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