE(bond)) {
5484 	case BOND_MODE_ROUNDROBIN:
5485 		slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5486 		break;
5487 
5488 	case BOND_MODE_ACTIVEBACKUP:
5489 		slave = bond_xmit_activebackup_slave_get(bond);
5490 		break;
5491 
5492 	case BOND_MODE_8023AD:
5493 	case BOND_MODE_XOR:
5494 		slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5495 		break;
5496 
5497 	default:
5498 		/* Should never happen. Mode guarded by bond_xdp_check() */
5499 		netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5500 		WARN_ON_ONCE(1);
5501 		return NULL;
5502 	}
5503 
5504 	if (slave)
5505 		return slave->dev;
5506 
5507 	return NULL;
5508 }
5509 
bond_xdp_xmit(struct net_device * bond_dev,int n,struct xdp_frame ** frames,u32 flags)5510 static int bond_xdp_xmit(struct net_device *bond_dev,
5511 			 int n, struct xdp_frame **frames, u32 flags)
5512 {
5513 	int nxmit, err = -ENXIO;
5514 
5515 	rcu_read_lock();
5516 
5517 	for (nxmit = 0; nxmit < n; nxmit++) {
5518 		struct xdp_frame *frame = frames[nxmit];
5519 		struct xdp_frame *frames1[] = {frame};
5520 		struct net_device *slave_dev;
5521 		struct xdp_buff xdp;
5522 
5523 		xdp_convert_frame_to_buff(frame, &xdp);
5524 
5525 		slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5526 		if (!slave_dev) {
5527 			err = -ENXIO;
5528 			break;
5529 		}
5530 
5531 		err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5532 		if (err < 1)
5533 			break;
5534 	}
5535 
5536 	rcu_read_unlock();
5537 
5538 	/* If error happened on the first frame then we can pass the error up, otherwise
5539 	 * report the number of frames that were xmitted.
5540 	 */
5541 	if (err < 0)
5542 		return (nxmit == 0 ? err : nxmit);
5543 
5544 	return nxmit;
5545 }
5546 
bond_xdp_set(struct net_device * dev,struct bpf_prog * prog,struct netlink_ext_ack * extack)5547 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5548 			struct netlink_ext_ack *extack)
5549 {
5550 	struct bonding *bond = netdev_priv(dev);
5551 	struct list_head *iter;
5552 	struct slave *slave, *rollback_slave;
5553 	struct bpf_prog *old_prog;
5554 	struct netdev_bpf xdp = {
5555 		.command = XDP_SETUP_PROG,
5556 		.flags   = 0,
5557 		.prog    = prog,
5558 		.extack  = extack,
5559 	};
5560 	int err;
5561 
5562 	ASSERT_RTNL();
5563 
5564 	if (!bond_xdp_check(bond))
5565 		return -EOPNOTSUPP;
5566 
5567 	old_prog = bond->xdp_prog;
5568 	bond->xdp_prog = prog;
5569 
5570 	bond_for_each_slave(bond, slave, iter) {
5571 		struct net_device *slave_dev = slave->dev;
5572 
5573 		if (!slave_dev->netdev_ops->ndo_bpf ||
5574 		    !slave_dev->netdev_ops->ndo_xdp_xmit) {
5575 			SLAVE_NL_ERR(dev, slave_dev, extack,
5576 				     "Slave device does not support XDP");
5577 			err = -EOPNOTSUPP;
5578 			goto err;
5579 		}
5580 
5581 		if (dev_xdp_prog_count(slave_dev) > 0) {
5582 			SLAVE_NL_ERR(dev, slave_dev, extack,
5583 				     "Slave has XDP program loaded, please unload before enslaving");
5584 			err = -EOPNOTSUPP;
5585 			goto err;
5586 		}
5587 
5588 		err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5589 		if (err < 0) {
5590 			/* ndo_bpf() sets extack error message */
5591 			slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5592 			goto err;
5593 		}
5594 		if (prog)
5595 			bpf_prog_inc(prog);
5596 	}
5597 
5598 	if (prog) {
5599 		static_branch_inc(&bpf_master_redirect_enabled_key);
5600 	} else if (old_prog) {
5601 		bpf_prog_put(old_prog);
5602 		static_branch_dec(&bpf_master_redirect_enabled_key);
5603 	}
5604 
5605 	return 0;
5606 
5607 err:
5608 	/* unwind the program changes */
5609 	bond->xdp_prog = old_prog;
5610 	xdp.prog = old_prog;
5611 	xdp.extack = NULL; /* do not overwrite original error */
5612 
5613 	bond_for_each_slave(bond, rollback_slave, iter) {
5614 		struct net_device *slave_dev = rollback_slave->dev;
5615 		int err_unwind;
5616 
5617 		if (slave == rollback_slave)
5618 			break;
5619 
5620 		err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5621 		if (err_unwind < 0)
5622 			slave_err(dev, slave_dev,
5623 				  "Error %d when unwinding XDP program change\n", err_unwind);
5624 		else if (xdp.prog)
5625 			bpf_prog_inc(xdp.prog);
5626 	}
5627 	return err;
5628 }
5629 
bond_xdp(struct net_device * dev,struct netdev_bpf * xdp)5630 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5631 {
5632 	switch (xdp->command) {
5633 	case XDP_SETUP_PROG:
5634 		return bond_xdp_set(dev, xdp->prog, xdp->extack);
5635 	default:
5636 		return -EINVAL;
5637 	}
5638 }
5639 
bond_mode_bcast_speed(struct slave * slave,u32 speed)5640 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5641 {
5642 	if (speed == 0 || speed == SPEED_UNKNOWN)
5643 		speed = slave->speed;
5644 	else
5645 		speed = min(speed, slave->speed);
5646 
5647 	return speed;
5648 }
5649 
5650 /* Set the BOND_PHC_INDEX flag to notify user space */
bond_set_phc_index_flag(struct kernel_hwtstamp_config * kernel_cfg)5651 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5652 {
5653 	struct ifreq *ifr = kernel_cfg->ifr;
5654 	struct hwtstamp_config cfg;
5655 
5656 	if (kernel_cfg->copied_to_user) {
5657 		/* Lower device has a legacy implementation */
5658 		if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
5659 			return -EFAULT;
5660 
5661 		cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5662 		if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)))
5663 			return -EFAULT;
5664 	} else {
5665 		kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5666 	}
5667 
5668 	return 0;
5669 }
5670 
bond_hwtstamp_get(struct net_device * dev,struct kernel_hwtstamp_config * cfg)5671 static int bond_hwtstamp_get(struct net_device *dev,
5672 			     struct kernel_hwtstamp_config *cfg)
5673 {
5674 	struct bonding *bond = netdev_priv(dev);
5675 	struct net_device *real_dev;
5676 	int err;
5677 
5678 	real_dev = bond_option_active_slave_get_rcu(bond);
5679 	if (!real_dev)
5680 		return -EOPNOTSUPP;
5681 
5682 	err = generic_hwtstamp_get_lower(real_dev, cfg);
5683 	if (err)
5684 		return err;
5685 
5686 	return bond_set_phc_index_flag(cfg);
5687 }
5688 
bond_hwtstamp_set(struct net_device * dev,struct kernel_hwtstamp_config * cfg,struct netlink_ext_ack * extack)5689 static int bond_hwtstamp_set(struct net_device *dev,
5690 			     struct kernel_hwtstamp_config *cfg,
5691 			     struct netlink_ext_ack *extack)
5692 {
5693 	struct bonding *bond = netdev_priv(dev);
5694 	struct net_device *real_dev;
5695 	int err;
5696 
5697 	if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5698 		return -EOPNOTSUPP;
5699 
5700 	real_dev = bond_option_active_slave_get_rcu(bond);
5701 	if (!real_dev)
5702 		return -EOPNOTSUPP;
5703 
5704 	err = generic_hwtstamp_set_lower(real_dev, cfg, extack);
5705 	if (err)
5706 		return err;
5707 
5708 	return bond_set_phc_index_flag(cfg);
5709 }
5710 
bond_ethtool_get_link_ksettings(struct net_device * bond_dev,struct ethtool_link_ksettings * cmd)5711 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5712 					   struct ethtool_link_ksettings *cmd)
5713 {
5714 	struct bonding *bond = netdev_priv(bond_dev);
5715 	struct list_head *iter;
5716 	struct slave *slave;
5717 	u32 speed = 0;
5718 
5719 	cmd->base.duplex = DUPLEX_UNKNOWN;
5720 	cmd->base.port = PORT_OTHER;
5721 
5722 	/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5723 	 * do not need to check mode.  Though link speed might not represent
5724 	 * the true receive or transmit bandwidth (not all modes are symmetric)
5725 	 * this is an accurate maximum.
5726 	 */
5727 	bond_for_each_slave(bond, slave, iter) {
5728 		if (bond_slave_can_tx(slave)) {
5729 			bond_update_speed_duplex(slave);
5730 			if (slave->speed != SPEED_UNKNOWN) {
5731 				if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5732 					speed = bond_mode_bcast_speed(slave,
5733 								      speed);
5734 				else
5735 					speed += slave->speed;
5736 			}
5737 			if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5738 			    slave->duplex != DUPLEX_UNKNOWN)
5739 				cmd->base.duplex = slave->duplex;
5740 		}
5741 	}
5742 	cmd->base.speed = speed ? : SPEED_UNKNOWN;
5743 
5744 	return 0;
5745 }
5746 
bond_ethtool_get_drvinfo(struct net_device * bond_dev,struct ethtool_drvinfo * drvinfo)5747 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5748 				     struct ethtool_drvinfo *drvinfo)
5749 {
5750 	strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5751 	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5752 		 BOND_ABI_VERSION);
5753 }
5754 
bond_ethtool_get_ts_info(struct net_device * bond_dev,struct ethtool_ts_info * info)5755 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5756 				    struct ethtool_ts_info *info)
5757 {
5758 	struct bonding *bond = netdev_priv(bond_dev);
5759 	struct ethtool_ts_info ts_info;
5760 	const struct ethtool_ops *ops;
5761 	struct net_device *real_dev;
5762 	bool sw_tx_support = false;
5763 	struct phy_device *phydev;
5764 	struct list_head *iter;
5765 	struct slave *slave;
5766 	int ret = 0;
5767 
5768 	rcu_read_lock();
5769 	real_dev = bond_option_active_slave_get_rcu(bond);
5770 	dev_hold(real_dev);
5771 	rcu_read_unlock();
5772 
5773 	if (real_dev) {
5774 		ops = real_dev->ethtool_ops;
5775 		phydev = real_dev->phydev;
5776 
5777 		if (phy_has_tsinfo(phydev)) {
5778 			ret = phy_ts_info(phydev, info);
5779 			goto out;
5780 		} else if (ops->get_ts_info) {
5781 			ret = ops->get_ts_info(real_dev, info);
5782 			goto out;
5783 		}
5784 	} else {
5785 		/* Check if all slaves support software tx timestamping */
5786 		rcu_read_lock();
5787 		bond_for_each_slave_rcu(bond, slave, iter) {
5788 			ret = -1;
5789 			ops = slave->dev->ethtool_ops;
5790 			phydev = slave->dev->phydev;
5791 
5792 			if (phy_has_tsinfo(phydev))
5793 				ret = phy_ts_info(phydev, &ts_info);
5794 			else if (ops->get_ts_info)
5795 				ret = ops->get_ts_info(slave->dev, &ts_info);
5796 
5797 			if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5798 				sw_tx_support = true;
5799 				continue;
5800 			}
5801 
5802 			sw_tx_support = false;
5803 			break;
5804 		}
5805 		rcu_read_unlock();
5806 	}
5807 
5808 	ret = 0;
5809 	info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5810 				SOF_TIMESTAMPING_SOFTWARE;
5811 	if (sw_tx_support)
5812 		info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5813 
5814 	info->phc_index = -1;
5815 
5816 out:
5817 	dev_put(real_dev);
5818 	return ret;
5819 }
5820 
5821 static const struct ethtool_ops bond_ethtool_ops = {
5822 	.get_drvinfo		= bond_ethtool_get_drvinfo,
5823 	.get_link		= ethtool_op_get_link,
5824 	.get_link_ksettings	= bond_ethtool_get_link_ksettings,
5825 	.get_ts_info		= bond_ethtool_get_ts_info,
5826 };
5827 
5828 static const struct net_device_ops bond_netdev_ops = {
5829 	.ndo_init		= bond_init,
5830 	.ndo_uninit		= bond_uninit,
5831 	.ndo_open		= bond_open,
5832 	.ndo_stop		= bond_close,
5833 	.ndo_start_xmit		= bond_start_xmit,
5834 	.ndo_select_queue	= bond_select_queue,
5835 	.ndo_get_stats64	= bond_get_stats,
5836 	.ndo_eth_ioctl		= bond_eth_ioctl,
5837 	.ndo_siocbond		= bond_do_ioctl,
5838 	.ndo_siocdevprivate	= bond_siocdevprivate,
5839 	.ndo_change_rx_flags	= bond_change_rx_flags,
5840 	.ndo_set_rx_mode	= bond_set_rx_mode,
5841 	.ndo_change_mtu		= bond_change_mtu,
5842 	.ndo_set_mac_address	= bond_set_mac_address,
5843 	.ndo_neigh_setup	= bond_neigh_setup,
5844 	.ndo_vlan_rx_add_vid	= bond_vlan_rx_add_vid,
5845 	.ndo_vlan_rx_kill_vid	= bond_vlan_rx_kill_vid,
5846 #ifdef CONFIG_NET_POLL_CONTROLLER
5847 	.ndo_netpoll_setup	= bond_netpoll_setup,
5848 	.ndo_netpoll_cleanup	= bond_netpoll_cleanup,
5849 	.ndo_poll_controller	= bond_poll_controller,
5850 #endif
5851 	.ndo_add_slave		= bond_enslave,
5852 	.ndo_del_slave		= bond_release,
5853 	.ndo_fix_features	= bond_fix_features,
5854 	.ndo_features_check	= passthru_features_check,
5855 	.ndo_get_xmit_slave	= bond_xmit_get_slave,
5856 	.ndo_sk_get_lower_dev	= bond_sk_get_lower_dev,
5857 	.ndo_bpf		= bond_xdp,
5858 	.ndo_xdp_xmit           = bond_xdp_xmit,
5859 	.ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5860 	.ndo_hwtstamp_get	= bond_hwtstamp_get,
5861 	.ndo_hwtstamp_set	= bond_hwtstamp_set,
5862 };
5863 
5864 static const struct device_type bond_type = {
5865 	.name = "bond",
5866 };
5867 
bond_destructor(struct net_device * bond_dev)5868 static void bond_destructor(struct net_device *bond_dev)
5869 {
5870 	struct bonding *bond = netdev_priv(bond_dev);
5871 
5872 	if (bond->wq)
5873 		destroy_workqueue(bond->wq);
5874 
5875 	free_percpu(bond->rr_tx_counter);
5876 }
5877 
bond_setup(struct net_device * bond_dev)5878 void bond_setup(struct net_device *bond_dev)
5879 {
5880 	struct bonding *bond = netdev_priv(bond_dev);
5881 
5882 	spin_lock_init(&bond->mode_lock);
5883 	bond->params = bonding_defaults;
5884 
5885 	/* Initialize pointers */
5886 	bond->dev = bond_dev;
5887 
5888 	/* Initialize the device entry points */
5889 	ether_setup(bond_dev);
5890 	bond_dev->max_mtu = ETH_MAX_MTU;
5891 	bond_dev->netdev_ops = &bond_netdev_ops;
5892 	bond_dev->ethtool_ops = &bond_ethtool_ops;
5893 
5894 	bond_dev->needs_free_netdev = true;
5895 	bond_dev->priv_destructor = bond_destructor;
5896 
5897 	SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5898 
5899 	/* Initialize the device options */
5900 	bond_dev->flags |= IFF_MASTER;
5901 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5902 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5903 
5904 #ifdef CONFIG_XFRM_OFFLOAD
5905 	/* set up xfrm device ops (only supported in active-backup right now) */
5906 	bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5907 	INIT_LIST_HEAD(&bond->ipsec_list);
5908 	spin_lock_init(&bond->ipsec_lock);
5909 #endif /* CONFIG_XFRM_OFFLOAD */
5910 
5911 	/* don't acquire bond device's netif_tx_lock when transmitting */
5912 	bond_dev->features |= NETIF_F_LLTX;
5913 
5914 	/* By default, we declare the bond to be fully
5915 	 * VLAN hardware accelerated capable. Special
5916 	 * care is taken in the various xmit functions
5917 	 * when there are slaves that are not hw accel
5918 	 * capable
5919 	 */
5920 
5921 	/* Don't allow bond devices to change network namespaces. */
5922 	bond_dev->features |= NETIF_F_NETNS_LOCAL;
5923 
5924 	bond_dev->hw_features = BOND_VLAN_FEATURES |
5925 				NETIF_F_HW_VLAN_CTAG_RX |
5926 				NETIF_F_HW_VLAN_CTAG_FILTER |
5927 				NETIF_F_HW_VLAN_STAG_RX |
5928 				NETIF_F_HW_VLAN_STAG_FILTER;
5929 
5930 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5931 	bond_dev->features |= bond_dev->hw_features;
5932 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5933 #ifdef CONFIG_XFRM_OFFLOAD
5934 	bond_dev->hw_features |= BOND_XFRM_FEATURES;
5935 	/* Only enable XFRM features if this is an active-backup config */
5936 	if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5937 		bond_dev->features |= BOND_XFRM_FEATURES;
5938 #endif /* CONFIG_XFRM_OFFLOAD */
5939 }
5940 
5941 /* Destroy a bonding device.
5942  * Must be under rtnl_lock when this function is called.
5943  */
bond_uninit(struct net_device * bond_dev)5944 static void bond_uninit(struct net_device *bond_dev)
5945 {
5946 	struct bonding *bond = netdev_priv(bond_dev);
5947 	struct list_head *iter;
5948 	struct slave *slave;
5949 
5950 	bond_netpoll_cleanup(bond_dev);
5951 
5952 	/* Release the bonded slaves */
5953 	bond_for_each_slave(bond, slave, iter)
5954 		__bond_release_one(bond_dev, slave->dev, true, true);
5955 	netdev_info(bond_dev, "Released all slaves\n");
5956 
5957 	bond_set_slave_arr(bond, NULL, NULL);
5958 
5959 	list_del(&bond->bond_list);
5960 
5961 	bond_debug_unregister(bond);
5962 }
5963 
5964 /*------------------------- Module initialization ---------------------------*/
5965 
bond_check_params(struct bond_params * params)5966 static int __init bond_check_params(struct bond_params *params)
5967 {
5968 	int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5969 	struct bond_opt_value newval;
5970 	const struct bond_opt_value *valptr;
5971 	int arp_all_targets_value = 0;
5972 	u16 ad_actor_sys_prio = 0;
5973 	u16 ad_user_port_key = 0;
5974 	__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5975 	int arp_ip_count;
5976 	int bond_mode	= BOND_MODE_ROUNDROBIN;
5977 	int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5978 	int lacp_fast = 0;
5979 	int tlb_dynamic_lb;
5980 
5981 	/* Convert string parameters. */
5982 	if (mode) {
5983 		bond_opt_initstr(&newval, mode);
5984 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5985 		if (!valptr) {
5986 			pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5987 			return -EINVAL;
5988 		}
5989 		bond_mode = valptr->value;
5990 	}
5991 
5992 	if (xmit_hash_policy) {
5993 		if (bond_mode == BOND_MODE_ROUNDROBIN ||
5994 		    bond_mode == BOND_MODE_ACTIVEBACKUP ||
5995 		    bond_mode == BOND_MODE_BROADCAST) {
5996 			pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5997 				bond_mode_name(bond_mode));
5998 		} else {
5999 			bond_opt_initstr(&newval, xmit_hash_policy);
6000 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
6001 						&newval);
6002 			if (!valptr) {
6003 				pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
6004 				       xmit_hash_policy);
6005 				return -EINVAL;
6006 			}
6007 			xmit_hashtype = valptr->value;
6008 		}
6009 	}
6010 
6011 	if (lacp_rate) {
6012 		if (bond_mode != BOND_MODE_8023AD) {
6013 			pr_info("lacp_rate param is irrelevant in mode %s\n",
6014 				bond_mode_name(bond_mode));
6015 		} else {
6016 			bond_opt_initstr(&newval, lacp_rate);
6017 			valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
6018 						&newval);
6019 			if (!valptr) {
6020 				pr_err("Error: Invalid lacp rate \"%s\"\n",
6021 				       lacp_rate);
6022 				return -EINVAL;
6023 			}
6024 			lacp_fast = valptr->value;
6025 		}
6026 	}
6027 
6028 	if (ad_select) {
6029 		bond_opt_initstr(&newval, ad_select);
6030 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6031 					&newval);
6032 		if (!valptr) {
6033 			pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6034 			return -EINVAL;
6035 		}
6036 		params->ad_select = valptr->value;
6037 		if (bond_mode != BOND_MODE_8023AD)
6038 			pr_warn("ad_select param only affects 802.3ad mode\n");
6039 	} else {
6040 		params->ad_select = BOND_AD_STABLE;
6041 	}
6042 
6043 	if (max_bonds < 0) {
6044 		pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6045 			max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6046 		max_bonds = BOND_DEFAULT_MAX_BONDS;
6047 	}
6048 
6049 	if (miimon < 0) {
6050 		pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6051 			miimon, INT_MAX);
6052 		miimon = 0;
6053 	}
6054 
6055 	if (updelay < 0) {
6056 		pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6057 			updelay, INT_MAX);
6058 		updelay = 0;
6059 	}
6060 
6061 	if (downdelay < 0) {
6062 		pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6063 			downdelay, INT_MAX);
6064 		downdelay = 0;
6065 	}
6066 
6067 	if ((use_carrier != 0) && (use_carrier != 1)) {
6068 		pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6069 			use_carrier);
6070 		use_carrier = 1;
6071 	}
6072 
6073 	if (num_peer_notif < 0 || num_peer_notif > 255) {
6074 		pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6075 			num_peer_notif);
6076 		num_peer_notif = 1;
6077 	}
6078 
6079 	/* reset values for 802.3ad/TLB/ALB */
6080 	if (!bond_mode_uses_arp(bond_mode)) {
6081 		if (!miimon) {
6082 			pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
6083 			pr_warn("Forcing miimon to 100msec\n");
6084 			miimon = BOND_DEFAULT_MIIMON;
6085 		}
6086 	}
6087 
6088 	if (tx_queues < 1 || tx_queues > 255) {
6089 		pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6090 			tx_queues, BOND_DEFAULT_TX_QUEUES);
6091 		tx_queues = BOND_DEFAULT_TX_QUEUES;
6092 	}
6093 
6094 	if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6095 		pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6096 			all_slaves_active);
6097 		all_slaves_active = 0;
6098 	}
6099 
6100 	if (resend_igmp < 0 || resend_igmp > 255) {
6101 		pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6102 			resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6103 		resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6104 	}
6105 
6106 	bond_opt_initval(&newval, packets_per_slave);
6107 	if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6108 		pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6109 			packets_per_slave, USHRT_MAX);
6110 		packets_per_slave = 1;
6111 	}
6112 
6113 	if (bond_mode == BOND_MODE_ALB) {
6114 		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",
6115 			  updelay);
6116 	}
6117 
6118 	if (!miimon) {
6119 		if (updelay || downdelay) {
6120 			/* just warn the user the up/down delay will have
6121 			 * no effect since miimon is zero...
6122 			 */
6123 			pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
6124 				updelay, downdelay);
6125 		}
6126 	} else {
6127 		/* don't allow arp monitoring */
6128 		if (arp_interval) {
6129 			pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6130 				miimon, arp_interval);
6131 			arp_interval = 0;
6132 		}
6133 
6134 		if ((updelay % miimon) != 0) {
6135 			pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6136 				updelay, miimon, (updelay / miimon) * miimon);
6137 		}
6138 
6139 		updelay /= miimon;
6140 
6141 		if ((downdelay % miimon) != 0) {
6142 			pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6143 				downdelay, miimon,
6144 				(downdelay / miimon) * miimon);
6145 		}
6146 
6147 		downdelay /= miimon;
6148 	}
6149 
6150 	if (arp_interval < 0) {
6151 		pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6152 			arp_interval, INT_MAX);
6153 		arp_interval = 0;
6154 	}
6155 
6156 	for (arp_ip_count = 0, i = 0;
6157 	     (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6158 		__be32 ip;
6159 
6160 		/* not a complete check, but good enough to catch mistakes */
6161 		if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6162 		    !bond_is_ip_target_ok(ip)) {
6163 			pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6164 				arp_ip_target[i]);
6165 			arp_interval = 0;
6166 		} else {
6167 			if (bond_get_targets_ip(arp_target, ip) == -1)
6168 				arp_target[arp_ip_count++] = ip;
6169 			else
6170 				pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6171 					&ip);
6172 		}
6173 	}
6174 
6175 	if (arp_interval && !arp_ip_count) {
6176 		/* don't allow arping if no arp_ip_target given... */
6177 		pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6178 			arp_interval);
6179 		arp_interval = 0;
6180 	}
6181 
6182 	if (arp_validate) {
6183 		if (!arp_interval) {
6184 			pr_err("arp_validate requires arp_interval\n");
6185 			return -EINVAL;
6186 		}
6187 
6188 		bond_opt_initstr(&newval, arp_validate);
6189 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6190 					&newval);
6191 		if (!valptr) {
6192 			pr_err("Error: invalid arp_validate \"%s\"\n",
6193 			       arp_validate);
6194 			return -EINVAL;
6195 		}
6196 		arp_validate_value = valptr->value;
6197 	} else {
6198 		arp_validate_value = 0;
6199 	}
6200 
6201 	if (arp_all_targets) {
6202 		bond_opt_initstr(&newval, arp_all_targets);
6203 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6204 					&newval);
6205 		if (!valptr) {
6206 			pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6207 			       arp_all_targets);
6208 			arp_all_targets_value = 0;
6209 		} else {
6210 			arp_all_targets_value = valptr->value;
6211 		}
6212 	}
6213 
6214 	if (miimon) {
6215 		pr_info("MII link monitoring set to %d ms\n", miimon);
6216 	} else if (arp_interval) {
6217 		valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6218 					  arp_validate_value);
6219 		pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6220 			arp_interval, valptr->string, arp_ip_count);
6221 
6222 		for (i = 0; i < arp_ip_count; i++)
6223 			pr_cont(" %s", arp_ip_target[i]);
6224 
6225 		pr_cont("\n");
6226 
6227 	} else if (max_bonds) {
6228 		/* miimon and arp_interval not set, we need one so things
6229 		 * work as expected, see bonding.txt for details
6230 		 */
6231 		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");
6232 	}
6233 
6234 	if (primary && !bond_mode_uses_primary(bond_mode)) {
6235 		/* currently, using a primary only makes sense
6236 		 * in active backup, TLB or ALB modes
6237 		 */
6238 		pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6239 			primary, bond_mode_name(bond_mode));
6240 		primary = NULL;
6241 	}
6242 
6243 	if (primary && primary_reselect) {
6244 		bond_opt_initstr(&newval, primary_reselect);
6245 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6246 					&newval);
6247 		if (!valptr) {
6248 			pr_err("Error: Invalid primary_reselect \"%s\"\n",
6249 			       primary_reselect);
6250 			return -EINVAL;
6251 		}
6252 		primary_reselect_value = valptr->value;
6253 	} else {
6254 		primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6255 	}
6256 
6257 	if (fail_over_mac) {
6258 		bond_opt_initstr(&newval, fail_over_mac);
6259 		valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6260 					&newval);
6261 		if (!valptr) {
6262 			pr_err("Error: invalid fail_over_mac \"%s\"\n",
6263 			       fail_over_mac);
6264 			return -EINVAL;
6265 		}
6266 		fail_over_mac_value = valptr->value;
6267 		if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6268 			pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6269 	} else {
6270 		fail_over_mac_value = BOND_FOM_NONE;
6271 	}
6272 
6273 	bond_opt_initstr(&newval, "default");
6274 	valptr = bond_opt_parse(
6275 			bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6276 				     &newval);
6277 	if (!valptr) {
6278 		pr_err("Error: No ad_actor_sys_prio default value");
6279 		return -EINVAL;
6280 	}
6281 	ad_actor_sys_prio = valptr->value;
6282 
6283 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6284 				&newval);
6285 	if (!valptr) {
6286 		pr_err("Error: No ad_user_port_key default value");
6287 		return -EINVAL;
6288 	}
6289 	ad_user_port_key = valptr->value;
6290 
6291 	bond_opt_initstr(&newval, "default");
6292 	valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6293 	if (!valptr) {
6294 		pr_err("Error: No tlb_dynamic_lb default value");
6295 		return -EINVAL;
6296 	}
6297 	tlb_dynamic_lb = valptr->value;
6298 
6299 	if (lp_interval == 0) {
6300 		pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6301 			INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6302 		lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6303 	}
6304 
6305 	/* fill params struct with the proper values */
6306 	params->mode = bond_mode;
6307 	params->xmit_policy = xmit_hashtype;
6308 	params->miimon = miimon;
6309 	params->num_peer_notif = num_peer_notif;
6310 	params->arp_interval = arp_interval;
6311 	params->arp_validate = arp_validate_value;
6312 	params->arp_all_targets = arp_all_targets_value;
6313 	params->missed_max = 2;
6314 	params->updelay = updelay;
6315 	params->downdelay = downdelay;
6316 	params->peer_notif_delay = 0;
6317 	params->use_carrier = use_carrier;
6318 	params->lacp_active = 1;
6319 	params->lacp_fast = lacp_fast;
6320 	params->primary[0] = 0;
6321 	params->primary_reselect = primary_reselect_value;
6322 	params->fail_over_mac = fail_over_mac_value;
6323 	params->tx_queues = tx_queues;
6324 	params->all_slaves_active = all_slaves_active;
6325 	params->resend_igmp = resend_igmp;
6326 	params->min_links = min_links;
6327 	params->lp_interval = lp_interval;
6328 	params->packets_per_slave = packets_per_slave;
6329 	params->tlb_dynamic_lb = tlb_dynamic_lb;
6330 	params->ad_actor_sys_prio = ad_actor_sys_prio;
6331 	eth_zero_addr(params->ad_actor_system);
6332 	params->ad_user_port_key = ad_user_port_key;
6333 	if (packets_per_slave > 0) {
6334 		params->reciprocal_packets_per_slave =
6335 			reciprocal_value(packets_per_slave);
6336 	} else {
6337 		/* reciprocal_packets_per_slave is unused if
6338 		 * packets_per_slave is 0 or 1, just initialize it
6339 		 */
6340 		params->reciprocal_packets_per_slave =
6341 			(struct reciprocal_value) { 0 };
6342 	}
6343 
6344 	if (primary)
6345 		strscpy_pad(params->primary, primary, sizeof(params->primary));
6346 
6347 	memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6348 #if IS_ENABLED(CONFIG_IPV6)
6349 	memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6350 #endif
6351 
6352 	return 0;
6353 }
6354 
6355 /* Called from registration process */
bond_init(struct net_device * bond_dev)6356 static int bond_init(struct net_device *bond_dev)
6357 {
6358 	struct bonding *bond = netdev_priv(bond_dev);
6359 	struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6360 
6361 	netdev_dbg(bond_dev, "Begin bond_init\n");
6362 
6363 	bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6364 	if (!bond->wq)
6365 		return -ENOMEM;
6366 
6367 	bond->notifier_ctx = false;
6368 
6369 	spin_lock_init(&bond->stats_lock);
6370 	netdev_lockdep_set_classes(bond_dev);
6371 
6372 	list_add_tail(&bond->bond_list, &bn->dev_list);
6373 
6374 	bond_prepare_sysfs_group(bond);
6375 
6376 	bond_debug_register(bond);
6377 
6378 	/* Ensure valid dev_addr */
6379 	if (is_zero_ether_addr(bond_dev->dev_addr) &&
6380 	    bond_dev->addr_assign_type == NET_ADDR_PERM)
6381 		eth_hw_addr_random(bond_dev);
6382 
6383 	return 0;
6384 }
6385 
bond_get_num_tx_queues(void)6386 unsigned int bond_get_num_tx_queues(void)
6387 {
6388 	return tx_queues;
6389 }
6390 
6391 /* Create a new bond based on the specified name and bonding parameters.
6392  * If name is NULL, obtain a suitable "bond%d" name for us.
6393  * Caller must NOT hold rtnl_lock; we need to release it here before we
6394  * set up our sysfs entries.
6395  */
bond_create(struct net * net,const char * name)6396 int bond_create(struct net *net, const char *name)
6397 {
6398 	struct net_device *bond_dev;
6399 	struct bonding *bond;
6400 	int res = -ENOMEM;
6401 
6402 	rtnl_lock();
6403 
6404 	bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6405 				   name ? name : "bond%d", NET_NAME_UNKNOWN,
6406 				   bond_setup, tx_queues);
6407 	if (!bond_dev)
6408 		goto out;
6409 
6410 	bond = netdev_priv(bond_dev);
6411 	dev_net_set(bond_dev, net);
6412 	bond_dev->rtnl_link_ops = &bond_link_ops;
6413 
6414 	res = register_netdevice(bond_dev);
6415 	if (res < 0) {
6416 		free_netdev(bond_dev);
6417 		goto out;
6418 	}
6419 
6420 	netif_carrier_off(bond_dev);
6421 
6422 	bond_work_init_all(bond);
6423 
6424 out:
6425 	rtnl_unlock();
6426 	return res;
6427 }
6428 
bond_net_init(struct net * net)6429 static int __net_init bond_net_init(struct net *net)
6430 {
6431 	struct bond_net *bn = net_generic(net, bond_net_id);
6432 
6433 	bn->net = net;
6434 	INIT_LIST_HEAD(&bn->dev_list);
6435 
6436 	bond_create_proc_dir(bn);
6437 	bond_create_sysfs(bn);
6438 
6439 	return 0;
6440 }
6441 
bond_net_exit_batch(struct list_head * net_list)6442 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6443 {
6444 	struct bond_net *bn;
6445 	struct net *net;
6446 	LIST_HEAD(list);
6447 
6448 	list_for_each_entry(net, net_list, exit_list) {
6449 		bn = net_generic(net, bond_net_id);
6450 		bond_destroy_sysfs(bn);
6451 	}
6452 
6453 	/* Kill off any bonds created after unregistering bond rtnl ops */
6454 	rtnl_lock();
6455 	list_for_each_entry(net, net_list, exit_list) {
6456 		struct bonding *bond, *tmp_bond;
6457 
6458 		bn = net_generic(net, bond_net_id);
6459 		list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6460 			unregister_netdevice_queue(bond->dev, &list);
6461 	}
6462 	unregister_netdevice_many(&list);
6463 	rtnl_unlock();
6464 
6465 	list_for_each_entry(net, net_list, exit_list) {
6466 		bn = net_generic(net, bond_net_id);
6467 		bond_destroy_proc_dir(bn);
6468 	}
6469 }
6470 
6471 static struct pernet_operations bond_net_ops = {
6472 	.init = bond_net_init,
6473 	.exit_batch = bond_net_exit_batch,
6474 	.id   = &bond_net_id,
6475 	.size = sizeof(struct bond_net),
6476 };
6477 
bonding_init(void)6478 static int __init bonding_init(void)
6479 {
6480 	int i;
6481 	int res;
6482 
6483 	res = bond_check_params(&bonding_defaults);
6484 	if (res)
6485 		goto out;
6486 
6487 	res = register_pernet_subsys(&bond_net_ops);
6488 	if (res)
6489 		goto out;
6490 
6491 	res = bond_netlink_init();
6492 	if (res)
6493 		goto err_link;
6494 
6495 	bond_create_debugfs();
6496 
6497 	for (i = 0; i < max_bonds; i++) {
6498 		res = bond_create(&init_net, NULL);
6499 		if (res)
6500 			goto err;
6501 	}
6502 
6503 	skb_flow_dissector_init(&flow_keys_bonding,
6504 				flow_keys_bonding_keys,
6505 				ARRAY_SIZE(flow_keys_bonding_keys));
6506 
6507 	register_netdevice_notifier(&bond_netdev_notifier);
6508 out:
6509 	return res;
6510 err:
6511 	bond_destroy_debugfs();
6512 	bond_netlink_fini();
6513 err_link:
6514 	unregister_pernet_subsys(&bond_net_ops);
6515 	goto out;
6516 
6517 }
6518 
bonding_exit(void)6519 static void __exit bonding_exit(void)
6520 {
6521 	unregister_netdevice_notifier(&bond_netdev_notifier);
6522 
6523 	bond_destroy_debugfs();
6524 
6525 	bond_netlink_fini();
6526 	unregister_pernet_subsys(&bond_net_ops);
6527 
6528 #ifdef CONFIG_NET_POLL_CONTROLLER
6529 	/* Make sure we don't have an imbalance on our netpoll blocking */
6530 	WARN_ON(atomic_read(&netpoll_block_tx));
6531 #endif
6532 }
6533 
6534 module_init(bonding_init);
6535 module_exit(bonding_exit);
6536 MODULE_LICENSE("GPL");
6537 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6538 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
6539