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
101static int max_bonds = BOND_DEFAULT_MAX_BONDS;
102static int tx_queues = BOND_DEFAULT_TX_QUEUES;
103static int num_peer_notif = 1;
104static int miimon;
105static int updelay;
106static int downdelay;
107static int use_carrier = 1;
108static char *mode;
109static char *primary;
110static char *primary_reselect;
111static char *lacp_rate;
112static int min_links;
113static char *ad_select;
114static char *xmit_hash_policy;
115static int arp_interval;
116static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
117static char *arp_validate;
118static char *arp_all_targets;
119static char *fail_over_mac;
120static int all_slaves_active;
121static struct bond_params bonding_defaults;
122static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
123static int packets_per_slave = 1;
124static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
125
126module_param(max_bonds, int, 0);
127MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
128module_param(tx_queues, int, 0);
129MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
130module_param_named(num_grat_arp, num_peer_notif, int, 0644);
131MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
132 "failover event (alias of num_unsol_na)");
133module_param_named(num_unsol_na, num_peer_notif, int, 0644);
134MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
135 "failover event (alias of num_grat_arp)");
136module_param(miimon, int, 0);
137MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
138module_param(updelay, int, 0);
139MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
140module_param(downdelay, int, 0);
141MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
142 "in milliseconds");
143module_param(use_carrier, int, 0);
144MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
145 "0 for off, 1 for on (default)");
146module_param(mode, charp, 0);
147MODULE_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");
151module_param(primary, charp, 0);
152MODULE_PARM_DESC(primary, "Primary network device to use");
153module_param(primary_reselect, charp, 0);
154MODULE_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");
161module_param(lacp_rate, charp, 0);
162MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163 "0 for slow, 1 for fast");
164module_param(ad_select, charp, 0);
165MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166 "0 for stable (default), 1 for bandwidth, "
167 "2 for count");
168module_param(min_links, int, 0);
169MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
170
171module_param(xmit_hash_policy, charp, 0);
172MODULE_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");
176module_param(arp_interval, int, 0);
177MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178module_param_array(arp_ip_target, charp, NULL, 0);
179MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180module_param(arp_validate, charp, 0);
181MODULE_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");
184module_param(arp_all_targets, charp, 0);
185MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186module_param(fail_over_mac, charp, 0);
187MODULE_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");
190module_param(all_slaves_active, int, 0);
191MODULE_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.");
194module_param(resend_igmp, int, 0);
195MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
196 "link failure");
197module_param(packets_per_slave, int, 0);
198MODULE_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.");
201module_param(lp_interval, uint, 0);
202MODULE_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
209atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210#endif
211
212unsigned int bond_net_id __read_mostly;
213
214static 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
257static struct flow_dissector flow_keys_bonding __read_mostly;
258
259/*-------------------------- Forward declarations ---------------------------*/
260
261static int bond_init(struct net_device *bond_dev);
262static void bond_uninit(struct net_device *bond_dev);
263static void bond_get_stats(struct net_device *bond_dev,
264 struct rtnl_link_stats64 *stats);
265static void bond_slave_arr_handler(struct work_struct *work);
266static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
267 int mod);
268static void bond_netdev_notify_work(struct work_struct *work);
269
270/*---------------------------- General routines -----------------------------*/
271
272const 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 */
297netdev_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, queue_mapping: qdisc_skb_cb(skb)->slave_dev_queue_mapping);
305
306 if (unlikely(netpoll_tx_running(bond->dev)))
307 return bond_netpoll_send_skb(slave: bond_get_slave_by_dev(bond, slave_dev), skb);
308
309 return dev_queue_xmit(skb);
310}
311
312static 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
325static 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 */
368static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
369 __be16 proto, u16 vid)
370{
371 struct bonding *bond = netdev_priv(dev: 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(dev: slave->dev, proto, vid);
378 if (res)
379 goto unwind;
380 }
381
382 return 0;
383
384unwind:
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(dev: 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 */
402static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
403 __be16 proto, u16 vid)
404{
405 struct bonding *bond = netdev_priv(dev: bond_dev);
406 struct list_head *iter;
407 struct slave *slave;
408
409 bond_for_each_slave(bond, slave, iter)
410 vlan_vid_del(dev: slave->dev, proto, vid);
411
412 if (bond_is_lb(bond))
413 bond_alb_clear_vlan(bond, vlan_id: 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 **/
426static 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(dev: 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(dev: 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(size: 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(list: &ipsec->list);
465 spin_lock_bh(lock: &bond->ipsec_lock);
466 list_add(new: &ipsec->list, head: &bond->ipsec_list);
467 spin_unlock_bh(lock: &bond->ipsec_lock);
468 } else {
469 kfree(objp: ipsec);
470 }
471 rcu_read_unlock();
472 return err;
473}
474
475static 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(dev: slave->dev)) {
489 spin_lock_bh(lock: &bond->ipsec_lock);
490 if (!list_empty(head: &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(lock: &bond->ipsec_lock);
495 goto out;
496 }
497
498 spin_lock_bh(lock: &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(lock: &bond->ipsec_lock);
507out:
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 **/
515static 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(dev: 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(dev: 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);
545out:
546 spin_lock_bh(lock: &bond->ipsec_lock);
547 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
548 if (ipsec->xs == xs) {
549 list_del(entry: &ipsec->list);
550 kfree(objp: ipsec);
551 break;
552 }
553 }
554 spin_unlock_bh(lock: &bond->ipsec_lock);
555 rcu_read_unlock();
556}
557
558static 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(lock: &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(dev: 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(lock: &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 **/
596static 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(dev: 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(dev: real_dev)) {
622 err = false;
623 goto out;
624 }
625
626 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
627out:
628 rcu_read_unlock();
629 return err;
630}
631
632static 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 */
647int 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(dev: bond->dev)) {
661 netif_carrier_on(dev: bond->dev);
662 return 1;
663 }
664 return 0;
665 }
666 }
667
668down:
669 if (netif_carrier_ok(dev: bond->dev)) {
670 netif_carrier_off(dev: 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 */
682static 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(dev: slave_dev, link_ksettings: &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
710const 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 */
741static 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(dev: slave_dev))
750 return 0;
751
752 if (bond->params.use_carrier)
753 return netif_carrier_ok(dev: 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(dest: ifr.ifr_name, src: slave_dev->name, IFNAMSIZ);
777 mii = if_mii(rq: &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 */
796static 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(dev: curr_active->dev, inc);
806 } else {
807 struct slave *slave;
808
809 bond_for_each_slave(bond, slave, iter) {
810 err = dev_set_promiscuity(dev: 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 */
819static 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(dev: curr_active->dev, inc);
829 } else {
830 struct slave *slave;
831
832 bond_for_each_slave(bond, slave, iter) {
833 err = dev_set_allmulti(dev: 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 */
845static 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(wq: bond->wq, dwork: &bond->mcast_work, delay: 1);
852 return;
853 }
854 call_netdevice_notifiers(val: NETDEV_RESEND_IGMP, dev: bond->dev);
855
856 if (bond->igmp_retrans > 1) {
857 bond->igmp_retrans--;
858 queue_delayed_work(wq: bond->wq, dwork: &bond->mcast_work, HZ/5);
859 }
860 rtnl_unlock();
861}
862
863/* Flush bond's hardware addresses from slave */
864static void bond_hw_addr_flush(struct net_device *bond_dev,
865 struct net_device *slave_dev)
866{
867 struct bonding *bond = netdev_priv(dev: bond_dev);
868
869 dev_uc_unsync(to: slave_dev, from: bond_dev);
870 dev_mc_unsync(to: slave_dev, from: bond_dev);
871
872 if (BOND_MODE(bond) == BOND_MODE_8023AD)
873 dev_mc_del(dev: slave_dev, addr: 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 */
883static 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(dev: old_active->dev, inc: -1);
889
890 if (bond->dev->flags & IFF_ALLMULTI)
891 dev_set_allmulti(dev: old_active->dev, inc: -1);
892
893 if (bond->dev->flags & IFF_UP)
894 bond_hw_addr_flush(bond_dev: bond->dev, slave_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(dev: new_active->dev, inc: 1);
901
902 if (bond->dev->flags & IFF_ALLMULTI)
903 dev_set_allmulti(dev: new_active->dev, inc: 1);
904
905 if (bond->dev->flags & IFF_UP) {
906 netif_addr_lock_bh(dev: bond->dev);
907 dev_uc_sync(to: new_active->dev, from: bond->dev);
908 dev_mc_sync(to: new_active->dev, from: bond->dev);
909 netif_addr_unlock_bh(dev: 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 */
921static 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(dev: bond_dev, addr: slave_dev->dev_addr, NULL);
929 if (err)
930 return err;
931
932 __dev_addr_set(dev: bond_dev, addr: slave_dev->dev_addr, len: slave_dev->addr_len);
933 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
934 call_netdevice_notifiers(val: NETDEV_CHANGEADDR, dev: bond_dev);
935 return 0;
936}
937
938static 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(addr1: bond->dev->dev_addr, addr2: 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 */
961static 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: bond->dev, slave_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(dst: tmp_mac, src: new_active->dev->dev_addr,
991 len: new_active->dev->addr_len);
992 bond_hw_addr_copy(dst: ss.__data,
993 src: old_active->dev->dev_addr,
994 len: old_active->dev->addr_len);
995 ss.ss_family = new_active->dev->type;
996 } else {
997 bond_hw_addr_copy(dst: ss.__data, src: bond->dev->dev_addr,
998 len: bond->dev->addr_len);
999 ss.ss_family = bond->dev->type;
1000 }
1001
1002 rv = dev_set_mac_address(dev: new_active->dev,
1003 sa: (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(dst: ss.__data, src: tmp_mac,
1014 len: new_active->dev->addr_len);
1015 ss.ss_family = old_active->dev->type;
1016
1017 rv = dev_set_mac_address(dev: old_active->dev,
1018 sa: (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);
1022out:
1023 break;
1024 default:
1025 netdev_err(dev: bond->dev, format: "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 */
1043static 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
1074link_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(dev: bond->dev, format: "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 */
1101static 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
1124static 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(dev: 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 */
1160void 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(slave: 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(slave: new_active, BOND_LINK_UP);
1190
1191 if (bond_is_lb(bond))
1192 bond_alb_handle_link_change(bond, slave: 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_slave: new_active);
1204 if (old_active)
1205 bond_set_slave_inactive_flags(slave: old_active,
1206 BOND_SLAVE_NOTIFY_NOW);
1207 if (new_active)
1208 bond_set_slave_active_flags(slave: 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(slave: 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(slave: 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(dev: 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(val: NETDEV_BONDING_FAILOVER, dev: bond->dev);
1238 if (should_notify_peers) {
1239 bond->send_peer_notif--;
1240 call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS,
1241 dev: 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(dev: 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(wq: bond->wq, dwork: &bond->mcast_work, delay: 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 */
1274void 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, new_active: best_slave);
1284 rv = bond_set_carrier(bond);
1285 if (!rv)
1286 return;
1287
1288 if (netif_carrier_ok(dev: bond->dev))
1289 netdev_info(dev: bond->dev, format: "active interface up!\n");
1290 else
1291 netdev_info(dev: bond->dev, format: "now running without any active interface!\n");
1292 }
1293}
1294
1295#ifdef CONFIG_NET_POLL_CONTROLLER
1296static inline int slave_enable_netpoll(struct slave *slave)
1297{
1298 struct netpoll *np;
1299 int err = 0;
1300
1301 np = kzalloc(size: sizeof(*np), GFP_KERNEL);
1302 err = -ENOMEM;
1303 if (!np)
1304 goto out;
1305
1306 err = __netpoll_setup(np, ndev: slave->dev);
1307 if (err) {
1308 kfree(objp: np);
1309 goto out;
1310 }
1311 slave->np = np;
1312out:
1313 return err;
1314}
1315static 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
1327static void bond_poll_controller(struct net_device *bond_dev)
1328{
1329 struct bonding *bond = netdev_priv(dev: 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: &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(dev: slave->dev);
1352 }
1353}
1354
1355static void bond_netpoll_cleanup(struct net_device *bond_dev)
1356{
1357 struct bonding *bond = netdev_priv(dev: 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
1366static 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(bond_dev: dev);
1377 break;
1378 }
1379 }
1380 return err;
1381}
1382#else
1383static inline int slave_enable_netpoll(struct slave *slave)
1384{
1385 return 0;
1386}
1387static inline void slave_disable_netpoll(struct slave *slave)
1388{
1389}
1390static void bond_netpoll_cleanup(struct net_device *bond_dev)
1391{
1392}
1393#endif
1394
1395/*---------------------------------- IOCTL ----------------------------------*/
1396
1397static 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(all: features,
1412 one: 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
1431static 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(all: vlan_features,
1455 one: slave->dev->vlan_features, BOND_VLAN_FEATURES);
1456
1457 enc_features = netdev_increment_features(all: enc_features,
1458 one: slave->dev->hw_enc_features,
1459 BOND_ENC_FEATURES);
1460
1461#ifdef CONFIG_XFRM_OFFLOAD
1462 xfrm_features = netdev_increment_features(all: xfrm_features,
1463 one: slave->dev->hw_enc_features,
1464 BOND_XFRM_FEATURES);
1465#endif /* CONFIG_XFRM_OFFLOAD */
1466
1467 mpls_features = netdev_increment_features(all: mpls_features,
1468 one: 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
1480done:
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(dev: bond_dev, segs: tso_max_segs);
1490 netif_set_tso_max_size(dev: bond_dev, size: 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(dev: bond_dev);
1498}
1499
1500static void bond_setup_by_slave(struct net_device *bond_dev,
1501 struct net_device *slave_dev)
1502{
1503 bond_dev->header_ops = slave_dev->header_ops;
1504
1505 bond_dev->type = slave_dev->type;
1506 bond_dev->hard_header_len = slave_dev->hard_header_len;
1507 bond_dev->needed_headroom = slave_dev->needed_headroom;
1508 bond_dev->addr_len = slave_dev->addr_len;
1509
1510 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1511 slave_dev->addr_len);
1512
1513 if (slave_dev->flags & IFF_POINTOPOINT) {
1514 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1515 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1516 }
1517}
1518
1519/* On bonding slaves other than the currently active slave, suppress
1520 * duplicates except for alb non-mcast/bcast.
1521 */
1522static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1523 struct slave *slave,
1524 struct bonding *bond)
1525{
1526 if (bond_is_slave_inactive(slave)) {
1527 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1528 skb->pkt_type != PACKET_BROADCAST &&
1529 skb->pkt_type != PACKET_MULTICAST)
1530 return false;
1531 return true;
1532 }
1533 return false;
1534}
1535
1536static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1537{
1538 struct sk_buff *skb = *pskb;
1539 struct slave *slave;
1540 struct bonding *bond;
1541 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1542 struct slave *);
1543 int ret = RX_HANDLER_ANOTHER;
1544
1545 skb = skb_share_check(skb, GFP_ATOMIC);
1546 if (unlikely(!skb))
1547 return RX_HANDLER_CONSUMED;
1548
1549 *pskb = skb;
1550
1551 slave = bond_slave_get_rcu(skb->dev);
1552 bond = slave->bond;
1553
1554 recv_probe = READ_ONCE(bond->recv_probe);
1555 if (recv_probe) {
1556 ret = recv_probe(skb, bond, slave);
1557 if (ret == RX_HANDLER_CONSUMED) {
1558 consume_skb(skb);
1559 return ret;
1560 }
1561 }
1562
1563 /*
1564 * For packets determined by bond_should_deliver_exact_match() call to
1565 * be suppressed we want to make an exception for link-local packets.
1566 * This is necessary for e.g. LLDP daemons to be able to monitor
1567 * inactive slave links without being forced to bind to them
1568 * explicitly.
1569 *
1570 * At the same time, packets that are passed to the bonding master
1571 * (including link-local ones) can have their originating interface
1572 * determined via PACKET_ORIGDEV socket option.
1573 */
1574 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1575 if (is_link_local_ether_addr(addr: eth_hdr(skb)->h_dest))
1576 return RX_HANDLER_PASS;
1577 return RX_HANDLER_EXACT;
1578 }
1579
1580 skb->dev = bond->dev;
1581
1582 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1583 netif_is_bridge_port(dev: bond->dev) &&
1584 skb->pkt_type == PACKET_HOST) {
1585
1586 if (unlikely(skb_cow_head(skb,
1587 skb->data - skb_mac_header(skb)))) {
1588 kfree_skb(skb);
1589 return RX_HANDLER_CONSUMED;
1590 }
1591 bond_hw_addr_copy(dst: eth_hdr(skb)->h_dest, src: bond->dev->dev_addr,
1592 len: bond->dev->addr_len);
1593 }
1594
1595 return ret;
1596}
1597
1598static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1599{
1600 switch (BOND_MODE(bond)) {
1601 case BOND_MODE_ROUNDROBIN:
1602 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1603 case BOND_MODE_ACTIVEBACKUP:
1604 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1605 case BOND_MODE_BROADCAST:
1606 return NETDEV_LAG_TX_TYPE_BROADCAST;
1607 case BOND_MODE_XOR:
1608 case BOND_MODE_8023AD:
1609 return NETDEV_LAG_TX_TYPE_HASH;
1610 default:
1611 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1612 }
1613}
1614
1615static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1616 enum netdev_lag_tx_type type)
1617{
1618 if (type != NETDEV_LAG_TX_TYPE_HASH)
1619 return NETDEV_LAG_HASH_NONE;
1620
1621 switch (bond->params.xmit_policy) {
1622 case BOND_XMIT_POLICY_LAYER2:
1623 return NETDEV_LAG_HASH_L2;
1624 case BOND_XMIT_POLICY_LAYER34:
1625 return NETDEV_LAG_HASH_L34;
1626 case BOND_XMIT_POLICY_LAYER23:
1627 return NETDEV_LAG_HASH_L23;
1628 case BOND_XMIT_POLICY_ENCAP23:
1629 return NETDEV_LAG_HASH_E23;
1630 case BOND_XMIT_POLICY_ENCAP34:
1631 return NETDEV_LAG_HASH_E34;
1632 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1633 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1634 default:
1635 return NETDEV_LAG_HASH_UNKNOWN;
1636 }
1637}
1638
1639static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1640 struct netlink_ext_ack *extack)
1641{
1642 struct netdev_lag_upper_info lag_upper_info;
1643 enum netdev_lag_tx_type type;
1644 int err;
1645
1646 type = bond_lag_tx_type(bond);
1647 lag_upper_info.tx_type = type;
1648 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1649
1650 err = netdev_master_upper_dev_link(dev: slave->dev, upper_dev: bond->dev, upper_priv: slave,
1651 upper_info: &lag_upper_info, extack);
1652 if (err)
1653 return err;
1654
1655 slave->dev->flags |= IFF_SLAVE;
1656 return 0;
1657}
1658
1659static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1660{
1661 netdev_upper_dev_unlink(dev: slave->dev, upper_dev: bond->dev);
1662 slave->dev->flags &= ~IFF_SLAVE;
1663}
1664
1665static void slave_kobj_release(struct kobject *kobj)
1666{
1667 struct slave *slave = to_slave(kobj);
1668 struct bonding *bond = bond_get_bond_by_slave(slave);
1669
1670 cancel_delayed_work_sync(dwork: &slave->notify_work);
1671 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1672 kfree(SLAVE_AD_INFO(slave));
1673
1674 kfree(objp: slave);
1675}
1676
1677static struct kobj_type slave_ktype = {
1678 .release = slave_kobj_release,
1679#ifdef CONFIG_SYSFS
1680 .sysfs_ops = &slave_sysfs_ops,
1681#endif
1682};
1683
1684static int bond_kobj_init(struct slave *slave)
1685{
1686 int err;
1687
1688 err = kobject_init_and_add(kobj: &slave->kobj, ktype: &slave_ktype,
1689 parent: &(slave->dev->dev.kobj), fmt: "bonding_slave");
1690 if (err)
1691 kobject_put(kobj: &slave->kobj);
1692
1693 return err;
1694}
1695
1696static struct slave *bond_alloc_slave(struct bonding *bond,
1697 struct net_device *slave_dev)
1698{
1699 struct slave *slave = NULL;
1700
1701 slave = kzalloc(size: sizeof(*slave), GFP_KERNEL);
1702 if (!slave)
1703 return NULL;
1704
1705 slave->bond = bond;
1706 slave->dev = slave_dev;
1707 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1708
1709 if (bond_kobj_init(slave))
1710 return NULL;
1711
1712 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1713 SLAVE_AD_INFO(slave) = kzalloc(size: sizeof(struct ad_slave_info),
1714 GFP_KERNEL);
1715 if (!SLAVE_AD_INFO(slave)) {
1716 kobject_put(kobj: &slave->kobj);
1717 return NULL;
1718 }
1719 }
1720
1721 return slave;
1722}
1723
1724static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1725{
1726 info->bond_mode = BOND_MODE(bond);
1727 info->miimon = bond->params.miimon;
1728 info->num_slaves = bond->slave_cnt;
1729}
1730
1731static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1732{
1733 strcpy(p: info->slave_name, q: slave->dev->name);
1734 info->link = slave->link;
1735 info->state = bond_slave_state(slave);
1736 info->link_failure_count = slave->link_failure_count;
1737}
1738
1739static void bond_netdev_notify_work(struct work_struct *_work)
1740{
1741 struct slave *slave = container_of(_work, struct slave,
1742 notify_work.work);
1743
1744 if (rtnl_trylock()) {
1745 struct netdev_bonding_info binfo;
1746
1747 bond_fill_ifslave(slave, info: &binfo.slave);
1748 bond_fill_ifbond(bond: slave->bond, info: &binfo.master);
1749 netdev_bonding_info_change(dev: slave->dev, bonding_info: &binfo);
1750 rtnl_unlock();
1751 } else {
1752 queue_delayed_work(wq: slave->bond->wq, dwork: &slave->notify_work, delay: 1);
1753 }
1754}
1755
1756void bond_queue_slave_event(struct slave *slave)
1757{
1758 queue_delayed_work(wq: slave->bond->wq, dwork: &slave->notify_work, delay: 0);
1759}
1760
1761void bond_lower_state_changed(struct slave *slave)
1762{
1763 struct netdev_lag_lower_state_info info;
1764
1765 info.link_up = slave->link == BOND_LINK_UP ||
1766 slave->link == BOND_LINK_FAIL;
1767 info.tx_enabled = bond_is_active_slave(slave);
1768 netdev_lower_state_changed(lower_dev: slave->dev, lower_state_info: &info);
1769}
1770
1771#define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1772 if (extack) \
1773 NL_SET_ERR_MSG(extack, errmsg); \
1774 else \
1775 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1776} while (0)
1777
1778#define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1779 if (extack) \
1780 NL_SET_ERR_MSG(extack, errmsg); \
1781 else \
1782 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1783} while (0)
1784
1785/* The bonding driver uses ether_setup() to convert a master bond device
1786 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1787 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1788 * if they were set
1789 */
1790static void bond_ether_setup(struct net_device *bond_dev)
1791{
1792 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1793
1794 ether_setup(dev: bond_dev);
1795 bond_dev->flags |= IFF_MASTER | flags;
1796 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1797}
1798
1799void bond_xdp_set_features(struct net_device *bond_dev)
1800{
1801 struct bonding *bond = netdev_priv(dev: bond_dev);
1802 xdp_features_t val = NETDEV_XDP_ACT_MASK;
1803 struct list_head *iter;
1804 struct slave *slave;
1805
1806 ASSERT_RTNL();
1807
1808 if (!bond_xdp_check(bond)) {
1809 xdp_clear_features_flag(dev: bond_dev);
1810 return;
1811 }
1812
1813 bond_for_each_slave(bond, slave, iter)
1814 val &= slave->dev->xdp_features;
1815
1816 xdp_set_features_flag(dev: bond_dev, val);
1817}
1818
1819/* enslave device <slave> to bond device <master> */
1820int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1821 struct netlink_ext_ack *extack)
1822{
1823 struct bonding *bond = netdev_priv(dev: bond_dev);
1824 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1825 struct slave *new_slave = NULL, *prev_slave;
1826 struct sockaddr_storage ss;
1827 int link_reporting;
1828 int res = 0, i;
1829
1830 if (slave_dev->flags & IFF_MASTER &&
1831 !netif_is_bond_master(dev: slave_dev)) {
1832 BOND_NL_ERR(bond_dev, extack,
1833 "Device type (master device) cannot be enslaved");
1834 return -EPERM;
1835 }
1836
1837 if (!bond->params.use_carrier &&
1838 slave_dev->ethtool_ops->get_link == NULL &&
1839 slave_ops->ndo_eth_ioctl == NULL) {
1840 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1841 }
1842
1843 /* already in-use? */
1844 if (netdev_is_rx_handler_busy(dev: slave_dev)) {
1845 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1846 "Device is in use and cannot be enslaved");
1847 return -EBUSY;
1848 }
1849
1850 if (bond_dev == slave_dev) {
1851 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1852 return -EPERM;
1853 }
1854
1855 /* vlan challenged mutual exclusion */
1856 /* no need to lock since we're protected by rtnl_lock */
1857 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1858 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1859 if (vlan_uses_dev(dev: bond_dev)) {
1860 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1861 "Can not enslave VLAN challenged device to VLAN enabled bond");
1862 return -EPERM;
1863 } else {
1864 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1865 }
1866 } else {
1867 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1868 }
1869
1870 if (slave_dev->features & NETIF_F_HW_ESP)
1871 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1872
1873 /* Old ifenslave binaries are no longer supported. These can
1874 * be identified with moderate accuracy by the state of the slave:
1875 * the current ifenslave will set the interface down prior to
1876 * enslaving it; the old ifenslave will not.
1877 */
1878 if (slave_dev->flags & IFF_UP) {
1879 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1880 "Device can not be enslaved while up");
1881 return -EPERM;
1882 }
1883
1884 /* set bonding device ether type by slave - bonding netdevices are
1885 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1886 * there is a need to override some of the type dependent attribs/funcs.
1887 *
1888 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1889 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1890 */
1891 if (!bond_has_slaves(bond)) {
1892 if (bond_dev->type != slave_dev->type) {
1893 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1894 bond_dev->type, slave_dev->type);
1895
1896 res = call_netdevice_notifiers(val: NETDEV_PRE_TYPE_CHANGE,
1897 dev: bond_dev);
1898 res = notifier_to_errno(ret: res);
1899 if (res) {
1900 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1901 return -EBUSY;
1902 }
1903
1904 /* Flush unicast and multicast addresses */
1905 dev_uc_flush(dev: bond_dev);
1906 dev_mc_flush(dev: bond_dev);
1907
1908 if (slave_dev->type != ARPHRD_ETHER)
1909 bond_setup_by_slave(bond_dev, slave_dev);
1910 else
1911 bond_ether_setup(bond_dev);
1912
1913 call_netdevice_notifiers(val: NETDEV_POST_TYPE_CHANGE,
1914 dev: bond_dev);
1915 }
1916 } else if (bond_dev->type != slave_dev->type) {
1917 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1918 "Device type is different from other slaves");
1919 return -EINVAL;
1920 }
1921
1922 if (slave_dev->type == ARPHRD_INFINIBAND &&
1923 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1924 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1925 "Only active-backup mode is supported for infiniband slaves");
1926 res = -EOPNOTSUPP;
1927 goto err_undo_flags;
1928 }
1929
1930 if (!slave_ops->ndo_set_mac_address ||
1931 slave_dev->type == ARPHRD_INFINIBAND) {
1932 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1933 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1934 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1935 if (!bond_has_slaves(bond)) {
1936 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1937 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1938 } else {
1939 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1940 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1941 res = -EOPNOTSUPP;
1942 goto err_undo_flags;
1943 }
1944 }
1945 }
1946
1947 call_netdevice_notifiers(val: NETDEV_JOIN, dev: slave_dev);
1948
1949 /* If this is the first slave, then we need to set the master's hardware
1950 * address to be the same as the slave's.
1951 */
1952 if (!bond_has_slaves(bond) &&
1953 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1954 res = bond_set_dev_addr(bond_dev: bond->dev, slave_dev);
1955 if (res)
1956 goto err_undo_flags;
1957 }
1958
1959 new_slave = bond_alloc_slave(bond, slave_dev);
1960 if (!new_slave) {
1961 res = -ENOMEM;
1962 goto err_undo_flags;
1963 }
1964
1965 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1966 * is set via sysfs or module option if desired.
1967 */
1968 new_slave->queue_id = 0;
1969
1970 /* Save slave's original mtu and then set it to match the bond */
1971 new_slave->original_mtu = slave_dev->mtu;
1972 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1973 if (res) {
1974 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1975 goto err_free;
1976 }
1977
1978 /* Save slave's original ("permanent") mac address for modes
1979 * that need it, and for restoring it upon release, and then
1980 * set it to the master's address
1981 */
1982 bond_hw_addr_copy(dst: new_slave->perm_hwaddr, src: slave_dev->dev_addr,
1983 len: slave_dev->addr_len);
1984
1985 if (!bond->params.fail_over_mac ||
1986 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1987 /* Set slave to master's mac address. The application already
1988 * set the master's mac address to that of the first slave
1989 */
1990 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1991 ss.ss_family = slave_dev->type;
1992 res = dev_set_mac_address(dev: slave_dev, sa: (struct sockaddr *)&ss,
1993 extack);
1994 if (res) {
1995 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1996 goto err_restore_mtu;
1997 }
1998 }
1999
2000 /* set no_addrconf flag before open to prevent IPv6 addrconf */
2001 slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2002
2003 /* open the slave since the application closed it */
2004 res = dev_open(dev: slave_dev, extack);
2005 if (res) {
2006 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2007 goto err_restore_mac;
2008 }
2009
2010 slave_dev->priv_flags |= IFF_BONDING;
2011 /* initialize slave stats */
2012 dev_get_stats(dev: new_slave->dev, storage: &new_slave->slave_stats);
2013
2014 if (bond_is_lb(bond)) {
2015 /* bond_alb_init_slave() must be called before all other stages since
2016 * it might fail and we do not want to have to undo everything
2017 */
2018 res = bond_alb_init_slave(bond, slave: new_slave);
2019 if (res)
2020 goto err_close;
2021 }
2022
2023 res = vlan_vids_add_by_dev(dev: slave_dev, by_dev: bond_dev);
2024 if (res) {
2025 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2026 goto err_close;
2027 }
2028
2029 prev_slave = bond_last_slave(bond);
2030
2031 new_slave->delay = 0;
2032 new_slave->link_failure_count = 0;
2033
2034 if (bond_update_speed_duplex(slave: new_slave) &&
2035 bond_needs_speed_duplex(bond))
2036 new_slave->link = BOND_LINK_DOWN;
2037
2038 new_slave->last_rx = jiffies -
2039 (msecs_to_jiffies(m: bond->params.arp_interval) + 1);
2040 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2041 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2042
2043 new_slave->last_tx = new_slave->last_rx;
2044
2045 if (bond->params.miimon && !bond->params.use_carrier) {
2046 link_reporting = bond_check_dev_link(bond, slave_dev, reporting: 1);
2047
2048 if ((link_reporting == -1) && !bond->params.arp_interval) {
2049 /* miimon is set but a bonded network driver
2050 * does not support ETHTOOL/MII and
2051 * arp_interval is not set. Note: if
2052 * use_carrier is enabled, we will never go
2053 * here (because netif_carrier is always
2054 * supported); thus, we don't need to change
2055 * the messages for netif_carrier.
2056 */
2057 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");
2058 } else if (link_reporting == -1) {
2059 /* unable get link status using mii/ethtool */
2060 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");
2061 }
2062 }
2063
2064 /* check for initial state */
2065 new_slave->link = BOND_LINK_NOCHANGE;
2066 if (bond->params.miimon) {
2067 if (bond_check_dev_link(bond, slave_dev, reporting: 0) == BMSR_LSTATUS) {
2068 if (bond->params.updelay) {
2069 bond_set_slave_link_state(slave: new_slave,
2070 BOND_LINK_BACK,
2071 BOND_SLAVE_NOTIFY_NOW);
2072 new_slave->delay = bond->params.updelay;
2073 } else {
2074 bond_set_slave_link_state(slave: new_slave,
2075 BOND_LINK_UP,
2076 BOND_SLAVE_NOTIFY_NOW);
2077 }
2078 } else {
2079 bond_set_slave_link_state(slave: new_slave, BOND_LINK_DOWN,
2080 BOND_SLAVE_NOTIFY_NOW);
2081 }
2082 } else if (bond->params.arp_interval) {
2083 bond_set_slave_link_state(slave: new_slave,
2084 state: (netif_carrier_ok(dev: slave_dev) ?
2085 BOND_LINK_UP : BOND_LINK_DOWN),
2086 BOND_SLAVE_NOTIFY_NOW);
2087 } else {
2088 bond_set_slave_link_state(slave: new_slave, BOND_LINK_UP,
2089 BOND_SLAVE_NOTIFY_NOW);
2090 }
2091
2092 if (new_slave->link != BOND_LINK_DOWN)
2093 new_slave->last_link_up = jiffies;
2094 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2095 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2096 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2097
2098 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2099 /* if there is a primary slave, remember it */
2100 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2101 rcu_assign_pointer(bond->primary_slave, new_slave);
2102 bond->force_primary = true;
2103 }
2104 }
2105
2106 switch (BOND_MODE(bond)) {
2107 case BOND_MODE_ACTIVEBACKUP:
2108 bond_set_slave_inactive_flags(slave: new_slave,
2109 BOND_SLAVE_NOTIFY_NOW);
2110 break;
2111 case BOND_MODE_8023AD:
2112 /* in 802.3ad mode, the internal mechanism
2113 * will activate the slaves in the selected
2114 * aggregator
2115 */
2116 bond_set_slave_inactive_flags(slave: new_slave, BOND_SLAVE_NOTIFY_NOW);
2117 /* if this is the first slave */
2118 if (!prev_slave) {
2119 SLAVE_AD_INFO(new_slave)->id = 1;
2120 /* Initialize AD with the number of times that the AD timer is called in 1 second
2121 * can be called only after the mac address of the bond is set
2122 */
2123 bond_3ad_initialize(bond);
2124 } else {
2125 SLAVE_AD_INFO(new_slave)->id =
2126 SLAVE_AD_INFO(prev_slave)->id + 1;
2127 }
2128
2129 bond_3ad_bind_slave(slave: new_slave);
2130 break;
2131 case BOND_MODE_TLB:
2132 case BOND_MODE_ALB:
2133 bond_set_active_slave(slave: new_slave);
2134 bond_set_slave_inactive_flags(slave: new_slave, BOND_SLAVE_NOTIFY_NOW);
2135 break;
2136 default:
2137 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2138
2139 /* always active in trunk mode */
2140 bond_set_active_slave(slave: new_slave);
2141
2142 /* In trunking mode there is little meaning to curr_active_slave
2143 * anyway (it holds no special properties of the bond device),
2144 * so we can change it without calling change_active_interface()
2145 */
2146 if (!rcu_access_pointer(bond->curr_active_slave) &&
2147 new_slave->link == BOND_LINK_UP)
2148 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2149
2150 break;
2151 } /* switch(bond_mode) */
2152
2153#ifdef CONFIG_NET_POLL_CONTROLLER
2154 if (bond->dev->npinfo) {
2155 if (slave_enable_netpoll(slave: new_slave)) {
2156 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2157 res = -EBUSY;
2158 goto err_detach;
2159 }
2160 }
2161#endif
2162
2163 if (!(bond_dev->features & NETIF_F_LRO))
2164 dev_disable_lro(dev: slave_dev);
2165
2166 res = netdev_rx_handler_register(dev: slave_dev, rx_handler: bond_handle_frame,
2167 rx_handler_data: new_slave);
2168 if (res) {
2169 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2170 goto err_detach;
2171 }
2172
2173 res = bond_master_upper_dev_link(bond, slave: new_slave, extack);
2174 if (res) {
2175 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2176 goto err_unregister;
2177 }
2178
2179 bond_lower_state_changed(slave: new_slave);
2180
2181 res = bond_sysfs_slave_add(slave: new_slave);
2182 if (res) {
2183 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2184 goto err_upper_unlink;
2185 }
2186
2187 /* If the mode uses primary, then the following is handled by
2188 * bond_change_active_slave().
2189 */
2190 if (!bond_uses_primary(bond)) {
2191 /* set promiscuity level to new slave */
2192 if (bond_dev->flags & IFF_PROMISC) {
2193 res = dev_set_promiscuity(dev: slave_dev, inc: 1);
2194 if (res)
2195 goto err_sysfs_del;
2196 }
2197
2198 /* set allmulti level to new slave */
2199 if (bond_dev->flags & IFF_ALLMULTI) {
2200 res = dev_set_allmulti(dev: slave_dev, inc: 1);
2201 if (res) {
2202 if (bond_dev->flags & IFF_PROMISC)
2203 dev_set_promiscuity(dev: slave_dev, inc: -1);
2204 goto err_sysfs_del;
2205 }
2206 }
2207
2208 if (bond_dev->flags & IFF_UP) {
2209 netif_addr_lock_bh(dev: bond_dev);
2210 dev_mc_sync_multiple(to: slave_dev, from: bond_dev);
2211 dev_uc_sync_multiple(to: slave_dev, from: bond_dev);
2212 netif_addr_unlock_bh(dev: bond_dev);
2213
2214 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2215 dev_mc_add(dev: slave_dev, addr: lacpdu_mcast_addr);
2216 }
2217 }
2218
2219 bond->slave_cnt++;
2220 bond_compute_features(bond);
2221 bond_set_carrier(bond);
2222
2223 if (bond_uses_primary(bond)) {
2224 block_netpoll_tx();
2225 bond_select_active_slave(bond);
2226 unblock_netpoll_tx();
2227 }
2228
2229 if (bond_mode_can_use_xmit_hash(bond))
2230 bond_update_slave_arr(bond, NULL);
2231
2232
2233 if (!slave_dev->netdev_ops->ndo_bpf ||
2234 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2235 if (bond->xdp_prog) {
2236 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2237 "Slave does not support XDP");
2238 res = -EOPNOTSUPP;
2239 goto err_sysfs_del;
2240 }
2241 } else if (bond->xdp_prog) {
2242 struct netdev_bpf xdp = {
2243 .command = XDP_SETUP_PROG,
2244 .flags = 0,
2245 .prog = bond->xdp_prog,
2246 .extack = extack,
2247 };
2248
2249 if (dev_xdp_prog_count(dev: slave_dev) > 0) {
2250 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2251 "Slave has XDP program loaded, please unload before enslaving");
2252 res = -EOPNOTSUPP;
2253 goto err_sysfs_del;
2254 }
2255
2256 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2257 if (res < 0) {
2258 /* ndo_bpf() sets extack error message */
2259 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2260 goto err_sysfs_del;
2261 }
2262 if (bond->xdp_prog)
2263 bpf_prog_inc(prog: bond->xdp_prog);
2264 }
2265
2266 bond_xdp_set_features(bond_dev);
2267
2268 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2269 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2270 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2271
2272 /* enslave is successful */
2273 bond_queue_slave_event(slave: new_slave);
2274 return 0;
2275
2276/* Undo stages on error */
2277err_sysfs_del:
2278 bond_sysfs_slave_del(slave: new_slave);
2279
2280err_upper_unlink:
2281 bond_upper_dev_unlink(bond, slave: new_slave);
2282
2283err_unregister:
2284 netdev_rx_handler_unregister(dev: slave_dev);
2285
2286err_detach:
2287 vlan_vids_del_by_dev(dev: slave_dev, by_dev: bond_dev);
2288 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2289 RCU_INIT_POINTER(bond->primary_slave, NULL);
2290 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2291 block_netpoll_tx();
2292 bond_change_active_slave(bond, NULL);
2293 bond_select_active_slave(bond);
2294 unblock_netpoll_tx();
2295 }
2296 /* either primary_slave or curr_active_slave might've changed */
2297 synchronize_rcu();
2298 slave_disable_netpoll(slave: new_slave);
2299
2300err_close:
2301 if (!netif_is_bond_master(dev: slave_dev))
2302 slave_dev->priv_flags &= ~IFF_BONDING;
2303 dev_close(dev: slave_dev);
2304
2305err_restore_mac:
2306 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2307 if (!bond->params.fail_over_mac ||
2308 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2309 /* XXX TODO - fom follow mode needs to change master's
2310 * MAC if this slave's MAC is in use by the bond, or at
2311 * least print a warning.
2312 */
2313 bond_hw_addr_copy(dst: ss.__data, src: new_slave->perm_hwaddr,
2314 len: new_slave->dev->addr_len);
2315 ss.ss_family = slave_dev->type;
2316 dev_set_mac_address(dev: slave_dev, sa: (struct sockaddr *)&ss, NULL);
2317 }
2318
2319err_restore_mtu:
2320 dev_set_mtu(slave_dev, new_slave->original_mtu);
2321
2322err_free:
2323 kobject_put(kobj: &new_slave->kobj);
2324
2325err_undo_flags:
2326 /* Enslave of first slave has failed and we need to fix master's mac */
2327 if (!bond_has_slaves(bond)) {
2328 if (ether_addr_equal_64bits(addr1: bond_dev->dev_addr,
2329 addr2: slave_dev->dev_addr))
2330 eth_hw_addr_random(dev: bond_dev);
2331 if (bond_dev->type != ARPHRD_ETHER) {
2332 dev_close(dev: bond_dev);
2333 bond_ether_setup(bond_dev);
2334 }
2335 }
2336
2337 return res;
2338}
2339
2340/* Try to release the slave device <slave> from the bond device <master>
2341 * It is legal to access curr_active_slave without a lock because all the function
2342 * is RTNL-locked. If "all" is true it means that the function is being called
2343 * while destroying a bond interface and all slaves are being released.
2344 *
2345 * The rules for slave state should be:
2346 * for Active/Backup:
2347 * Active stays on all backups go down
2348 * for Bonded connections:
2349 * The first up interface should be left on and all others downed.
2350 */
2351static int __bond_release_one(struct net_device *bond_dev,
2352 struct net_device *slave_dev,
2353 bool all, bool unregister)
2354{
2355 struct bonding *bond = netdev_priv(dev: bond_dev);
2356 struct slave *slave, *oldcurrent;
2357 struct sockaddr_storage ss;
2358 int old_flags = bond_dev->flags;
2359 netdev_features_t old_features = bond_dev->features;
2360
2361 /* slave is not a slave or master is not master of this slave */
2362 if (!(slave_dev->flags & IFF_SLAVE) ||
2363 !netdev_has_upper_dev(dev: slave_dev, upper_dev: bond_dev)) {
2364 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2365 return -EINVAL;
2366 }
2367
2368 block_netpoll_tx();
2369
2370 slave = bond_get_slave_by_dev(bond, slave_dev);
2371 if (!slave) {
2372 /* not a slave of this bond */
2373 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2374 unblock_netpoll_tx();
2375 return -EINVAL;
2376 }
2377
2378 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2379
2380 bond_sysfs_slave_del(slave);
2381
2382 /* recompute stats just before removing the slave */
2383 bond_get_stats(bond_dev: bond->dev, stats: &bond->bond_stats);
2384
2385 if (bond->xdp_prog) {
2386 struct netdev_bpf xdp = {
2387 .command = XDP_SETUP_PROG,
2388 .flags = 0,
2389 .prog = NULL,
2390 .extack = NULL,
2391 };
2392 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2393 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2394 }
2395
2396 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2397 * for this slave anymore.
2398 */
2399 netdev_rx_handler_unregister(dev: slave_dev);
2400
2401 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2402 bond_3ad_unbind_slave(slave);
2403
2404 bond_upper_dev_unlink(bond, slave);
2405
2406 if (bond_mode_can_use_xmit_hash(bond))
2407 bond_update_slave_arr(bond, skipslave: slave);
2408
2409 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2410 bond_is_active_slave(slave) ? "active" : "backup");
2411
2412 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2413
2414 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2415
2416 if (!all && (!bond->params.fail_over_mac ||
2417 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2418 if (ether_addr_equal_64bits(addr1: bond_dev->dev_addr, addr2: slave->perm_hwaddr) &&
2419 bond_has_slaves(bond))
2420 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",
2421 slave->perm_hwaddr);
2422 }
2423
2424 if (rtnl_dereference(bond->primary_slave) == slave)
2425 RCU_INIT_POINTER(bond->primary_slave, NULL);
2426
2427 if (oldcurrent == slave)
2428 bond_change_active_slave(bond, NULL);
2429
2430 if (bond_is_lb(bond)) {
2431 /* Must be called only after the slave has been
2432 * detached from the list and the curr_active_slave
2433 * has been cleared (if our_slave == old_current),
2434 * but before a new active slave is selected.
2435 */
2436 bond_alb_deinit_slave(bond, slave);
2437 }
2438
2439 if (all) {
2440 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2441 } else if (oldcurrent == slave) {
2442 /* Note that we hold RTNL over this sequence, so there
2443 * is no concern that another slave add/remove event
2444 * will interfere.
2445 */
2446 bond_select_active_slave(bond);
2447 }
2448
2449 bond_set_carrier(bond);
2450 if (!bond_has_slaves(bond))
2451 eth_hw_addr_random(dev: bond_dev);
2452
2453 unblock_netpoll_tx();
2454 synchronize_rcu();
2455 bond->slave_cnt--;
2456
2457 if (!bond_has_slaves(bond)) {
2458 call_netdevice_notifiers(val: NETDEV_CHANGEADDR, dev: bond->dev);
2459 call_netdevice_notifiers(val: NETDEV_RELEASE, dev: bond->dev);
2460 }
2461
2462 bond_compute_features(bond);
2463 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2464 (old_features & NETIF_F_VLAN_CHALLENGED))
2465 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2466
2467 vlan_vids_del_by_dev(dev: slave_dev, by_dev: bond_dev);
2468
2469 /* If the mode uses primary, then this case was handled above by
2470 * bond_change_active_slave(..., NULL)
2471 */
2472 if (!bond_uses_primary(bond)) {
2473 /* unset promiscuity level from slave
2474 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2475 * of the IFF_PROMISC flag in the bond_dev, but we need the
2476 * value of that flag before that change, as that was the value
2477 * when this slave was attached, so we cache at the start of the
2478 * function and use it here. Same goes for ALLMULTI below
2479 */
2480 if (old_flags & IFF_PROMISC)
2481 dev_set_promiscuity(dev: slave_dev, inc: -1);
2482
2483 /* unset allmulti level from slave */
2484 if (old_flags & IFF_ALLMULTI)
2485 dev_set_allmulti(dev: slave_dev, inc: -1);
2486
2487 if (old_flags & IFF_UP)
2488 bond_hw_addr_flush(bond_dev, slave_dev);
2489 }
2490
2491 slave_disable_netpoll(slave);
2492
2493 /* close slave before restoring its mac address */
2494 dev_close(dev: slave_dev);
2495
2496 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2497
2498 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2499 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2500 /* restore original ("permanent") mac address */
2501 bond_hw_addr_copy(dst: ss.__data, src: slave->perm_hwaddr,
2502 len: slave->dev->addr_len);
2503 ss.ss_family = slave_dev->type;
2504 dev_set_mac_address(dev: slave_dev, sa: (struct sockaddr *)&ss, NULL);
2505 }
2506
2507 if (unregister)
2508 __dev_set_mtu(slave_dev, slave->original_mtu);
2509 else
2510 dev_set_mtu(slave_dev, slave->original_mtu);
2511
2512 if (!netif_is_bond_master(dev: slave_dev))
2513 slave_dev->priv_flags &= ~IFF_BONDING;
2514
2515 bond_xdp_set_features(bond_dev);
2516 kobject_put(kobj: &slave->kobj);
2517
2518 return 0;
2519}
2520
2521/* A wrapper used because of ndo_del_link */
2522int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2523{
2524 return __bond_release_one(bond_dev, slave_dev, all: false, unregister: false);
2525}
2526
2527/* First release a slave and then destroy the bond if no more slaves are left.
2528 * Must be under rtnl_lock when this function is called.
2529 */
2530static int bond_release_and_destroy(struct net_device *bond_dev,
2531 struct net_device *slave_dev)
2532{
2533 struct bonding *bond = netdev_priv(dev: bond_dev);
2534 int ret;
2535
2536 ret = __bond_release_one(bond_dev, slave_dev, all: false, unregister: true);
2537 if (ret == 0 && !bond_has_slaves(bond) &&
2538 bond_dev->reg_state != NETREG_UNREGISTERING) {
2539 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2540 netdev_info(dev: bond_dev, format: "Destroying bond\n");
2541 bond_remove_proc_entry(bond);
2542 unregister_netdevice(dev: bond_dev);
2543 }
2544 return ret;
2545}
2546
2547static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2548{
2549 struct bonding *bond = netdev_priv(dev: bond_dev);
2550
2551 bond_fill_ifbond(bond, info);
2552}
2553
2554static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2555{
2556 struct bonding *bond = netdev_priv(dev: bond_dev);
2557 struct list_head *iter;
2558 int i = 0, res = -ENODEV;
2559 struct slave *slave;
2560
2561 bond_for_each_slave(bond, slave, iter) {
2562 if (i++ == (int)info->slave_id) {
2563 res = 0;
2564 bond_fill_ifslave(slave, info);
2565 break;
2566 }
2567 }
2568
2569 return res;
2570}
2571
2572/*-------------------------------- Monitoring -------------------------------*/
2573
2574/* called with rcu_read_lock() */
2575static int bond_miimon_inspect(struct bonding *bond)
2576{
2577 bool ignore_updelay = false;
2578 int link_state, commit = 0;
2579 struct list_head *iter;
2580 struct slave *slave;
2581
2582 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2583 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2584 } else {
2585 struct bond_up_slave *usable_slaves;
2586
2587 usable_slaves = rcu_dereference(bond->usable_slaves);
2588
2589 if (usable_slaves && usable_slaves->count == 0)
2590 ignore_updelay = true;
2591 }
2592
2593 bond_for_each_slave_rcu(bond, slave, iter) {
2594 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2595
2596 link_state = bond_check_dev_link(bond, slave_dev: slave->dev, reporting: 0);
2597
2598 switch (slave->link) {
2599 case BOND_LINK_UP:
2600 if (link_state)
2601 continue;
2602
2603 bond_propose_link_state(slave, BOND_LINK_FAIL);
2604 commit++;
2605 slave->delay = bond->params.downdelay;
2606 if (slave->delay) {
2607 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2608 (BOND_MODE(bond) ==
2609 BOND_MODE_ACTIVEBACKUP) ?
2610 (bond_is_active_slave(slave) ?
2611 "active " : "backup ") : "",
2612 bond->params.downdelay * bond->params.miimon);
2613 }
2614 fallthrough;
2615 case BOND_LINK_FAIL:
2616 if (link_state) {
2617 /* recovered before downdelay expired */
2618 bond_propose_link_state(slave, BOND_LINK_UP);
2619 slave->last_link_up = jiffies;
2620 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2621 (bond->params.downdelay - slave->delay) *
2622 bond->params.miimon);
2623 commit++;
2624 continue;
2625 }
2626
2627 if (slave->delay <= 0) {
2628 bond_propose_link_state(slave, BOND_LINK_DOWN);
2629 commit++;
2630 continue;
2631 }
2632
2633 slave->delay--;
2634 break;
2635
2636 case BOND_LINK_DOWN:
2637 if (!link_state)
2638 continue;
2639
2640 bond_propose_link_state(slave, BOND_LINK_BACK);
2641 commit++;
2642 slave->delay = bond->params.updelay;
2643
2644 if (slave->delay) {
2645 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2646 ignore_updelay ? 0 :
2647 bond->params.updelay *
2648 bond->params.miimon);
2649 }
2650 fallthrough;
2651 case BOND_LINK_BACK:
2652 if (!link_state) {
2653 bond_propose_link_state(slave, BOND_LINK_DOWN);
2654 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2655 (bond->params.updelay - slave->delay) *
2656 bond->params.miimon);
2657 commit++;
2658 continue;
2659 }
2660
2661 if (ignore_updelay)
2662 slave->delay = 0;
2663
2664 if (slave->delay <= 0) {
2665 bond_propose_link_state(slave, BOND_LINK_UP);
2666 commit++;
2667 ignore_updelay = false;
2668 continue;
2669 }
2670
2671 slave->delay--;
2672 break;
2673 }
2674 }
2675
2676 return commit;
2677}
2678
2679static void bond_miimon_link_change(struct bonding *bond,
2680 struct slave *slave,
2681 char link)
2682{
2683 switch (BOND_MODE(bond)) {
2684 case BOND_MODE_8023AD:
2685 bond_3ad_handle_link_change(slave, link);
2686 break;
2687 case BOND_MODE_TLB:
2688 case BOND_MODE_ALB:
2689 bond_alb_handle_link_change(bond, slave, link);
2690 break;
2691 case BOND_MODE_XOR:
2692 bond_update_slave_arr(bond, NULL);
2693 break;
2694 }
2695}
2696
2697static void bond_miimon_commit(struct bonding *bond)
2698{
2699 struct slave *slave, *primary, *active;
2700 bool do_failover = false;
2701 struct list_head *iter;
2702
2703 ASSERT_RTNL();
2704
2705 bond_for_each_slave(bond, slave, iter) {
2706 switch (slave->link_new_state) {
2707 case BOND_LINK_NOCHANGE:
2708 /* For 802.3ad mode, check current slave speed and
2709 * duplex again in case its port was disabled after
2710 * invalid speed/duplex reporting but recovered before
2711 * link monitoring could make a decision on the actual
2712 * link status
2713 */
2714 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2715 slave->link == BOND_LINK_UP)
2716 bond_3ad_adapter_speed_duplex_changed(slave);
2717 continue;
2718
2719 case BOND_LINK_UP:
2720 if (bond_update_speed_duplex(slave) &&
2721 bond_needs_speed_duplex(bond)) {
2722 slave->link = BOND_LINK_DOWN;
2723 if (net_ratelimit())
2724 slave_warn(bond->dev, slave->dev,
2725 "failed to get link speed/duplex\n");
2726 continue;
2727 }
2728 bond_set_slave_link_state(slave, BOND_LINK_UP,
2729 BOND_SLAVE_NOTIFY_NOW);
2730 slave->last_link_up = jiffies;
2731
2732 primary = rtnl_dereference(bond->primary_slave);
2733 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2734 /* prevent it from being the active one */
2735 bond_set_backup_slave(slave);
2736 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2737 /* make it immediately active */
2738 bond_set_active_slave(slave);
2739 }
2740
2741 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2742 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2743 slave->duplex ? "full" : "half");
2744
2745 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2746
2747 active = rtnl_dereference(bond->curr_active_slave);
2748 if (!active || slave == primary || slave->prio > active->prio)
2749 do_failover = true;
2750
2751 continue;
2752
2753 case BOND_LINK_DOWN:
2754 if (slave->link_failure_count < UINT_MAX)
2755 slave->link_failure_count++;
2756
2757 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2758 BOND_SLAVE_NOTIFY_NOW);
2759
2760 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2761 BOND_MODE(bond) == BOND_MODE_8023AD)
2762 bond_set_slave_inactive_flags(slave,
2763 BOND_SLAVE_NOTIFY_NOW);
2764
2765 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2766
2767 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2768
2769 if (slave == rcu_access_pointer(bond->curr_active_slave))
2770 do_failover = true;
2771
2772 continue;
2773
2774 default:
2775 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2776 slave->link_new_state);
2777 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2778
2779 continue;
2780 }
2781 }
2782
2783 if (do_failover) {
2784 block_netpoll_tx();
2785 bond_select_active_slave(bond);
2786 unblock_netpoll_tx();
2787 }
2788
2789 bond_set_carrier(bond);
2790}
2791
2792/* bond_mii_monitor
2793 *
2794 * Really a wrapper that splits the mii monitor into two phases: an
2795 * inspection, then (if inspection indicates something needs to be done)
2796 * an acquisition of appropriate locks followed by a commit phase to
2797 * implement whatever link state changes are indicated.
2798 */
2799static void bond_mii_monitor(struct work_struct *work)
2800{
2801 struct bonding *bond = container_of(work, struct bonding,
2802 mii_work.work);
2803 bool should_notify_peers = false;
2804 bool commit;
2805 unsigned long delay;
2806 struct slave *slave;
2807 struct list_head *iter;
2808
2809 delay = msecs_to_jiffies(m: bond->params.miimon);
2810
2811 if (!bond_has_slaves(bond))
2812 goto re_arm;
2813
2814 rcu_read_lock();
2815 should_notify_peers = bond_should_notify_peers(bond);
2816 commit = !!bond_miimon_inspect(bond);
2817 if (bond->send_peer_notif) {
2818 rcu_read_unlock();
2819 if (rtnl_trylock()) {
2820 bond->send_peer_notif--;
2821 rtnl_unlock();
2822 }
2823 } else {
2824 rcu_read_unlock();
2825 }
2826
2827 if (commit) {
2828 /* Race avoidance with bond_close cancel of workqueue */
2829 if (!rtnl_trylock()) {
2830 delay = 1;
2831 should_notify_peers = false;
2832 goto re_arm;
2833 }
2834
2835 bond_for_each_slave(bond, slave, iter) {
2836 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2837 }
2838 bond_miimon_commit(bond);
2839
2840 rtnl_unlock(); /* might sleep, hold no other locks */
2841 }
2842
2843re_arm:
2844 if (bond->params.miimon)
2845 queue_delayed_work(wq: bond->wq, dwork: &bond->mii_work, delay);
2846
2847 if (should_notify_peers) {
2848 if (!rtnl_trylock())
2849 return;
2850 call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS, dev: bond->dev);
2851 rtnl_unlock();
2852 }
2853}
2854
2855static int bond_upper_dev_walk(struct net_device *upper,
2856 struct netdev_nested_priv *priv)
2857{
2858 __be32 ip = *(__be32 *)priv->data;
2859
2860 return ip == bond_confirm_addr(dev: upper, dst: 0, local: ip);
2861}
2862
2863static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2864{
2865 struct netdev_nested_priv priv = {
2866 .data = (void *)&ip,
2867 };
2868 bool ret = false;
2869
2870 if (ip == bond_confirm_addr(dev: bond->dev, dst: 0, local: ip))
2871 return true;
2872
2873 rcu_read_lock();
2874 if (netdev_walk_all_upper_dev_rcu(dev: bond->dev, fn: bond_upper_dev_walk, priv: &priv))
2875 ret = true;
2876 rcu_read_unlock();
2877
2878 return ret;
2879}
2880
2881#define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
2882
2883static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2884 struct sk_buff *skb)
2885{
2886 struct net_device *bond_dev = slave->bond->dev;
2887 struct net_device *slave_dev = slave->dev;
2888 struct bond_vlan_tag *outer_tag = tags;
2889
2890 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
2891 return true;
2892
2893 tags++;
2894
2895 /* Go through all the tags backwards and add them to the packet */
2896 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
2897 if (!tags->vlan_id) {
2898 tags++;
2899 continue;
2900 }
2901
2902 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2903 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2904 skb = vlan_insert_tag_set_proto(skb, vlan_proto: tags->vlan_proto,
2905 vlan_tci: tags->vlan_id);
2906 if (!skb) {
2907 net_err_ratelimited("failed to insert inner VLAN tag\n");
2908 return false;
2909 }
2910
2911 tags++;
2912 }
2913 /* Set the outer tag */
2914 if (outer_tag->vlan_id) {
2915 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2916 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2917 __vlan_hwaccel_put_tag(skb, vlan_proto: outer_tag->vlan_proto,
2918 vlan_tci: outer_tag->vlan_id);
2919 }
2920
2921 return true;
2922}
2923
2924/* We go to the (large) trouble of VLAN tagging ARP frames because
2925 * switches in VLAN mode (especially if ports are configured as
2926 * "native" to a VLAN) might not pass non-tagged frames.
2927 */
2928static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2929 __be32 src_ip, struct bond_vlan_tag *tags)
2930{
2931 struct net_device *bond_dev = slave->bond->dev;
2932 struct net_device *slave_dev = slave->dev;
2933 struct sk_buff *skb;
2934
2935 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2936 arp_op, &dest_ip, &src_ip);
2937
2938 skb = arp_create(type: arp_op, ETH_P_ARP, dest_ip, dev: slave_dev, src_ip,
2939 NULL, src_hw: slave_dev->dev_addr, NULL);
2940
2941 if (!skb) {
2942 net_err_ratelimited("ARP packet allocation failed\n");
2943 return;
2944 }
2945
2946 if (bond_handle_vlan(slave, tags, skb)) {
2947 slave_update_last_tx(slave);
2948 arp_xmit(skb);
2949 }
2950
2951 return;
2952}
2953
2954/* Validate the device path between the @start_dev and the @end_dev.
2955 * The path is valid if the @end_dev is reachable through device
2956 * stacking.
2957 * When the path is validated, collect any vlan information in the
2958 * path.
2959 */
2960struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2961 struct net_device *end_dev,
2962 int level)
2963{
2964 struct bond_vlan_tag *tags;
2965 struct net_device *upper;
2966 struct list_head *iter;
2967
2968 if (start_dev == end_dev) {
2969 tags = kcalloc(n: level + 1, size: sizeof(*tags), GFP_ATOMIC);
2970 if (!tags)
2971 return ERR_PTR(error: -ENOMEM);
2972 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
2973 return tags;
2974 }
2975
2976 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2977 tags = bond_verify_device_path(start_dev: upper, end_dev, level: level + 1);
2978 if (IS_ERR_OR_NULL(ptr: tags)) {
2979 if (IS_ERR(ptr: tags))
2980 return tags;
2981 continue;
2982 }
2983 if (is_vlan_dev(dev: upper)) {
2984 tags[level].vlan_proto = vlan_dev_vlan_proto(dev: upper);
2985 tags[level].vlan_id = vlan_dev_vlan_id(dev: upper);
2986 }
2987
2988 return tags;
2989 }
2990
2991 return NULL;
2992}
2993
2994static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2995{
2996 struct rtable *rt;
2997 struct bond_vlan_tag *tags;
2998 __be32 *targets = bond->params.arp_targets, addr;
2999 int i;
3000
3001 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3002 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3003 __func__, &targets[i]);
3004 tags = NULL;
3005
3006 /* Find out through which dev should the packet go */
3007 rt = ip_route_output(net: dev_net(dev: bond->dev), daddr: targets[i], saddr: 0,
3008 RTO_ONLINK, oif: 0);
3009 if (IS_ERR(ptr: rt)) {
3010 /* there's no route to target - try to send arp
3011 * probe to generate any traffic (arp_validate=0)
3012 */
3013 if (bond->params.arp_validate)
3014 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3015 bond->dev->name,
3016 &targets[i]);
3017 bond_arp_send(slave, ARPOP_REQUEST, dest_ip: targets[i],
3018 src_ip: 0, tags);
3019 continue;
3020 }
3021
3022 /* bond device itself */
3023 if (rt->dst.dev == bond->dev)
3024 goto found;
3025
3026 rcu_read_lock();
3027 tags = bond_verify_device_path(start_dev: bond->dev, end_dev: rt->dst.dev, level: 0);
3028 rcu_read_unlock();
3029
3030 if (!IS_ERR_OR_NULL(ptr: tags))
3031 goto found;
3032
3033 /* Not our device - skip */
3034 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3035 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3036
3037 ip_rt_put(rt);
3038 continue;
3039
3040found:
3041 addr = bond_confirm_addr(dev: rt->dst.dev, dst: targets[i], local: 0);
3042 ip_rt_put(rt);
3043 bond_arp_send(slave, ARPOP_REQUEST, dest_ip: targets[i], src_ip: addr, tags);
3044 kfree(objp: tags);
3045 }
3046}
3047
3048static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3049{
3050 int i;
3051
3052 if (!sip || !bond_has_this_ip(bond, ip: tip)) {
3053 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3054 __func__, &sip, &tip);
3055 return;
3056 }
3057
3058 i = bond_get_targets_ip(targets: bond->params.arp_targets, ip: sip);
3059 if (i == -1) {
3060 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3061 __func__, &sip);
3062 return;
3063 }
3064 slave->last_rx = jiffies;
3065 slave->target_last_arp_rx[i] = jiffies;
3066}
3067
3068static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3069 struct slave *slave)
3070{
3071 struct arphdr *arp = (struct arphdr *)skb->data;
3072 struct slave *curr_active_slave, *curr_arp_slave;
3073 unsigned char *arp_ptr;
3074 __be32 sip, tip;
3075 unsigned int alen;
3076
3077 alen = arp_hdr_len(dev: bond->dev);
3078
3079 if (alen > skb_headlen(skb)) {
3080 arp = kmalloc(size: alen, GFP_ATOMIC);
3081 if (!arp)
3082 goto out_unlock;
3083 if (skb_copy_bits(skb, offset: 0, to: arp, len: alen) < 0)
3084 goto out_unlock;
3085 }
3086
3087 if (arp->ar_hln != bond->dev->addr_len ||
3088 skb->pkt_type == PACKET_OTHERHOST ||
3089 skb->pkt_type == PACKET_LOOPBACK ||
3090 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3091 arp->ar_pro != htons(ETH_P_IP) ||
3092 arp->ar_pln != 4)
3093 goto out_unlock;
3094
3095 arp_ptr = (unsigned char *)(arp + 1);
3096 arp_ptr += bond->dev->addr_len;
3097 memcpy(&sip, arp_ptr, 4);
3098 arp_ptr += 4 + bond->dev->addr_len;
3099 memcpy(&tip, arp_ptr, 4);
3100
3101 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3102 __func__, slave->dev->name, bond_slave_state(slave),
3103 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3104 &sip, &tip);
3105
3106 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3107 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3108
3109 /* We 'trust' the received ARP enough to validate it if:
3110 *
3111 * (a) the slave receiving the ARP is active (which includes the
3112 * current ARP slave, if any), or
3113 *
3114 * (b) the receiving slave isn't active, but there is a currently
3115 * active slave and it received valid arp reply(s) after it became
3116 * the currently active slave, or
3117 *
3118 * (c) there is an ARP slave that sent an ARP during the prior ARP
3119 * interval, and we receive an ARP reply on any slave. We accept
3120 * these because switch FDB update delays may deliver the ARP
3121 * reply to a slave other than the sender of the ARP request.
3122 *
3123 * Note: for (b), backup slaves are receiving the broadcast ARP
3124 * request, not a reply. This request passes from the sending
3125 * slave through the L2 switch(es) to the receiving slave. Since
3126 * this is checking the request, sip/tip are swapped for
3127 * validation.
3128 *
3129 * This is done to avoid endless looping when we can't reach the
3130 * arp_ip_target and fool ourselves with our own arp requests.
3131 */
3132 if (bond_is_active_slave(slave))
3133 bond_validate_arp(bond, slave, sip, tip);
3134 else if (curr_active_slave &&
3135 time_after(slave_last_rx(bond, curr_active_slave),
3136 curr_active_slave->last_link_up))
3137 bond_validate_arp(bond, slave, sip: tip, tip: sip);
3138 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3139 bond_time_in_interval(bond, last_act: slave_last_tx(slave: curr_arp_slave), mod: 1))
3140 bond_validate_arp(bond, slave, sip, tip);
3141
3142out_unlock:
3143 if (arp != (struct arphdr *)skb->data)
3144 kfree(objp: arp);
3145 return RX_HANDLER_ANOTHER;
3146}
3147
3148#if IS_ENABLED(CONFIG_IPV6)
3149static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3150 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3151{
3152 struct net_device *bond_dev = slave->bond->dev;
3153 struct net_device *slave_dev = slave->dev;
3154 struct in6_addr mcaddr;
3155 struct sk_buff *skb;
3156
3157 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3158 daddr, saddr);
3159
3160 skb = ndisc_ns_create(dev: slave_dev, solicit: daddr, saddr, nonce: 0);
3161 if (!skb) {
3162 net_err_ratelimited("NS packet allocation failed\n");
3163 return;
3164 }
3165
3166 addrconf_addr_solict_mult(addr: daddr, solicited: &mcaddr);
3167 if (bond_handle_vlan(slave, tags, skb)) {
3168 slave_update_last_tx(slave);
3169 ndisc_send_skb(skb, daddr: &mcaddr, saddr);
3170 }
3171}
3172
3173static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3174{
3175 struct in6_addr *targets = bond->params.ns_targets;
3176 struct bond_vlan_tag *tags;
3177 struct dst_entry *dst;
3178 struct in6_addr saddr;
3179 struct flowi6 fl6;
3180 int i;
3181
3182 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(a: &targets[i]); i++) {
3183 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3184 __func__, &targets[i]);
3185 tags = NULL;
3186
3187 /* Find out through which dev should the packet go */
3188 memset(&fl6, 0, sizeof(struct flowi6));
3189 fl6.daddr = targets[i];
3190 fl6.flowi6_oif = bond->dev->ifindex;
3191
3192 dst = ip6_route_output(net: dev_net(dev: bond->dev), NULL, fl6: &fl6);
3193 if (dst->error) {
3194 dst_release(dst);
3195 /* there's no route to target - try to send arp
3196 * probe to generate any traffic (arp_validate=0)
3197 */
3198 if (bond->params.arp_validate)
3199 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3200 bond->dev->name,
3201 &targets[i]);
3202 bond_ns_send(slave, daddr: &targets[i], saddr: &in6addr_any, tags);
3203 continue;
3204 }
3205
3206 /* bond device itself */
3207 if (dst->dev == bond->dev)
3208 goto found;
3209
3210 rcu_read_lock();
3211 tags = bond_verify_device_path(start_dev: bond->dev, end_dev: dst->dev, level: 0);
3212 rcu_read_unlock();
3213
3214 if (!IS_ERR_OR_NULL(ptr: tags))
3215 goto found;
3216
3217 /* Not our device - skip */
3218 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3219 &targets[i], dst->dev ? dst->dev->name : "NULL");
3220
3221 dst_release(dst);
3222 continue;
3223
3224found:
3225 if (!ipv6_dev_get_saddr(net: dev_net(dev: dst->dev), dev: dst->dev, daddr: &targets[i], srcprefs: 0, saddr: &saddr))
3226 bond_ns_send(slave, daddr: &targets[i], saddr: &saddr, tags);
3227 else
3228 bond_ns_send(slave, daddr: &targets[i], saddr: &in6addr_any, tags);
3229
3230 dst_release(dst);
3231 kfree(objp: tags);
3232 }
3233}
3234
3235static int bond_confirm_addr6(struct net_device *dev,
3236 struct netdev_nested_priv *priv)
3237{
3238 struct in6_addr *addr = (struct in6_addr *)priv->data;
3239
3240 return ipv6_chk_addr(net: dev_net(dev), addr, dev, strict: 0);
3241}
3242
3243static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3244{
3245 struct netdev_nested_priv priv = {
3246 .data = addr,
3247 };
3248 int ret = false;
3249
3250 if (bond_confirm_addr6(dev: bond->dev, priv: &priv))
3251 return true;
3252
3253 rcu_read_lock();
3254 if (netdev_walk_all_upper_dev_rcu(dev: bond->dev, fn: bond_confirm_addr6, priv: &priv))
3255 ret = true;
3256 rcu_read_unlock();
3257
3258 return ret;
3259}
3260
3261static void bond_validate_na(struct bonding *bond, struct slave *slave,
3262 struct in6_addr *saddr, struct in6_addr *daddr)
3263{
3264 int i;
3265
3266 /* Ignore NAs that:
3267 * 1. Source address is unspecified address.
3268 * 2. Dest address is neither all-nodes multicast address nor
3269 * exist on bond interface.
3270 */
3271 if (ipv6_addr_any(a: saddr) ||
3272 (!ipv6_addr_equal(a1: daddr, a2: &in6addr_linklocal_allnodes) &&
3273 !bond_has_this_ip6(bond, addr: daddr))) {
3274 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3275 __func__, saddr, daddr);
3276 return;
3277 }
3278
3279 i = bond_get_targets_ip6(targets: bond->params.ns_targets, ip: saddr);
3280 if (i == -1) {
3281 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3282 __func__, saddr);
3283 return;
3284 }
3285 slave->last_rx = jiffies;
3286 slave->target_last_arp_rx[i] = jiffies;
3287}
3288
3289static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3290 struct slave *slave)
3291{
3292 struct slave *curr_active_slave, *curr_arp_slave;
3293 struct in6_addr *saddr, *daddr;
3294 struct {
3295 struct ipv6hdr ip6;
3296 struct icmp6hdr icmp6;
3297 } *combined, _combined;
3298
3299 if (skb->pkt_type == PACKET_OTHERHOST ||
3300 skb->pkt_type == PACKET_LOOPBACK)
3301 goto out;
3302
3303 combined = skb_header_pointer(skb, offset: 0, len: sizeof(_combined), buffer: &_combined);
3304 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3305 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3306 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3307 goto out;
3308
3309 saddr = &combined->ip6.saddr;
3310 daddr = &combined->ip6.daddr;
3311
3312 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3313 __func__, slave->dev->name, bond_slave_state(slave),
3314 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3315 saddr, daddr);
3316
3317 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3318 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3319
3320 /* We 'trust' the received ARP enough to validate it if:
3321 * see bond_arp_rcv().
3322 */
3323 if (bond_is_active_slave(slave))
3324 bond_validate_na(bond, slave, saddr, daddr);
3325 else if (curr_active_slave &&
3326 time_after(slave_last_rx(bond, curr_active_slave),
3327 curr_active_slave->last_link_up))
3328 bond_validate_na(bond, slave, saddr: daddr, daddr: saddr);
3329 else if (curr_arp_slave &&
3330 bond_time_in_interval(bond, last_act: slave_last_tx(slave: curr_arp_slave), mod: 1))
3331 bond_validate_na(bond, slave, saddr, daddr);
3332
3333out:
3334 return RX_HANDLER_ANOTHER;
3335}
3336#endif
3337
3338int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3339 struct slave *slave)
3340{
3341#if IS_ENABLED(CONFIG_IPV6)
3342 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3343#endif
3344 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3345
3346 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3347 __func__, skb->dev->name);
3348
3349 /* Use arp validate logic for both ARP and NS */
3350 if (!slave_do_arp_validate(bond, slave)) {
3351 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3352#if IS_ENABLED(CONFIG_IPV6)
3353 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3354#endif
3355 !slave_do_arp_validate_only(bond))
3356 slave->last_rx = jiffies;
3357 return RX_HANDLER_ANOTHER;
3358 } else if (is_arp) {
3359 return bond_arp_rcv(skb, bond, slave);
3360#if IS_ENABLED(CONFIG_IPV6)
3361 } else if (is_ipv6) {
3362 return bond_na_rcv(skb, bond, slave);
3363#endif
3364 } else {
3365 return RX_HANDLER_ANOTHER;
3366 }
3367}
3368
3369static void bond_send_validate(struct bonding *bond, struct slave *slave)
3370{
3371 bond_arp_send_all(bond, slave);
3372#if IS_ENABLED(CONFIG_IPV6)
3373 bond_ns_send_all(bond, slave);
3374#endif
3375}
3376
3377/* function to verify if we're in the arp_interval timeslice, returns true if
3378 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3379 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3380 */
3381static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3382 int mod)
3383{
3384 int delta_in_ticks = msecs_to_jiffies(m: bond->params.arp_interval);
3385
3386 return time_in_range(jiffies,
3387 last_act - delta_in_ticks,
3388 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3389}
3390
3391/* This function is called regularly to monitor each slave's link
3392 * ensuring that traffic is being sent and received when arp monitoring
3393 * is used in load-balancing mode. if the adapter has been dormant, then an
3394 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3395 * arp monitoring in active backup mode.
3396 */
3397static void bond_loadbalance_arp_mon(struct bonding *bond)
3398{
3399 struct slave *slave, *oldcurrent;
3400 struct list_head *iter;
3401 int do_failover = 0, slave_state_changed = 0;
3402
3403 if (!bond_has_slaves(bond))
3404 goto re_arm;
3405
3406 rcu_read_lock();
3407
3408 oldcurrent = rcu_dereference(bond->curr_active_slave);
3409 /* see if any of the previous devices are up now (i.e. they have
3410 * xmt and rcv traffic). the curr_active_slave does not come into
3411 * the picture unless it is null. also, slave->last_link_up is not
3412 * needed here because we send an arp on each slave and give a slave
3413 * as long as it needs to get the tx/rx within the delta.
3414 * TODO: what about up/down delay in arp mode? it wasn't here before
3415 * so it can wait
3416 */
3417 bond_for_each_slave_rcu(bond, slave, iter) {
3418 unsigned long last_tx = slave_last_tx(slave);
3419
3420 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3421
3422 if (slave->link != BOND_LINK_UP) {
3423 if (bond_time_in_interval(bond, last_act: last_tx, mod: 1) &&
3424 bond_time_in_interval(bond, last_act: slave->last_rx, mod: 1)) {
3425
3426 bond_propose_link_state(slave, BOND_LINK_UP);
3427 slave_state_changed = 1;
3428
3429 /* primary_slave has no meaning in round-robin
3430 * mode. the window of a slave being up and
3431 * curr_active_slave being null after enslaving
3432 * is closed.
3433 */
3434 if (!oldcurrent) {
3435 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3436 do_failover = 1;
3437 } else {
3438 slave_info(bond->dev, slave->dev, "interface is now up\n");
3439 }
3440 }
3441 } else {
3442 /* slave->link == BOND_LINK_UP */
3443
3444 /* not all switches will respond to an arp request
3445 * when the source ip is 0, so don't take the link down
3446 * if we don't know our ip yet
3447 */
3448 if (!bond_time_in_interval(bond, last_act: last_tx, mod: bond->params.missed_max) ||
3449 !bond_time_in_interval(bond, last_act: slave->last_rx, mod: bond->params.missed_max)) {
3450
3451 bond_propose_link_state(slave, BOND_LINK_DOWN);
3452 slave_state_changed = 1;
3453
3454 if (slave->link_failure_count < UINT_MAX)
3455 slave->link_failure_count++;
3456
3457 slave_info(bond->dev, slave->dev, "interface is now down\n");
3458
3459 if (slave == oldcurrent)
3460 do_failover = 1;
3461 }
3462 }
3463
3464 /* note: if switch is in round-robin mode, all links
3465 * must tx arp to ensure all links rx an arp - otherwise
3466 * links may oscillate or not come up at all; if switch is
3467 * in something like xor mode, there is nothing we can
3468 * do - all replies will be rx'ed on same link causing slaves
3469 * to be unstable during low/no traffic periods
3470 */
3471 if (bond_slave_is_up(slave))
3472 bond_send_validate(bond, slave);
3473 }
3474
3475 rcu_read_unlock();
3476
3477 if (do_failover || slave_state_changed) {
3478 if (!rtnl_trylock())
3479 goto re_arm;
3480
3481 bond_for_each_slave(bond, slave, iter) {
3482 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3483 slave->link = slave->link_new_state;
3484 }
3485
3486 if (slave_state_changed) {
3487 bond_slave_state_change(bond);
3488 if (BOND_MODE(bond) == BOND_MODE_XOR)
3489 bond_update_slave_arr(bond, NULL);
3490 }
3491 if (do_failover) {
3492 block_netpoll_tx();
3493 bond_select_active_slave(bond);
3494 unblock_netpoll_tx();
3495 }
3496 rtnl_unlock();
3497 }
3498
3499re_arm:
3500 if (bond->params.arp_interval)
3501 queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work,
3502 delay: msecs_to_jiffies(m: bond->params.arp_interval));
3503}
3504
3505/* Called to inspect slaves for active-backup mode ARP monitor link state
3506 * changes. Sets proposed link state in slaves to specify what action
3507 * should take place for the slave. Returns 0 if no changes are found, >0
3508 * if changes to link states must be committed.
3509 *
3510 * Called with rcu_read_lock held.
3511 */
3512static int bond_ab_arp_inspect(struct bonding *bond)
3513{
3514 unsigned long last_tx, last_rx;
3515 struct list_head *iter;
3516 struct slave *slave;
3517 int commit = 0;
3518
3519 bond_for_each_slave_rcu(bond, slave, iter) {
3520 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3521 last_rx = slave_last_rx(bond, slave);
3522
3523 if (slave->link != BOND_LINK_UP) {
3524 if (bond_time_in_interval(bond, last_act: last_rx, mod: 1)) {
3525 bond_propose_link_state(slave, BOND_LINK_UP);
3526 commit++;
3527 } else if (slave->link == BOND_LINK_BACK) {
3528 bond_propose_link_state(slave, BOND_LINK_FAIL);
3529 commit++;
3530 }
3531 continue;
3532 }
3533
3534 /* Give slaves 2*delta after being enslaved or made
3535 * active. This avoids bouncing, as the last receive
3536 * times need a full ARP monitor cycle to be updated.
3537 */
3538 if (bond_time_in_interval(bond, last_act: slave->last_link_up, mod: 2))
3539 continue;
3540
3541 /* Backup slave is down if:
3542 * - No current_arp_slave AND
3543 * - more than (missed_max+1)*delta since last receive AND
3544 * - the bond has an IP address
3545 *
3546 * Note: a non-null current_arp_slave indicates
3547 * the curr_active_slave went down and we are
3548 * searching for a new one; under this condition
3549 * we only take the curr_active_slave down - this
3550 * gives each slave a chance to tx/rx traffic
3551 * before being taken out
3552 */
3553 if (!bond_is_active_slave(slave) &&
3554 !rcu_access_pointer(bond->current_arp_slave) &&
3555 !bond_time_in_interval(bond, last_act: last_rx, mod: bond->params.missed_max + 1)) {
3556 bond_propose_link_state(slave, BOND_LINK_DOWN);
3557 commit++;
3558 }
3559
3560 /* Active slave is down if:
3561 * - more than missed_max*delta since transmitting OR
3562 * - (more than missed_max*delta since receive AND
3563 * the bond has an IP address)
3564 */
3565 last_tx = slave_last_tx(slave);
3566 if (bond_is_active_slave(slave) &&
3567 (!bond_time_in_interval(bond, last_act: last_tx, mod: bond->params.missed_max) ||
3568 !bond_time_in_interval(bond, last_act: last_rx, mod: bond->params.missed_max))) {
3569 bond_propose_link_state(slave, BOND_LINK_DOWN);
3570 commit++;
3571 }
3572 }
3573
3574 return commit;
3575}
3576
3577/* Called to commit link state changes noted by inspection step of
3578 * active-backup mode ARP monitor.
3579 *
3580 * Called with RTNL hold.
3581 */
3582static void bond_ab_arp_commit(struct bonding *bond)
3583{
3584 bool do_failover = false;
3585 struct list_head *iter;
3586 unsigned long last_tx;
3587 struct slave *slave;
3588
3589 bond_for_each_slave(bond, slave, iter) {
3590 switch (slave->link_new_state) {
3591 case BOND_LINK_NOCHANGE:
3592 continue;
3593
3594 case BOND_LINK_UP:
3595 last_tx = slave_last_tx(slave);
3596 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3597 (!rtnl_dereference(bond->curr_active_slave) &&
3598 bond_time_in_interval(bond, last_act: last_tx, mod: 1))) {
3599 struct slave *current_arp_slave;
3600
3601 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3602 bond_set_slave_link_state(slave, BOND_LINK_UP,
3603 BOND_SLAVE_NOTIFY_NOW);
3604 if (current_arp_slave) {
3605 bond_set_slave_inactive_flags(
3606 slave: current_arp_slave,
3607 BOND_SLAVE_NOTIFY_NOW);
3608 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3609 }
3610
3611 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3612
3613 if (!rtnl_dereference(bond->curr_active_slave) ||
3614 slave == rtnl_dereference(bond->primary_slave) ||
3615 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3616 do_failover = true;
3617
3618 }
3619
3620 continue;
3621
3622 case BOND_LINK_DOWN:
3623 if (slave->link_failure_count < UINT_MAX)
3624 slave->link_failure_count++;
3625
3626 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3627 BOND_SLAVE_NOTIFY_NOW);
3628 bond_set_slave_inactive_flags(slave,
3629 BOND_SLAVE_NOTIFY_NOW);
3630
3631 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3632
3633 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3634 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3635 do_failover = true;
3636 }
3637
3638 continue;
3639
3640 case BOND_LINK_FAIL:
3641 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3642 BOND_SLAVE_NOTIFY_NOW);
3643 bond_set_slave_inactive_flags(slave,
3644 BOND_SLAVE_NOTIFY_NOW);
3645
3646 /* A slave has just been enslaved and has become
3647 * the current active slave.
3648 */
3649 if (rtnl_dereference(bond->curr_active_slave))
3650 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3651 continue;
3652
3653 default:
3654 slave_err(bond->dev, slave->dev,
3655 "impossible: link_new_state %d on slave\n",
3656 slave->link_new_state);
3657 continue;
3658 }
3659 }
3660
3661 if (do_failover) {
3662 block_netpoll_tx();
3663 bond_select_active_slave(bond);
3664 unblock_netpoll_tx();
3665 }
3666
3667 bond_set_carrier(bond);
3668}
3669
3670/* Send ARP probes for active-backup mode ARP monitor.
3671 *
3672 * Called with rcu_read_lock held.
3673 */
3674static bool bond_ab_arp_probe(struct bonding *bond)
3675{
3676 struct slave *slave, *before = NULL, *new_slave = NULL,
3677 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3678 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3679 struct list_head *iter;
3680 bool found = false;
3681 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3682
3683 if (curr_arp_slave && curr_active_slave)
3684 netdev_info(dev: bond->dev, format: "PROBE: c_arp %s && cas %s BAD\n",
3685 curr_arp_slave->dev->name,
3686 curr_active_slave->dev->name);
3687
3688 if (curr_active_slave) {
3689 bond_send_validate(bond, slave: curr_active_slave);
3690 return should_notify_rtnl;
3691 }
3692
3693 /* if we don't have a curr_active_slave, search for the next available
3694 * backup slave from the current_arp_slave and make it the candidate
3695 * for becoming the curr_active_slave
3696 */
3697
3698 if (!curr_arp_slave) {
3699 curr_arp_slave = bond_first_slave_rcu(bond);
3700 if (!curr_arp_slave)
3701 return should_notify_rtnl;
3702 }
3703
3704 bond_for_each_slave_rcu(bond, slave, iter) {
3705 if (!found && !before && bond_slave_is_up(slave))
3706 before = slave;
3707
3708 if (found && !new_slave && bond_slave_is_up(slave))
3709 new_slave = slave;
3710 /* if the link state is up at this point, we
3711 * mark it down - this can happen if we have
3712 * simultaneous link failures and
3713 * reselect_active_interface doesn't make this
3714 * one the current slave so it is still marked
3715 * up when it is actually down
3716 */
3717 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3718 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3719 BOND_SLAVE_NOTIFY_LATER);
3720 if (slave->link_failure_count < UINT_MAX)
3721 slave->link_failure_count++;
3722
3723 bond_set_slave_inactive_flags(slave,
3724 BOND_SLAVE_NOTIFY_LATER);
3725
3726 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3727 }
3728 if (slave == curr_arp_slave)
3729 found = true;
3730 }
3731
3732 if (!new_slave && before)
3733 new_slave = before;
3734
3735 if (!new_slave)
3736 goto check_state;
3737
3738 bond_set_slave_link_state(slave: new_slave, BOND_LINK_BACK,
3739 BOND_SLAVE_NOTIFY_LATER);
3740 bond_set_slave_active_flags(slave: new_slave, BOND_SLAVE_NOTIFY_LATER);
3741 bond_send_validate(bond, slave: new_slave);
3742 new_slave->last_link_up = jiffies;
3743 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3744
3745check_state:
3746 bond_for_each_slave_rcu(bond, slave, iter) {
3747 if (slave->should_notify || slave->should_notify_link) {
3748 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3749 break;
3750 }
3751 }
3752 return should_notify_rtnl;
3753}
3754
3755static void bond_activebackup_arp_mon(struct bonding *bond)
3756{
3757 bool should_notify_peers = false;
3758 bool should_notify_rtnl = false;
3759 int delta_in_ticks;
3760
3761 delta_in_ticks = msecs_to_jiffies(m: bond->params.arp_interval);
3762
3763 if (!bond_has_slaves(bond))
3764 goto re_arm;
3765
3766 rcu_read_lock();
3767
3768 should_notify_peers = bond_should_notify_peers(bond);
3769
3770 if (bond_ab_arp_inspect(bond)) {
3771 rcu_read_unlock();
3772
3773 /* Race avoidance with bond_close flush of workqueue */
3774 if (!rtnl_trylock()) {
3775 delta_in_ticks = 1;
3776 should_notify_peers = false;
3777 goto re_arm;
3778 }
3779
3780 bond_ab_arp_commit(bond);
3781
3782 rtnl_unlock();
3783 rcu_read_lock();
3784 }
3785
3786 should_notify_rtnl = bond_ab_arp_probe(bond);
3787 rcu_read_unlock();
3788
3789re_arm:
3790 if (bond->params.arp_interval)
3791 queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work, delay: delta_in_ticks);
3792
3793 if (should_notify_peers || should_notify_rtnl) {
3794 if (!rtnl_trylock())
3795 return;
3796
3797 if (should_notify_peers) {
3798 bond->send_peer_notif--;
3799 call_netdevice_notifiers(val: NETDEV_NOTIFY_PEERS,
3800 dev: bond->dev);
3801 }
3802 if (should_notify_rtnl) {
3803 bond_slave_state_notify(bond);
3804 bond_slave_link_notify(bond);
3805 }
3806
3807 rtnl_unlock();
3808 }
3809}
3810
3811static void bond_arp_monitor(struct work_struct *work)
3812{
3813 struct bonding *bond = container_of(work, struct bonding,
3814 arp_work.work);
3815
3816 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3817 bond_activebackup_arp_mon(bond);
3818 else
3819 bond_loadbalance_arp_mon(bond);
3820}
3821
3822/*-------------------------- netdev event handling --------------------------*/
3823
3824/* Change device name */
3825static int bond_event_changename(struct bonding *bond)
3826{
3827 bond_remove_proc_entry(bond);
3828 bond_create_proc_entry(bond);
3829
3830 bond_debug_reregister(bond);
3831
3832 return NOTIFY_DONE;
3833}
3834
3835static int bond_master_netdev_event(unsigned long event,
3836 struct net_device *bond_dev)
3837{
3838 struct bonding *event_bond = netdev_priv(dev: bond_dev);
3839
3840 netdev_dbg(bond_dev, "%s called\n", __func__);
3841
3842 switch (event) {
3843 case NETDEV_CHANGENAME:
3844 return bond_event_changename(bond: event_bond);
3845 case NETDEV_UNREGISTER:
3846 bond_remove_proc_entry(bond: event_bond);
3847#ifdef CONFIG_XFRM_OFFLOAD
3848 xfrm_dev_state_flush(net: dev_net(dev: bond_dev), dev: bond_dev, task_valid: true);
3849#endif /* CONFIG_XFRM_OFFLOAD */
3850 break;
3851 case NETDEV_REGISTER:
3852 bond_create_proc_entry(bond: event_bond);
3853 break;
3854 default:
3855 break;
3856 }
3857
3858 return NOTIFY_DONE;
3859}
3860
3861static int bond_slave_netdev_event(unsigned long event,
3862 struct net_device *slave_dev)
3863{
3864 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3865 struct bonding *bond;
3866 struct net_device *bond_dev;
3867
3868 /* A netdev event can be generated while enslaving a device
3869 * before netdev_rx_handler_register is called in which case
3870 * slave will be NULL
3871 */
3872 if (!slave) {
3873 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3874 return NOTIFY_DONE;
3875 }
3876
3877 bond_dev = slave->bond->dev;
3878 bond = slave->bond;
3879 primary = rtnl_dereference(bond->primary_slave);
3880
3881 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3882
3883 switch (event) {
3884 case NETDEV_UNREGISTER:
3885 if (bond_dev->type != ARPHRD_ETHER)
3886 bond_release_and_destroy(bond_dev, slave_dev);
3887 else
3888 __bond_release_one(bond_dev, slave_dev, all: false, unregister: true);
3889 break;
3890 case NETDEV_UP:
3891 case NETDEV_CHANGE:
3892 /* For 802.3ad mode only:
3893 * Getting invalid Speed/Duplex values here will put slave
3894 * in weird state. Mark it as link-fail if the link was
3895 * previously up or link-down if it hasn't yet come up, and
3896 * let link-monitoring (miimon) set it right when correct
3897 * speeds/duplex are available.
3898 */
3899 if (bond_update_speed_duplex(slave) &&
3900 BOND_MODE(bond) == BOND_MODE_8023AD) {
3901 if (slave->last_link_up)
3902 slave->link = BOND_LINK_FAIL;
3903 else
3904 slave->link = BOND_LINK_DOWN;
3905 }
3906
3907 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3908 bond_3ad_adapter_speed_duplex_changed(slave);
3909 fallthrough;
3910 case NETDEV_DOWN:
3911 /* Refresh slave-array if applicable!
3912 * If the setup does not use miimon or arpmon (mode-specific!),
3913 * then these events will not cause the slave-array to be
3914 * refreshed. This will cause xmit to use a slave that is not
3915 * usable. Avoid such situation by refeshing the array at these
3916 * events. If these (miimon/arpmon) parameters are configured
3917 * then array gets refreshed twice and that should be fine!
3918 */
3919 if (bond_mode_can_use_xmit_hash(bond))
3920 bond_update_slave_arr(bond, NULL);
3921 break;
3922 case NETDEV_CHANGEMTU:
3923 /* TODO: Should slaves be allowed to
3924 * independently alter their MTU? For
3925 * an active-backup bond, slaves need
3926 * not be the same type of device, so
3927 * MTUs may vary. For other modes,
3928 * slaves arguably should have the
3929 * same MTUs. To do this, we'd need to
3930 * take over the slave's change_mtu
3931 * function for the duration of their
3932 * servitude.
3933 */
3934 break;
3935 case NETDEV_CHANGENAME:
3936 /* we don't care if we don't have primary set */
3937 if (!bond_uses_primary(bond) ||
3938 !bond->params.primary[0])
3939 break;
3940
3941 if (slave == primary) {
3942 /* slave's name changed - he's no longer primary */
3943 RCU_INIT_POINTER(bond->primary_slave, NULL);
3944 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3945 /* we have a new primary slave */
3946 rcu_assign_pointer(bond->primary_slave, slave);
3947 } else { /* we didn't change primary - exit */
3948 break;
3949 }
3950
3951 netdev_info(dev: bond->dev, format: "Primary slave changed to %s, reselecting active slave\n",
3952 primary ? slave_dev->name : "none");
3953
3954 block_netpoll_tx();
3955 bond_select_active_slave(bond);
3956 unblock_netpoll_tx();
3957 break;
3958 case NETDEV_FEAT_CHANGE:
3959 if (!bond->notifier_ctx) {
3960 bond->notifier_ctx = true;
3961 bond_compute_features(bond);
3962 bond->notifier_ctx = false;
3963 }
3964 break;
3965 case NETDEV_RESEND_IGMP:
3966 /* Propagate to master device */
3967 call_netdevice_notifiers(val: event, dev: slave->bond->dev);
3968 break;
3969 case NETDEV_XDP_FEAT_CHANGE:
3970 bond_xdp_set_features(bond_dev);
3971 break;
3972 default:
3973 break;
3974 }
3975
3976 return NOTIFY_DONE;
3977}
3978
3979/* bond_netdev_event: handle netdev notifier chain events.
3980 *
3981 * This function receives events for the netdev chain. The caller (an
3982 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3983 * locks for us to safely manipulate the slave devices (RTNL lock,
3984 * dev_probe_lock).
3985 */
3986static int bond_netdev_event(struct notifier_block *this,
3987 unsigned long event, void *ptr)
3988{
3989 struct net_device *event_dev = netdev_notifier_info_to_dev(info: ptr);
3990
3991 netdev_dbg(event_dev, "%s received %s\n",
3992 __func__, netdev_cmd_to_name(event));
3993
3994 if (!(event_dev->priv_flags & IFF_BONDING))
3995 return NOTIFY_DONE;
3996
3997 if (event_dev->flags & IFF_MASTER) {
3998 int ret;
3999
4000 ret = bond_master_netdev_event(event, bond_dev: event_dev);
4001 if (ret != NOTIFY_DONE)
4002 return ret;
4003 }
4004
4005 if (event_dev->flags & IFF_SLAVE)
4006 return bond_slave_netdev_event(event, slave_dev: event_dev);
4007
4008 return NOTIFY_DONE;
4009}
4010
4011static struct notifier_block bond_netdev_notifier = {
4012 .notifier_call = bond_netdev_event,
4013};
4014
4015/*---------------------------- Hashing Policies -----------------------------*/
4016
4017/* Helper to access data in a packet, with or without a backing skb.
4018 * If skb is given the data is linearized if necessary via pskb_may_pull.
4019 */
4020static inline const void *bond_pull_data(struct sk_buff *skb,
4021 const void *data, int hlen, int n)
4022{
4023 if (likely(n <= hlen))
4024 return data;
4025 else if (skb && likely(pskb_may_pull(skb, n)))
4026 return skb->data;
4027
4028 return NULL;
4029}
4030
4031/* L2 hash helper */
4032static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4033{
4034 struct ethhdr *ep;
4035
4036 data = bond_pull_data(skb, data, hlen, n: mhoff + sizeof(struct ethhdr));
4037 if (!data)
4038 return 0;
4039
4040 ep = (struct ethhdr *)(data + mhoff);
4041 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4042}
4043
4044static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4045 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4046{
4047 const struct ipv6hdr *iph6;
4048 const struct iphdr *iph;
4049
4050 if (l2_proto == htons(ETH_P_IP)) {
4051 data = bond_pull_data(skb, data, hlen, n: *nhoff + sizeof(*iph));
4052 if (!data)
4053 return false;
4054
4055 iph = (const struct iphdr *)(data + *nhoff);
4056 iph_to_flow_copy_v4addrs(flow: fk, iph);
4057 *nhoff += iph->ihl << 2;
4058 if (!ip_is_fragment(iph))
4059 *ip_proto = iph->protocol;
4060 } else if (l2_proto == htons(ETH_P_IPV6)) {
4061 data = bond_pull_data(skb, data, hlen, n: *nhoff + sizeof(*iph6));
4062 if (!data)
4063 return false;
4064
4065 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4066 iph_to_flow_copy_v6addrs(flow: fk, iph: iph6);
4067 *nhoff += sizeof(*iph6);
4068 *ip_proto = iph6->nexthdr;
4069 } else {
4070 return false;
4071 }
4072
4073 if (l34 && *ip_proto >= 0)
4074 fk->ports.ports = __skb_flow_get_ports(skb, thoff: *nhoff, ip_proto: *ip_proto, data, hlen_proto: hlen);
4075
4076 return true;
4077}
4078
4079static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4080{
4081 u32 srcmac_vendor = 0, srcmac_dev = 0;
4082 struct ethhdr *mac_hdr;
4083 u16 vlan = 0;
4084 int i;
4085
4086 data = bond_pull_data(skb, data, hlen, n: mhoff + sizeof(struct ethhdr));
4087 if (!data)
4088 return 0;
4089 mac_hdr = (struct ethhdr *)(data + mhoff);
4090
4091 for (i = 0; i < 3; i++)
4092 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4093
4094 for (i = 3; i < ETH_ALEN; i++)
4095 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4096
4097 if (skb && skb_vlan_tag_present(skb))
4098 vlan = skb_vlan_tag_get(skb);
4099
4100 return vlan ^ srcmac_vendor ^ srcmac_dev;
4101}
4102
4103/* Extract the appropriate headers based on bond's xmit policy */
4104static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4105 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4106{
4107 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4108 int ip_proto = -1;
4109
4110 switch (bond->params.xmit_policy) {
4111 case BOND_XMIT_POLICY_ENCAP23:
4112 case BOND_XMIT_POLICY_ENCAP34:
4113 memset(fk, 0, sizeof(*fk));
4114 return __skb_flow_dissect(NULL, skb, flow_dissector: &flow_keys_bonding,
4115 target_container: fk, data, proto: l2_proto, nhoff, hlen, flags: 0);
4116 default:
4117 break;
4118 }
4119
4120 fk->ports.ports = 0;
4121 memset(&fk->icmp, 0, sizeof(fk->icmp));
4122 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, nhoff: &nhoff, ip_proto: &ip_proto, l34))
4123 return false;
4124
4125 /* ICMP error packets contains at least 8 bytes of the header
4126 * of the packet which generated the error. Use this information
4127 * to correlate ICMP error packets within the same flow which
4128 * generated the error.
4129 */
4130 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4131 skb_flow_get_icmp_tci(skb, key_icmp: &fk->icmp, data, thoff: nhoff, hlen);
4132 if (ip_proto == IPPROTO_ICMP) {
4133 if (!icmp_is_err(type: fk->icmp.type))
4134 return true;
4135
4136 nhoff += sizeof(struct icmphdr);
4137 } else if (ip_proto == IPPROTO_ICMPV6) {
4138 if (!icmpv6_is_err(type: fk->icmp.type))
4139 return true;
4140
4141 nhoff += sizeof(struct icmp6hdr);
4142 }
4143 return bond_flow_ip(skb, fk, data, hlen, l2_proto, nhoff: &nhoff, ip_proto: &ip_proto, l34);
4144 }
4145
4146 return true;
4147}
4148
4149static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4150{
4151 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4152 (__force u32)flow_get_u32_src(flow);
4153 hash ^= (hash >> 16);
4154 hash ^= (hash >> 8);
4155
4156 /* discard lowest hash bit to deal with the common even ports pattern */
4157 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4158 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4159 return hash >> 1;
4160
4161 return hash;
4162}
4163
4164/* Generate hash based on xmit policy. If @skb is given it is used to linearize
4165 * the data as required, but this function can be used without it if the data is
4166 * known to be linear (e.g. with xdp_buff).
4167 */
4168static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4169 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4170{
4171 struct flow_keys flow;
4172 u32 hash;
4173
4174 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4175 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4176
4177 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4178 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, fk: &flow))
4179 return bond_eth_hash(skb, data, mhoff, hlen);
4180
4181 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4182 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4183 hash = bond_eth_hash(skb, data, mhoff, hlen);
4184 } else {
4185 if (flow.icmp.id)
4186 memcpy(&hash, &flow.icmp, sizeof(hash));
4187 else
4188 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4189 }
4190
4191 return bond_ip_hash(hash, flow: &flow, xmit_policy: bond->params.xmit_policy);
4192}
4193
4194/**
4195 * bond_xmit_hash - generate a hash value based on the xmit policy
4196 * @bond: bonding device
4197 * @skb: buffer to use for headers
4198 *
4199 * This function will extract the necessary headers from the skb buffer and use
4200 * them to generate a hash based on the xmit_policy set in the bonding device
4201 */
4202u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4203{
4204 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4205 skb->l4_hash)
4206 return skb->hash;
4207
4208 return __bond_xmit_hash(bond, skb, data: skb->data, l2_proto: skb->protocol,
4209 mhoff: 0, nhoff: skb_network_offset(skb),
4210 hlen: skb_headlen(skb));
4211}
4212
4213/**
4214 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4215 * @bond: bonding device
4216 * @xdp: buffer to use for headers
4217 *
4218 * The XDP variant of bond_xmit_hash.
4219 */
4220static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4221{
4222 struct ethhdr *eth;
4223
4224 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4225 return 0;
4226
4227 eth = (struct ethhdr *)xdp->data;
4228
4229 return __bond_xmit_hash(bond, NULL, data: xdp->data, l2_proto: eth->h_proto, mhoff: 0,
4230 nhoff: sizeof(struct ethhdr), hlen: xdp->data_end - xdp->data);
4231}
4232
4233/*-------------------------- Device entry points ----------------------------*/
4234
4235void bond_work_init_all(struct bonding *bond)
4236{
4237 INIT_DELAYED_WORK(&bond->mcast_work,
4238 bond_resend_igmp_join_requests_delayed);
4239 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4240 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4241 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4242 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4243 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4244}
4245
4246static void bond_work_cancel_all(struct bonding *bond)
4247{
4248 cancel_delayed_work_sync(dwork: &bond->mii_work);
4249 cancel_delayed_work_sync(dwork: &bond->arp_work);
4250 cancel_delayed_work_sync(dwork: &bond->alb_work);
4251 cancel_delayed_work_sync(dwork: &bond->ad_work);
4252 cancel_delayed_work_sync(dwork: &bond->mcast_work);
4253 cancel_delayed_work_sync(dwork: &bond->slave_arr_work);
4254}
4255
4256static int bond_open(struct net_device *bond_dev)
4257{
4258 struct bonding *bond = netdev_priv(dev: bond_dev);
4259 struct list_head *iter;
4260 struct slave *slave;
4261
4262 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4263 bond->rr_tx_counter = alloc_percpu(u32);
4264 if (!bond->rr_tx_counter)
4265 return -ENOMEM;
4266 }
4267
4268 /* reset slave->backup and slave->inactive */
4269 if (bond_has_slaves(bond)) {
4270 bond_for_each_slave(bond, slave, iter) {
4271 if (bond_uses_primary(bond) &&
4272 slave != rcu_access_pointer(bond->curr_active_slave)) {
4273 bond_set_slave_inactive_flags(slave,
4274 BOND_SLAVE_NOTIFY_NOW);
4275 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4276 bond_set_slave_active_flags(slave,
4277 BOND_SLAVE_NOTIFY_NOW);
4278 }
4279 }
4280 }
4281
4282 if (bond_is_lb(bond)) {
4283 /* bond_alb_initialize must be called before the timer
4284 * is started.
4285 */
4286 if (bond_alb_initialize(bond, rlb_enabled: (BOND_MODE(bond) == BOND_MODE_ALB)))
4287 return -ENOMEM;
4288 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4289 queue_delayed_work(wq: bond->wq, dwork: &bond->alb_work, delay: 0);
4290 }
4291
4292 if (bond->params.miimon) /* link check interval, in milliseconds. */
4293 queue_delayed_work(wq: bond->wq, dwork: &bond->mii_work, delay: 0);
4294
4295 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4296 queue_delayed_work(wq: bond->wq, dwork: &bond->arp_work, delay: 0);
4297 bond->recv_probe = bond_rcv_validate;
4298 }
4299
4300 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4301 queue_delayed_work(wq: bond->wq, dwork: &bond->ad_work, delay: 0);
4302 /* register to receive LACPDUs */
4303 bond->recv_probe = bond_3ad_lacpdu_recv;
4304 bond_3ad_initiate_agg_selection(bond, timeout: 1);
4305
4306 bond_for_each_slave(bond, slave, iter)
4307 dev_mc_add(dev: slave->dev, addr: lacpdu_mcast_addr);
4308 }
4309
4310 if (bond_mode_can_use_xmit_hash(bond))
4311 bond_update_slave_arr(bond, NULL);
4312
4313 return 0;
4314}
4315
4316static int bond_close(struct net_device *bond_dev)
4317{
4318 struct bonding *bond = netdev_priv(dev: bond_dev);
4319 struct slave *slave;
4320
4321 bond_work_cancel_all(bond);
4322 bond->send_peer_notif = 0;
4323 if (bond_is_lb(bond))
4324 bond_alb_deinitialize(bond);
4325 bond->recv_probe = NULL;
4326
4327 if (bond_uses_primary(bond)) {
4328 rcu_read_lock();
4329 slave = rcu_dereference(bond->curr_active_slave);
4330 if (slave)
4331 bond_hw_addr_flush(bond_dev, slave_dev: slave->dev);
4332 rcu_read_unlock();
4333 } else {
4334 struct list_head *iter;
4335
4336 bond_for_each_slave(bond, slave, iter)
4337 bond_hw_addr_flush(bond_dev, slave_dev: slave->dev);
4338 }
4339
4340 return 0;
4341}
4342
4343/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4344 * that some drivers can provide 32bit values only.
4345 */
4346static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4347 const struct rtnl_link_stats64 *_new,
4348 const struct rtnl_link_stats64 *_old)
4349{
4350 const u64 *new = (const u64 *)_new;
4351 const u64 *old = (const u64 *)_old;
4352 u64 *res = (u64 *)_res;
4353 int i;
4354
4355 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4356 u64 nv = new[i];
4357 u64 ov = old[i];
4358 s64 delta = nv - ov;
4359
4360 /* detects if this particular field is 32bit only */
4361 if (((nv | ov) >> 32) == 0)
4362 delta = (s64)(s32)((u32)nv - (u32)ov);
4363
4364 /* filter anomalies, some drivers reset their stats
4365 * at down/up events.
4366 */
4367 if (delta > 0)
4368 res[i] += delta;
4369 }
4370}
4371
4372#ifdef CONFIG_LOCKDEP
4373static int bond_get_lowest_level_rcu(struct net_device *dev)
4374{
4375 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4376 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4377 int cur = 0, max = 0;
4378
4379 now = dev;
4380 iter = &dev->adj_list.lower;
4381
4382 while (1) {
4383 next = NULL;
4384 while (1) {
4385 ldev = netdev_next_lower_dev_rcu(dev: now, iter: &iter);
4386 if (!ldev)
4387 break;
4388
4389 next = ldev;
4390 niter = &ldev->adj_list.lower;
4391 dev_stack[cur] = now;
4392 iter_stack[cur++] = iter;
4393 if (max <= cur)
4394 max = cur;
4395 break;
4396 }
4397
4398 if (!next) {
4399 if (!cur)
4400 return max;
4401 next = dev_stack[--cur];
4402 niter = iter_stack[cur];
4403 }
4404
4405 now = next;
4406 iter = niter;
4407 }
4408
4409 return max;
4410}
4411#endif
4412
4413static void bond_get_stats(struct net_device *bond_dev,
4414 struct rtnl_link_stats64 *stats)
4415{
4416 struct bonding *bond = netdev_priv(dev: bond_dev);
4417 struct rtnl_link_stats64 temp;
4418 struct list_head *iter;
4419 struct slave *slave;
4420 int nest_level = 0;
4421
4422
4423 rcu_read_lock();
4424#ifdef CONFIG_LOCKDEP
4425 nest_level = bond_get_lowest_level_rcu(dev: bond_dev);
4426#endif
4427
4428 spin_lock_nested(&bond->stats_lock, nest_level);
4429 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4430
4431 bond_for_each_slave_rcu(bond, slave, iter) {
4432 const struct rtnl_link_stats64 *new =
4433 dev_get_stats(dev: slave->dev, storage: &temp);
4434
4435 bond_fold_stats(res: stats, new: new, old: &slave->slave_stats);
4436
4437 /* save off the slave stats for the next run */
4438 memcpy(&slave->slave_stats, new, sizeof(*new));
4439 }
4440
4441 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4442 spin_unlock(lock: &bond->stats_lock);
4443 rcu_read_unlock();
4444}
4445
4446static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4447{
4448 struct bonding *bond = netdev_priv(dev: bond_dev);
4449 struct mii_ioctl_data *mii = NULL;
4450
4451 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4452
4453 switch (cmd) {
4454 case SIOCGMIIPHY:
4455 mii = if_mii(rq: ifr);
4456 if (!mii)
4457 return -EINVAL;
4458
4459 mii->phy_id = 0;
4460 fallthrough;
4461 case SIOCGMIIREG:
4462 /* We do this again just in case we were called by SIOCGMIIREG
4463 * instead of SIOCGMIIPHY.
4464 */
4465 mii = if_mii(rq: ifr);
4466 if (!mii)
4467 return -EINVAL;
4468
4469 if (mii->reg_num == 1) {
4470 mii->val_out = 0;
4471 if (netif_carrier_ok(dev: bond->dev))
4472 mii->val_out = BMSR_LSTATUS;
4473 }
4474
4475 break;
4476 default:
4477 return -EOPNOTSUPP;
4478 }
4479
4480 return 0;
4481}
4482
4483static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4484{
4485 struct bonding *bond = netdev_priv(dev: bond_dev);
4486 struct net_device *slave_dev = NULL;
4487 struct ifbond k_binfo;
4488 struct ifbond __user *u_binfo = NULL;
4489 struct ifslave k_sinfo;
4490 struct ifslave __user *u_sinfo = NULL;
4491 struct bond_opt_value newval;
4492 struct net *net;
4493 int res = 0;
4494
4495 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4496
4497 switch (cmd) {
4498 case SIOCBONDINFOQUERY:
4499 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4500
4501 if (copy_from_user(to: &k_binfo, from: u_binfo, n: sizeof(ifbond)))
4502 return -EFAULT;
4503
4504 bond_info_query(bond_dev, info: &k_binfo);
4505 if (copy_to_user(to: u_binfo, from: &k_binfo, n: sizeof(ifbond)))
4506 return -EFAULT;
4507
4508 return 0;
4509 case SIOCBONDSLAVEINFOQUERY:
4510 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4511
4512 if (copy_from_user(to: &k_sinfo, from: u_sinfo, n: sizeof(ifslave)))
4513 return -EFAULT;
4514
4515 res = bond_slave_info_query(bond_dev, info: &k_sinfo);
4516 if (res == 0 &&
4517 copy_to_user(to: u_sinfo, from: &k_sinfo, n: sizeof(ifslave)))
4518 return -EFAULT;
4519
4520 return res;
4521 default:
4522 break;
4523 }
4524
4525 net = dev_net(dev: bond_dev);
4526
4527 if (!ns_capable(ns: net->user_ns, CAP_NET_ADMIN))
4528 return -EPERM;
4529
4530 slave_dev = __dev_get_by_name(net, name: ifr->ifr_slave);
4531
4532 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4533
4534 if (!slave_dev)
4535 return -ENODEV;
4536
4537 switch (cmd) {
4538 case SIOCBONDENSLAVE:
4539 res = bond_enslave(bond_dev, slave_dev, NULL);
4540 break;
4541 case SIOCBONDRELEASE:
4542 res = bond_release(bond_dev, slave_dev);
4543 break;
4544 case SIOCBONDSETHWADDR:
4545 res = bond_set_dev_addr(bond_dev, slave_dev);
4546 break;
4547 case SIOCBONDCHANGEACTIVE:
4548 bond_opt_initstr(&newval, slave_dev->name);
4549 res = __bond_opt_set_notify(bond, option: BOND_OPT_ACTIVE_SLAVE,
4550 val: &newval);
4551 break;
4552 default:
4553 res = -EOPNOTSUPP;
4554 }
4555
4556 return res;
4557}
4558
4559static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4560 void __user *data, int cmd)
4561{
4562 struct ifreq ifrdata = { .ifr_data = data };
4563
4564 switch (cmd) {
4565 case BOND_INFO_QUERY_OLD:
4566 return bond_do_ioctl(bond_dev, ifr: &ifrdata, SIOCBONDINFOQUERY);
4567 case BOND_SLAVE_INFO_QUERY_OLD:
4568 return bond_do_ioctl(bond_dev, ifr: &ifrdata, SIOCBONDSLAVEINFOQUERY);
4569 case BOND_ENSLAVE_OLD:
4570 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4571 case BOND_RELEASE_OLD:
4572 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4573 case BOND_SETHWADDR_OLD:
4574 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4575 case BOND_CHANGE_ACTIVE_OLD:
4576 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4577 }
4578
4579 return -EOPNOTSUPP;
4580}
4581
4582static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4583{
4584 struct bonding *bond = netdev_priv(dev: bond_dev);
4585
4586 if (change & IFF_PROMISC)
4587 bond_set_promiscuity(bond,
4588 inc: bond_dev->flags & IFF_PROMISC ? 1 : -1);
4589
4590 if (change & IFF_ALLMULTI)
4591 bond_set_allmulti(bond,
4592 inc: bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4593}
4594
4595static void bond_set_rx_mode(struct net_device *bond_dev)
4596{
4597 struct bonding *bond = netdev_priv(dev: bond_dev);
4598 struct list_head *iter;
4599 struct slave *slave;
4600
4601 rcu_read_lock();
4602 if (bond_uses_primary(bond)) {
4603 slave = rcu_dereference(bond->curr_active_slave);
4604 if (slave) {
4605 dev_uc_sync(to: slave->dev, from: bond_dev);
4606 dev_mc_sync(to: slave->dev, from: bond_dev);
4607 }
4608 } else {
4609 bond_for_each_slave_rcu(bond, slave, iter) {
4610 dev_uc_sync_multiple(to: slave->dev, from: bond_dev);
4611 dev_mc_sync_multiple(to: slave->dev, from: bond_dev);
4612 }
4613 }
4614 rcu_read_unlock();
4615}
4616
4617static int bond_neigh_init(struct neighbour *n)
4618{
4619 struct bonding *bond = netdev_priv(dev: n->dev);
4620 const struct net_device_ops *slave_ops;
4621 struct neigh_parms parms;
4622 struct slave *slave;
4623 int ret = 0;
4624
4625 rcu_read_lock();
4626 slave = bond_first_slave_rcu(bond);
4627 if (!slave)
4628 goto out;
4629 slave_ops = slave->dev->netdev_ops;
4630 if (!slave_ops->ndo_neigh_setup)
4631 goto out;
4632
4633 /* TODO: find another way [1] to implement this.
4634 * Passing a zeroed structure is fragile,
4635 * but at least we do not pass garbage.
4636 *
4637 * [1] One way would be that ndo_neigh_setup() never touch
4638 * struct neigh_parms, but propagate the new neigh_setup()
4639 * back to ___neigh_create() / neigh_parms_alloc()
4640 */
4641 memset(&parms, 0, sizeof(parms));
4642 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4643
4644 if (ret)
4645 goto out;
4646
4647 if (parms.neigh_setup)
4648 ret = parms.neigh_setup(n);
4649out:
4650 rcu_read_unlock();
4651 return ret;
4652}
4653
4654/* The bonding ndo_neigh_setup is called at init time beofre any
4655 * slave exists. So we must declare proxy setup function which will
4656 * be used at run time to resolve the actual slave neigh param setup.
4657 *
4658 * It's also called by master devices (such as vlans) to setup their
4659 * underlying devices. In that case - do nothing, we're already set up from
4660 * our init.
4661 */
4662static int bond_neigh_setup(struct net_device *dev,
4663 struct neigh_parms *parms)
4664{
4665 /* modify only our neigh_parms */
4666 if (parms->dev == dev)
4667 parms->neigh_setup = bond_neigh_init;
4668
4669 return 0;
4670}
4671
4672/* Change the MTU of all of a master's slaves to match the master */
4673static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4674{
4675 struct bonding *bond = netdev_priv(dev: bond_dev);
4676 struct slave *slave, *rollback_slave;
4677 struct list_head *iter;
4678 int res = 0;
4679
4680 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4681
4682 bond_for_each_slave(bond, slave, iter) {
4683 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4684 slave, slave->dev->netdev_ops->ndo_change_mtu);
4685
4686 res = dev_set_mtu(slave->dev, new_mtu);
4687
4688 if (res) {
4689 /* If we failed to set the slave's mtu to the new value
4690 * we must abort the operation even in ACTIVE_BACKUP
4691 * mode, because if we allow the backup slaves to have
4692 * different mtu values than the active slave we'll
4693 * need to change their mtu when doing a failover. That
4694 * means changing their mtu from timer context, which
4695 * is probably not a good idea.
4696 */
4697 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4698 res, new_mtu);
4699 goto unwind;
4700 }
4701 }
4702
4703 bond_dev->mtu = new_mtu;
4704
4705 return 0;
4706
4707unwind:
4708 /* unwind from head to the slave that failed */
4709 bond_for_each_slave(bond, rollback_slave, iter) {
4710 int tmp_res;
4711
4712 if (rollback_slave == slave)
4713 break;
4714
4715 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4716 if (tmp_res)
4717 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4718 tmp_res);
4719 }
4720
4721 return res;
4722}
4723
4724/* Change HW address
4725 *
4726 * Note that many devices must be down to change the HW address, and
4727 * downing the master releases all slaves. We can make bonds full of
4728 * bonding devices to test this, however.
4729 */
4730static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4731{
4732 struct bonding *bond = netdev_priv(dev: bond_dev);
4733 struct slave *slave, *rollback_slave;
4734 struct sockaddr_storage *ss = addr, tmp_ss;
4735 struct list_head *iter;
4736 int res = 0;
4737
4738 if (BOND_MODE(bond) == BOND_MODE_ALB)
4739 return bond_alb_set_mac_address(bond_dev, addr);
4740
4741
4742 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4743
4744 /* If fail_over_mac is enabled, do nothing and return success.
4745 * Returning an error causes ifenslave to fail.
4746 */
4747 if (bond->params.fail_over_mac &&
4748 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4749 return 0;
4750
4751 if (!is_valid_ether_addr(addr: ss->__data))
4752 return -EADDRNOTAVAIL;
4753
4754 bond_for_each_slave(bond, slave, iter) {
4755 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4756 __func__, slave);
4757 res = dev_set_mac_address(dev: slave->dev, sa: addr, NULL);
4758 if (res) {
4759 /* TODO: consider downing the slave
4760 * and retry ?
4761 * User should expect communications
4762 * breakage anyway until ARP finish
4763 * updating, so...
4764 */
4765 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4766 __func__, res);
4767 goto unwind;
4768 }
4769 }
4770
4771 /* success */
4772 dev_addr_set(dev: bond_dev, addr: ss->__data);
4773 return 0;
4774
4775unwind:
4776 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4777 tmp_ss.ss_family = bond_dev->type;
4778
4779 /* unwind from head to the slave that failed */
4780 bond_for_each_slave(bond, rollback_slave, iter) {
4781 int tmp_res;
4782
4783 if (rollback_slave == slave)
4784 break;
4785
4786 tmp_res = dev_set_mac_address(dev: rollback_slave->dev,
4787 sa: (struct sockaddr *)&tmp_ss, NULL);
4788 if (tmp_res) {
4789 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4790 __func__, tmp_res);
4791 }
4792 }
4793
4794 return res;
4795}
4796
4797/**
4798 * bond_get_slave_by_id - get xmit slave with slave_id
4799 * @bond: bonding device that is transmitting
4800 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4801 *
4802 * This function tries to get slave with slave_id but in case
4803 * it fails, it tries to find the first available slave for transmission.
4804 */
4805static struct slave *bond_get_slave_by_id(struct bonding *bond,
4806 int slave_id)
4807{
4808 struct list_head *iter;
4809 struct slave *slave;
4810 int i = slave_id;
4811
4812 /* Here we start from the slave with slave_id */
4813 bond_for_each_slave_rcu(bond, slave, iter) {
4814 if (--i < 0) {
4815 if (bond_slave_can_tx(slave))
4816 return slave;
4817 }
4818 }
4819
4820 /* Here we start from the first slave up to slave_id */
4821 i = slave_id;
4822 bond_for_each_slave_rcu(bond, slave, iter) {
4823 if (--i < 0)
4824 break;
4825 if (bond_slave_can_tx(slave))
4826 return slave;
4827 }
4828 /* no slave that can tx has been found */
4829 return NULL;
4830}
4831
4832/**
4833 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4834 * @bond: bonding device to use
4835 *
4836 * Based on the value of the bonding device's packets_per_slave parameter
4837 * this function generates a slave id, which is usually used as the next
4838 * slave to transmit through.
4839 */
4840static u32 bond_rr_gen_slave_id(struct bonding *bond)
4841{
4842 u32 slave_id;
4843 struct reciprocal_value reciprocal_packets_per_slave;
4844 int packets_per_slave = bond->params.packets_per_slave;
4845
4846 switch (packets_per_slave) {
4847 case 0:
4848 slave_id = get_random_u32();
4849 break;
4850 case 1:
4851 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4852 break;
4853 default:
4854 reciprocal_packets_per_slave =
4855 bond->params.reciprocal_packets_per_slave;
4856 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4857 slave_id = reciprocal_divide(a: slave_id,
4858 R: reciprocal_packets_per_slave);
4859 break;
4860 }
4861
4862 return slave_id;
4863}
4864
4865static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4866 struct sk_buff *skb)
4867{
4868 struct slave *slave;
4869 int slave_cnt;
4870 u32 slave_id;
4871
4872 /* Start with the curr_active_slave that joined the bond as the
4873 * default for sending IGMP traffic. For failover purposes one
4874 * needs to maintain some consistency for the interface that will
4875 * send the join/membership reports. The curr_active_slave found
4876 * will send all of this type of traffic.
4877 */
4878 if (skb->protocol == htons(ETH_P_IP)) {
4879 int noff = skb_network_offset(skb);
4880 struct iphdr *iph;
4881
4882 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4883 goto non_igmp;
4884
4885 iph = ip_hdr(skb);
4886 if (iph->protocol == IPPROTO_IGMP) {
4887 slave = rcu_dereference(bond->curr_active_slave);
4888 if (slave)
4889 return slave;
4890 return bond_get_slave_by_id(bond, slave_id: 0);
4891 }
4892 }
4893
4894non_igmp:
4895 slave_cnt = READ_ONCE(bond->slave_cnt);
4896 if (likely(slave_cnt)) {
4897 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4898 return bond_get_slave_by_id(bond, slave_id);
4899 }
4900 return NULL;
4901}
4902
4903static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4904 struct xdp_buff *xdp)
4905{
4906 struct slave *slave;
4907 int slave_cnt;
4908 u32 slave_id;
4909 const struct ethhdr *eth;
4910 void *data = xdp->data;
4911
4912 if (data + sizeof(struct ethhdr) > xdp->data_end)
4913 goto non_igmp;
4914
4915 eth = (struct ethhdr *)data;
4916 data += sizeof(struct ethhdr);
4917
4918 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4919 if (eth->h_proto == htons(ETH_P_IP)) {
4920 const struct iphdr *iph;
4921
4922 if (data + sizeof(struct iphdr) > xdp->data_end)
4923 goto non_igmp;
4924
4925 iph = (struct iphdr *)data;
4926
4927 if (iph->protocol == IPPROTO_IGMP) {
4928 slave = rcu_dereference(bond->curr_active_slave);
4929 if (slave)
4930 return slave;
4931 return bond_get_slave_by_id(bond, slave_id: 0);
4932 }
4933 }
4934
4935non_igmp:
4936 slave_cnt = READ_ONCE(bond->slave_cnt);
4937 if (likely(slave_cnt)) {
4938 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4939 return bond_get_slave_by_id(bond, slave_id);
4940 }
4941 return NULL;
4942}
4943
4944static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4945 struct net_device *bond_dev)
4946{
4947 struct bonding *bond = netdev_priv(dev: bond_dev);
4948 struct slave *slave;
4949
4950 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4951 if (likely(slave))
4952 return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
4953
4954 return bond_tx_drop(dev: bond_dev, skb);
4955}
4956
4957static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4958{
4959 return rcu_dereference(bond->curr_active_slave);
4960}
4961
4962/* In active-backup mode, we know that bond->curr_active_slave is always valid if
4963 * the bond has a usable interface.
4964 */
4965static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4966 struct net_device *bond_dev)
4967{
4968 struct bonding *bond = netdev_priv(dev: bond_dev);
4969 struct slave *slave;
4970
4971 slave = bond_xmit_activebackup_slave_get(bond);
4972 if (slave)
4973 return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
4974
4975 return bond_tx_drop(dev: bond_dev, skb);
4976}
4977
4978/* Use this to update slave_array when (a) it's not appropriate to update
4979 * slave_array right away (note that update_slave_array() may sleep)
4980 * and / or (b) RTNL is not held.
4981 */
4982void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4983{
4984 queue_delayed_work(wq: bond->wq, dwork: &bond->slave_arr_work, delay);
4985}
4986
4987/* Slave array work handler. Holds only RTNL */
4988static void bond_slave_arr_handler(struct work_struct *work)
4989{
4990 struct bonding *bond = container_of(work, struct bonding,
4991 slave_arr_work.work);
4992 int ret;
4993
4994 if (!rtnl_trylock())
4995 goto err;
4996
4997 ret = bond_update_slave_arr(bond, NULL);
4998 rtnl_unlock();
4999 if (ret) {
5000 pr_warn_ratelimited("Failed to update slave array from WT\n");
5001 goto err;
5002 }
5003 return;
5004
5005err:
5006 bond_slave_arr_work_rearm(bond, delay: 1);
5007}
5008
5009static void bond_skip_slave(struct bond_up_slave *slaves,
5010 struct slave *skipslave)
5011{
5012 int idx;
5013
5014 /* Rare situation where caller has asked to skip a specific
5015 * slave but allocation failed (most likely!). BTW this is
5016 * only possible when the call is initiated from
5017 * __bond_release_one(). In this situation; overwrite the
5018 * skipslave entry in the array with the last entry from the
5019 * array to avoid a situation where the xmit path may choose
5020 * this to-be-skipped slave to send a packet out.
5021 */
5022 for (idx = 0; slaves && idx < slaves->count; idx++) {
5023 if (skipslave == slaves->arr[idx]) {
5024 slaves->arr[idx] =
5025 slaves->arr[slaves->count - 1];
5026 slaves->count--;
5027 break;
5028 }
5029 }
5030}
5031
5032static void bond_set_slave_arr(struct bonding *bond,
5033 struct bond_up_slave *usable_slaves,
5034 struct bond_up_slave *all_slaves)
5035{
5036 struct bond_up_slave *usable, *all;
5037
5038 usable = rtnl_dereference(bond->usable_slaves);
5039 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5040 kfree_rcu(usable, rcu);
5041
5042 all = rtnl_dereference(bond->all_slaves);
5043 rcu_assign_pointer(bond->all_slaves, all_slaves);
5044 kfree_rcu(all, rcu);
5045}
5046
5047static void bond_reset_slave_arr(struct bonding *bond)
5048{
5049 bond_set_slave_arr(bond, NULL, NULL);
5050}
5051
5052/* Build the usable slaves array in control path for modes that use xmit-hash
5053 * to determine the slave interface -
5054 * (a) BOND_MODE_8023AD
5055 * (b) BOND_MODE_XOR
5056 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5057 *
5058 * The caller is expected to hold RTNL only and NO other lock!
5059 */
5060int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5061{
5062 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5063 struct slave *slave;
5064 struct list_head *iter;
5065 int agg_id = 0;
5066 int ret = 0;
5067
5068 might_sleep();
5069
5070 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5071 bond->slave_cnt), GFP_KERNEL);
5072 all_slaves = kzalloc(struct_size(all_slaves, arr,
5073 bond->slave_cnt), GFP_KERNEL);
5074 if (!usable_slaves || !all_slaves) {
5075 ret = -ENOMEM;
5076 goto out;
5077 }
5078 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5079 struct ad_info ad_info;
5080
5081 spin_lock_bh(lock: &bond->mode_lock);
5082 if (bond_3ad_get_active_agg_info(bond, ad_info: &ad_info)) {
5083 spin_unlock_bh(lock: &bond->mode_lock);
5084 pr_debug("bond_3ad_get_active_agg_info failed\n");
5085 /* No active aggragator means it's not safe to use
5086 * the previous array.
5087 */
5088 bond_reset_slave_arr(bond);
5089 goto out;
5090 }
5091 spin_unlock_bh(lock: &bond->mode_lock);
5092 agg_id = ad_info.aggregator_id;
5093 }
5094 bond_for_each_slave(bond, slave, iter) {
5095 if (skipslave == slave)
5096 continue;
5097
5098 all_slaves->arr[all_slaves->count++] = slave;
5099 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5100 struct aggregator *agg;
5101
5102 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5103 if (!agg || agg->aggregator_identifier != agg_id)
5104 continue;
5105 }
5106 if (!bond_slave_can_tx(slave))
5107 continue;
5108
5109 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5110 usable_slaves->count);
5111
5112 usable_slaves->arr[usable_slaves->count++] = slave;
5113 }
5114
5115 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5116 return ret;
5117out:
5118 if (ret != 0 && skipslave) {
5119 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5120 skipslave);
5121 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5122 skipslave);
5123 }
5124 kfree_rcu(all_slaves, rcu);
5125 kfree_rcu(usable_slaves, rcu);
5126
5127 return ret;
5128}
5129
5130static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5131 struct sk_buff *skb,
5132 struct bond_up_slave *slaves)
5133{
5134 struct slave *slave;
5135 unsigned int count;
5136 u32 hash;
5137
5138 hash = bond_xmit_hash(bond, skb);
5139 count = slaves ? READ_ONCE(slaves->count) : 0;
5140 if (unlikely(!count))
5141 return NULL;
5142
5143 slave = slaves->arr[hash % count];
5144 return slave;
5145}
5146
5147static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5148 struct xdp_buff *xdp)
5149{
5150 struct bond_up_slave *slaves;
5151 unsigned int count;
5152 u32 hash;
5153
5154 hash = bond_xmit_hash_xdp(bond, xdp);
5155 slaves = rcu_dereference(bond->usable_slaves);
5156 count = slaves ? READ_ONCE(slaves->count) : 0;
5157 if (unlikely(!count))
5158 return NULL;
5159
5160 return slaves->arr[hash % count];
5161}
5162
5163/* Use this Xmit function for 3AD as well as XOR modes. The current
5164 * usable slave array is formed in the control path. The xmit function
5165 * just calculates hash and sends the packet out.
5166 */
5167static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5168 struct net_device *dev)
5169{
5170 struct bonding *bond = netdev_priv(dev);
5171 struct bond_up_slave *slaves;
5172 struct slave *slave;
5173
5174 slaves = rcu_dereference(bond->usable_slaves);
5175 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5176 if (likely(slave))
5177 return bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
5178
5179 return bond_tx_drop(dev, skb);
5180}
5181
5182/* in broadcast mode, we send everything to all usable interfaces. */
5183static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5184 struct net_device *bond_dev)
5185{
5186 struct bonding *bond = netdev_priv(dev: bond_dev);
5187 struct slave *slave = NULL;
5188 struct list_head *iter;
5189 bool xmit_suc = false;
5190 bool skb_used = false;
5191
5192 bond_for_each_slave_rcu(bond, slave, iter) {
5193 struct sk_buff *skb2;
5194
5195 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5196 continue;
5197
5198 if (bond_is_last_slave(bond, slave)) {
5199 skb2 = skb;
5200 skb_used = true;
5201 } else {
5202 skb2 = skb_clone(skb, GFP_ATOMIC);
5203 if (!skb2) {
5204 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5205 bond_dev->name, __func__);
5206 continue;
5207 }
5208 }
5209
5210 if (bond_dev_queue_xmit(bond, skb: skb2, slave_dev: slave->dev) == NETDEV_TX_OK)
5211 xmit_suc = true;
5212 }
5213
5214 if (!skb_used)
5215 dev_kfree_skb_any(skb);
5216
5217 if (xmit_suc)
5218 return NETDEV_TX_OK;
5219
5220 dev_core_stats_tx_dropped_inc(dev: bond_dev);
5221 return NET_XMIT_DROP;
5222}
5223
5224/*------------------------- Device initialization ---------------------------*/
5225
5226/* Lookup the slave that corresponds to a qid */
5227static inline int bond_slave_override(struct bonding *bond,
5228 struct sk_buff *skb)
5229{
5230 struct slave *slave = NULL;
5231 struct list_head *iter;
5232
5233 if (!skb_rx_queue_recorded(skb))
5234 return 1;
5235
5236 /* Find out if any slaves have the same mapping as this skb. */
5237 bond_for_each_slave_rcu(bond, slave, iter) {
5238 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5239 if (bond_slave_is_up(slave) &&
5240 slave->link == BOND_LINK_UP) {
5241 bond_dev_queue_xmit(bond, skb, slave_dev: slave->dev);
5242 return 0;
5243 }
5244 /* If the slave isn't UP, use default transmit policy. */
5245 break;
5246 }
5247 }
5248
5249 return 1;
5250}
5251
5252
5253static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5254 struct net_device *sb_dev)
5255{
5256 /* This helper function exists to help dev_pick_tx get the correct
5257 * destination queue. Using a helper function skips a call to
5258 * skb_tx_hash and will put the skbs in the queue we expect on their
5259 * way down to the bonding driver.
5260 */
5261 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5262
5263 /* Save the original txq to restore before passing to the driver */
5264 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5265
5266 if (unlikely(txq >= dev->real_num_tx_queues)) {
5267 do {
5268 txq -= dev->real_num_tx_queues;
5269 } while (txq >= dev->real_num_tx_queues);
5270 }
5271 return txq;
5272}
5273
5274static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5275 struct sk_buff *skb,
5276 bool all_slaves)
5277{
5278 struct bonding *bond = netdev_priv(dev: master_dev);
5279 struct bond_up_slave *slaves;
5280 struct slave *slave = NULL;
5281
5282 switch (BOND_MODE(bond)) {
5283 case BOND_MODE_ROUNDROBIN:
5284 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5285 break;
5286 case BOND_MODE_ACTIVEBACKUP:
5287 slave = bond_xmit_activebackup_slave_get(bond);
5288 break;
5289 case BOND_MODE_8023AD:
5290 case BOND_MODE_XOR:
5291 if (all_slaves)
5292 slaves = rcu_dereference(bond->all_slaves);
5293 else
5294 slaves = rcu_dereference(bond->usable_slaves);
5295 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5296 break;
5297 case BOND_MODE_BROADCAST:
5298 break;
5299 case BOND_MODE_ALB:
5300 slave = bond_xmit_alb_slave_get(bond, skb);
5301 break;
5302 case BOND_MODE_TLB:
5303 slave = bond_xmit_tlb_slave_get(bond, skb);
5304 break;
5305 default:
5306 /* Should never happen, mode already checked */
5307 WARN_ONCE(true, "Unknown bonding mode");
5308 break;
5309 }
5310
5311 if (slave)
5312 return slave->dev;
5313 return NULL;
5314}
5315
5316static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5317{
5318 switch (sk->sk_family) {
5319#if IS_ENABLED(CONFIG_IPV6)
5320 case AF_INET6:
5321 if (ipv6_only_sock(sk) ||
5322 ipv6_addr_type(addr: &sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5323 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5324 flow->addrs.v6addrs.src = inet6_sk(sk: sk)->saddr;
5325 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5326 break;
5327 }
5328 fallthrough;
5329#endif
5330 default: /* AF_INET */
5331 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5332 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5333 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5334 break;
5335 }
5336
5337 flow->ports.src = inet_sk(sk)->inet_sport;
5338 flow->ports.dst = inet_sk(sk)->inet_dport;
5339}
5340
5341/**
5342 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5343 * @sk: socket to use for headers
5344 *
5345 * This function will extract the necessary field from the socket and use
5346 * them to generate a hash based on the LAYER34 xmit_policy.
5347 * Assumes that sk is a TCP or UDP socket.
5348 */
5349static u32 bond_sk_hash_l34(struct sock *sk)
5350{
5351 struct flow_keys flow;
5352 u32 hash;
5353
5354 bond_sk_to_flow(sk, flow: &flow);
5355
5356 /* L4 */
5357 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5358 /* L3 */
5359 return bond_ip_hash(hash, flow: &flow, BOND_XMIT_POLICY_LAYER34);
5360}
5361
5362static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5363 struct sock *sk)
5364{
5365 struct bond_up_slave *slaves;
5366 struct slave *slave;
5367 unsigned int count;
5368 u32 hash;
5369
5370 slaves = rcu_dereference(bond->usable_slaves);
5371 count = slaves ? READ_ONCE(slaves->count) : 0;
5372 if (unlikely(!count))
5373 return NULL;
5374
5375 hash = bond_sk_hash_l34(sk);
5376 slave = slaves->arr[hash % count];
5377
5378 return slave->dev;
5379}
5380
5381static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5382 struct sock *sk)
5383{
5384 struct bonding *bond = netdev_priv(dev);
5385 struct net_device *lower = NULL;
5386
5387 rcu_read_lock();
5388 if (bond_sk_check(bond))
5389 lower = __bond_sk_get_lower_dev(bond, sk);
5390 rcu_read_unlock();
5391
5392 return lower;
5393}
5394
5395#if IS_ENABLED(CONFIG_TLS_DEVICE)
5396static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5397 struct net_device *dev)
5398{
5399 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5400
5401 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5402 * was true, if tls_device_down is running in parallel, but it's OK,
5403 * because bond_get_slave_by_dev has a NULL check.
5404 */
5405 if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5406 return bond_dev_queue_xmit(bond, skb, slave_dev: tls_netdev);
5407 return bond_tx_drop(dev, skb);
5408}
5409#endif
5410
5411static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5412{
5413 struct bonding *bond = netdev_priv(dev);
5414
5415 if (bond_should_override_tx_queue(bond) &&
5416 !bond_slave_override(bond, skb))
5417 return NETDEV_TX_OK;
5418
5419#if IS_ENABLED(CONFIG_TLS_DEVICE)
5420 if (tls_is_skb_tx_device_offloaded(skb))
5421 return bond_tls_device_xmit(bond, skb, dev);
5422#endif
5423
5424 switch (BOND_MODE(bond)) {
5425 case BOND_MODE_ROUNDROBIN:
5426 return bond_xmit_roundrobin(skb, bond_dev: dev);
5427 case BOND_MODE_ACTIVEBACKUP:
5428 return bond_xmit_activebackup(skb, bond_dev: dev);
5429 case BOND_MODE_8023AD:
5430 case BOND_MODE_XOR:
5431 return bond_3ad_xor_xmit(skb, dev);
5432 case BOND_MODE_BROADCAST:
5433 return bond_xmit_broadcast(skb, bond_dev: dev);
5434 case BOND_MODE_ALB:
5435 return bond_alb_xmit(skb, bond_dev: dev);
5436 case BOND_MODE_TLB:
5437 return bond_tlb_xmit(skb, bond_dev: dev);
5438 default:
5439 /* Should never happen, mode already checked */
5440 netdev_err(dev, format: "Unknown bonding mode %d\n", BOND_MODE(bond));
5441 WARN_ON_ONCE(1);
5442 return bond_tx_drop(dev, skb);
5443 }
5444}
5445
5446static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5447{
5448 struct bonding *bond = netdev_priv(dev);
5449 netdev_tx_t ret = NETDEV_TX_OK;
5450
5451 /* If we risk deadlock from transmitting this in the
5452 * netpoll path, tell netpoll to queue the frame for later tx
5453 */
5454 if (unlikely(is_netpoll_tx_blocked(dev)))
5455 return NETDEV_TX_BUSY;
5456
5457 rcu_read_lock();
5458 if (bond_has_slaves(bond))
5459 ret = __bond_start_xmit(skb, dev);
5460 else
5461 ret = bond_tx_drop(dev, skb);
5462 rcu_read_unlock();
5463
5464 return ret;
5465}
5466
5467static struct net_device *
5468bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5469{
5470 struct bonding *bond = netdev_priv(dev: bond_dev);
5471 struct slave *slave;
5472
5473 /* Caller needs to hold rcu_read_lock() */
5474
5475 switch (BOND_MODE(bond)) {
5476 case BOND_MODE_ROUNDROBIN:
5477 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5478 break;
5479
5480 case BOND_MODE_ACTIVEBACKUP:
5481 slave = bond_xmit_activebackup_slave_get(bond);
5482 break;
5483
5484 case BOND_MODE_8023AD:
5485 case BOND_MODE_XOR:
5486 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5487 break;
5488
5489 default:
5490 /* Should never happen. Mode guarded by bond_xdp_check() */
5491 netdev_err(dev: bond_dev, format: "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5492 WARN_ON_ONCE(1);
5493 return NULL;
5494 }
5495
5496 if (slave)
5497 return slave->dev;
5498
5499 return NULL;
5500}
5501
5502static int bond_xdp_xmit(struct net_device *bond_dev,
5503 int n, struct xdp_frame **frames, u32 flags)
5504{
5505 int nxmit, err = -ENXIO;
5506
5507 rcu_read_lock();
5508
5509 for (nxmit = 0; nxmit < n; nxmit++) {
5510 struct xdp_frame *frame = frames[nxmit];
5511 struct xdp_frame *frames1[] = {frame};
5512 struct net_device *slave_dev;
5513 struct xdp_buff xdp;
5514
5515 xdp_convert_frame_to_buff(frame, xdp: &xdp);
5516
5517 slave_dev = bond_xdp_get_xmit_slave(bond_dev, xdp: &xdp);
5518 if (!slave_dev) {
5519 err = -ENXIO;
5520 break;
5521 }
5522
5523 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5524 if (err < 1)
5525 break;
5526 }
5527
5528 rcu_read_unlock();
5529
5530 /* If error happened on the first frame then we can pass the error up, otherwise
5531 * report the number of frames that were xmitted.
5532 */
5533 if (err < 0)
5534 return (nxmit == 0 ? err : nxmit);
5535
5536 return nxmit;
5537}
5538
5539static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5540 struct netlink_ext_ack *extack)
5541{
5542 struct bonding *bond = netdev_priv(dev);
5543 struct list_head *iter;
5544 struct slave *slave, *rollback_slave;
5545 struct bpf_prog *old_prog;
5546 struct netdev_bpf xdp = {
5547 .command = XDP_SETUP_PROG,
5548 .flags = 0,
5549 .prog = prog,
5550 .extack = extack,
5551 };
5552 int err;
5553
5554 ASSERT_RTNL();
5555
5556 if (!bond_xdp_check(bond))
5557 return -EOPNOTSUPP;
5558
5559 old_prog = bond->xdp_prog;
5560 bond->xdp_prog = prog;
5561
5562 bond_for_each_slave(bond, slave, iter) {
5563 struct net_device *slave_dev = slave->dev;
5564
5565 if (!slave_dev->netdev_ops->ndo_bpf ||
5566 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5567 SLAVE_NL_ERR(dev, slave_dev, extack,
5568 "Slave device does not support XDP");
5569 err = -EOPNOTSUPP;
5570 goto err;
5571 }
5572
5573 if (dev_xdp_prog_count(dev: slave_dev) > 0) {
5574 SLAVE_NL_ERR(dev, slave_dev, extack,
5575 "Slave has XDP program loaded, please unload before enslaving");
5576 err = -EOPNOTSUPP;
5577 goto err;
5578 }
5579
5580 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5581 if (err < 0) {
5582 /* ndo_bpf() sets extack error message */
5583 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5584 goto err;
5585 }
5586 if (prog)
5587 bpf_prog_inc(prog);
5588 }
5589
5590 if (prog) {
5591 static_branch_inc(&bpf_master_redirect_enabled_key);
5592 } else if (old_prog) {
5593 bpf_prog_put(prog: old_prog);
5594 static_branch_dec(&bpf_master_redirect_enabled_key);
5595 }
5596
5597 return 0;
5598
5599err:
5600 /* unwind the program changes */
5601 bond->xdp_prog = old_prog;
5602 xdp.prog = old_prog;
5603 xdp.extack = NULL; /* do not overwrite original error */
5604
5605 bond_for_each_slave(bond, rollback_slave, iter) {
5606 struct net_device *slave_dev = rollback_slave->dev;
5607 int err_unwind;
5608
5609 if (slave == rollback_slave)
5610 break;
5611
5612 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5613 if (err_unwind < 0)
5614 slave_err(dev, slave_dev,
5615 "Error %d when unwinding XDP program change\n", err_unwind);
5616 else if (xdp.prog)
5617 bpf_prog_inc(prog: xdp.prog);
5618 }
5619 return err;
5620}
5621
5622static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5623{
5624 switch (xdp->command) {
5625 case XDP_SETUP_PROG:
5626 return bond_xdp_set(dev, prog: xdp->prog, extack: xdp->extack);
5627 default:
5628 return -EINVAL;
5629 }
5630}
5631
5632static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5633{
5634 if (speed == 0 || speed == SPEED_UNKNOWN)
5635 speed = slave->speed;
5636 else
5637 speed = min(speed, slave->speed);
5638
5639 return speed;
5640}
5641
5642/* Set the BOND_PHC_INDEX flag to notify user space */
5643static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5644{
5645 struct ifreq *ifr = kernel_cfg->ifr;
5646 struct hwtstamp_config cfg;
5647
5648 if (kernel_cfg->copied_to_user) {
5649 /* Lower device has a legacy implementation */
5650 if (copy_from_user(to: &cfg, from: ifr->ifr_data, n: sizeof(cfg)))
5651 return -EFAULT;
5652
5653 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5654 if (copy_to_user(to: ifr->ifr_data, from: &cfg, n: sizeof(cfg)))
5655 return -EFAULT;
5656 } else {
5657 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5658 }
5659
5660 return 0;
5661}
5662
5663static int bond_hwtstamp_get(struct net_device *dev,
5664 struct kernel_hwtstamp_config *cfg)
5665{
5666 struct bonding *bond = netdev_priv(dev);
5667 struct net_device *real_dev;
5668 int err;
5669
5670 real_dev = bond_option_active_slave_get_rcu(bond);
5671 if (!real_dev)
5672 return -EOPNOTSUPP;
5673
5674 err = generic_hwtstamp_get_lower(dev: real_dev, kernel_cfg: cfg);
5675 if (err)
5676 return err;
5677
5678 return bond_set_phc_index_flag(kernel_cfg: cfg);
5679}
5680
5681static int bond_hwtstamp_set(struct net_device *dev,
5682 struct kernel_hwtstamp_config *cfg,
5683 struct netlink_ext_ack *extack)
5684{
5685 struct bonding *bond = netdev_priv(dev);
5686 struct net_device *real_dev;
5687 int err;
5688
5689 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5690 return -EOPNOTSUPP;
5691
5692 real_dev = bond_option_active_slave_get_rcu(bond);
5693 if (!real_dev)
5694 return -EOPNOTSUPP;
5695
5696 err = generic_hwtstamp_set_lower(dev: real_dev, kernel_cfg: cfg, extack);
5697 if (err)
5698 return err;
5699
5700 return bond_set_phc_index_flag(kernel_cfg: cfg);
5701}
5702
5703static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5704 struct ethtool_link_ksettings *cmd)
5705{
5706 struct bonding *bond = netdev_priv(dev: bond_dev);
5707 struct list_head *iter;
5708 struct slave *slave;
5709 u32 speed = 0;
5710
5711 cmd->base.duplex = DUPLEX_UNKNOWN;
5712 cmd->base.port = PORT_OTHER;
5713
5714 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5715 * do not need to check mode. Though link speed might not represent
5716 * the true receive or transmit bandwidth (not all modes are symmetric)
5717 * this is an accurate maximum.
5718 */
5719 bond_for_each_slave(bond, slave, iter) {
5720 if (bond_slave_can_tx(slave)) {
5721 bond_update_speed_duplex(slave);
5722 if (slave->speed != SPEED_UNKNOWN) {
5723 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5724 speed = bond_mode_bcast_speed(slave,
5725 speed);
5726 else
5727 speed += slave->speed;
5728 }
5729 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5730 slave->duplex != DUPLEX_UNKNOWN)
5731 cmd->base.duplex = slave->duplex;
5732 }
5733 }
5734 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5735
5736 return 0;
5737}
5738
5739static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5740 struct ethtool_drvinfo *drvinfo)
5741{
5742 strscpy(p: drvinfo->driver, DRV_NAME, size: sizeof(drvinfo->driver));
5743 snprintf(buf: drvinfo->fw_version, size: sizeof(drvinfo->fw_version), fmt: "%d",
5744 BOND_ABI_VERSION);
5745}
5746
5747static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5748 struct ethtool_ts_info *info)
5749{
5750 struct bonding *bond = netdev_priv(dev: bond_dev);
5751 struct ethtool_ts_info ts_info;
5752 const struct ethtool_ops *ops;
5753 struct net_device *real_dev;
5754 bool sw_tx_support = false;
5755 struct phy_device *phydev;
5756 struct list_head *iter;
5757 struct slave *slave;
5758 int ret = 0;
5759
5760 rcu_read_lock();
5761 real_dev = bond_option_active_slave_get_rcu(bond);
5762 dev_hold(dev: real_dev);
5763 rcu_read_unlock();
5764
5765 if (real_dev) {
5766 ops = real_dev->ethtool_ops;
5767 phydev = real_dev->phydev;
5768
5769 if (phy_has_tsinfo(phydev)) {
5770 ret = phy_ts_info(phydev, tsinfo: info);
5771 goto out;
5772 } else if (ops->get_ts_info) {
5773 ret = ops->get_ts_info(real_dev, info);
5774 goto out;
5775 }
5776 } else {
5777 /* Check if all slaves support software tx timestamping */
5778 rcu_read_lock();
5779 bond_for_each_slave_rcu(bond, slave, iter) {
5780 ret = -1;
5781 ops = slave->dev->ethtool_ops;
5782 phydev = slave->dev->phydev;
5783
5784 if (phy_has_tsinfo(phydev))
5785 ret = phy_ts_info(phydev, tsinfo: &ts_info);
5786 else if (ops->get_ts_info)
5787 ret = ops->get_ts_info(slave->dev, &ts_info);
5788
5789 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5790 sw_tx_support = true;
5791 continue;
5792 }
5793
5794 sw_tx_support = false;
5795 break;
5796 }
5797 rcu_read_unlock();
5798 }
5799
5800 ret = 0;
5801 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5802 SOF_TIMESTAMPING_SOFTWARE;
5803 if (sw_tx_support)
5804 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5805
5806 info->phc_index = -1;
5807
5808out:
5809 dev_put(dev: real_dev);
5810 return ret;
5811}
5812
5813static const struct ethtool_ops bond_ethtool_ops = {
5814 .get_drvinfo = bond_ethtool_get_drvinfo,
5815 .get_link = ethtool_op_get_link,
5816 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5817 .get_ts_info = bond_ethtool_get_ts_info,
5818};
5819
5820static const struct net_device_ops bond_netdev_ops = {
5821 .ndo_init = bond_init,
5822 .ndo_uninit = bond_uninit,
5823 .ndo_open = bond_open,
5824 .ndo_stop = bond_close,
5825 .ndo_start_xmit = bond_start_xmit,
5826 .ndo_select_queue = bond_select_queue,
5827 .ndo_get_stats64 = bond_get_stats,
5828 .ndo_eth_ioctl = bond_eth_ioctl,
5829 .ndo_siocbond = bond_do_ioctl,
5830 .ndo_siocdevprivate = bond_siocdevprivate,
5831 .ndo_change_rx_flags = bond_change_rx_flags,
5832 .ndo_set_rx_mode = bond_set_rx_mode,
5833 .ndo_change_mtu = bond_change_mtu,
5834 .ndo_set_mac_address = bond_set_mac_address,
5835 .ndo_neigh_setup = bond_neigh_setup,
5836 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5837 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5838#ifdef CONFIG_NET_POLL_CONTROLLER
5839 .ndo_netpoll_setup = bond_netpoll_setup,
5840 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5841 .ndo_poll_controller = bond_poll_controller,
5842#endif
5843 .ndo_add_slave = bond_enslave,
5844 .ndo_del_slave = bond_release,
5845 .ndo_fix_features = bond_fix_features,
5846 .ndo_features_check = passthru_features_check,
5847 .ndo_get_xmit_slave = bond_xmit_get_slave,
5848 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5849 .ndo_bpf = bond_xdp,
5850 .ndo_xdp_xmit = bond_xdp_xmit,
5851 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5852 .ndo_hwtstamp_get = bond_hwtstamp_get,
5853 .ndo_hwtstamp_set = bond_hwtstamp_set,
5854};
5855
5856static const struct device_type bond_type = {
5857 .name = "bond",
5858};
5859
5860static void bond_destructor(struct net_device *bond_dev)
5861{
5862 struct bonding *bond = netdev_priv(dev: bond_dev);
5863
5864 if (bond->wq)
5865 destroy_workqueue(wq: bond->wq);
5866
5867 free_percpu(pdata: bond->rr_tx_counter);
5868}
5869
5870void bond_setup(struct net_device *bond_dev)
5871{
5872 struct bonding *bond = netdev_priv(dev: bond_dev);
5873
5874 spin_lock_init(&bond->mode_lock);
5875 bond->params = bonding_defaults;
5876
5877 /* Initialize pointers */
5878 bond->dev = bond_dev;
5879
5880 /* Initialize the device entry points */
5881 ether_setup(dev: bond_dev);
5882 bond_dev->max_mtu = ETH_MAX_MTU;
5883 bond_dev->netdev_ops = &bond_netdev_ops;
5884 bond_dev->ethtool_ops = &bond_ethtool_ops;
5885
5886 bond_dev->needs_free_netdev = true;
5887 bond_dev->priv_destructor = bond_destructor;
5888
5889 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5890
5891 /* Initialize the device options */
5892 bond_dev->flags |= IFF_MASTER;
5893 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5894 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5895
5896#ifdef CONFIG_XFRM_OFFLOAD
5897 /* set up xfrm device ops (only supported in active-backup right now) */
5898 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5899 INIT_LIST_HEAD(list: &bond->ipsec_list);
5900 spin_lock_init(&bond->ipsec_lock);
5901#endif /* CONFIG_XFRM_OFFLOAD */
5902
5903 /* don't acquire bond device's netif_tx_lock when transmitting */
5904 bond_dev->features |= NETIF_F_LLTX;
5905
5906 /* By default, we declare the bond to be fully
5907 * VLAN hardware accelerated capable. Special
5908 * care is taken in the various xmit functions
5909 * when there are slaves that are not hw accel
5910 * capable
5911 */
5912
5913 /* Don't allow bond devices to change network namespaces. */
5914 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5915
5916 bond_dev->hw_features = BOND_VLAN_FEATURES |
5917 NETIF_F_HW_VLAN_CTAG_RX |
5918 NETIF_F_HW_VLAN_CTAG_FILTER |
5919 NETIF_F_HW_VLAN_STAG_RX |
5920 NETIF_F_HW_VLAN_STAG_FILTER;
5921
5922 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5923 bond_dev->features |= bond_dev->hw_features;
5924 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5925#ifdef CONFIG_XFRM_OFFLOAD
5926 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5927 /* Only enable XFRM features if this is an active-backup config */
5928 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5929 bond_dev->features |= BOND_XFRM_FEATURES;
5930#endif /* CONFIG_XFRM_OFFLOAD */
5931
5932 if (bond_xdp_check(bond))
5933 bond_dev->xdp_features = NETDEV_XDP_ACT_MASK;
5934}
5935
5936/* Destroy a bonding device.
5937 * Must be under rtnl_lock when this function is called.
5938 */
5939static void bond_uninit(struct net_device *bond_dev)
5940{
5941 struct bonding *bond = netdev_priv(dev: bond_dev);
5942 struct list_head *iter;
5943 struct slave *slave;
5944
5945 bond_netpoll_cleanup(bond_dev);
5946
5947 /* Release the bonded slaves */
5948 bond_for_each_slave(bond, slave, iter)
5949 __bond_release_one(bond_dev, slave_dev: slave->dev, all: true, unregister: true);
5950 netdev_info(dev: bond_dev, format: "Released all slaves\n");
5951
5952 bond_set_slave_arr(bond, NULL, NULL);
5953
5954 list_del(entry: &bond->bond_list);
5955
5956 bond_debug_unregister(bond);
5957}
5958
5959/*------------------------- Module initialization ---------------------------*/
5960
5961static int __init bond_check_params(struct bond_params *params)
5962{
5963 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5964 struct bond_opt_value newval;
5965 const struct bond_opt_value *valptr;
5966 int arp_all_targets_value = 0;
5967 u16 ad_actor_sys_prio = 0;
5968 u16 ad_user_port_key = 0;
5969 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5970 int arp_ip_count;
5971 int bond_mode = BOND_MODE_ROUNDROBIN;
5972 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5973 int lacp_fast = 0;
5974 int tlb_dynamic_lb;
5975
5976 /* Convert string parameters. */
5977 if (mode) {
5978 bond_opt_initstr(&newval, mode);
5979 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_MODE), val: &newval);
5980 if (!valptr) {
5981 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5982 return -EINVAL;
5983 }
5984 bond_mode = valptr->value;
5985 }
5986
5987 if (xmit_hash_policy) {
5988 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5989 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5990 bond_mode == BOND_MODE_BROADCAST) {
5991 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5992 bond_mode_name(bond_mode));
5993 } else {
5994 bond_opt_initstr(&newval, xmit_hash_policy);
5995 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_XMIT_HASH),
5996 val: &newval);
5997 if (!valptr) {
5998 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5999 xmit_hash_policy);
6000 return -EINVAL;
6001 }
6002 xmit_hashtype = valptr->value;
6003 }
6004 }
6005
6006 if (lacp_rate) {
6007 if (bond_mode != BOND_MODE_8023AD) {
6008 pr_info("lacp_rate param is irrelevant in mode %s\n",
6009 bond_mode_name(bond_mode));
6010 } else {
6011 bond_opt_initstr(&newval, lacp_rate);
6012 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_LACP_RATE),
6013 val: &newval);
6014 if (!valptr) {
6015 pr_err("Error: Invalid lacp rate \"%s\"\n",
6016 lacp_rate);
6017 return -EINVAL;
6018 }
6019 lacp_fast = valptr->value;
6020 }
6021 }
6022
6023 if (ad_select) {
6024 bond_opt_initstr(&newval, ad_select);
6025 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_AD_SELECT),
6026 val: &newval);
6027 if (!valptr) {
6028 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6029 return -EINVAL;
6030 }
6031 params->ad_select = valptr->value;
6032 if (bond_mode != BOND_MODE_8023AD)
6033 pr_warn("ad_select param only affects 802.3ad mode\n");
6034 } else {
6035 params->ad_select = BOND_AD_STABLE;
6036 }
6037
6038 if (max_bonds < 0) {
6039 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6040 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6041 max_bonds = BOND_DEFAULT_MAX_BONDS;
6042 }
6043
6044 if (miimon < 0) {
6045 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6046 miimon, INT_MAX);
6047 miimon = 0;
6048 }
6049
6050 if (updelay < 0) {
6051 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6052 updelay, INT_MAX);
6053 updelay = 0;
6054 }
6055
6056 if (downdelay < 0) {
6057 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6058 downdelay, INT_MAX);
6059 downdelay = 0;
6060 }
6061
6062 if ((use_carrier != 0) && (use_carrier != 1)) {
6063 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6064 use_carrier);
6065 use_carrier = 1;
6066 }
6067
6068 if (num_peer_notif < 0 || num_peer_notif > 255) {
6069 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6070 num_peer_notif);
6071 num_peer_notif = 1;
6072 }
6073
6074 /* reset values for 802.3ad/TLB/ALB */
6075 if (!bond_mode_uses_arp(mode: bond_mode)) {
6076 if (!miimon) {
6077 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");
6078 pr_warn("Forcing miimon to 100msec\n");
6079 miimon = BOND_DEFAULT_MIIMON;
6080 }
6081 }
6082
6083 if (tx_queues < 1 || tx_queues > 255) {
6084 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6085 tx_queues, BOND_DEFAULT_TX_QUEUES);
6086 tx_queues = BOND_DEFAULT_TX_QUEUES;
6087 }
6088
6089 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6090 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6091 all_slaves_active);
6092 all_slaves_active = 0;
6093 }
6094
6095 if (resend_igmp < 0 || resend_igmp > 255) {
6096 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6097 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6098 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6099 }
6100
6101 bond_opt_initval(&newval, packets_per_slave);
6102 if (!bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_PACKETS_PER_SLAVE), val: &newval)) {
6103 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6104 packets_per_slave, USHRT_MAX);
6105 packets_per_slave = 1;
6106 }
6107
6108 if (bond_mode == BOND_MODE_ALB) {
6109 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",
6110 updelay);
6111 }
6112
6113 if (!miimon) {
6114 if (updelay || downdelay) {
6115 /* just warn the user the up/down delay will have
6116 * no effect since miimon is zero...
6117 */
6118 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",
6119 updelay, downdelay);
6120 }
6121 } else {
6122 /* don't allow arp monitoring */
6123 if (arp_interval) {
6124 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6125 miimon, arp_interval);
6126 arp_interval = 0;
6127 }
6128
6129 if ((updelay % miimon) != 0) {
6130 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6131 updelay, miimon, (updelay / miimon) * miimon);
6132 }
6133
6134 updelay /= miimon;
6135
6136 if ((downdelay % miimon) != 0) {
6137 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6138 downdelay, miimon,
6139 (downdelay / miimon) * miimon);
6140 }
6141
6142 downdelay /= miimon;
6143 }
6144
6145 if (arp_interval < 0) {
6146 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6147 arp_interval, INT_MAX);
6148 arp_interval = 0;
6149 }
6150
6151 for (arp_ip_count = 0, i = 0;
6152 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6153 __be32 ip;
6154
6155 /* not a complete check, but good enough to catch mistakes */
6156 if (!in4_pton(src: arp_ip_target[i], srclen: -1, dst: (u8 *)&ip, delim: -1, NULL) ||
6157 !bond_is_ip_target_ok(addr: ip)) {
6158 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6159 arp_ip_target[i]);
6160 arp_interval = 0;
6161 } else {
6162 if (bond_get_targets_ip(targets: arp_target, ip) == -1)
6163 arp_target[arp_ip_count++] = ip;
6164 else
6165 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6166 &ip);
6167 }
6168 }
6169
6170 if (arp_interval && !arp_ip_count) {
6171 /* don't allow arping if no arp_ip_target given... */
6172 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6173 arp_interval);
6174 arp_interval = 0;
6175 }
6176
6177 if (arp_validate) {
6178 if (!arp_interval) {
6179 pr_err("arp_validate requires arp_interval\n");
6180 return -EINVAL;
6181 }
6182
6183 bond_opt_initstr(&newval, arp_validate);
6184 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_ARP_VALIDATE),
6185 val: &newval);
6186 if (!valptr) {
6187 pr_err("Error: invalid arp_validate \"%s\"\n",
6188 arp_validate);
6189 return -EINVAL;
6190 }
6191 arp_validate_value = valptr->value;
6192 } else {
6193 arp_validate_value = 0;
6194 }
6195
6196 if (arp_all_targets) {
6197 bond_opt_initstr(&newval, arp_all_targets);
6198 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_ARP_ALL_TARGETS),
6199 val: &newval);
6200 if (!valptr) {
6201 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6202 arp_all_targets);
6203 arp_all_targets_value = 0;
6204 } else {
6205 arp_all_targets_value = valptr->value;
6206 }
6207 }
6208
6209 if (miimon) {
6210 pr_info("MII link monitoring set to %d ms\n", miimon);
6211 } else if (arp_interval) {
6212 valptr = bond_opt_get_val(option: BOND_OPT_ARP_VALIDATE,
6213 val: arp_validate_value);
6214 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6215 arp_interval, valptr->string, arp_ip_count);
6216
6217 for (i = 0; i < arp_ip_count; i++)
6218 pr_cont(" %s", arp_ip_target[i]);
6219
6220 pr_cont("\n");
6221
6222 } else if (max_bonds) {
6223 /* miimon and arp_interval not set, we need one so things
6224 * work as expected, see bonding.txt for details
6225 */
6226 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");
6227 }
6228
6229 if (primary && !bond_mode_uses_primary(mode: bond_mode)) {
6230 /* currently, using a primary only makes sense
6231 * in active backup, TLB or ALB modes
6232 */
6233 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6234 primary, bond_mode_name(bond_mode));
6235 primary = NULL;
6236 }
6237
6238 if (primary && primary_reselect) {
6239 bond_opt_initstr(&newval, primary_reselect);
6240 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_PRIMARY_RESELECT),
6241 val: &newval);
6242 if (!valptr) {
6243 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6244 primary_reselect);
6245 return -EINVAL;
6246 }
6247 primary_reselect_value = valptr->value;
6248 } else {
6249 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6250 }
6251
6252 if (fail_over_mac) {
6253 bond_opt_initstr(&newval, fail_over_mac);
6254 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_FAIL_OVER_MAC),
6255 val: &newval);
6256 if (!valptr) {
6257 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6258 fail_over_mac);
6259 return -EINVAL;
6260 }
6261 fail_over_mac_value = valptr->value;
6262 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6263 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6264 } else {
6265 fail_over_mac_value = BOND_FOM_NONE;
6266 }
6267
6268 bond_opt_initstr(&newval, "default");
6269 valptr = bond_opt_parse(
6270 opt: bond_opt_get(option: BOND_OPT_AD_ACTOR_SYS_PRIO),
6271 val: &newval);
6272 if (!valptr) {
6273 pr_err("Error: No ad_actor_sys_prio default value");
6274 return -EINVAL;
6275 }
6276 ad_actor_sys_prio = valptr->value;
6277
6278 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_AD_USER_PORT_KEY),
6279 val: &newval);
6280 if (!valptr) {
6281 pr_err("Error: No ad_user_port_key default value");
6282 return -EINVAL;
6283 }
6284 ad_user_port_key = valptr->value;
6285
6286 bond_opt_initstr(&newval, "default");
6287 valptr = bond_opt_parse(opt: bond_opt_get(option: BOND_OPT_TLB_DYNAMIC_LB), val: &newval);
6288 if (!valptr) {
6289 pr_err("Error: No tlb_dynamic_lb default value");
6290 return -EINVAL;
6291 }
6292 tlb_dynamic_lb = valptr->value;
6293
6294 if (lp_interval == 0) {
6295 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6296 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6297 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6298 }
6299
6300 /* fill params struct with the proper values */
6301 params->mode = bond_mode;
6302 params->xmit_policy = xmit_hashtype;
6303 params->miimon = miimon;
6304 params->num_peer_notif = num_peer_notif;
6305 params->arp_interval = arp_interval;
6306 params->arp_validate = arp_validate_value;
6307 params->arp_all_targets = arp_all_targets_value;
6308 params->missed_max = 2;
6309 params->updelay = updelay;
6310 params->downdelay = downdelay;
6311 params->peer_notif_delay = 0;
6312 params->use_carrier = use_carrier;
6313 params->lacp_active = 1;
6314 params->lacp_fast = lacp_fast;
6315 params->primary[0] = 0;
6316 params->primary_reselect = primary_reselect_value;
6317 params->fail_over_mac = fail_over_mac_value;
6318 params->tx_queues = tx_queues;
6319 params->all_slaves_active = all_slaves_active;
6320 params->resend_igmp = resend_igmp;
6321 params->min_links = min_links;
6322 params->lp_interval = lp_interval;
6323 params->packets_per_slave = packets_per_slave;
6324 params->tlb_dynamic_lb = tlb_dynamic_lb;
6325 params->ad_actor_sys_prio = ad_actor_sys_prio;
6326 eth_zero_addr(addr: params->ad_actor_system);
6327 params->ad_user_port_key = ad_user_port_key;
6328 if (packets_per_slave > 0) {
6329 params->reciprocal_packets_per_slave =
6330 reciprocal_value(d: packets_per_slave);
6331 } else {
6332 /* reciprocal_packets_per_slave is unused if
6333 * packets_per_slave is 0 or 1, just initialize it
6334 */
6335 params->reciprocal_packets_per_slave =
6336 (struct reciprocal_value) { 0 };
6337 }
6338
6339 if (primary)
6340 strscpy_pad(params->primary, primary, sizeof(params->primary));
6341
6342 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6343#if IS_ENABLED(CONFIG_IPV6)
6344 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6345#endif
6346
6347 return 0;
6348}
6349
6350/* Called from registration process */
6351static int bond_init(struct net_device *bond_dev)
6352{
6353 struct bonding *bond = netdev_priv(dev: bond_dev);
6354 struct bond_net *bn = net_generic(net: dev_net(dev: bond_dev), id: bond_net_id);
6355
6356 netdev_dbg(bond_dev, "Begin bond_init\n");
6357
6358 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6359 if (!bond->wq)
6360 return -ENOMEM;
6361
6362 bond->notifier_ctx = false;
6363
6364 spin_lock_init(&bond->stats_lock);
6365 netdev_lockdep_set_classes(bond_dev);
6366
6367 list_add_tail(new: &bond->bond_list, head: &bn->dev_list);
6368
6369 bond_prepare_sysfs_group(bond);
6370
6371 bond_debug_register(bond);
6372
6373 /* Ensure valid dev_addr */
6374 if (is_zero_ether_addr(addr: bond_dev->dev_addr) &&
6375 bond_dev->addr_assign_type == NET_ADDR_PERM)
6376 eth_hw_addr_random(dev: bond_dev);
6377
6378 return 0;
6379}
6380
6381unsigned int bond_get_num_tx_queues(void)
6382{
6383 return tx_queues;
6384}
6385
6386/* Create a new bond based on the specified name and bonding parameters.
6387 * If name is NULL, obtain a suitable "bond%d" name for us.
6388 * Caller must NOT hold rtnl_lock; we need to release it here before we
6389 * set up our sysfs entries.
6390 */
6391int bond_create(struct net *net, const char *name)
6392{
6393 struct net_device *bond_dev;
6394 struct bonding *bond;
6395 int res = -ENOMEM;
6396
6397 rtnl_lock();
6398
6399 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6400 name ? name : "bond%d", NET_NAME_UNKNOWN,
6401 bond_setup, tx_queues);
6402 if (!bond_dev)
6403 goto out;
6404
6405 bond = netdev_priv(dev: bond_dev);
6406 dev_net_set(dev: bond_dev, net);
6407 bond_dev->rtnl_link_ops = &bond_link_ops;
6408
6409 res = register_netdevice(dev: bond_dev);
6410 if (res < 0) {
6411 free_netdev(dev: bond_dev);
6412 goto out;
6413 }
6414
6415 netif_carrier_off(dev: bond_dev);
6416
6417 bond_work_init_all(bond);
6418
6419out:
6420 rtnl_unlock();
6421 return res;
6422}
6423
6424static int __net_init bond_net_init(struct net *net)
6425{
6426 struct bond_net *bn = net_generic(net, id: bond_net_id);
6427
6428 bn->net = net;
6429 INIT_LIST_HEAD(list: &bn->dev_list);
6430
6431 bond_create_proc_dir(bn);
6432 bond_create_sysfs(net: bn);
6433
6434 return 0;
6435}
6436
6437static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6438{
6439 struct bond_net *bn;
6440 struct net *net;
6441 LIST_HEAD(list);
6442
6443 list_for_each_entry(net, net_list, exit_list) {
6444 bn = net_generic(net, id: bond_net_id);
6445 bond_destroy_sysfs(net: bn);
6446 }
6447
6448 /* Kill off any bonds created after unregistering bond rtnl ops */
6449 rtnl_lock();
6450 list_for_each_entry(net, net_list, exit_list) {
6451 struct bonding *bond, *tmp_bond;
6452
6453 bn = net_generic(net, id: bond_net_id);
6454 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6455 unregister_netdevice_queue(dev: bond->dev, head: &list);
6456 }
6457 unregister_netdevice_many(head: &list);
6458 rtnl_unlock();
6459
6460 list_for_each_entry(net, net_list, exit_list) {
6461 bn = net_generic(net, id: bond_net_id);
6462 bond_destroy_proc_dir(bn);
6463 }
6464}
6465
6466static struct pernet_operations bond_net_ops = {
6467 .init = bond_net_init,
6468 .exit_batch = bond_net_exit_batch,
6469 .id = &bond_net_id,
6470 .size = sizeof(struct bond_net),
6471};
6472
6473static int __init bonding_init(void)
6474{
6475 int i;
6476 int res;
6477
6478 res = bond_check_params(params: &bonding_defaults);
6479 if (res)
6480 goto out;
6481
6482 res = register_pernet_subsys(&bond_net_ops);
6483 if (res)
6484 goto out;
6485
6486 res = bond_netlink_init();
6487 if (res)
6488 goto err_link;
6489
6490 bond_create_debugfs();
6491
6492 for (i = 0; i < max_bonds; i++) {
6493 res = bond_create(net: &init_net, NULL);
6494 if (res)
6495 goto err;
6496 }
6497
6498 skb_flow_dissector_init(flow_dissector: &flow_keys_bonding,
6499 key: flow_keys_bonding_keys,
6500 ARRAY_SIZE(flow_keys_bonding_keys));
6501
6502 register_netdevice_notifier(nb: &bond_netdev_notifier);
6503out:
6504 return res;
6505err:
6506 bond_destroy_debugfs();
6507 bond_netlink_fini();
6508err_link:
6509 unregister_pernet_subsys(&bond_net_ops);
6510 goto out;
6511
6512}
6513
6514static void __exit bonding_exit(void)
6515{
6516 unregister_netdevice_notifier(nb: &bond_netdev_notifier);
6517
6518 bond_destroy_debugfs();
6519
6520 bond_netlink_fini();
6521 unregister_pernet_subsys(&bond_net_ops);
6522
6523#ifdef CONFIG_NET_POLL_CONTROLLER
6524 /* Make sure we don't have an imbalance on our netpoll blocking */
6525 WARN_ON(atomic_read(&netpoll_block_tx));
6526#endif
6527}
6528
6529module_init(bonding_init);
6530module_exit(bonding_exit);
6531MODULE_LICENSE("GPL");
6532MODULE_DESCRIPTION(DRV_DESCRIPTION);
6533MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
6534

source code of linux/drivers/net/bonding/bond_main.c