Loading...
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/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
42#include <net/ip.h>
43#include <linux/ip.h>
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
54#include <linux/io.h>
55#include <asm/dma.h>
56#include <linux/uaccess.h>
57#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
60#include <linux/igmp.h>
61#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
65#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
72#include <linux/jiffies.h>
73#include <linux/preempt.h>
74#include <net/route.h>
75#include <net/net_namespace.h>
76#include <net/netns/generic.h>
77#include <net/pkt_sched.h>
78#include <linux/rculist.h>
79#include <net/flow_dissector.h>
80#include <net/switchdev.h>
81#include <net/bonding.h>
82#include <net/bond_3ad.h>
83#include <net/bond_alb.h>
84
85#include "bonding_priv.h"
86
87/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
90
91static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93static int num_peer_notif = 1;
94static int miimon;
95static int updelay;
96static int downdelay;
97static int use_carrier = 1;
98static char *mode;
99static char *primary;
100static char *primary_reselect;
101static char *lacp_rate;
102static int min_links;
103static char *ad_select;
104static char *xmit_hash_policy;
105static int arp_interval;
106static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107static char *arp_validate;
108static char *arp_all_targets;
109static char *fail_over_mac;
110static int all_slaves_active;
111static struct bond_params bonding_defaults;
112static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113static int packets_per_slave = 1;
114static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115
116module_param(max_bonds, int, 0);
117MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118module_param(tx_queues, int, 0);
119MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126module_param(miimon, int, 0);
127MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128module_param(updelay, int, 0);
129MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130module_param(downdelay, int, 0);
131MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
133module_param(use_carrier, int, 0);
134MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136module_param(mode, charp, 0);
137MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141module_param(primary, charp, 0);
142MODULE_PARM_DESC(primary, "Primary network device to use");
143module_param(primary_reselect, charp, 0);
144MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "once it comes up; "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
148 "better, "
149 "2 for only on active slave "
150 "failure");
151module_param(lacp_rate, charp, 0);
152MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154module_param(ad_select, charp, 0);
155MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
157 "2 for count");
158module_param(min_links, int, 0);
159MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160
161module_param(xmit_hash_policy, charp, 0);
162MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166module_param(arp_interval, int, 0);
167MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168module_param_array(arp_ip_target, charp, NULL, 0);
169MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170module_param(arp_validate, charp, 0);
171MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174module_param(arp_all_targets, charp, 0);
175MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176module_param(fail_over_mac, charp, 0);
177MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180module_param(all_slaves_active, int, 0);
181MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184module_param(resend_igmp, int, 0);
185MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
187module_param(packets_per_slave, int, 0);
188MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191module_param(lp_interval, uint, 0);
192MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
195
196/*----------------------------- Global variables ----------------------------*/
197
198#ifdef CONFIG_NET_POLL_CONTROLLER
199atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200#endif
201
202unsigned int bond_net_id __read_mostly;
203
204/*-------------------------- Forward declarations ---------------------------*/
205
206static int bond_init(struct net_device *bond_dev);
207static void bond_uninit(struct net_device *bond_dev);
208static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210static void bond_slave_arr_handler(struct work_struct *work);
211static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 int mod);
213
214/*---------------------------- General routines -----------------------------*/
215
216const char *bond_mode_name(int mode)
217{
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
226 };
227
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
229 return "unknown";
230
231 return names[mode];
232}
233
234/*---------------------------------- VLAN -----------------------------------*/
235
236/**
237 * bond_dev_queue_xmit - Prepare skb for xmit.
238 *
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
242 */
243void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
245{
246 skb->dev = slave_dev;
247
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
251
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
254 else
255 dev_queue_xmit(skb);
256}
257
258/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259 * We don't protect the slave list iteration with a lock because:
260 * a. This operation is performed in IOCTL context,
261 * b. The operation is protected by the RTNL semaphore in the 8021q code,
262 * c. Holding a lock with BH disabled while directly calling a base driver
263 * entry point is generally a BAD idea.
264 *
265 * The design of synchronization/protection for this operation in the 8021q
266 * module is good for one or more VLAN devices over a single physical device
267 * and cannot be extended for a teaming solution like bonding, so there is a
268 * potential race condition here where a net device from the vlan group might
269 * be referenced (either by a base driver or the 8021q code) while it is being
270 * removed from the system. However, it turns out we're not making matters
271 * worse, and if it works for regular VLAN usage it will work here too.
272*/
273
274/**
275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276 * @bond_dev: bonding net device that got called
277 * @vid: vlan id being added
278 */
279static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 __be16 proto, u16 vid)
281{
282 struct bonding *bond = netdev_priv(bond_dev);
283 struct slave *slave, *rollback_slave;
284 struct list_head *iter;
285 int res;
286
287 bond_for_each_slave(bond, slave, iter) {
288 res = vlan_vid_add(slave->dev, proto, vid);
289 if (res)
290 goto unwind;
291 }
292
293 return 0;
294
295unwind:
296 /* unwind to the slave that failed */
297 bond_for_each_slave(bond, rollback_slave, iter) {
298 if (rollback_slave == slave)
299 break;
300
301 vlan_vid_del(rollback_slave->dev, proto, vid);
302 }
303
304 return res;
305}
306
307/**
308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309 * @bond_dev: bonding net device that got called
310 * @vid: vlan id being removed
311 */
312static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 __be16 proto, u16 vid)
314{
315 struct bonding *bond = netdev_priv(bond_dev);
316 struct list_head *iter;
317 struct slave *slave;
318
319 bond_for_each_slave(bond, slave, iter)
320 vlan_vid_del(slave->dev, proto, vid);
321
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vid);
324
325 return 0;
326}
327
328/*------------------------------- Link status -------------------------------*/
329
330/* Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
333 *
334 * Returns zero if carrier state does not change, nonzero if it does.
335 */
336int bond_set_carrier(struct bonding *bond)
337{
338 struct list_head *iter;
339 struct slave *slave;
340
341 if (!bond_has_slaves(bond))
342 goto down;
343
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
346
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
351 return 1;
352 }
353 return 0;
354 }
355 }
356
357down:
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
360 return 1;
361 }
362 return 0;
363}
364
365/* Get link speed and duplex from the slave's base driver
366 * using ethtool. If for some reason the call fails or the
367 * values are invalid, set speed and duplex to -1,
368 * and return. Return 1 if speed or duplex settings are
369 * UNKNOWN; 0 otherwise.
370 */
371static int bond_update_speed_duplex(struct slave *slave)
372{
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_link_ksettings ecmd;
375 int res;
376
377 slave->speed = SPEED_UNKNOWN;
378 slave->duplex = DUPLEX_UNKNOWN;
379
380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
381 if (res < 0)
382 return 1;
383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
384 return 1;
385 switch (ecmd.base.duplex) {
386 case DUPLEX_FULL:
387 case DUPLEX_HALF:
388 break;
389 default:
390 return 1;
391 }
392
393 slave->speed = ecmd.base.speed;
394 slave->duplex = ecmd.base.duplex;
395
396 return 0;
397}
398
399const char *bond_slave_link_status(s8 link)
400{
401 switch (link) {
402 case BOND_LINK_UP:
403 return "up";
404 case BOND_LINK_FAIL:
405 return "going down";
406 case BOND_LINK_DOWN:
407 return "down";
408 case BOND_LINK_BACK:
409 return "going back";
410 default:
411 return "unknown";
412 }
413}
414
415/* if <dev> supports MII link status reporting, check its link status.
416 *
417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418 * depending upon the setting of the use_carrier parameter.
419 *
420 * Return either BMSR_LSTATUS, meaning that the link is up (or we
421 * can't tell and just pretend it is), or 0, meaning that the link is
422 * down.
423 *
424 * If reporting is non-zero, instead of faking link up, return -1 if
425 * both ETHTOOL and MII ioctls fail (meaning the device does not
426 * support them). If use_carrier is set, return whatever it says.
427 * It'd be nice if there was a good way to tell if a driver supports
428 * netif_carrier, but there really isn't.
429 */
430static int bond_check_dev_link(struct bonding *bond,
431 struct net_device *slave_dev, int reporting)
432{
433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 int (*ioctl)(struct net_device *, struct ifreq *, int);
435 struct ifreq ifr;
436 struct mii_ioctl_data *mii;
437
438 if (!reporting && !netif_running(slave_dev))
439 return 0;
440
441 if (bond->params.use_carrier)
442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
443
444 /* Try to get link status using Ethtool first. */
445 if (slave_dev->ethtool_ops->get_link)
446 return slave_dev->ethtool_ops->get_link(slave_dev) ?
447 BMSR_LSTATUS : 0;
448
449 /* Ethtool can't be used, fallback to MII ioctls. */
450 ioctl = slave_ops->ndo_do_ioctl;
451 if (ioctl) {
452 /* TODO: set pointer to correct ioctl on a per team member
453 * bases to make this more efficient. that is, once
454 * we determine the correct ioctl, we will always
455 * call it and not the others for that team
456 * member.
457 */
458
459 /* We cannot assume that SIOCGMIIPHY will also read a
460 * register; not all network drivers (e.g., e100)
461 * support that.
462 */
463
464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
466 mii = if_mii(&ifr);
467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 mii->reg_num = MII_BMSR;
469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 return mii->val_out & BMSR_LSTATUS;
471 }
472 }
473
474 /* If reporting, report that either there's no dev->do_ioctl,
475 * or both SIOCGMIIREG and get_link failed (meaning that we
476 * cannot report link status). If not reporting, pretend
477 * we're ok.
478 */
479 return reporting ? -1 : BMSR_LSTATUS;
480}
481
482/*----------------------------- Multicast list ------------------------------*/
483
484/* Push the promiscuity flag down to appropriate slaves */
485static int bond_set_promiscuity(struct bonding *bond, int inc)
486{
487 struct list_head *iter;
488 int err = 0;
489
490 if (bond_uses_primary(bond)) {
491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
492
493 if (curr_active)
494 err = dev_set_promiscuity(curr_active->dev, inc);
495 } else {
496 struct slave *slave;
497
498 bond_for_each_slave(bond, slave, iter) {
499 err = dev_set_promiscuity(slave->dev, inc);
500 if (err)
501 return err;
502 }
503 }
504 return err;
505}
506
507/* Push the allmulti flag down to all slaves */
508static int bond_set_allmulti(struct bonding *bond, int inc)
509{
510 struct list_head *iter;
511 int err = 0;
512
513 if (bond_uses_primary(bond)) {
514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
515
516 if (curr_active)
517 err = dev_set_allmulti(curr_active->dev, inc);
518 } else {
519 struct slave *slave;
520
521 bond_for_each_slave(bond, slave, iter) {
522 err = dev_set_allmulti(slave->dev, inc);
523 if (err)
524 return err;
525 }
526 }
527 return err;
528}
529
530/* Retrieve the list of registered multicast addresses for the bonding
531 * device and retransmit an IGMP JOIN request to the current active
532 * slave.
533 */
534static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
535{
536 struct bonding *bond = container_of(work, struct bonding,
537 mcast_work.work);
538
539 if (!rtnl_trylock()) {
540 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
541 return;
542 }
543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
544
545 if (bond->igmp_retrans > 1) {
546 bond->igmp_retrans--;
547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
548 }
549 rtnl_unlock();
550}
551
552/* Flush bond's hardware addresses from slave */
553static void bond_hw_addr_flush(struct net_device *bond_dev,
554 struct net_device *slave_dev)
555{
556 struct bonding *bond = netdev_priv(bond_dev);
557
558 dev_uc_unsync(slave_dev, bond_dev);
559 dev_mc_unsync(slave_dev, bond_dev);
560
561 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 /* del lacpdu mc addr from mc list */
563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
564
565 dev_mc_del(slave_dev, lacpdu_multicast);
566 }
567}
568
569/*--------------------------- Active slave change ---------------------------*/
570
571/* Update the hardware address list and promisc/allmulti for the new and
572 * old active slaves (if any). Modes that are not using primary keep all
573 * slaves up date at all times; only the modes that use primary need to call
574 * this function to swap these settings during a failover.
575 */
576static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 struct slave *old_active)
578{
579 if (old_active) {
580 if (bond->dev->flags & IFF_PROMISC)
581 dev_set_promiscuity(old_active->dev, -1);
582
583 if (bond->dev->flags & IFF_ALLMULTI)
584 dev_set_allmulti(old_active->dev, -1);
585
586 bond_hw_addr_flush(bond->dev, old_active->dev);
587 }
588
589 if (new_active) {
590 /* FIXME: Signal errors upstream. */
591 if (bond->dev->flags & IFF_PROMISC)
592 dev_set_promiscuity(new_active->dev, 1);
593
594 if (bond->dev->flags & IFF_ALLMULTI)
595 dev_set_allmulti(new_active->dev, 1);
596
597 netif_addr_lock_bh(bond->dev);
598 dev_uc_sync(new_active->dev, bond->dev);
599 dev_mc_sync(new_active->dev, bond->dev);
600 netif_addr_unlock_bh(bond->dev);
601 }
602}
603
604/**
605 * bond_set_dev_addr - clone slave's address to bond
606 * @bond_dev: bond net device
607 * @slave_dev: slave net device
608 *
609 * Should be called with RTNL held.
610 */
611static void bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
613{
614 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
615 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
616 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
617 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
618 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
619}
620
621static struct slave *bond_get_old_active(struct bonding *bond,
622 struct slave *new_active)
623{
624 struct slave *slave;
625 struct list_head *iter;
626
627 bond_for_each_slave(bond, slave, iter) {
628 if (slave == new_active)
629 continue;
630
631 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
632 return slave;
633 }
634
635 return NULL;
636}
637
638/* bond_do_fail_over_mac
639 *
640 * Perform special MAC address swapping for fail_over_mac settings
641 *
642 * Called with RTNL
643 */
644static void bond_do_fail_over_mac(struct bonding *bond,
645 struct slave *new_active,
646 struct slave *old_active)
647{
648 u8 tmp_mac[MAX_ADDR_LEN];
649 struct sockaddr_storage ss;
650 int rv;
651
652 switch (bond->params.fail_over_mac) {
653 case BOND_FOM_ACTIVE:
654 if (new_active)
655 bond_set_dev_addr(bond->dev, new_active->dev);
656 break;
657 case BOND_FOM_FOLLOW:
658 /* if new_active && old_active, swap them
659 * if just old_active, do nothing (going to no active slave)
660 * if just new_active, set new_active to bond's MAC
661 */
662 if (!new_active)
663 return;
664
665 if (!old_active)
666 old_active = bond_get_old_active(bond, new_active);
667
668 if (old_active) {
669 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
670 new_active->dev->addr_len);
671 bond_hw_addr_copy(ss.__data,
672 old_active->dev->dev_addr,
673 old_active->dev->addr_len);
674 ss.ss_family = new_active->dev->type;
675 } else {
676 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
677 bond->dev->addr_len);
678 ss.ss_family = bond->dev->type;
679 }
680
681 rv = dev_set_mac_address(new_active->dev,
682 (struct sockaddr *)&ss);
683 if (rv) {
684 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
685 -rv, new_active->dev->name);
686 goto out;
687 }
688
689 if (!old_active)
690 goto out;
691
692 bond_hw_addr_copy(ss.__data, tmp_mac,
693 new_active->dev->addr_len);
694 ss.ss_family = old_active->dev->type;
695
696 rv = dev_set_mac_address(old_active->dev,
697 (struct sockaddr *)&ss);
698 if (rv)
699 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
700 -rv, new_active->dev->name);
701out:
702 break;
703 default:
704 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
705 bond->params.fail_over_mac);
706 break;
707 }
708
709}
710
711static struct slave *bond_choose_primary_or_current(struct bonding *bond)
712{
713 struct slave *prim = rtnl_dereference(bond->primary_slave);
714 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
715
716 if (!prim || prim->link != BOND_LINK_UP) {
717 if (!curr || curr->link != BOND_LINK_UP)
718 return NULL;
719 return curr;
720 }
721
722 if (bond->force_primary) {
723 bond->force_primary = false;
724 return prim;
725 }
726
727 if (!curr || curr->link != BOND_LINK_UP)
728 return prim;
729
730 /* At this point, prim and curr are both up */
731 switch (bond->params.primary_reselect) {
732 case BOND_PRI_RESELECT_ALWAYS:
733 return prim;
734 case BOND_PRI_RESELECT_BETTER:
735 if (prim->speed < curr->speed)
736 return curr;
737 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
738 return curr;
739 return prim;
740 case BOND_PRI_RESELECT_FAILURE:
741 return curr;
742 default:
743 netdev_err(bond->dev, "impossible primary_reselect %d\n",
744 bond->params.primary_reselect);
745 return curr;
746 }
747}
748
749/**
750 * bond_find_best_slave - select the best available slave to be the active one
751 * @bond: our bonding struct
752 */
753static struct slave *bond_find_best_slave(struct bonding *bond)
754{
755 struct slave *slave, *bestslave = NULL;
756 struct list_head *iter;
757 int mintime = bond->params.updelay;
758
759 slave = bond_choose_primary_or_current(bond);
760 if (slave)
761 return slave;
762
763 bond_for_each_slave(bond, slave, iter) {
764 if (slave->link == BOND_LINK_UP)
765 return slave;
766 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
767 slave->delay < mintime) {
768 mintime = slave->delay;
769 bestslave = slave;
770 }
771 }
772
773 return bestslave;
774}
775
776static bool bond_should_notify_peers(struct bonding *bond)
777{
778 struct slave *slave;
779
780 rcu_read_lock();
781 slave = rcu_dereference(bond->curr_active_slave);
782 rcu_read_unlock();
783
784 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
785 slave ? slave->dev->name : "NULL");
786
787 if (!slave || !bond->send_peer_notif ||
788 !netif_carrier_ok(bond->dev) ||
789 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
790 return false;
791
792 return true;
793}
794
795/**
796 * change_active_interface - change the active slave into the specified one
797 * @bond: our bonding struct
798 * @new: the new slave to make the active one
799 *
800 * Set the new slave to the bond's settings and unset them on the old
801 * curr_active_slave.
802 * Setting include flags, mc-list, promiscuity, allmulti, etc.
803 *
804 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
805 * because it is apparently the best available slave we have, even though its
806 * updelay hasn't timed out yet.
807 *
808 * Caller must hold RTNL.
809 */
810void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
811{
812 struct slave *old_active;
813
814 ASSERT_RTNL();
815
816 old_active = rtnl_dereference(bond->curr_active_slave);
817
818 if (old_active == new_active)
819 return;
820
821 if (new_active) {
822 new_active->last_link_up = jiffies;
823
824 if (new_active->link == BOND_LINK_BACK) {
825 if (bond_uses_primary(bond)) {
826 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
827 new_active->dev->name,
828 (bond->params.updelay - new_active->delay) * bond->params.miimon);
829 }
830
831 new_active->delay = 0;
832 bond_set_slave_link_state(new_active, BOND_LINK_UP,
833 BOND_SLAVE_NOTIFY_NOW);
834
835 if (BOND_MODE(bond) == BOND_MODE_8023AD)
836 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
837
838 if (bond_is_lb(bond))
839 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
840 } else {
841 if (bond_uses_primary(bond)) {
842 netdev_info(bond->dev, "making interface %s the new active one\n",
843 new_active->dev->name);
844 }
845 }
846 }
847
848 if (bond_uses_primary(bond))
849 bond_hw_addr_swap(bond, new_active, old_active);
850
851 if (bond_is_lb(bond)) {
852 bond_alb_handle_active_change(bond, new_active);
853 if (old_active)
854 bond_set_slave_inactive_flags(old_active,
855 BOND_SLAVE_NOTIFY_NOW);
856 if (new_active)
857 bond_set_slave_active_flags(new_active,
858 BOND_SLAVE_NOTIFY_NOW);
859 } else {
860 rcu_assign_pointer(bond->curr_active_slave, new_active);
861 }
862
863 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
864 if (old_active)
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
867
868 if (new_active) {
869 bool should_notify_peers = false;
870
871 bond_set_slave_active_flags(new_active,
872 BOND_SLAVE_NOTIFY_NOW);
873
874 if (bond->params.fail_over_mac)
875 bond_do_fail_over_mac(bond, new_active,
876 old_active);
877
878 if (netif_running(bond->dev)) {
879 bond->send_peer_notif =
880 bond->params.num_peer_notif;
881 should_notify_peers =
882 bond_should_notify_peers(bond);
883 }
884
885 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
886 if (should_notify_peers)
887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
888 bond->dev);
889 }
890 }
891
892 /* resend IGMP joins since active slave has changed or
893 * all were sent on curr_active_slave.
894 * resend only if bond is brought up with the affected
895 * bonding modes and the retransmission is enabled
896 */
897 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
898 ((bond_uses_primary(bond) && new_active) ||
899 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
900 bond->igmp_retrans = bond->params.resend_igmp;
901 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
902 }
903}
904
905/**
906 * bond_select_active_slave - select a new active slave, if needed
907 * @bond: our bonding struct
908 *
909 * This functions should be called when one of the following occurs:
910 * - The old curr_active_slave has been released or lost its link.
911 * - The primary_slave has got its link back.
912 * - A slave has got its link back and there's no old curr_active_slave.
913 *
914 * Caller must hold RTNL.
915 */
916void bond_select_active_slave(struct bonding *bond)
917{
918 struct slave *best_slave;
919 int rv;
920
921 ASSERT_RTNL();
922
923 best_slave = bond_find_best_slave(bond);
924 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
925 bond_change_active_slave(bond, best_slave);
926 rv = bond_set_carrier(bond);
927 if (!rv)
928 return;
929
930 if (netif_carrier_ok(bond->dev))
931 netdev_info(bond->dev, "first active interface up!\n");
932 else
933 netdev_info(bond->dev, "now running without any active interface!\n");
934 }
935}
936
937#ifdef CONFIG_NET_POLL_CONTROLLER
938static inline int slave_enable_netpoll(struct slave *slave)
939{
940 struct netpoll *np;
941 int err = 0;
942
943 np = kzalloc(sizeof(*np), GFP_KERNEL);
944 err = -ENOMEM;
945 if (!np)
946 goto out;
947
948 err = __netpoll_setup(np, slave->dev);
949 if (err) {
950 kfree(np);
951 goto out;
952 }
953 slave->np = np;
954out:
955 return err;
956}
957static inline void slave_disable_netpoll(struct slave *slave)
958{
959 struct netpoll *np = slave->np;
960
961 if (!np)
962 return;
963
964 slave->np = NULL;
965 __netpoll_free_async(np);
966}
967
968static void bond_poll_controller(struct net_device *bond_dev)
969{
970 struct bonding *bond = netdev_priv(bond_dev);
971 struct slave *slave = NULL;
972 struct list_head *iter;
973 struct ad_info ad_info;
974 struct netpoll_info *ni;
975 const struct net_device_ops *ops;
976
977 if (BOND_MODE(bond) == BOND_MODE_8023AD)
978 if (bond_3ad_get_active_agg_info(bond, &ad_info))
979 return;
980
981 bond_for_each_slave_rcu(bond, slave, iter) {
982 ops = slave->dev->netdev_ops;
983 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
984 continue;
985
986 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
987 struct aggregator *agg =
988 SLAVE_AD_INFO(slave)->port.aggregator;
989
990 if (agg &&
991 agg->aggregator_identifier != ad_info.aggregator_id)
992 continue;
993 }
994
995 ni = rcu_dereference_bh(slave->dev->npinfo);
996 if (down_trylock(&ni->dev_lock))
997 continue;
998 ops->ndo_poll_controller(slave->dev);
999 up(&ni->dev_lock);
1000 }
1001}
1002
1003static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004{
1005 struct bonding *bond = netdev_priv(bond_dev);
1006 struct list_head *iter;
1007 struct slave *slave;
1008
1009 bond_for_each_slave(bond, slave, iter)
1010 if (bond_slave_is_up(slave))
1011 slave_disable_netpoll(slave);
1012}
1013
1014static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1015{
1016 struct bonding *bond = netdev_priv(dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1019 int err = 0;
1020
1021 bond_for_each_slave(bond, slave, iter) {
1022 err = slave_enable_netpoll(slave);
1023 if (err) {
1024 bond_netpoll_cleanup(dev);
1025 break;
1026 }
1027 }
1028 return err;
1029}
1030#else
1031static inline int slave_enable_netpoll(struct slave *slave)
1032{
1033 return 0;
1034}
1035static inline void slave_disable_netpoll(struct slave *slave)
1036{
1037}
1038static void bond_netpoll_cleanup(struct net_device *bond_dev)
1039{
1040}
1041#endif
1042
1043/*---------------------------------- IOCTL ----------------------------------*/
1044
1045static netdev_features_t bond_fix_features(struct net_device *dev,
1046 netdev_features_t features)
1047{
1048 struct bonding *bond = netdev_priv(dev);
1049 struct list_head *iter;
1050 netdev_features_t mask;
1051 struct slave *slave;
1052
1053 mask = features;
1054
1055 features &= ~NETIF_F_ONE_FOR_ALL;
1056 features |= NETIF_F_ALL_FOR_ALL;
1057
1058 bond_for_each_slave(bond, slave, iter) {
1059 features = netdev_increment_features(features,
1060 slave->dev->features,
1061 mask);
1062 }
1063 features = netdev_add_tso_features(features, mask);
1064
1065 return features;
1066}
1067
1068#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1069 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1070 NETIF_F_HIGHDMA | NETIF_F_LRO)
1071
1072#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1074
1075static void bond_compute_features(struct bonding *bond)
1076{
1077 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1078 IFF_XMIT_DST_RELEASE_PERM;
1079 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1080 netdev_features_t enc_features = BOND_ENC_FEATURES;
1081 struct net_device *bond_dev = bond->dev;
1082 struct list_head *iter;
1083 struct slave *slave;
1084 unsigned short max_hard_header_len = ETH_HLEN;
1085 unsigned int gso_max_size = GSO_MAX_SIZE;
1086 u16 gso_max_segs = GSO_MAX_SEGS;
1087
1088 if (!bond_has_slaves(bond))
1089 goto done;
1090 vlan_features &= NETIF_F_ALL_FOR_ALL;
1091
1092 bond_for_each_slave(bond, slave, iter) {
1093 vlan_features = netdev_increment_features(vlan_features,
1094 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1095
1096 enc_features = netdev_increment_features(enc_features,
1097 slave->dev->hw_enc_features,
1098 BOND_ENC_FEATURES);
1099 dst_release_flag &= slave->dev->priv_flags;
1100 if (slave->dev->hard_header_len > max_hard_header_len)
1101 max_hard_header_len = slave->dev->hard_header_len;
1102
1103 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1104 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1105 }
1106 bond_dev->hard_header_len = max_hard_header_len;
1107
1108done:
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 bond_dev->gso_max_segs = gso_max_segs;
1112 netif_set_gso_max_size(bond_dev, gso_max_size);
1113
1114 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1115 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1116 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1117 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1118
1119 netdev_change_features(bond_dev);
1120}
1121
1122static void bond_setup_by_slave(struct net_device *bond_dev,
1123 struct net_device *slave_dev)
1124{
1125 bond_dev->header_ops = slave_dev->header_ops;
1126
1127 bond_dev->type = slave_dev->type;
1128 bond_dev->hard_header_len = slave_dev->hard_header_len;
1129 bond_dev->addr_len = slave_dev->addr_len;
1130
1131 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1132 slave_dev->addr_len);
1133}
1134
1135/* On bonding slaves other than the currently active slave, suppress
1136 * duplicates except for alb non-mcast/bcast.
1137 */
1138static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1139 struct slave *slave,
1140 struct bonding *bond)
1141{
1142 if (bond_is_slave_inactive(slave)) {
1143 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1144 skb->pkt_type != PACKET_BROADCAST &&
1145 skb->pkt_type != PACKET_MULTICAST)
1146 return false;
1147 return true;
1148 }
1149 return false;
1150}
1151
1152static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1153{
1154 struct sk_buff *skb = *pskb;
1155 struct slave *slave;
1156 struct bonding *bond;
1157 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1158 struct slave *);
1159 int ret = RX_HANDLER_ANOTHER;
1160
1161 skb = skb_share_check(skb, GFP_ATOMIC);
1162 if (unlikely(!skb))
1163 return RX_HANDLER_CONSUMED;
1164
1165 *pskb = skb;
1166
1167 slave = bond_slave_get_rcu(skb->dev);
1168 bond = slave->bond;
1169
1170 recv_probe = READ_ONCE(bond->recv_probe);
1171 if (recv_probe) {
1172 ret = recv_probe(skb, bond, slave);
1173 if (ret == RX_HANDLER_CONSUMED) {
1174 consume_skb(skb);
1175 return ret;
1176 }
1177 }
1178
1179 /* don't change skb->dev for link-local packets */
1180 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1181 return RX_HANDLER_PASS;
1182 if (bond_should_deliver_exact_match(skb, slave, bond))
1183 return RX_HANDLER_EXACT;
1184
1185 skb->dev = bond->dev;
1186
1187 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189 skb->pkt_type == PACKET_HOST) {
1190
1191 if (unlikely(skb_cow_head(skb,
1192 skb->data - skb_mac_header(skb)))) {
1193 kfree_skb(skb);
1194 return RX_HANDLER_CONSUMED;
1195 }
1196 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1197 bond->dev->addr_len);
1198 }
1199
1200 return ret;
1201}
1202
1203static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1204{
1205 switch (BOND_MODE(bond)) {
1206 case BOND_MODE_ROUNDROBIN:
1207 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1208 case BOND_MODE_ACTIVEBACKUP:
1209 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1210 case BOND_MODE_BROADCAST:
1211 return NETDEV_LAG_TX_TYPE_BROADCAST;
1212 case BOND_MODE_XOR:
1213 case BOND_MODE_8023AD:
1214 return NETDEV_LAG_TX_TYPE_HASH;
1215 default:
1216 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1217 }
1218}
1219
1220static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1221 struct netlink_ext_ack *extack)
1222{
1223 struct netdev_lag_upper_info lag_upper_info;
1224
1225 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1226
1227 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1228 &lag_upper_info, extack);
1229}
1230
1231static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232{
1233 netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 slave->dev->flags &= ~IFF_SLAVE;
1235}
1236
1237static struct slave *bond_alloc_slave(struct bonding *bond)
1238{
1239 struct slave *slave = NULL;
1240
1241 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1242 if (!slave)
1243 return NULL;
1244
1245 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1246 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1247 GFP_KERNEL);
1248 if (!SLAVE_AD_INFO(slave)) {
1249 kfree(slave);
1250 return NULL;
1251 }
1252 }
1253 return slave;
1254}
1255
1256static void bond_free_slave(struct slave *slave)
1257{
1258 struct bonding *bond = bond_get_bond_by_slave(slave);
1259
1260 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1261 kfree(SLAVE_AD_INFO(slave));
1262
1263 kfree(slave);
1264}
1265
1266static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1267{
1268 info->bond_mode = BOND_MODE(bond);
1269 info->miimon = bond->params.miimon;
1270 info->num_slaves = bond->slave_cnt;
1271}
1272
1273static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1274{
1275 strcpy(info->slave_name, slave->dev->name);
1276 info->link = slave->link;
1277 info->state = bond_slave_state(slave);
1278 info->link_failure_count = slave->link_failure_count;
1279}
1280
1281static void bond_netdev_notify(struct net_device *dev,
1282 struct netdev_bonding_info *info)
1283{
1284 rtnl_lock();
1285 netdev_bonding_info_change(dev, info);
1286 rtnl_unlock();
1287}
1288
1289static void bond_netdev_notify_work(struct work_struct *_work)
1290{
1291 struct netdev_notify_work *w =
1292 container_of(_work, struct netdev_notify_work, work.work);
1293
1294 bond_netdev_notify(w->dev, &w->bonding_info);
1295 dev_put(w->dev);
1296 kfree(w);
1297}
1298
1299void bond_queue_slave_event(struct slave *slave)
1300{
1301 struct bonding *bond = slave->bond;
1302 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1303
1304 if (!nnw)
1305 return;
1306
1307 dev_hold(slave->dev);
1308 nnw->dev = slave->dev;
1309 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1310 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1311 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1312
1313 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1314}
1315
1316void bond_lower_state_changed(struct slave *slave)
1317{
1318 struct netdev_lag_lower_state_info info;
1319
1320 info.link_up = slave->link == BOND_LINK_UP ||
1321 slave->link == BOND_LINK_FAIL;
1322 info.tx_enabled = bond_is_active_slave(slave);
1323 netdev_lower_state_changed(slave->dev, &info);
1324}
1325
1326/* enslave device <slave> to bond device <master> */
1327int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1328 struct netlink_ext_ack *extack)
1329{
1330 struct bonding *bond = netdev_priv(bond_dev);
1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 struct slave *new_slave = NULL, *prev_slave;
1333 struct sockaddr_storage ss;
1334 int link_reporting;
1335 int res = 0, i;
1336
1337 if (!bond->params.use_carrier &&
1338 slave_dev->ethtool_ops->get_link == NULL &&
1339 slave_ops->ndo_do_ioctl == NULL) {
1340 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341 slave_dev->name);
1342 }
1343
1344 /* already in-use? */
1345 if (netdev_is_rx_handler_busy(slave_dev)) {
1346 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1347 netdev_err(bond_dev,
1348 "Error: Device is in use and cannot be enslaved\n");
1349 return -EBUSY;
1350 }
1351
1352 if (bond_dev == slave_dev) {
1353 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1354 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1355 return -EPERM;
1356 }
1357
1358 /* vlan challenged mutual exclusion */
1359 /* no need to lock since we're protected by rtnl_lock */
1360 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1361 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1362 slave_dev->name);
1363 if (vlan_uses_dev(bond_dev)) {
1364 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1365 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366 slave_dev->name, bond_dev->name);
1367 return -EPERM;
1368 } else {
1369 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370 slave_dev->name, slave_dev->name,
1371 bond_dev->name);
1372 }
1373 } else {
1374 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1375 slave_dev->name);
1376 }
1377
1378 /* Old ifenslave binaries are no longer supported. These can
1379 * be identified with moderate accuracy by the state of the slave:
1380 * the current ifenslave will set the interface down prior to
1381 * enslaving it; the old ifenslave will not.
1382 */
1383 if (slave_dev->flags & IFF_UP) {
1384 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1385 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1386 slave_dev->name);
1387 return -EPERM;
1388 }
1389
1390 /* set bonding device ether type by slave - bonding netdevices are
1391 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1392 * there is a need to override some of the type dependent attribs/funcs.
1393 *
1394 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1395 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396 */
1397 if (!bond_has_slaves(bond)) {
1398 if (bond_dev->type != slave_dev->type) {
1399 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1400 bond_dev->type, slave_dev->type);
1401
1402 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403 bond_dev);
1404 res = notifier_to_errno(res);
1405 if (res) {
1406 netdev_err(bond_dev, "refused to change device type\n");
1407 return -EBUSY;
1408 }
1409
1410 /* Flush unicast and multicast addresses */
1411 dev_uc_flush(bond_dev);
1412 dev_mc_flush(bond_dev);
1413
1414 if (slave_dev->type != ARPHRD_ETHER)
1415 bond_setup_by_slave(bond_dev, slave_dev);
1416 else {
1417 ether_setup(bond_dev);
1418 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 }
1420
1421 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1422 bond_dev);
1423 }
1424 } else if (bond_dev->type != slave_dev->type) {
1425 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1426 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1427 slave_dev->name, slave_dev->type, bond_dev->type);
1428 return -EINVAL;
1429 }
1430
1431 if (slave_dev->type == ARPHRD_INFINIBAND &&
1432 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1433 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1434 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1435 slave_dev->type);
1436 res = -EOPNOTSUPP;
1437 goto err_undo_flags;
1438 }
1439
1440 if (!slave_ops->ndo_set_mac_address ||
1441 slave_dev->type == ARPHRD_INFINIBAND) {
1442 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1443 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1444 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1445 if (!bond_has_slaves(bond)) {
1446 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1447 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1448 } else {
1449 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1450 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1451 res = -EOPNOTSUPP;
1452 goto err_undo_flags;
1453 }
1454 }
1455 }
1456
1457 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1458
1459 /* If this is the first slave, then we need to set the master's hardware
1460 * address to be the same as the slave's.
1461 */
1462 if (!bond_has_slaves(bond) &&
1463 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1464 bond_set_dev_addr(bond->dev, slave_dev);
1465
1466 new_slave = bond_alloc_slave(bond);
1467 if (!new_slave) {
1468 res = -ENOMEM;
1469 goto err_undo_flags;
1470 }
1471
1472 new_slave->bond = bond;
1473 new_slave->dev = slave_dev;
1474 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1475 * is set via sysfs or module option if desired.
1476 */
1477 new_slave->queue_id = 0;
1478
1479 /* Save slave's original mtu and then set it to match the bond */
1480 new_slave->original_mtu = slave_dev->mtu;
1481 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1482 if (res) {
1483 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1484 goto err_free;
1485 }
1486
1487 /* Save slave's original ("permanent") mac address for modes
1488 * that need it, and for restoring it upon release, and then
1489 * set it to the master's address
1490 */
1491 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1492 slave_dev->addr_len);
1493
1494 if (!bond->params.fail_over_mac ||
1495 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1496 /* Set slave to master's mac address. The application already
1497 * set the master's mac address to that of the first slave
1498 */
1499 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1500 ss.ss_family = slave_dev->type;
1501 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1502 if (res) {
1503 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1504 goto err_restore_mtu;
1505 }
1506 }
1507
1508 /* set slave flag before open to prevent IPv6 addrconf */
1509 slave_dev->flags |= IFF_SLAVE;
1510
1511 /* open the slave since the application closed it */
1512 res = dev_open(slave_dev);
1513 if (res) {
1514 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1515 goto err_restore_mac;
1516 }
1517
1518 slave_dev->priv_flags |= IFF_BONDING;
1519 /* initialize slave stats */
1520 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1521
1522 if (bond_is_lb(bond)) {
1523 /* bond_alb_init_slave() must be called before all other stages since
1524 * it might fail and we do not want to have to undo everything
1525 */
1526 res = bond_alb_init_slave(bond, new_slave);
1527 if (res)
1528 goto err_close;
1529 }
1530
1531 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1532 if (res) {
1533 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1534 slave_dev->name);
1535 goto err_close;
1536 }
1537
1538 prev_slave = bond_last_slave(bond);
1539
1540 new_slave->delay = 0;
1541 new_slave->link_failure_count = 0;
1542
1543 if (bond_update_speed_duplex(new_slave) &&
1544 bond_needs_speed_duplex(bond))
1545 new_slave->link = BOND_LINK_DOWN;
1546
1547 new_slave->last_rx = jiffies -
1548 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1549 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1550 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1551
1552 if (bond->params.miimon && !bond->params.use_carrier) {
1553 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1554
1555 if ((link_reporting == -1) && !bond->params.arp_interval) {
1556 /* miimon is set but a bonded network driver
1557 * does not support ETHTOOL/MII and
1558 * arp_interval is not set. Note: if
1559 * use_carrier is enabled, we will never go
1560 * here (because netif_carrier is always
1561 * supported); thus, we don't need to change
1562 * the messages for netif_carrier.
1563 */
1564 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1565 slave_dev->name);
1566 } else if (link_reporting == -1) {
1567 /* unable get link status using mii/ethtool */
1568 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1569 slave_dev->name);
1570 }
1571 }
1572
1573 /* check for initial state */
1574 new_slave->link = BOND_LINK_NOCHANGE;
1575 if (bond->params.miimon) {
1576 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1577 if (bond->params.updelay) {
1578 bond_set_slave_link_state(new_slave,
1579 BOND_LINK_BACK,
1580 BOND_SLAVE_NOTIFY_NOW);
1581 new_slave->delay = bond->params.updelay;
1582 } else {
1583 bond_set_slave_link_state(new_slave,
1584 BOND_LINK_UP,
1585 BOND_SLAVE_NOTIFY_NOW);
1586 }
1587 } else {
1588 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1589 BOND_SLAVE_NOTIFY_NOW);
1590 }
1591 } else if (bond->params.arp_interval) {
1592 bond_set_slave_link_state(new_slave,
1593 (netif_carrier_ok(slave_dev) ?
1594 BOND_LINK_UP : BOND_LINK_DOWN),
1595 BOND_SLAVE_NOTIFY_NOW);
1596 } else {
1597 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1598 BOND_SLAVE_NOTIFY_NOW);
1599 }
1600
1601 if (new_slave->link != BOND_LINK_DOWN)
1602 new_slave->last_link_up = jiffies;
1603 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1604 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1605 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1606
1607 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1608 /* if there is a primary slave, remember it */
1609 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1610 rcu_assign_pointer(bond->primary_slave, new_slave);
1611 bond->force_primary = true;
1612 }
1613 }
1614
1615 switch (BOND_MODE(bond)) {
1616 case BOND_MODE_ACTIVEBACKUP:
1617 bond_set_slave_inactive_flags(new_slave,
1618 BOND_SLAVE_NOTIFY_NOW);
1619 break;
1620 case BOND_MODE_8023AD:
1621 /* in 802.3ad mode, the internal mechanism
1622 * will activate the slaves in the selected
1623 * aggregator
1624 */
1625 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1626 /* if this is the first slave */
1627 if (!prev_slave) {
1628 SLAVE_AD_INFO(new_slave)->id = 1;
1629 /* Initialize AD with the number of times that the AD timer is called in 1 second
1630 * can be called only after the mac address of the bond is set
1631 */
1632 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1633 } else {
1634 SLAVE_AD_INFO(new_slave)->id =
1635 SLAVE_AD_INFO(prev_slave)->id + 1;
1636 }
1637
1638 bond_3ad_bind_slave(new_slave);
1639 break;
1640 case BOND_MODE_TLB:
1641 case BOND_MODE_ALB:
1642 bond_set_active_slave(new_slave);
1643 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1644 break;
1645 default:
1646 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1647
1648 /* always active in trunk mode */
1649 bond_set_active_slave(new_slave);
1650
1651 /* In trunking mode there is little meaning to curr_active_slave
1652 * anyway (it holds no special properties of the bond device),
1653 * so we can change it without calling change_active_interface()
1654 */
1655 if (!rcu_access_pointer(bond->curr_active_slave) &&
1656 new_slave->link == BOND_LINK_UP)
1657 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1658
1659 break;
1660 } /* switch(bond_mode) */
1661
1662#ifdef CONFIG_NET_POLL_CONTROLLER
1663 if (bond->dev->npinfo) {
1664 if (slave_enable_netpoll(new_slave)) {
1665 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1666 res = -EBUSY;
1667 goto err_detach;
1668 }
1669 }
1670#endif
1671
1672 if (!(bond_dev->features & NETIF_F_LRO))
1673 dev_disable_lro(slave_dev);
1674
1675 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1676 new_slave);
1677 if (res) {
1678 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1679 goto err_detach;
1680 }
1681
1682 res = bond_master_upper_dev_link(bond, new_slave, extack);
1683 if (res) {
1684 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1685 goto err_unregister;
1686 }
1687
1688 res = bond_sysfs_slave_add(new_slave);
1689 if (res) {
1690 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1691 goto err_upper_unlink;
1692 }
1693
1694 /* If the mode uses primary, then the following is handled by
1695 * bond_change_active_slave().
1696 */
1697 if (!bond_uses_primary(bond)) {
1698 /* set promiscuity level to new slave */
1699 if (bond_dev->flags & IFF_PROMISC) {
1700 res = dev_set_promiscuity(slave_dev, 1);
1701 if (res)
1702 goto err_sysfs_del;
1703 }
1704
1705 /* set allmulti level to new slave */
1706 if (bond_dev->flags & IFF_ALLMULTI) {
1707 res = dev_set_allmulti(slave_dev, 1);
1708 if (res) {
1709 if (bond_dev->flags & IFF_PROMISC)
1710 dev_set_promiscuity(slave_dev, -1);
1711 goto err_sysfs_del;
1712 }
1713 }
1714
1715 netif_addr_lock_bh(bond_dev);
1716 dev_mc_sync_multiple(slave_dev, bond_dev);
1717 dev_uc_sync_multiple(slave_dev, bond_dev);
1718 netif_addr_unlock_bh(bond_dev);
1719
1720 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1721 /* add lacpdu mc addr to mc list */
1722 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1723
1724 dev_mc_add(slave_dev, lacpdu_multicast);
1725 }
1726 }
1727
1728 bond->slave_cnt++;
1729 bond_compute_features(bond);
1730 bond_set_carrier(bond);
1731
1732 if (bond_uses_primary(bond)) {
1733 block_netpoll_tx();
1734 bond_select_active_slave(bond);
1735 unblock_netpoll_tx();
1736 }
1737
1738 if (bond_mode_uses_xmit_hash(bond))
1739 bond_update_slave_arr(bond, NULL);
1740
1741 bond->nest_level = dev_get_nest_level(bond_dev);
1742
1743 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1744 slave_dev->name,
1745 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1746 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1747
1748 /* enslave is successful */
1749 bond_queue_slave_event(new_slave);
1750 return 0;
1751
1752/* Undo stages on error */
1753err_sysfs_del:
1754 bond_sysfs_slave_del(new_slave);
1755
1756err_upper_unlink:
1757 bond_upper_dev_unlink(bond, new_slave);
1758
1759err_unregister:
1760 netdev_rx_handler_unregister(slave_dev);
1761
1762err_detach:
1763 vlan_vids_del_by_dev(slave_dev, bond_dev);
1764 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1765 RCU_INIT_POINTER(bond->primary_slave, NULL);
1766 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1767 block_netpoll_tx();
1768 bond_change_active_slave(bond, NULL);
1769 bond_select_active_slave(bond);
1770 unblock_netpoll_tx();
1771 }
1772 /* either primary_slave or curr_active_slave might've changed */
1773 synchronize_rcu();
1774 slave_disable_netpoll(new_slave);
1775
1776err_close:
1777 slave_dev->priv_flags &= ~IFF_BONDING;
1778 dev_close(slave_dev);
1779
1780err_restore_mac:
1781 slave_dev->flags &= ~IFF_SLAVE;
1782 if (!bond->params.fail_over_mac ||
1783 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1784 /* XXX TODO - fom follow mode needs to change master's
1785 * MAC if this slave's MAC is in use by the bond, or at
1786 * least print a warning.
1787 */
1788 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1789 new_slave->dev->addr_len);
1790 ss.ss_family = slave_dev->type;
1791 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1792 }
1793
1794err_restore_mtu:
1795 dev_set_mtu(slave_dev, new_slave->original_mtu);
1796
1797err_free:
1798 bond_free_slave(new_slave);
1799
1800err_undo_flags:
1801 /* Enslave of first slave has failed and we need to fix master's mac */
1802 if (!bond_has_slaves(bond)) {
1803 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1804 slave_dev->dev_addr))
1805 eth_hw_addr_random(bond_dev);
1806 if (bond_dev->type != ARPHRD_ETHER) {
1807 dev_close(bond_dev);
1808 ether_setup(bond_dev);
1809 bond_dev->flags |= IFF_MASTER;
1810 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1811 }
1812 }
1813
1814 return res;
1815}
1816
1817/* Try to release the slave device <slave> from the bond device <master>
1818 * It is legal to access curr_active_slave without a lock because all the function
1819 * is RTNL-locked. If "all" is true it means that the function is being called
1820 * while destroying a bond interface and all slaves are being released.
1821 *
1822 * The rules for slave state should be:
1823 * for Active/Backup:
1824 * Active stays on all backups go down
1825 * for Bonded connections:
1826 * The first up interface should be left on and all others downed.
1827 */
1828static int __bond_release_one(struct net_device *bond_dev,
1829 struct net_device *slave_dev,
1830 bool all, bool unregister)
1831{
1832 struct bonding *bond = netdev_priv(bond_dev);
1833 struct slave *slave, *oldcurrent;
1834 struct sockaddr_storage ss;
1835 int old_flags = bond_dev->flags;
1836 netdev_features_t old_features = bond_dev->features;
1837
1838 /* slave is not a slave or master is not master of this slave */
1839 if (!(slave_dev->flags & IFF_SLAVE) ||
1840 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1841 netdev_dbg(bond_dev, "cannot release %s\n",
1842 slave_dev->name);
1843 return -EINVAL;
1844 }
1845
1846 block_netpoll_tx();
1847
1848 slave = bond_get_slave_by_dev(bond, slave_dev);
1849 if (!slave) {
1850 /* not a slave of this bond */
1851 netdev_info(bond_dev, "%s not enslaved\n",
1852 slave_dev->name);
1853 unblock_netpoll_tx();
1854 return -EINVAL;
1855 }
1856
1857 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1858
1859 bond_sysfs_slave_del(slave);
1860
1861 /* recompute stats just before removing the slave */
1862 bond_get_stats(bond->dev, &bond->bond_stats);
1863
1864 bond_upper_dev_unlink(bond, slave);
1865 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1866 * for this slave anymore.
1867 */
1868 netdev_rx_handler_unregister(slave_dev);
1869
1870 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1871 bond_3ad_unbind_slave(slave);
1872
1873 if (bond_mode_uses_xmit_hash(bond))
1874 bond_update_slave_arr(bond, slave);
1875
1876 netdev_info(bond_dev, "Releasing %s interface %s\n",
1877 bond_is_active_slave(slave) ? "active" : "backup",
1878 slave_dev->name);
1879
1880 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1881
1882 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1883
1884 if (!all && (!bond->params.fail_over_mac ||
1885 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1886 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1887 bond_has_slaves(bond))
1888 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1889 slave_dev->name, slave->perm_hwaddr,
1890 bond_dev->name, slave_dev->name);
1891 }
1892
1893 if (rtnl_dereference(bond->primary_slave) == slave)
1894 RCU_INIT_POINTER(bond->primary_slave, NULL);
1895
1896 if (oldcurrent == slave)
1897 bond_change_active_slave(bond, NULL);
1898
1899 if (bond_is_lb(bond)) {
1900 /* Must be called only after the slave has been
1901 * detached from the list and the curr_active_slave
1902 * has been cleared (if our_slave == old_current),
1903 * but before a new active slave is selected.
1904 */
1905 bond_alb_deinit_slave(bond, slave);
1906 }
1907
1908 if (all) {
1909 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1910 } else if (oldcurrent == slave) {
1911 /* Note that we hold RTNL over this sequence, so there
1912 * is no concern that another slave add/remove event
1913 * will interfere.
1914 */
1915 bond_select_active_slave(bond);
1916 }
1917
1918 if (!bond_has_slaves(bond)) {
1919 bond_set_carrier(bond);
1920 eth_hw_addr_random(bond_dev);
1921 }
1922
1923 unblock_netpoll_tx();
1924 synchronize_rcu();
1925 bond->slave_cnt--;
1926
1927 if (!bond_has_slaves(bond)) {
1928 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1929 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1930 }
1931
1932 bond_compute_features(bond);
1933 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1934 (old_features & NETIF_F_VLAN_CHALLENGED))
1935 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1936 slave_dev->name, bond_dev->name);
1937
1938 vlan_vids_del_by_dev(slave_dev, bond_dev);
1939
1940 /* If the mode uses primary, then this case was handled above by
1941 * bond_change_active_slave(..., NULL)
1942 */
1943 if (!bond_uses_primary(bond)) {
1944 /* unset promiscuity level from slave
1945 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1946 * of the IFF_PROMISC flag in the bond_dev, but we need the
1947 * value of that flag before that change, as that was the value
1948 * when this slave was attached, so we cache at the start of the
1949 * function and use it here. Same goes for ALLMULTI below
1950 */
1951 if (old_flags & IFF_PROMISC)
1952 dev_set_promiscuity(slave_dev, -1);
1953
1954 /* unset allmulti level from slave */
1955 if (old_flags & IFF_ALLMULTI)
1956 dev_set_allmulti(slave_dev, -1);
1957
1958 bond_hw_addr_flush(bond_dev, slave_dev);
1959 }
1960
1961 slave_disable_netpoll(slave);
1962
1963 /* close slave before restoring its mac address */
1964 dev_close(slave_dev);
1965
1966 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1967 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1968 /* restore original ("permanent") mac address */
1969 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1970 slave->dev->addr_len);
1971 ss.ss_family = slave_dev->type;
1972 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1973 }
1974
1975 if (unregister)
1976 __dev_set_mtu(slave_dev, slave->original_mtu);
1977 else
1978 dev_set_mtu(slave_dev, slave->original_mtu);
1979
1980 slave_dev->priv_flags &= ~IFF_BONDING;
1981
1982 bond_free_slave(slave);
1983
1984 return 0;
1985}
1986
1987/* A wrapper used because of ndo_del_link */
1988int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1989{
1990 return __bond_release_one(bond_dev, slave_dev, false, false);
1991}
1992
1993/* First release a slave and then destroy the bond if no more slaves are left.
1994 * Must be under rtnl_lock when this function is called.
1995 */
1996static int bond_release_and_destroy(struct net_device *bond_dev,
1997 struct net_device *slave_dev)
1998{
1999 struct bonding *bond = netdev_priv(bond_dev);
2000 int ret;
2001
2002 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2003 if (ret == 0 && !bond_has_slaves(bond)) {
2004 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2005 netdev_info(bond_dev, "Destroying bond %s\n",
2006 bond_dev->name);
2007 bond_remove_proc_entry(bond);
2008 unregister_netdevice(bond_dev);
2009 }
2010 return ret;
2011}
2012
2013static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2014{
2015 struct bonding *bond = netdev_priv(bond_dev);
2016 bond_fill_ifbond(bond, info);
2017}
2018
2019static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2020{
2021 struct bonding *bond = netdev_priv(bond_dev);
2022 struct list_head *iter;
2023 int i = 0, res = -ENODEV;
2024 struct slave *slave;
2025
2026 bond_for_each_slave(bond, slave, iter) {
2027 if (i++ == (int)info->slave_id) {
2028 res = 0;
2029 bond_fill_ifslave(slave, info);
2030 break;
2031 }
2032 }
2033
2034 return res;
2035}
2036
2037/*-------------------------------- Monitoring -------------------------------*/
2038
2039/* called with rcu_read_lock() */
2040static int bond_miimon_inspect(struct bonding *bond)
2041{
2042 int link_state, commit = 0;
2043 struct list_head *iter;
2044 struct slave *slave;
2045 bool ignore_updelay;
2046
2047 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2048
2049 bond_for_each_slave_rcu(bond, slave, iter) {
2050 slave->new_link = BOND_LINK_NOCHANGE;
2051 slave->link_new_state = slave->link;
2052
2053 link_state = bond_check_dev_link(bond, slave->dev, 0);
2054
2055 switch (slave->link) {
2056 case BOND_LINK_UP:
2057 if (link_state)
2058 continue;
2059
2060 bond_propose_link_state(slave, BOND_LINK_FAIL);
2061 commit++;
2062 slave->delay = bond->params.downdelay;
2063 if (slave->delay) {
2064 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2065 (BOND_MODE(bond) ==
2066 BOND_MODE_ACTIVEBACKUP) ?
2067 (bond_is_active_slave(slave) ?
2068 "active " : "backup ") : "",
2069 slave->dev->name,
2070 bond->params.downdelay * bond->params.miimon);
2071 }
2072 /*FALLTHRU*/
2073 case BOND_LINK_FAIL:
2074 if (link_state) {
2075 /* recovered before downdelay expired */
2076 bond_propose_link_state(slave, BOND_LINK_UP);
2077 slave->last_link_up = jiffies;
2078 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2079 (bond->params.downdelay - slave->delay) *
2080 bond->params.miimon,
2081 slave->dev->name);
2082 commit++;
2083 continue;
2084 }
2085
2086 if (slave->delay <= 0) {
2087 slave->new_link = BOND_LINK_DOWN;
2088 commit++;
2089 continue;
2090 }
2091
2092 slave->delay--;
2093 break;
2094
2095 case BOND_LINK_DOWN:
2096 if (!link_state)
2097 continue;
2098
2099 bond_propose_link_state(slave, BOND_LINK_BACK);
2100 commit++;
2101 slave->delay = bond->params.updelay;
2102
2103 if (slave->delay) {
2104 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2105 slave->dev->name,
2106 ignore_updelay ? 0 :
2107 bond->params.updelay *
2108 bond->params.miimon);
2109 }
2110 /*FALLTHRU*/
2111 case BOND_LINK_BACK:
2112 if (!link_state) {
2113 bond_propose_link_state(slave, BOND_LINK_DOWN);
2114 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2115 (bond->params.updelay - slave->delay) *
2116 bond->params.miimon,
2117 slave->dev->name);
2118 commit++;
2119 continue;
2120 }
2121
2122 if (ignore_updelay)
2123 slave->delay = 0;
2124
2125 if (slave->delay <= 0) {
2126 slave->new_link = BOND_LINK_UP;
2127 commit++;
2128 ignore_updelay = false;
2129 continue;
2130 }
2131
2132 slave->delay--;
2133 break;
2134 }
2135 }
2136
2137 return commit;
2138}
2139
2140static void bond_miimon_commit(struct bonding *bond)
2141{
2142 struct list_head *iter;
2143 struct slave *slave, *primary;
2144
2145 bond_for_each_slave(bond, slave, iter) {
2146 switch (slave->new_link) {
2147 case BOND_LINK_NOCHANGE:
2148 continue;
2149
2150 case BOND_LINK_UP:
2151 if (bond_update_speed_duplex(slave) &&
2152 bond_needs_speed_duplex(bond)) {
2153 slave->link = BOND_LINK_DOWN;
2154 if (net_ratelimit())
2155 netdev_warn(bond->dev,
2156 "failed to get link speed/duplex for %s\n",
2157 slave->dev->name);
2158 continue;
2159 }
2160 bond_set_slave_link_state(slave, BOND_LINK_UP,
2161 BOND_SLAVE_NOTIFY_NOW);
2162 slave->last_link_up = jiffies;
2163
2164 primary = rtnl_dereference(bond->primary_slave);
2165 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2166 /* prevent it from being the active one */
2167 bond_set_backup_slave(slave);
2168 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2169 /* make it immediately active */
2170 bond_set_active_slave(slave);
2171 } else if (slave != primary) {
2172 /* prevent it from being the active one */
2173 bond_set_backup_slave(slave);
2174 }
2175
2176 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2177 slave->dev->name,
2178 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2179 slave->duplex ? "full" : "half");
2180
2181 /* notify ad that the link status has changed */
2182 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2183 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2184
2185 if (bond_is_lb(bond))
2186 bond_alb_handle_link_change(bond, slave,
2187 BOND_LINK_UP);
2188
2189 if (BOND_MODE(bond) == BOND_MODE_XOR)
2190 bond_update_slave_arr(bond, NULL);
2191
2192 if (!bond->curr_active_slave || slave == primary)
2193 goto do_failover;
2194
2195 continue;
2196
2197 case BOND_LINK_DOWN:
2198 if (slave->link_failure_count < UINT_MAX)
2199 slave->link_failure_count++;
2200
2201 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2202 BOND_SLAVE_NOTIFY_NOW);
2203
2204 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2205 BOND_MODE(bond) == BOND_MODE_8023AD)
2206 bond_set_slave_inactive_flags(slave,
2207 BOND_SLAVE_NOTIFY_NOW);
2208
2209 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2210 slave->dev->name);
2211
2212 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2213 bond_3ad_handle_link_change(slave,
2214 BOND_LINK_DOWN);
2215
2216 if (bond_is_lb(bond))
2217 bond_alb_handle_link_change(bond, slave,
2218 BOND_LINK_DOWN);
2219
2220 if (BOND_MODE(bond) == BOND_MODE_XOR)
2221 bond_update_slave_arr(bond, NULL);
2222
2223 if (slave == rcu_access_pointer(bond->curr_active_slave))
2224 goto do_failover;
2225
2226 continue;
2227
2228 default:
2229 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2230 slave->new_link, slave->dev->name);
2231 slave->new_link = BOND_LINK_NOCHANGE;
2232
2233 continue;
2234 }
2235
2236do_failover:
2237 block_netpoll_tx();
2238 bond_select_active_slave(bond);
2239 unblock_netpoll_tx();
2240 }
2241
2242 bond_set_carrier(bond);
2243}
2244
2245/* bond_mii_monitor
2246 *
2247 * Really a wrapper that splits the mii monitor into two phases: an
2248 * inspection, then (if inspection indicates something needs to be done)
2249 * an acquisition of appropriate locks followed by a commit phase to
2250 * implement whatever link state changes are indicated.
2251 */
2252static void bond_mii_monitor(struct work_struct *work)
2253{
2254 struct bonding *bond = container_of(work, struct bonding,
2255 mii_work.work);
2256 bool should_notify_peers = false;
2257 unsigned long delay;
2258 struct slave *slave;
2259 struct list_head *iter;
2260
2261 delay = msecs_to_jiffies(bond->params.miimon);
2262
2263 if (!bond_has_slaves(bond))
2264 goto re_arm;
2265
2266 rcu_read_lock();
2267
2268 should_notify_peers = bond_should_notify_peers(bond);
2269
2270 if (bond_miimon_inspect(bond)) {
2271 rcu_read_unlock();
2272
2273 /* Race avoidance with bond_close cancel of workqueue */
2274 if (!rtnl_trylock()) {
2275 delay = 1;
2276 should_notify_peers = false;
2277 goto re_arm;
2278 }
2279
2280 bond_for_each_slave(bond, slave, iter) {
2281 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2282 }
2283 bond_miimon_commit(bond);
2284
2285 rtnl_unlock(); /* might sleep, hold no other locks */
2286 } else
2287 rcu_read_unlock();
2288
2289re_arm:
2290 if (bond->params.miimon)
2291 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2292
2293 if (should_notify_peers) {
2294 if (!rtnl_trylock())
2295 return;
2296 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2297 rtnl_unlock();
2298 }
2299}
2300
2301static int bond_upper_dev_walk(struct net_device *upper, void *data)
2302{
2303 __be32 ip = *((__be32 *)data);
2304
2305 return ip == bond_confirm_addr(upper, 0, ip);
2306}
2307
2308static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2309{
2310 bool ret = false;
2311
2312 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2313 return true;
2314
2315 rcu_read_lock();
2316 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2317 ret = true;
2318 rcu_read_unlock();
2319
2320 return ret;
2321}
2322
2323/* We go to the (large) trouble of VLAN tagging ARP frames because
2324 * switches in VLAN mode (especially if ports are configured as
2325 * "native" to a VLAN) might not pass non-tagged frames.
2326 */
2327static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2328 __be32 dest_ip, __be32 src_ip,
2329 struct bond_vlan_tag *tags)
2330{
2331 struct sk_buff *skb;
2332 struct bond_vlan_tag *outer_tag = tags;
2333
2334 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2335 arp_op, slave_dev->name, &dest_ip, &src_ip);
2336
2337 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2338 NULL, slave_dev->dev_addr, NULL);
2339
2340 if (!skb) {
2341 net_err_ratelimited("ARP packet allocation failed\n");
2342 return;
2343 }
2344
2345 if (!tags || tags->vlan_proto == VLAN_N_VID)
2346 goto xmit;
2347
2348 tags++;
2349
2350 /* Go through all the tags backwards and add them to the packet */
2351 while (tags->vlan_proto != VLAN_N_VID) {
2352 if (!tags->vlan_id) {
2353 tags++;
2354 continue;
2355 }
2356
2357 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2358 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2359 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2360 tags->vlan_id);
2361 if (!skb) {
2362 net_err_ratelimited("failed to insert inner VLAN tag\n");
2363 return;
2364 }
2365
2366 tags++;
2367 }
2368 /* Set the outer tag */
2369 if (outer_tag->vlan_id) {
2370 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2371 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2372 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2373 outer_tag->vlan_id);
2374 }
2375
2376xmit:
2377 arp_xmit(skb);
2378}
2379
2380/* Validate the device path between the @start_dev and the @end_dev.
2381 * The path is valid if the @end_dev is reachable through device
2382 * stacking.
2383 * When the path is validated, collect any vlan information in the
2384 * path.
2385 */
2386struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2387 struct net_device *end_dev,
2388 int level)
2389{
2390 struct bond_vlan_tag *tags;
2391 struct net_device *upper;
2392 struct list_head *iter;
2393
2394 if (start_dev == end_dev) {
2395 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2396 if (!tags)
2397 return ERR_PTR(-ENOMEM);
2398 tags[level].vlan_proto = VLAN_N_VID;
2399 return tags;
2400 }
2401
2402 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2403 tags = bond_verify_device_path(upper, end_dev, level + 1);
2404 if (IS_ERR_OR_NULL(tags)) {
2405 if (IS_ERR(tags))
2406 return tags;
2407 continue;
2408 }
2409 if (is_vlan_dev(upper)) {
2410 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2411 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2412 }
2413
2414 return tags;
2415 }
2416
2417 return NULL;
2418}
2419
2420static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2421{
2422 struct rtable *rt;
2423 struct bond_vlan_tag *tags;
2424 __be32 *targets = bond->params.arp_targets, addr;
2425 int i;
2426
2427 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2428 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2429 tags = NULL;
2430
2431 /* Find out through which dev should the packet go */
2432 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2433 RTO_ONLINK, 0);
2434 if (IS_ERR(rt)) {
2435 /* there's no route to target - try to send arp
2436 * probe to generate any traffic (arp_validate=0)
2437 */
2438 if (bond->params.arp_validate)
2439 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2440 bond->dev->name,
2441 &targets[i]);
2442 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2443 0, tags);
2444 continue;
2445 }
2446
2447 /* bond device itself */
2448 if (rt->dst.dev == bond->dev)
2449 goto found;
2450
2451 rcu_read_lock();
2452 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2453 rcu_read_unlock();
2454
2455 if (!IS_ERR_OR_NULL(tags))
2456 goto found;
2457
2458 /* Not our device - skip */
2459 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2460 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2461
2462 ip_rt_put(rt);
2463 continue;
2464
2465found:
2466 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2467 ip_rt_put(rt);
2468 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2469 addr, tags);
2470 kfree(tags);
2471 }
2472}
2473
2474static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2475{
2476 int i;
2477
2478 if (!sip || !bond_has_this_ip(bond, tip)) {
2479 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2480 &sip, &tip);
2481 return;
2482 }
2483
2484 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2485 if (i == -1) {
2486 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2487 &sip);
2488 return;
2489 }
2490 slave->last_rx = jiffies;
2491 slave->target_last_arp_rx[i] = jiffies;
2492}
2493
2494int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2495 struct slave *slave)
2496{
2497 struct arphdr *arp = (struct arphdr *)skb->data;
2498 struct slave *curr_active_slave, *curr_arp_slave;
2499 unsigned char *arp_ptr;
2500 __be32 sip, tip;
2501 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2502 unsigned int alen;
2503
2504 if (!slave_do_arp_validate(bond, slave)) {
2505 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2506 !slave_do_arp_validate_only(bond))
2507 slave->last_rx = jiffies;
2508 return RX_HANDLER_ANOTHER;
2509 } else if (!is_arp) {
2510 return RX_HANDLER_ANOTHER;
2511 }
2512
2513 alen = arp_hdr_len(bond->dev);
2514
2515 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2516 skb->dev->name);
2517
2518 if (alen > skb_headlen(skb)) {
2519 arp = kmalloc(alen, GFP_ATOMIC);
2520 if (!arp)
2521 goto out_unlock;
2522 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2523 goto out_unlock;
2524 }
2525
2526 if (arp->ar_hln != bond->dev->addr_len ||
2527 skb->pkt_type == PACKET_OTHERHOST ||
2528 skb->pkt_type == PACKET_LOOPBACK ||
2529 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2530 arp->ar_pro != htons(ETH_P_IP) ||
2531 arp->ar_pln != 4)
2532 goto out_unlock;
2533
2534 arp_ptr = (unsigned char *)(arp + 1);
2535 arp_ptr += bond->dev->addr_len;
2536 memcpy(&sip, arp_ptr, 4);
2537 arp_ptr += 4 + bond->dev->addr_len;
2538 memcpy(&tip, arp_ptr, 4);
2539
2540 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2541 slave->dev->name, bond_slave_state(slave),
2542 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2543 &sip, &tip);
2544
2545 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2546 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2547
2548 /* We 'trust' the received ARP enough to validate it if:
2549 *
2550 * (a) the slave receiving the ARP is active (which includes the
2551 * current ARP slave, if any), or
2552 *
2553 * (b) the receiving slave isn't active, but there is a currently
2554 * active slave and it received valid arp reply(s) after it became
2555 * the currently active slave, or
2556 *
2557 * (c) there is an ARP slave that sent an ARP during the prior ARP
2558 * interval, and we receive an ARP reply on any slave. We accept
2559 * these because switch FDB update delays may deliver the ARP
2560 * reply to a slave other than the sender of the ARP request.
2561 *
2562 * Note: for (b), backup slaves are receiving the broadcast ARP
2563 * request, not a reply. This request passes from the sending
2564 * slave through the L2 switch(es) to the receiving slave. Since
2565 * this is checking the request, sip/tip are swapped for
2566 * validation.
2567 *
2568 * This is done to avoid endless looping when we can't reach the
2569 * arp_ip_target and fool ourselves with our own arp requests.
2570 */
2571 if (bond_is_active_slave(slave))
2572 bond_validate_arp(bond, slave, sip, tip);
2573 else if (curr_active_slave &&
2574 time_after(slave_last_rx(bond, curr_active_slave),
2575 curr_active_slave->last_link_up))
2576 bond_validate_arp(bond, slave, tip, sip);
2577 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2578 bond_time_in_interval(bond,
2579 dev_trans_start(curr_arp_slave->dev), 1))
2580 bond_validate_arp(bond, slave, sip, tip);
2581
2582out_unlock:
2583 if (arp != (struct arphdr *)skb->data)
2584 kfree(arp);
2585 return RX_HANDLER_ANOTHER;
2586}
2587
2588/* function to verify if we're in the arp_interval timeslice, returns true if
2589 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2590 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2591 */
2592static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2593 int mod)
2594{
2595 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2596
2597 return time_in_range(jiffies,
2598 last_act - delta_in_ticks,
2599 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2600}
2601
2602/* This function is called regularly to monitor each slave's link
2603 * ensuring that traffic is being sent and received when arp monitoring
2604 * is used in load-balancing mode. if the adapter has been dormant, then an
2605 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2606 * arp monitoring in active backup mode.
2607 */
2608static void bond_loadbalance_arp_mon(struct bonding *bond)
2609{
2610 struct slave *slave, *oldcurrent;
2611 struct list_head *iter;
2612 int do_failover = 0, slave_state_changed = 0;
2613
2614 if (!bond_has_slaves(bond))
2615 goto re_arm;
2616
2617 rcu_read_lock();
2618
2619 oldcurrent = rcu_dereference(bond->curr_active_slave);
2620 /* see if any of the previous devices are up now (i.e. they have
2621 * xmt and rcv traffic). the curr_active_slave does not come into
2622 * the picture unless it is null. also, slave->last_link_up is not
2623 * needed here because we send an arp on each slave and give a slave
2624 * as long as it needs to get the tx/rx within the delta.
2625 * TODO: what about up/down delay in arp mode? it wasn't here before
2626 * so it can wait
2627 */
2628 bond_for_each_slave_rcu(bond, slave, iter) {
2629 unsigned long trans_start = dev_trans_start(slave->dev);
2630
2631 slave->new_link = BOND_LINK_NOCHANGE;
2632
2633 if (slave->link != BOND_LINK_UP) {
2634 if (bond_time_in_interval(bond, trans_start, 1) &&
2635 bond_time_in_interval(bond, slave->last_rx, 1)) {
2636
2637 slave->new_link = BOND_LINK_UP;
2638 slave_state_changed = 1;
2639
2640 /* primary_slave has no meaning in round-robin
2641 * mode. the window of a slave being up and
2642 * curr_active_slave being null after enslaving
2643 * is closed.
2644 */
2645 if (!oldcurrent) {
2646 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2647 slave->dev->name);
2648 do_failover = 1;
2649 } else {
2650 netdev_info(bond->dev, "interface %s is now up\n",
2651 slave->dev->name);
2652 }
2653 }
2654 } else {
2655 /* slave->link == BOND_LINK_UP */
2656
2657 /* not all switches will respond to an arp request
2658 * when the source ip is 0, so don't take the link down
2659 * if we don't know our ip yet
2660 */
2661 if (!bond_time_in_interval(bond, trans_start, 2) ||
2662 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2663
2664 slave->new_link = BOND_LINK_DOWN;
2665 slave_state_changed = 1;
2666
2667 if (slave->link_failure_count < UINT_MAX)
2668 slave->link_failure_count++;
2669
2670 netdev_info(bond->dev, "interface %s is now down\n",
2671 slave->dev->name);
2672
2673 if (slave == oldcurrent)
2674 do_failover = 1;
2675 }
2676 }
2677
2678 /* note: if switch is in round-robin mode, all links
2679 * must tx arp to ensure all links rx an arp - otherwise
2680 * links may oscillate or not come up at all; if switch is
2681 * in something like xor mode, there is nothing we can
2682 * do - all replies will be rx'ed on same link causing slaves
2683 * to be unstable during low/no traffic periods
2684 */
2685 if (bond_slave_is_up(slave))
2686 bond_arp_send_all(bond, slave);
2687 }
2688
2689 rcu_read_unlock();
2690
2691 if (do_failover || slave_state_changed) {
2692 if (!rtnl_trylock())
2693 goto re_arm;
2694
2695 bond_for_each_slave(bond, slave, iter) {
2696 if (slave->new_link != BOND_LINK_NOCHANGE)
2697 slave->link = slave->new_link;
2698 }
2699
2700 if (slave_state_changed) {
2701 bond_slave_state_change(bond);
2702 if (BOND_MODE(bond) == BOND_MODE_XOR)
2703 bond_update_slave_arr(bond, NULL);
2704 }
2705 if (do_failover) {
2706 block_netpoll_tx();
2707 bond_select_active_slave(bond);
2708 unblock_netpoll_tx();
2709 }
2710 rtnl_unlock();
2711 }
2712
2713re_arm:
2714 if (bond->params.arp_interval)
2715 queue_delayed_work(bond->wq, &bond->arp_work,
2716 msecs_to_jiffies(bond->params.arp_interval));
2717}
2718
2719/* Called to inspect slaves for active-backup mode ARP monitor link state
2720 * changes. Sets new_link in slaves to specify what action should take
2721 * place for the slave. Returns 0 if no changes are found, >0 if changes
2722 * to link states must be committed.
2723 *
2724 * Called with rcu_read_lock held.
2725 */
2726static int bond_ab_arp_inspect(struct bonding *bond)
2727{
2728 unsigned long trans_start, last_rx;
2729 struct list_head *iter;
2730 struct slave *slave;
2731 int commit = 0;
2732
2733 bond_for_each_slave_rcu(bond, slave, iter) {
2734 slave->new_link = BOND_LINK_NOCHANGE;
2735 last_rx = slave_last_rx(bond, slave);
2736
2737 if (slave->link != BOND_LINK_UP) {
2738 if (bond_time_in_interval(bond, last_rx, 1)) {
2739 slave->new_link = BOND_LINK_UP;
2740 commit++;
2741 }
2742 continue;
2743 }
2744
2745 /* Give slaves 2*delta after being enslaved or made
2746 * active. This avoids bouncing, as the last receive
2747 * times need a full ARP monitor cycle to be updated.
2748 */
2749 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2750 continue;
2751
2752 /* Backup slave is down if:
2753 * - No current_arp_slave AND
2754 * - more than 3*delta since last receive AND
2755 * - the bond has an IP address
2756 *
2757 * Note: a non-null current_arp_slave indicates
2758 * the curr_active_slave went down and we are
2759 * searching for a new one; under this condition
2760 * we only take the curr_active_slave down - this
2761 * gives each slave a chance to tx/rx traffic
2762 * before being taken out
2763 */
2764 if (!bond_is_active_slave(slave) &&
2765 !rcu_access_pointer(bond->current_arp_slave) &&
2766 !bond_time_in_interval(bond, last_rx, 3)) {
2767 slave->new_link = BOND_LINK_DOWN;
2768 commit++;
2769 }
2770
2771 /* Active slave is down if:
2772 * - more than 2*delta since transmitting OR
2773 * - (more than 2*delta since receive AND
2774 * the bond has an IP address)
2775 */
2776 trans_start = dev_trans_start(slave->dev);
2777 if (bond_is_active_slave(slave) &&
2778 (!bond_time_in_interval(bond, trans_start, 2) ||
2779 !bond_time_in_interval(bond, last_rx, 2))) {
2780 slave->new_link = BOND_LINK_DOWN;
2781 commit++;
2782 }
2783 }
2784
2785 return commit;
2786}
2787
2788/* Called to commit link state changes noted by inspection step of
2789 * active-backup mode ARP monitor.
2790 *
2791 * Called with RTNL hold.
2792 */
2793static void bond_ab_arp_commit(struct bonding *bond)
2794{
2795 unsigned long trans_start;
2796 struct list_head *iter;
2797 struct slave *slave;
2798
2799 bond_for_each_slave(bond, slave, iter) {
2800 switch (slave->new_link) {
2801 case BOND_LINK_NOCHANGE:
2802 continue;
2803
2804 case BOND_LINK_UP:
2805 trans_start = dev_trans_start(slave->dev);
2806 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2807 (!rtnl_dereference(bond->curr_active_slave) &&
2808 bond_time_in_interval(bond, trans_start, 1))) {
2809 struct slave *current_arp_slave;
2810
2811 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2812 bond_set_slave_link_state(slave, BOND_LINK_UP,
2813 BOND_SLAVE_NOTIFY_NOW);
2814 if (current_arp_slave) {
2815 bond_set_slave_inactive_flags(
2816 current_arp_slave,
2817 BOND_SLAVE_NOTIFY_NOW);
2818 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2819 }
2820
2821 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2822 slave->dev->name);
2823
2824 if (!rtnl_dereference(bond->curr_active_slave) ||
2825 slave == rtnl_dereference(bond->primary_slave))
2826 goto do_failover;
2827
2828 }
2829
2830 continue;
2831
2832 case BOND_LINK_DOWN:
2833 if (slave->link_failure_count < UINT_MAX)
2834 slave->link_failure_count++;
2835
2836 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2837 BOND_SLAVE_NOTIFY_NOW);
2838 bond_set_slave_inactive_flags(slave,
2839 BOND_SLAVE_NOTIFY_NOW);
2840
2841 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2842 slave->dev->name);
2843
2844 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2845 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2846 goto do_failover;
2847 }
2848
2849 continue;
2850
2851 default:
2852 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2853 slave->new_link, slave->dev->name);
2854 continue;
2855 }
2856
2857do_failover:
2858 block_netpoll_tx();
2859 bond_select_active_slave(bond);
2860 unblock_netpoll_tx();
2861 }
2862
2863 bond_set_carrier(bond);
2864}
2865
2866/* Send ARP probes for active-backup mode ARP monitor.
2867 *
2868 * Called with rcu_read_lock held.
2869 */
2870static bool bond_ab_arp_probe(struct bonding *bond)
2871{
2872 struct slave *slave, *before = NULL, *new_slave = NULL,
2873 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2874 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2875 struct list_head *iter;
2876 bool found = false;
2877 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2878
2879 if (curr_arp_slave && curr_active_slave)
2880 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2881 curr_arp_slave->dev->name,
2882 curr_active_slave->dev->name);
2883
2884 if (curr_active_slave) {
2885 bond_arp_send_all(bond, curr_active_slave);
2886 return should_notify_rtnl;
2887 }
2888
2889 /* if we don't have a curr_active_slave, search for the next available
2890 * backup slave from the current_arp_slave and make it the candidate
2891 * for becoming the curr_active_slave
2892 */
2893
2894 if (!curr_arp_slave) {
2895 curr_arp_slave = bond_first_slave_rcu(bond);
2896 if (!curr_arp_slave)
2897 return should_notify_rtnl;
2898 }
2899
2900 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2901
2902 bond_for_each_slave_rcu(bond, slave, iter) {
2903 if (!found && !before && bond_slave_is_up(slave))
2904 before = slave;
2905
2906 if (found && !new_slave && bond_slave_is_up(slave))
2907 new_slave = slave;
2908 /* if the link state is up at this point, we
2909 * mark it down - this can happen if we have
2910 * simultaneous link failures and
2911 * reselect_active_interface doesn't make this
2912 * one the current slave so it is still marked
2913 * up when it is actually down
2914 */
2915 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2916 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2917 BOND_SLAVE_NOTIFY_LATER);
2918 if (slave->link_failure_count < UINT_MAX)
2919 slave->link_failure_count++;
2920
2921 bond_set_slave_inactive_flags(slave,
2922 BOND_SLAVE_NOTIFY_LATER);
2923
2924 netdev_info(bond->dev, "backup interface %s is now down\n",
2925 slave->dev->name);
2926 }
2927 if (slave == curr_arp_slave)
2928 found = true;
2929 }
2930
2931 if (!new_slave && before)
2932 new_slave = before;
2933
2934 if (!new_slave)
2935 goto check_state;
2936
2937 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2938 BOND_SLAVE_NOTIFY_LATER);
2939 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2940 bond_arp_send_all(bond, new_slave);
2941 new_slave->last_link_up = jiffies;
2942 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2943
2944check_state:
2945 bond_for_each_slave_rcu(bond, slave, iter) {
2946 if (slave->should_notify || slave->should_notify_link) {
2947 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2948 break;
2949 }
2950 }
2951 return should_notify_rtnl;
2952}
2953
2954static void bond_activebackup_arp_mon(struct bonding *bond)
2955{
2956 bool should_notify_peers = false;
2957 bool should_notify_rtnl = false;
2958 int delta_in_ticks;
2959
2960 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2961
2962 if (!bond_has_slaves(bond))
2963 goto re_arm;
2964
2965 rcu_read_lock();
2966
2967 should_notify_peers = bond_should_notify_peers(bond);
2968
2969 if (bond_ab_arp_inspect(bond)) {
2970 rcu_read_unlock();
2971
2972 /* Race avoidance with bond_close flush of workqueue */
2973 if (!rtnl_trylock()) {
2974 delta_in_ticks = 1;
2975 should_notify_peers = false;
2976 goto re_arm;
2977 }
2978
2979 bond_ab_arp_commit(bond);
2980
2981 rtnl_unlock();
2982 rcu_read_lock();
2983 }
2984
2985 should_notify_rtnl = bond_ab_arp_probe(bond);
2986 rcu_read_unlock();
2987
2988re_arm:
2989 if (bond->params.arp_interval)
2990 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2991
2992 if (should_notify_peers || should_notify_rtnl) {
2993 if (!rtnl_trylock())
2994 return;
2995
2996 if (should_notify_peers)
2997 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2998 bond->dev);
2999 if (should_notify_rtnl) {
3000 bond_slave_state_notify(bond);
3001 bond_slave_link_notify(bond);
3002 }
3003
3004 rtnl_unlock();
3005 }
3006}
3007
3008static void bond_arp_monitor(struct work_struct *work)
3009{
3010 struct bonding *bond = container_of(work, struct bonding,
3011 arp_work.work);
3012
3013 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3014 bond_activebackup_arp_mon(bond);
3015 else
3016 bond_loadbalance_arp_mon(bond);
3017}
3018
3019/*-------------------------- netdev event handling --------------------------*/
3020
3021/* Change device name */
3022static int bond_event_changename(struct bonding *bond)
3023{
3024 bond_remove_proc_entry(bond);
3025 bond_create_proc_entry(bond);
3026
3027 bond_debug_reregister(bond);
3028
3029 return NOTIFY_DONE;
3030}
3031
3032static int bond_master_netdev_event(unsigned long event,
3033 struct net_device *bond_dev)
3034{
3035 struct bonding *event_bond = netdev_priv(bond_dev);
3036
3037 switch (event) {
3038 case NETDEV_CHANGENAME:
3039 return bond_event_changename(event_bond);
3040 case NETDEV_UNREGISTER:
3041 bond_remove_proc_entry(event_bond);
3042 break;
3043 case NETDEV_REGISTER:
3044 bond_create_proc_entry(event_bond);
3045 break;
3046 case NETDEV_NOTIFY_PEERS:
3047 if (event_bond->send_peer_notif)
3048 event_bond->send_peer_notif--;
3049 break;
3050 default:
3051 break;
3052 }
3053
3054 return NOTIFY_DONE;
3055}
3056
3057static int bond_slave_netdev_event(unsigned long event,
3058 struct net_device *slave_dev)
3059{
3060 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3061 struct bonding *bond;
3062 struct net_device *bond_dev;
3063
3064 /* A netdev event can be generated while enslaving a device
3065 * before netdev_rx_handler_register is called in which case
3066 * slave will be NULL
3067 */
3068 if (!slave)
3069 return NOTIFY_DONE;
3070 bond_dev = slave->bond->dev;
3071 bond = slave->bond;
3072 primary = rtnl_dereference(bond->primary_slave);
3073
3074 switch (event) {
3075 case NETDEV_UNREGISTER:
3076 if (bond_dev->type != ARPHRD_ETHER)
3077 bond_release_and_destroy(bond_dev, slave_dev);
3078 else
3079 __bond_release_one(bond_dev, slave_dev, false, true);
3080 break;
3081 case NETDEV_UP:
3082 case NETDEV_CHANGE:
3083 /* For 802.3ad mode only:
3084 * Getting invalid Speed/Duplex values here will put slave
3085 * in weird state. So mark it as link-down for the time
3086 * being and let link-monitoring (miimon) set it right when
3087 * correct speeds/duplex are available.
3088 */
3089 if (bond_update_speed_duplex(slave) &&
3090 BOND_MODE(bond) == BOND_MODE_8023AD)
3091 slave->link = BOND_LINK_DOWN;
3092
3093 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3094 bond_3ad_adapter_speed_duplex_changed(slave);
3095 /* Fallthrough */
3096 case NETDEV_DOWN:
3097 /* Refresh slave-array if applicable!
3098 * If the setup does not use miimon or arpmon (mode-specific!),
3099 * then these events will not cause the slave-array to be
3100 * refreshed. This will cause xmit to use a slave that is not
3101 * usable. Avoid such situation by refeshing the array at these
3102 * events. If these (miimon/arpmon) parameters are configured
3103 * then array gets refreshed twice and that should be fine!
3104 */
3105 if (bond_mode_uses_xmit_hash(bond))
3106 bond_update_slave_arr(bond, NULL);
3107 break;
3108 case NETDEV_CHANGEMTU:
3109 /* TODO: Should slaves be allowed to
3110 * independently alter their MTU? For
3111 * an active-backup bond, slaves need
3112 * not be the same type of device, so
3113 * MTUs may vary. For other modes,
3114 * slaves arguably should have the
3115 * same MTUs. To do this, we'd need to
3116 * take over the slave's change_mtu
3117 * function for the duration of their
3118 * servitude.
3119 */
3120 break;
3121 case NETDEV_CHANGENAME:
3122 /* we don't care if we don't have primary set */
3123 if (!bond_uses_primary(bond) ||
3124 !bond->params.primary[0])
3125 break;
3126
3127 if (slave == primary) {
3128 /* slave's name changed - he's no longer primary */
3129 RCU_INIT_POINTER(bond->primary_slave, NULL);
3130 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3131 /* we have a new primary slave */
3132 rcu_assign_pointer(bond->primary_slave, slave);
3133 } else { /* we didn't change primary - exit */
3134 break;
3135 }
3136
3137 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3138 primary ? slave_dev->name : "none");
3139
3140 block_netpoll_tx();
3141 bond_select_active_slave(bond);
3142 unblock_netpoll_tx();
3143 break;
3144 case NETDEV_FEAT_CHANGE:
3145 bond_compute_features(bond);
3146 break;
3147 case NETDEV_RESEND_IGMP:
3148 /* Propagate to master device */
3149 call_netdevice_notifiers(event, slave->bond->dev);
3150 break;
3151 default:
3152 break;
3153 }
3154
3155 return NOTIFY_DONE;
3156}
3157
3158/* bond_netdev_event: handle netdev notifier chain events.
3159 *
3160 * This function receives events for the netdev chain. The caller (an
3161 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3162 * locks for us to safely manipulate the slave devices (RTNL lock,
3163 * dev_probe_lock).
3164 */
3165static int bond_netdev_event(struct notifier_block *this,
3166 unsigned long event, void *ptr)
3167{
3168 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3169
3170 netdev_dbg(event_dev, "event: %lx\n", event);
3171
3172 if (!(event_dev->priv_flags & IFF_BONDING))
3173 return NOTIFY_DONE;
3174
3175 if (event_dev->flags & IFF_MASTER) {
3176 netdev_dbg(event_dev, "IFF_MASTER\n");
3177 return bond_master_netdev_event(event, event_dev);
3178 }
3179
3180 if (event_dev->flags & IFF_SLAVE) {
3181 netdev_dbg(event_dev, "IFF_SLAVE\n");
3182 return bond_slave_netdev_event(event, event_dev);
3183 }
3184
3185 return NOTIFY_DONE;
3186}
3187
3188static struct notifier_block bond_netdev_notifier = {
3189 .notifier_call = bond_netdev_event,
3190};
3191
3192/*---------------------------- Hashing Policies -----------------------------*/
3193
3194/* L2 hash helper */
3195static inline u32 bond_eth_hash(struct sk_buff *skb)
3196{
3197 struct ethhdr *ep, hdr_tmp;
3198
3199 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3200 if (ep)
3201 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3202 return 0;
3203}
3204
3205/* Extract the appropriate headers based on bond's xmit policy */
3206static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3207 struct flow_keys *fk)
3208{
3209 const struct ipv6hdr *iph6;
3210 const struct iphdr *iph;
3211 int noff, proto = -1;
3212
3213 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3214 return skb_flow_dissect_flow_keys(skb, fk, 0);
3215
3216 fk->ports.ports = 0;
3217 noff = skb_network_offset(skb);
3218 if (skb->protocol == htons(ETH_P_IP)) {
3219 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3220 return false;
3221 iph = ip_hdr(skb);
3222 iph_to_flow_copy_v4addrs(fk, iph);
3223 noff += iph->ihl << 2;
3224 if (!ip_is_fragment(iph))
3225 proto = iph->protocol;
3226 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3227 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3228 return false;
3229 iph6 = ipv6_hdr(skb);
3230 iph_to_flow_copy_v6addrs(fk, iph6);
3231 noff += sizeof(*iph6);
3232 proto = iph6->nexthdr;
3233 } else {
3234 return false;
3235 }
3236 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3237 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3238
3239 return true;
3240}
3241
3242/**
3243 * bond_xmit_hash - generate a hash value based on the xmit policy
3244 * @bond: bonding device
3245 * @skb: buffer to use for headers
3246 *
3247 * This function will extract the necessary headers from the skb buffer and use
3248 * them to generate a hash based on the xmit_policy set in the bonding device
3249 */
3250u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3251{
3252 struct flow_keys flow;
3253 u32 hash;
3254
3255 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3256 skb->l4_hash)
3257 return skb->hash;
3258
3259 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3260 !bond_flow_dissect(bond, skb, &flow))
3261 return bond_eth_hash(skb);
3262
3263 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3264 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3265 hash = bond_eth_hash(skb);
3266 else
3267 hash = (__force u32)flow.ports.ports;
3268 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3269 (__force u32)flow_get_u32_src(&flow);
3270 hash ^= (hash >> 16);
3271 hash ^= (hash >> 8);
3272
3273 return hash >> 1;
3274}
3275
3276/*-------------------------- Device entry points ----------------------------*/
3277
3278void bond_work_init_all(struct bonding *bond)
3279{
3280 INIT_DELAYED_WORK(&bond->mcast_work,
3281 bond_resend_igmp_join_requests_delayed);
3282 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3283 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3284 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3285 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3286 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3287}
3288
3289static void bond_work_cancel_all(struct bonding *bond)
3290{
3291 cancel_delayed_work_sync(&bond->mii_work);
3292 cancel_delayed_work_sync(&bond->arp_work);
3293 cancel_delayed_work_sync(&bond->alb_work);
3294 cancel_delayed_work_sync(&bond->ad_work);
3295 cancel_delayed_work_sync(&bond->mcast_work);
3296 cancel_delayed_work_sync(&bond->slave_arr_work);
3297}
3298
3299static int bond_open(struct net_device *bond_dev)
3300{
3301 struct bonding *bond = netdev_priv(bond_dev);
3302 struct list_head *iter;
3303 struct slave *slave;
3304
3305 /* reset slave->backup and slave->inactive */
3306 if (bond_has_slaves(bond)) {
3307 bond_for_each_slave(bond, slave, iter) {
3308 if (bond_uses_primary(bond) &&
3309 slave != rcu_access_pointer(bond->curr_active_slave)) {
3310 bond_set_slave_inactive_flags(slave,
3311 BOND_SLAVE_NOTIFY_NOW);
3312 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3313 bond_set_slave_active_flags(slave,
3314 BOND_SLAVE_NOTIFY_NOW);
3315 }
3316 }
3317 }
3318
3319 if (bond_is_lb(bond)) {
3320 /* bond_alb_initialize must be called before the timer
3321 * is started.
3322 */
3323 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3324 return -ENOMEM;
3325 if (bond->params.tlb_dynamic_lb)
3326 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3327 }
3328
3329 if (bond->params.miimon) /* link check interval, in milliseconds. */
3330 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3331
3332 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3333 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3334 bond->recv_probe = bond_arp_rcv;
3335 }
3336
3337 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3338 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3339 /* register to receive LACPDUs */
3340 bond->recv_probe = bond_3ad_lacpdu_recv;
3341 bond_3ad_initiate_agg_selection(bond, 1);
3342 }
3343
3344 if (bond_mode_uses_xmit_hash(bond))
3345 bond_update_slave_arr(bond, NULL);
3346
3347 return 0;
3348}
3349
3350static int bond_close(struct net_device *bond_dev)
3351{
3352 struct bonding *bond = netdev_priv(bond_dev);
3353
3354 bond_work_cancel_all(bond);
3355 bond->send_peer_notif = 0;
3356 if (bond_is_lb(bond))
3357 bond_alb_deinitialize(bond);
3358 bond->recv_probe = NULL;
3359
3360 return 0;
3361}
3362
3363/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3364 * that some drivers can provide 32bit values only.
3365 */
3366static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3367 const struct rtnl_link_stats64 *_new,
3368 const struct rtnl_link_stats64 *_old)
3369{
3370 const u64 *new = (const u64 *)_new;
3371 const u64 *old = (const u64 *)_old;
3372 u64 *res = (u64 *)_res;
3373 int i;
3374
3375 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3376 u64 nv = new[i];
3377 u64 ov = old[i];
3378 s64 delta = nv - ov;
3379
3380 /* detects if this particular field is 32bit only */
3381 if (((nv | ov) >> 32) == 0)
3382 delta = (s64)(s32)((u32)nv - (u32)ov);
3383
3384 /* filter anomalies, some drivers reset their stats
3385 * at down/up events.
3386 */
3387 if (delta > 0)
3388 res[i] += delta;
3389 }
3390}
3391
3392static void bond_get_stats(struct net_device *bond_dev,
3393 struct rtnl_link_stats64 *stats)
3394{
3395 struct bonding *bond = netdev_priv(bond_dev);
3396 struct rtnl_link_stats64 temp;
3397 struct list_head *iter;
3398 struct slave *slave;
3399
3400 spin_lock(&bond->stats_lock);
3401 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3402
3403 rcu_read_lock();
3404 bond_for_each_slave_rcu(bond, slave, iter) {
3405 const struct rtnl_link_stats64 *new =
3406 dev_get_stats(slave->dev, &temp);
3407
3408 bond_fold_stats(stats, new, &slave->slave_stats);
3409
3410 /* save off the slave stats for the next run */
3411 memcpy(&slave->slave_stats, new, sizeof(*new));
3412 }
3413 rcu_read_unlock();
3414
3415 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3416 spin_unlock(&bond->stats_lock);
3417}
3418
3419static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3420{
3421 struct bonding *bond = netdev_priv(bond_dev);
3422 struct net_device *slave_dev = NULL;
3423 struct ifbond k_binfo;
3424 struct ifbond __user *u_binfo = NULL;
3425 struct ifslave k_sinfo;
3426 struct ifslave __user *u_sinfo = NULL;
3427 struct mii_ioctl_data *mii = NULL;
3428 struct bond_opt_value newval;
3429 struct net *net;
3430 int res = 0;
3431
3432 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3433
3434 switch (cmd) {
3435 case SIOCGMIIPHY:
3436 mii = if_mii(ifr);
3437 if (!mii)
3438 return -EINVAL;
3439
3440 mii->phy_id = 0;
3441 /* Fall Through */
3442 case SIOCGMIIREG:
3443 /* We do this again just in case we were called by SIOCGMIIREG
3444 * instead of SIOCGMIIPHY.
3445 */
3446 mii = if_mii(ifr);
3447 if (!mii)
3448 return -EINVAL;
3449
3450 if (mii->reg_num == 1) {
3451 mii->val_out = 0;
3452 if (netif_carrier_ok(bond->dev))
3453 mii->val_out = BMSR_LSTATUS;
3454 }
3455
3456 return 0;
3457 case BOND_INFO_QUERY_OLD:
3458 case SIOCBONDINFOQUERY:
3459 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3460
3461 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3462 return -EFAULT;
3463
3464 bond_info_query(bond_dev, &k_binfo);
3465 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3466 return -EFAULT;
3467
3468 return 0;
3469 case BOND_SLAVE_INFO_QUERY_OLD:
3470 case SIOCBONDSLAVEINFOQUERY:
3471 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3472
3473 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3474 return -EFAULT;
3475
3476 res = bond_slave_info_query(bond_dev, &k_sinfo);
3477 if (res == 0 &&
3478 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3479 return -EFAULT;
3480
3481 return res;
3482 default:
3483 break;
3484 }
3485
3486 net = dev_net(bond_dev);
3487
3488 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3489 return -EPERM;
3490
3491 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3492
3493 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3494
3495 if (!slave_dev)
3496 return -ENODEV;
3497
3498 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3499 switch (cmd) {
3500 case BOND_ENSLAVE_OLD:
3501 case SIOCBONDENSLAVE:
3502 res = bond_enslave(bond_dev, slave_dev, NULL);
3503 break;
3504 case BOND_RELEASE_OLD:
3505 case SIOCBONDRELEASE:
3506 res = bond_release(bond_dev, slave_dev);
3507 break;
3508 case BOND_SETHWADDR_OLD:
3509 case SIOCBONDSETHWADDR:
3510 bond_set_dev_addr(bond_dev, slave_dev);
3511 res = 0;
3512 break;
3513 case BOND_CHANGE_ACTIVE_OLD:
3514 case SIOCBONDCHANGEACTIVE:
3515 bond_opt_initstr(&newval, slave_dev->name);
3516 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3517 &newval);
3518 break;
3519 default:
3520 res = -EOPNOTSUPP;
3521 }
3522
3523 return res;
3524}
3525
3526static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3527{
3528 struct bonding *bond = netdev_priv(bond_dev);
3529
3530 if (change & IFF_PROMISC)
3531 bond_set_promiscuity(bond,
3532 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3533
3534 if (change & IFF_ALLMULTI)
3535 bond_set_allmulti(bond,
3536 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3537}
3538
3539static void bond_set_rx_mode(struct net_device *bond_dev)
3540{
3541 struct bonding *bond = netdev_priv(bond_dev);
3542 struct list_head *iter;
3543 struct slave *slave;
3544
3545 rcu_read_lock();
3546 if (bond_uses_primary(bond)) {
3547 slave = rcu_dereference(bond->curr_active_slave);
3548 if (slave) {
3549 dev_uc_sync(slave->dev, bond_dev);
3550 dev_mc_sync(slave->dev, bond_dev);
3551 }
3552 } else {
3553 bond_for_each_slave_rcu(bond, slave, iter) {
3554 dev_uc_sync_multiple(slave->dev, bond_dev);
3555 dev_mc_sync_multiple(slave->dev, bond_dev);
3556 }
3557 }
3558 rcu_read_unlock();
3559}
3560
3561static int bond_neigh_init(struct neighbour *n)
3562{
3563 struct bonding *bond = netdev_priv(n->dev);
3564 const struct net_device_ops *slave_ops;
3565 struct neigh_parms parms;
3566 struct slave *slave;
3567 int ret;
3568
3569 slave = bond_first_slave(bond);
3570 if (!slave)
3571 return 0;
3572 slave_ops = slave->dev->netdev_ops;
3573 if (!slave_ops->ndo_neigh_setup)
3574 return 0;
3575
3576 parms.neigh_setup = NULL;
3577 parms.neigh_cleanup = NULL;
3578 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3579 if (ret)
3580 return ret;
3581
3582 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3583 * after the last slave has been detached. Assumes that all slaves
3584 * utilize the same neigh_cleanup (true at this writing as only user
3585 * is ipoib).
3586 */
3587 n->parms->neigh_cleanup = parms.neigh_cleanup;
3588
3589 if (!parms.neigh_setup)
3590 return 0;
3591
3592 return parms.neigh_setup(n);
3593}
3594
3595/* The bonding ndo_neigh_setup is called at init time beofre any
3596 * slave exists. So we must declare proxy setup function which will
3597 * be used at run time to resolve the actual slave neigh param setup.
3598 *
3599 * It's also called by master devices (such as vlans) to setup their
3600 * underlying devices. In that case - do nothing, we're already set up from
3601 * our init.
3602 */
3603static int bond_neigh_setup(struct net_device *dev,
3604 struct neigh_parms *parms)
3605{
3606 /* modify only our neigh_parms */
3607 if (parms->dev == dev)
3608 parms->neigh_setup = bond_neigh_init;
3609
3610 return 0;
3611}
3612
3613/* Change the MTU of all of a master's slaves to match the master */
3614static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3615{
3616 struct bonding *bond = netdev_priv(bond_dev);
3617 struct slave *slave, *rollback_slave;
3618 struct list_head *iter;
3619 int res = 0;
3620
3621 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3622
3623 bond_for_each_slave(bond, slave, iter) {
3624 netdev_dbg(bond_dev, "s %p c_m %p\n",
3625 slave, slave->dev->netdev_ops->ndo_change_mtu);
3626
3627 res = dev_set_mtu(slave->dev, new_mtu);
3628
3629 if (res) {
3630 /* If we failed to set the slave's mtu to the new value
3631 * we must abort the operation even in ACTIVE_BACKUP
3632 * mode, because if we allow the backup slaves to have
3633 * different mtu values than the active slave we'll
3634 * need to change their mtu when doing a failover. That
3635 * means changing their mtu from timer context, which
3636 * is probably not a good idea.
3637 */
3638 netdev_dbg(bond_dev, "err %d %s\n", res,
3639 slave->dev->name);
3640 goto unwind;
3641 }
3642 }
3643
3644 bond_dev->mtu = new_mtu;
3645
3646 return 0;
3647
3648unwind:
3649 /* unwind from head to the slave that failed */
3650 bond_for_each_slave(bond, rollback_slave, iter) {
3651 int tmp_res;
3652
3653 if (rollback_slave == slave)
3654 break;
3655
3656 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3657 if (tmp_res) {
3658 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3659 tmp_res, rollback_slave->dev->name);
3660 }
3661 }
3662
3663 return res;
3664}
3665
3666/* Change HW address
3667 *
3668 * Note that many devices must be down to change the HW address, and
3669 * downing the master releases all slaves. We can make bonds full of
3670 * bonding devices to test this, however.
3671 */
3672static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3673{
3674 struct bonding *bond = netdev_priv(bond_dev);
3675 struct slave *slave, *rollback_slave;
3676 struct sockaddr_storage *ss = addr, tmp_ss;
3677 struct list_head *iter;
3678 int res = 0;
3679
3680 if (BOND_MODE(bond) == BOND_MODE_ALB)
3681 return bond_alb_set_mac_address(bond_dev, addr);
3682
3683
3684 netdev_dbg(bond_dev, "bond=%p\n", bond);
3685
3686 /* If fail_over_mac is enabled, do nothing and return success.
3687 * Returning an error causes ifenslave to fail.
3688 */
3689 if (bond->params.fail_over_mac &&
3690 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3691 return 0;
3692
3693 if (!is_valid_ether_addr(ss->__data))
3694 return -EADDRNOTAVAIL;
3695
3696 bond_for_each_slave(bond, slave, iter) {
3697 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3698 res = dev_set_mac_address(slave->dev, addr);
3699 if (res) {
3700 /* TODO: consider downing the slave
3701 * and retry ?
3702 * User should expect communications
3703 * breakage anyway until ARP finish
3704 * updating, so...
3705 */
3706 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3707 goto unwind;
3708 }
3709 }
3710
3711 /* success */
3712 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3713 return 0;
3714
3715unwind:
3716 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3717 tmp_ss.ss_family = bond_dev->type;
3718
3719 /* unwind from head to the slave that failed */
3720 bond_for_each_slave(bond, rollback_slave, iter) {
3721 int tmp_res;
3722
3723 if (rollback_slave == slave)
3724 break;
3725
3726 tmp_res = dev_set_mac_address(rollback_slave->dev,
3727 (struct sockaddr *)&tmp_ss);
3728 if (tmp_res) {
3729 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3730 tmp_res, rollback_slave->dev->name);
3731 }
3732 }
3733
3734 return res;
3735}
3736
3737/**
3738 * bond_xmit_slave_id - transmit skb through slave with slave_id
3739 * @bond: bonding device that is transmitting
3740 * @skb: buffer to transmit
3741 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3742 *
3743 * This function tries to transmit through slave with slave_id but in case
3744 * it fails, it tries to find the first available slave for transmission.
3745 * The skb is consumed in all cases, thus the function is void.
3746 */
3747static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3748{
3749 struct list_head *iter;
3750 struct slave *slave;
3751 int i = slave_id;
3752
3753 /* Here we start from the slave with slave_id */
3754 bond_for_each_slave_rcu(bond, slave, iter) {
3755 if (--i < 0) {
3756 if (bond_slave_can_tx(slave)) {
3757 bond_dev_queue_xmit(bond, skb, slave->dev);
3758 return;
3759 }
3760 }
3761 }
3762
3763 /* Here we start from the first slave up to slave_id */
3764 i = slave_id;
3765 bond_for_each_slave_rcu(bond, slave, iter) {
3766 if (--i < 0)
3767 break;
3768 if (bond_slave_can_tx(slave)) {
3769 bond_dev_queue_xmit(bond, skb, slave->dev);
3770 return;
3771 }
3772 }
3773 /* no slave that can tx has been found */
3774 bond_tx_drop(bond->dev, skb);
3775}
3776
3777/**
3778 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3779 * @bond: bonding device to use
3780 *
3781 * Based on the value of the bonding device's packets_per_slave parameter
3782 * this function generates a slave id, which is usually used as the next
3783 * slave to transmit through.
3784 */
3785static u32 bond_rr_gen_slave_id(struct bonding *bond)
3786{
3787 u32 slave_id;
3788 struct reciprocal_value reciprocal_packets_per_slave;
3789 int packets_per_slave = bond->params.packets_per_slave;
3790
3791 switch (packets_per_slave) {
3792 case 0:
3793 slave_id = prandom_u32();
3794 break;
3795 case 1:
3796 slave_id = bond->rr_tx_counter;
3797 break;
3798 default:
3799 reciprocal_packets_per_slave =
3800 bond->params.reciprocal_packets_per_slave;
3801 slave_id = reciprocal_divide(bond->rr_tx_counter,
3802 reciprocal_packets_per_slave);
3803 break;
3804 }
3805 bond->rr_tx_counter++;
3806
3807 return slave_id;
3808}
3809
3810static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3811{
3812 struct bonding *bond = netdev_priv(bond_dev);
3813 struct iphdr *iph = ip_hdr(skb);
3814 struct slave *slave;
3815 u32 slave_id;
3816
3817 /* Start with the curr_active_slave that joined the bond as the
3818 * default for sending IGMP traffic. For failover purposes one
3819 * needs to maintain some consistency for the interface that will
3820 * send the join/membership reports. The curr_active_slave found
3821 * will send all of this type of traffic.
3822 */
3823 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3824 slave = rcu_dereference(bond->curr_active_slave);
3825 if (slave)
3826 bond_dev_queue_xmit(bond, skb, slave->dev);
3827 else
3828 bond_xmit_slave_id(bond, skb, 0);
3829 } else {
3830 int slave_cnt = READ_ONCE(bond->slave_cnt);
3831
3832 if (likely(slave_cnt)) {
3833 slave_id = bond_rr_gen_slave_id(bond);
3834 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3835 } else {
3836 bond_tx_drop(bond_dev, skb);
3837 }
3838 }
3839
3840 return NETDEV_TX_OK;
3841}
3842
3843/* In active-backup mode, we know that bond->curr_active_slave is always valid if
3844 * the bond has a usable interface.
3845 */
3846static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3847{
3848 struct bonding *bond = netdev_priv(bond_dev);
3849 struct slave *slave;
3850
3851 slave = rcu_dereference(bond->curr_active_slave);
3852 if (slave)
3853 bond_dev_queue_xmit(bond, skb, slave->dev);
3854 else
3855 bond_tx_drop(bond_dev, skb);
3856
3857 return NETDEV_TX_OK;
3858}
3859
3860/* Use this to update slave_array when (a) it's not appropriate to update
3861 * slave_array right away (note that update_slave_array() may sleep)
3862 * and / or (b) RTNL is not held.
3863 */
3864void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3865{
3866 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3867}
3868
3869/* Slave array work handler. Holds only RTNL */
3870static void bond_slave_arr_handler(struct work_struct *work)
3871{
3872 struct bonding *bond = container_of(work, struct bonding,
3873 slave_arr_work.work);
3874 int ret;
3875
3876 if (!rtnl_trylock())
3877 goto err;
3878
3879 ret = bond_update_slave_arr(bond, NULL);
3880 rtnl_unlock();
3881 if (ret) {
3882 pr_warn_ratelimited("Failed to update slave array from WT\n");
3883 goto err;
3884 }
3885 return;
3886
3887err:
3888 bond_slave_arr_work_rearm(bond, 1);
3889}
3890
3891/* Build the usable slaves array in control path for modes that use xmit-hash
3892 * to determine the slave interface -
3893 * (a) BOND_MODE_8023AD
3894 * (b) BOND_MODE_XOR
3895 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3896 *
3897 * The caller is expected to hold RTNL only and NO other lock!
3898 */
3899int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3900{
3901 struct slave *slave;
3902 struct list_head *iter;
3903 struct bond_up_slave *new_arr, *old_arr;
3904 int agg_id = 0;
3905 int ret = 0;
3906
3907#ifdef CONFIG_LOCKDEP
3908 WARN_ON(lockdep_is_held(&bond->mode_lock));
3909#endif
3910
3911 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3912 GFP_KERNEL);
3913 if (!new_arr) {
3914 ret = -ENOMEM;
3915 pr_err("Failed to build slave-array.\n");
3916 goto out;
3917 }
3918 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3919 struct ad_info ad_info;
3920
3921 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3922 pr_debug("bond_3ad_get_active_agg_info failed\n");
3923 kfree_rcu(new_arr, rcu);
3924 /* No active aggragator means it's not safe to use
3925 * the previous array.
3926 */
3927 old_arr = rtnl_dereference(bond->slave_arr);
3928 if (old_arr) {
3929 RCU_INIT_POINTER(bond->slave_arr, NULL);
3930 kfree_rcu(old_arr, rcu);
3931 }
3932 goto out;
3933 }
3934 agg_id = ad_info.aggregator_id;
3935 }
3936 bond_for_each_slave(bond, slave, iter) {
3937 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3938 struct aggregator *agg;
3939
3940 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3941 if (!agg || agg->aggregator_identifier != agg_id)
3942 continue;
3943 }
3944 if (!bond_slave_can_tx(slave))
3945 continue;
3946 if (skipslave == slave)
3947 continue;
3948 new_arr->arr[new_arr->count++] = slave;
3949 }
3950
3951 old_arr = rtnl_dereference(bond->slave_arr);
3952 rcu_assign_pointer(bond->slave_arr, new_arr);
3953 if (old_arr)
3954 kfree_rcu(old_arr, rcu);
3955out:
3956 if (ret != 0 && skipslave) {
3957 int idx;
3958
3959 /* Rare situation where caller has asked to skip a specific
3960 * slave but allocation failed (most likely!). BTW this is
3961 * only possible when the call is initiated from
3962 * __bond_release_one(). In this situation; overwrite the
3963 * skipslave entry in the array with the last entry from the
3964 * array to avoid a situation where the xmit path may choose
3965 * this to-be-skipped slave to send a packet out.
3966 */
3967 old_arr = rtnl_dereference(bond->slave_arr);
3968 for (idx = 0; idx < old_arr->count; idx++) {
3969 if (skipslave == old_arr->arr[idx]) {
3970 old_arr->arr[idx] =
3971 old_arr->arr[old_arr->count-1];
3972 old_arr->count--;
3973 break;
3974 }
3975 }
3976 }
3977 return ret;
3978}
3979
3980/* Use this Xmit function for 3AD as well as XOR modes. The current
3981 * usable slave array is formed in the control path. The xmit function
3982 * just calculates hash and sends the packet out.
3983 */
3984static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3985{
3986 struct bonding *bond = netdev_priv(dev);
3987 struct slave *slave;
3988 struct bond_up_slave *slaves;
3989 unsigned int count;
3990
3991 slaves = rcu_dereference(bond->slave_arr);
3992 count = slaves ? READ_ONCE(slaves->count) : 0;
3993 if (likely(count)) {
3994 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3995 bond_dev_queue_xmit(bond, skb, slave->dev);
3996 } else {
3997 bond_tx_drop(dev, skb);
3998 }
3999
4000 return NETDEV_TX_OK;
4001}
4002
4003/* in broadcast mode, we send everything to all usable interfaces. */
4004static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4005{
4006 struct bonding *bond = netdev_priv(bond_dev);
4007 struct slave *slave = NULL;
4008 struct list_head *iter;
4009
4010 bond_for_each_slave_rcu(bond, slave, iter) {
4011 if (bond_is_last_slave(bond, slave))
4012 break;
4013 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4014 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4015
4016 if (!skb2) {
4017 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4018 bond_dev->name, __func__);
4019 continue;
4020 }
4021 bond_dev_queue_xmit(bond, skb2, slave->dev);
4022 }
4023 }
4024 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4025 bond_dev_queue_xmit(bond, skb, slave->dev);
4026 else
4027 bond_tx_drop(bond_dev, skb);
4028
4029 return NETDEV_TX_OK;
4030}
4031
4032/*------------------------- Device initialization ---------------------------*/
4033
4034/* Lookup the slave that corresponds to a qid */
4035static inline int bond_slave_override(struct bonding *bond,
4036 struct sk_buff *skb)
4037{
4038 struct slave *slave = NULL;
4039 struct list_head *iter;
4040
4041 if (!skb->queue_mapping)
4042 return 1;
4043
4044 /* Find out if any slaves have the same mapping as this skb. */
4045 bond_for_each_slave_rcu(bond, slave, iter) {
4046 if (slave->queue_id == skb->queue_mapping) {
4047 if (bond_slave_is_up(slave) &&
4048 slave->link == BOND_LINK_UP) {
4049 bond_dev_queue_xmit(bond, skb, slave->dev);
4050 return 0;
4051 }
4052 /* If the slave isn't UP, use default transmit policy. */
4053 break;
4054 }
4055 }
4056
4057 return 1;
4058}
4059
4060
4061static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4062 void *accel_priv, select_queue_fallback_t fallback)
4063{
4064 /* This helper function exists to help dev_pick_tx get the correct
4065 * destination queue. Using a helper function skips a call to
4066 * skb_tx_hash and will put the skbs in the queue we expect on their
4067 * way down to the bonding driver.
4068 */
4069 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4070
4071 /* Save the original txq to restore before passing to the driver */
4072 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4073
4074 if (unlikely(txq >= dev->real_num_tx_queues)) {
4075 do {
4076 txq -= dev->real_num_tx_queues;
4077 } while (txq >= dev->real_num_tx_queues);
4078 }
4079 return txq;
4080}
4081
4082static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4083{
4084 struct bonding *bond = netdev_priv(dev);
4085
4086 if (bond_should_override_tx_queue(bond) &&
4087 !bond_slave_override(bond, skb))
4088 return NETDEV_TX_OK;
4089
4090 switch (BOND_MODE(bond)) {
4091 case BOND_MODE_ROUNDROBIN:
4092 return bond_xmit_roundrobin(skb, dev);
4093 case BOND_MODE_ACTIVEBACKUP:
4094 return bond_xmit_activebackup(skb, dev);
4095 case BOND_MODE_8023AD:
4096 case BOND_MODE_XOR:
4097 return bond_3ad_xor_xmit(skb, dev);
4098 case BOND_MODE_BROADCAST:
4099 return bond_xmit_broadcast(skb, dev);
4100 case BOND_MODE_ALB:
4101 return bond_alb_xmit(skb, dev);
4102 case BOND_MODE_TLB:
4103 return bond_tlb_xmit(skb, dev);
4104 default:
4105 /* Should never happen, mode already checked */
4106 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4107 WARN_ON_ONCE(1);
4108 bond_tx_drop(dev, skb);
4109 return NETDEV_TX_OK;
4110 }
4111}
4112
4113static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4114{
4115 struct bonding *bond = netdev_priv(dev);
4116 netdev_tx_t ret = NETDEV_TX_OK;
4117
4118 /* If we risk deadlock from transmitting this in the
4119 * netpoll path, tell netpoll to queue the frame for later tx
4120 */
4121 if (unlikely(is_netpoll_tx_blocked(dev)))
4122 return NETDEV_TX_BUSY;
4123
4124 rcu_read_lock();
4125 if (bond_has_slaves(bond))
4126 ret = __bond_start_xmit(skb, dev);
4127 else
4128 bond_tx_drop(dev, skb);
4129 rcu_read_unlock();
4130
4131 return ret;
4132}
4133
4134static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4135 struct ethtool_link_ksettings *cmd)
4136{
4137 struct bonding *bond = netdev_priv(bond_dev);
4138 unsigned long speed = 0;
4139 struct list_head *iter;
4140 struct slave *slave;
4141
4142 cmd->base.duplex = DUPLEX_UNKNOWN;
4143 cmd->base.port = PORT_OTHER;
4144
4145 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4146 * do not need to check mode. Though link speed might not represent
4147 * the true receive or transmit bandwidth (not all modes are symmetric)
4148 * this is an accurate maximum.
4149 */
4150 bond_for_each_slave(bond, slave, iter) {
4151 if (bond_slave_can_tx(slave)) {
4152 if (slave->speed != SPEED_UNKNOWN)
4153 speed += slave->speed;
4154 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4155 slave->duplex != DUPLEX_UNKNOWN)
4156 cmd->base.duplex = slave->duplex;
4157 }
4158 }
4159 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4160
4161 return 0;
4162}
4163
4164static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4165 struct ethtool_drvinfo *drvinfo)
4166{
4167 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4168 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4169 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4170 BOND_ABI_VERSION);
4171}
4172
4173static const struct ethtool_ops bond_ethtool_ops = {
4174 .get_drvinfo = bond_ethtool_get_drvinfo,
4175 .get_link = ethtool_op_get_link,
4176 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4177};
4178
4179static const struct net_device_ops bond_netdev_ops = {
4180 .ndo_init = bond_init,
4181 .ndo_uninit = bond_uninit,
4182 .ndo_open = bond_open,
4183 .ndo_stop = bond_close,
4184 .ndo_start_xmit = bond_start_xmit,
4185 .ndo_select_queue = bond_select_queue,
4186 .ndo_get_stats64 = bond_get_stats,
4187 .ndo_do_ioctl = bond_do_ioctl,
4188 .ndo_change_rx_flags = bond_change_rx_flags,
4189 .ndo_set_rx_mode = bond_set_rx_mode,
4190 .ndo_change_mtu = bond_change_mtu,
4191 .ndo_set_mac_address = bond_set_mac_address,
4192 .ndo_neigh_setup = bond_neigh_setup,
4193 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4194 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4195#ifdef CONFIG_NET_POLL_CONTROLLER
4196 .ndo_netpoll_setup = bond_netpoll_setup,
4197 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4198 .ndo_poll_controller = bond_poll_controller,
4199#endif
4200 .ndo_add_slave = bond_enslave,
4201 .ndo_del_slave = bond_release,
4202 .ndo_fix_features = bond_fix_features,
4203 .ndo_features_check = passthru_features_check,
4204};
4205
4206static const struct device_type bond_type = {
4207 .name = "bond",
4208};
4209
4210static void bond_destructor(struct net_device *bond_dev)
4211{
4212 struct bonding *bond = netdev_priv(bond_dev);
4213 if (bond->wq)
4214 destroy_workqueue(bond->wq);
4215}
4216
4217void bond_setup(struct net_device *bond_dev)
4218{
4219 struct bonding *bond = netdev_priv(bond_dev);
4220
4221 spin_lock_init(&bond->mode_lock);
4222 spin_lock_init(&bond->stats_lock);
4223 bond->params = bonding_defaults;
4224
4225 /* Initialize pointers */
4226 bond->dev = bond_dev;
4227
4228 /* Initialize the device entry points */
4229 ether_setup(bond_dev);
4230 bond_dev->max_mtu = ETH_MAX_MTU;
4231 bond_dev->netdev_ops = &bond_netdev_ops;
4232 bond_dev->ethtool_ops = &bond_ethtool_ops;
4233
4234 bond_dev->needs_free_netdev = true;
4235 bond_dev->priv_destructor = bond_destructor;
4236
4237 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4238
4239 /* Initialize the device options */
4240 bond_dev->flags |= IFF_MASTER;
4241 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4242 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4243
4244 /* don't acquire bond device's netif_tx_lock when transmitting */
4245 bond_dev->features |= NETIF_F_LLTX;
4246
4247 /* By default, we declare the bond to be fully
4248 * VLAN hardware accelerated capable. Special
4249 * care is taken in the various xmit functions
4250 * when there are slaves that are not hw accel
4251 * capable
4252 */
4253
4254 /* Don't allow bond devices to change network namespaces. */
4255 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4256
4257 bond_dev->hw_features = BOND_VLAN_FEATURES |
4258 NETIF_F_HW_VLAN_CTAG_TX |
4259 NETIF_F_HW_VLAN_CTAG_RX |
4260 NETIF_F_HW_VLAN_CTAG_FILTER;
4261
4262 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4263 bond_dev->features |= bond_dev->hw_features;
4264}
4265
4266/* Destroy a bonding device.
4267 * Must be under rtnl_lock when this function is called.
4268 */
4269static void bond_uninit(struct net_device *bond_dev)
4270{
4271 struct bonding *bond = netdev_priv(bond_dev);
4272 struct list_head *iter;
4273 struct slave *slave;
4274 struct bond_up_slave *arr;
4275
4276 bond_netpoll_cleanup(bond_dev);
4277
4278 /* Release the bonded slaves */
4279 bond_for_each_slave(bond, slave, iter)
4280 __bond_release_one(bond_dev, slave->dev, true, true);
4281 netdev_info(bond_dev, "Released all slaves\n");
4282
4283 arr = rtnl_dereference(bond->slave_arr);
4284 if (arr) {
4285 RCU_INIT_POINTER(bond->slave_arr, NULL);
4286 kfree_rcu(arr, rcu);
4287 }
4288
4289 list_del(&bond->bond_list);
4290
4291 bond_debug_unregister(bond);
4292}
4293
4294/*------------------------- Module initialization ---------------------------*/
4295
4296static int bond_check_params(struct bond_params *params)
4297{
4298 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4299 struct bond_opt_value newval;
4300 const struct bond_opt_value *valptr;
4301 int arp_all_targets_value = 0;
4302 u16 ad_actor_sys_prio = 0;
4303 u16 ad_user_port_key = 0;
4304 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4305 int arp_ip_count;
4306 int bond_mode = BOND_MODE_ROUNDROBIN;
4307 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4308 int lacp_fast = 0;
4309 int tlb_dynamic_lb;
4310
4311 /* Convert string parameters. */
4312 if (mode) {
4313 bond_opt_initstr(&newval, mode);
4314 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4315 if (!valptr) {
4316 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4317 return -EINVAL;
4318 }
4319 bond_mode = valptr->value;
4320 }
4321
4322 if (xmit_hash_policy) {
4323 if ((bond_mode != BOND_MODE_XOR) &&
4324 (bond_mode != BOND_MODE_8023AD) &&
4325 (bond_mode != BOND_MODE_TLB)) {
4326 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4327 bond_mode_name(bond_mode));
4328 } else {
4329 bond_opt_initstr(&newval, xmit_hash_policy);
4330 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4331 &newval);
4332 if (!valptr) {
4333 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4334 xmit_hash_policy);
4335 return -EINVAL;
4336 }
4337 xmit_hashtype = valptr->value;
4338 }
4339 }
4340
4341 if (lacp_rate) {
4342 if (bond_mode != BOND_MODE_8023AD) {
4343 pr_info("lacp_rate param is irrelevant in mode %s\n",
4344 bond_mode_name(bond_mode));
4345 } else {
4346 bond_opt_initstr(&newval, lacp_rate);
4347 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4348 &newval);
4349 if (!valptr) {
4350 pr_err("Error: Invalid lacp rate \"%s\"\n",
4351 lacp_rate);
4352 return -EINVAL;
4353 }
4354 lacp_fast = valptr->value;
4355 }
4356 }
4357
4358 if (ad_select) {
4359 bond_opt_initstr(&newval, ad_select);
4360 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4361 &newval);
4362 if (!valptr) {
4363 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4364 return -EINVAL;
4365 }
4366 params->ad_select = valptr->value;
4367 if (bond_mode != BOND_MODE_8023AD)
4368 pr_warn("ad_select param only affects 802.3ad mode\n");
4369 } else {
4370 params->ad_select = BOND_AD_STABLE;
4371 }
4372
4373 if (max_bonds < 0) {
4374 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4375 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4376 max_bonds = BOND_DEFAULT_MAX_BONDS;
4377 }
4378
4379 if (miimon < 0) {
4380 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4381 miimon, INT_MAX);
4382 miimon = 0;
4383 }
4384
4385 if (updelay < 0) {
4386 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4387 updelay, INT_MAX);
4388 updelay = 0;
4389 }
4390
4391 if (downdelay < 0) {
4392 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4393 downdelay, INT_MAX);
4394 downdelay = 0;
4395 }
4396
4397 if ((use_carrier != 0) && (use_carrier != 1)) {
4398 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4399 use_carrier);
4400 use_carrier = 1;
4401 }
4402
4403 if (num_peer_notif < 0 || num_peer_notif > 255) {
4404 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4405 num_peer_notif);
4406 num_peer_notif = 1;
4407 }
4408
4409 /* reset values for 802.3ad/TLB/ALB */
4410 if (!bond_mode_uses_arp(bond_mode)) {
4411 if (!miimon) {
4412 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");
4413 pr_warn("Forcing miimon to 100msec\n");
4414 miimon = BOND_DEFAULT_MIIMON;
4415 }
4416 }
4417
4418 if (tx_queues < 1 || tx_queues > 255) {
4419 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4420 tx_queues, BOND_DEFAULT_TX_QUEUES);
4421 tx_queues = BOND_DEFAULT_TX_QUEUES;
4422 }
4423
4424 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4425 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4426 all_slaves_active);
4427 all_slaves_active = 0;
4428 }
4429
4430 if (resend_igmp < 0 || resend_igmp > 255) {
4431 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4432 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4433 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4434 }
4435
4436 bond_opt_initval(&newval, packets_per_slave);
4437 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4438 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4439 packets_per_slave, USHRT_MAX);
4440 packets_per_slave = 1;
4441 }
4442
4443 if (bond_mode == BOND_MODE_ALB) {
4444 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",
4445 updelay);
4446 }
4447
4448 if (!miimon) {
4449 if (updelay || downdelay) {
4450 /* just warn the user the up/down delay will have
4451 * no effect since miimon is zero...
4452 */
4453 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",
4454 updelay, downdelay);
4455 }
4456 } else {
4457 /* don't allow arp monitoring */
4458 if (arp_interval) {
4459 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4460 miimon, arp_interval);
4461 arp_interval = 0;
4462 }
4463
4464 if ((updelay % miimon) != 0) {
4465 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4466 updelay, miimon, (updelay / miimon) * miimon);
4467 }
4468
4469 updelay /= miimon;
4470
4471 if ((downdelay % miimon) != 0) {
4472 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4473 downdelay, miimon,
4474 (downdelay / miimon) * miimon);
4475 }
4476
4477 downdelay /= miimon;
4478 }
4479
4480 if (arp_interval < 0) {
4481 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4482 arp_interval, INT_MAX);
4483 arp_interval = 0;
4484 }
4485
4486 for (arp_ip_count = 0, i = 0;
4487 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4488 __be32 ip;
4489
4490 /* not a complete check, but good enough to catch mistakes */
4491 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4492 !bond_is_ip_target_ok(ip)) {
4493 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4494 arp_ip_target[i]);
4495 arp_interval = 0;
4496 } else {
4497 if (bond_get_targets_ip(arp_target, ip) == -1)
4498 arp_target[arp_ip_count++] = ip;
4499 else
4500 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4501 &ip);
4502 }
4503 }
4504
4505 if (arp_interval && !arp_ip_count) {
4506 /* don't allow arping if no arp_ip_target given... */
4507 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4508 arp_interval);
4509 arp_interval = 0;
4510 }
4511
4512 if (arp_validate) {
4513 if (!arp_interval) {
4514 pr_err("arp_validate requires arp_interval\n");
4515 return -EINVAL;
4516 }
4517
4518 bond_opt_initstr(&newval, arp_validate);
4519 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4520 &newval);
4521 if (!valptr) {
4522 pr_err("Error: invalid arp_validate \"%s\"\n",
4523 arp_validate);
4524 return -EINVAL;
4525 }
4526 arp_validate_value = valptr->value;
4527 } else {
4528 arp_validate_value = 0;
4529 }
4530
4531 if (arp_all_targets) {
4532 bond_opt_initstr(&newval, arp_all_targets);
4533 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4534 &newval);
4535 if (!valptr) {
4536 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4537 arp_all_targets);
4538 arp_all_targets_value = 0;
4539 } else {
4540 arp_all_targets_value = valptr->value;
4541 }
4542 }
4543
4544 if (miimon) {
4545 pr_info("MII link monitoring set to %d ms\n", miimon);
4546 } else if (arp_interval) {
4547 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4548 arp_validate_value);
4549 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4550 arp_interval, valptr->string, arp_ip_count);
4551
4552 for (i = 0; i < arp_ip_count; i++)
4553 pr_cont(" %s", arp_ip_target[i]);
4554
4555 pr_cont("\n");
4556
4557 } else if (max_bonds) {
4558 /* miimon and arp_interval not set, we need one so things
4559 * work as expected, see bonding.txt for details
4560 */
4561 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");
4562 }
4563
4564 if (primary && !bond_mode_uses_primary(bond_mode)) {
4565 /* currently, using a primary only makes sense
4566 * in active backup, TLB or ALB modes
4567 */
4568 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4569 primary, bond_mode_name(bond_mode));
4570 primary = NULL;
4571 }
4572
4573 if (primary && primary_reselect) {
4574 bond_opt_initstr(&newval, primary_reselect);
4575 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4576 &newval);
4577 if (!valptr) {
4578 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4579 primary_reselect);
4580 return -EINVAL;
4581 }
4582 primary_reselect_value = valptr->value;
4583 } else {
4584 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4585 }
4586
4587 if (fail_over_mac) {
4588 bond_opt_initstr(&newval, fail_over_mac);
4589 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4590 &newval);
4591 if (!valptr) {
4592 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4593 fail_over_mac);
4594 return -EINVAL;
4595 }
4596 fail_over_mac_value = valptr->value;
4597 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4598 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4599 } else {
4600 fail_over_mac_value = BOND_FOM_NONE;
4601 }
4602
4603 bond_opt_initstr(&newval, "default");
4604 valptr = bond_opt_parse(
4605 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4606 &newval);
4607 if (!valptr) {
4608 pr_err("Error: No ad_actor_sys_prio default value");
4609 return -EINVAL;
4610 }
4611 ad_actor_sys_prio = valptr->value;
4612
4613 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4614 &newval);
4615 if (!valptr) {
4616 pr_err("Error: No ad_user_port_key default value");
4617 return -EINVAL;
4618 }
4619 ad_user_port_key = valptr->value;
4620
4621 bond_opt_initstr(&newval, "default");
4622 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4623 if (!valptr) {
4624 pr_err("Error: No tlb_dynamic_lb default value");
4625 return -EINVAL;
4626 }
4627 tlb_dynamic_lb = valptr->value;
4628
4629 if (lp_interval == 0) {
4630 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4631 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4632 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4633 }
4634
4635 /* fill params struct with the proper values */
4636 params->mode = bond_mode;
4637 params->xmit_policy = xmit_hashtype;
4638 params->miimon = miimon;
4639 params->num_peer_notif = num_peer_notif;
4640 params->arp_interval = arp_interval;
4641 params->arp_validate = arp_validate_value;
4642 params->arp_all_targets = arp_all_targets_value;
4643 params->updelay = updelay;
4644 params->downdelay = downdelay;
4645 params->use_carrier = use_carrier;
4646 params->lacp_fast = lacp_fast;
4647 params->primary[0] = 0;
4648 params->primary_reselect = primary_reselect_value;
4649 params->fail_over_mac = fail_over_mac_value;
4650 params->tx_queues = tx_queues;
4651 params->all_slaves_active = all_slaves_active;
4652 params->resend_igmp = resend_igmp;
4653 params->min_links = min_links;
4654 params->lp_interval = lp_interval;
4655 params->packets_per_slave = packets_per_slave;
4656 params->tlb_dynamic_lb = tlb_dynamic_lb;
4657 params->ad_actor_sys_prio = ad_actor_sys_prio;
4658 eth_zero_addr(params->ad_actor_system);
4659 params->ad_user_port_key = ad_user_port_key;
4660 if (packets_per_slave > 0) {
4661 params->reciprocal_packets_per_slave =
4662 reciprocal_value(packets_per_slave);
4663 } else {
4664 /* reciprocal_packets_per_slave is unused if
4665 * packets_per_slave is 0 or 1, just initialize it
4666 */
4667 params->reciprocal_packets_per_slave =
4668 (struct reciprocal_value) { 0 };
4669 }
4670
4671 if (primary) {
4672 strncpy(params->primary, primary, IFNAMSIZ);
4673 params->primary[IFNAMSIZ - 1] = 0;
4674 }
4675
4676 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4677
4678 return 0;
4679}
4680
4681/* Called from registration process */
4682static int bond_init(struct net_device *bond_dev)
4683{
4684 struct bonding *bond = netdev_priv(bond_dev);
4685 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4686
4687 netdev_dbg(bond_dev, "Begin bond_init\n");
4688
4689 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4690 if (!bond->wq)
4691 return -ENOMEM;
4692
4693 netdev_lockdep_set_classes(bond_dev);
4694
4695 list_add_tail(&bond->bond_list, &bn->dev_list);
4696
4697 bond_prepare_sysfs_group(bond);
4698
4699 bond_debug_register(bond);
4700
4701 /* Ensure valid dev_addr */
4702 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4703 bond_dev->addr_assign_type == NET_ADDR_PERM)
4704 eth_hw_addr_random(bond_dev);
4705
4706 return 0;
4707}
4708
4709unsigned int bond_get_num_tx_queues(void)
4710{
4711 return tx_queues;
4712}
4713
4714/* Create a new bond based on the specified name and bonding parameters.
4715 * If name is NULL, obtain a suitable "bond%d" name for us.
4716 * Caller must NOT hold rtnl_lock; we need to release it here before we
4717 * set up our sysfs entries.
4718 */
4719int bond_create(struct net *net, const char *name)
4720{
4721 struct net_device *bond_dev;
4722 struct bonding *bond;
4723 struct alb_bond_info *bond_info;
4724 int res;
4725
4726 rtnl_lock();
4727
4728 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4729 name ? name : "bond%d", NET_NAME_UNKNOWN,
4730 bond_setup, tx_queues);
4731 if (!bond_dev) {
4732 pr_err("%s: eek! can't alloc netdev!\n", name);
4733 rtnl_unlock();
4734 return -ENOMEM;
4735 }
4736
4737 /*
4738 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4739 * It is set to 0 by default which is wrong.
4740 */
4741 bond = netdev_priv(bond_dev);
4742 bond_info = &(BOND_ALB_INFO(bond));
4743 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4744
4745 dev_net_set(bond_dev, net);
4746 bond_dev->rtnl_link_ops = &bond_link_ops;
4747
4748 res = register_netdevice(bond_dev);
4749
4750 netif_carrier_off(bond_dev);
4751
4752 bond_work_init_all(bond);
4753
4754 rtnl_unlock();
4755 if (res < 0)
4756 free_netdev(bond_dev);
4757 return res;
4758}
4759
4760static int __net_init bond_net_init(struct net *net)
4761{
4762 struct bond_net *bn = net_generic(net, bond_net_id);
4763
4764 bn->net = net;
4765 INIT_LIST_HEAD(&bn->dev_list);
4766
4767 bond_create_proc_dir(bn);
4768 bond_create_sysfs(bn);
4769
4770 return 0;
4771}
4772
4773static void __net_exit bond_net_exit(struct net *net)
4774{
4775 struct bond_net *bn = net_generic(net, bond_net_id);
4776 struct bonding *bond, *tmp_bond;
4777 LIST_HEAD(list);
4778
4779 bond_destroy_sysfs(bn);
4780
4781 /* Kill off any bonds created after unregistering bond rtnl ops */
4782 rtnl_lock();
4783 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4784 unregister_netdevice_queue(bond->dev, &list);
4785 unregister_netdevice_many(&list);
4786 rtnl_unlock();
4787
4788 bond_destroy_proc_dir(bn);
4789}
4790
4791static struct pernet_operations bond_net_ops = {
4792 .init = bond_net_init,
4793 .exit = bond_net_exit,
4794 .id = &bond_net_id,
4795 .size = sizeof(struct bond_net),
4796};
4797
4798static int __init bonding_init(void)
4799{
4800 int i;
4801 int res;
4802
4803 pr_info("%s", bond_version);
4804
4805 res = bond_check_params(&bonding_defaults);
4806 if (res)
4807 goto out;
4808
4809 res = register_pernet_subsys(&bond_net_ops);
4810 if (res)
4811 goto out;
4812
4813 res = bond_netlink_init();
4814 if (res)
4815 goto err_link;
4816
4817 bond_create_debugfs();
4818
4819 for (i = 0; i < max_bonds; i++) {
4820 res = bond_create(&init_net, NULL);
4821 if (res)
4822 goto err;
4823 }
4824
4825 register_netdevice_notifier(&bond_netdev_notifier);
4826out:
4827 return res;
4828err:
4829 bond_destroy_debugfs();
4830 bond_netlink_fini();
4831err_link:
4832 unregister_pernet_subsys(&bond_net_ops);
4833 goto out;
4834
4835}
4836
4837static void __exit bonding_exit(void)
4838{
4839 unregister_netdevice_notifier(&bond_netdev_notifier);
4840
4841 bond_destroy_debugfs();
4842
4843 bond_netlink_fini();
4844 unregister_pernet_subsys(&bond_net_ops);
4845
4846#ifdef CONFIG_NET_POLL_CONTROLLER
4847 /* Make sure we don't have an imbalance on our netpoll blocking */
4848 WARN_ON(atomic_read(&netpoll_block_tx));
4849#endif
4850}
4851
4852module_init(bonding_init);
4853module_exit(bonding_exit);
4854MODULE_LICENSE("GPL");
4855MODULE_VERSION(DRV_VERSION);
4856MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4857MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
1// SPDX-License-Identifier: GPL-1.0+
2/*
3 * originally based on the dummy device.
4 *
5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
6 * Based on dummy.c, and eql.c devices.
7 *
8 * bonding.c: an Ethernet Bonding driver
9 *
10 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Cisco 5500
12 * Sun Trunking (Solaris)
13 * Alteon AceDirector Trunks
14 * Linux Bonding
15 * and probably many L2 switches ...
16 *
17 * How it works:
18 * ifconfig bond0 ipaddress netmask up
19 * will setup a network device, with an ip address. No mac address
20 * will be assigned at this time. The hw mac address will come from
21 * the first slave bonded to the channel. All slaves will then use
22 * this hw mac address.
23 *
24 * ifconfig bond0 down
25 * will release all slaves, marking them as down.
26 *
27 * ifenslave bond0 eth0
28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
29 * a: be used as initial mac address
30 * b: if a hw mac address already is there, eth0's hw mac address
31 * will then be set from bond0.
32 *
33 */
34
35#include <linux/kernel.h>
36#include <linux/module.h>
37#include <linux/types.h>
38#include <linux/fcntl.h>
39#include <linux/filter.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
44#include <net/ip.h>
45#include <linux/ip.h>
46#include <linux/icmp.h>
47#include <linux/icmpv6.h>
48#include <linux/tcp.h>
49#include <linux/udp.h>
50#include <linux/slab.h>
51#include <linux/string.h>
52#include <linux/init.h>
53#include <linux/timer.h>
54#include <linux/socket.h>
55#include <linux/ctype.h>
56#include <linux/inet.h>
57#include <linux/bitops.h>
58#include <linux/io.h>
59#include <asm/dma.h>
60#include <linux/uaccess.h>
61#include <linux/errno.h>
62#include <linux/netdevice.h>
63#include <linux/inetdevice.h>
64#include <linux/igmp.h>
65#include <linux/etherdevice.h>
66#include <linux/skbuff.h>
67#include <net/sock.h>
68#include <linux/rtnetlink.h>
69#include <linux/smp.h>
70#include <linux/if_ether.h>
71#include <net/arp.h>
72#include <linux/mii.h>
73#include <linux/ethtool.h>
74#include <linux/if_vlan.h>
75#include <linux/if_bonding.h>
76#include <linux/phy.h>
77#include <linux/jiffies.h>
78#include <linux/preempt.h>
79#include <net/route.h>
80#include <net/net_namespace.h>
81#include <net/netns/generic.h>
82#include <net/pkt_sched.h>
83#include <linux/rculist.h>
84#include <net/flow_dissector.h>
85#include <net/xfrm.h>
86#include <net/bonding.h>
87#include <net/bond_3ad.h>
88#include <net/bond_alb.h>
89#if IS_ENABLED(CONFIG_TLS_DEVICE)
90#include <net/tls.h>
91#endif
92#include <net/ip6_route.h>
93#include <net/xdp.h>
94
95#include "bonding_priv.h"
96
97/*---------------------------- Module parameters ----------------------------*/
98
99/* monitor all links that often (in milliseconds). <=0 disables monitoring */
100
101static int max_bonds = BOND_DEFAULT_MAX_BONDS;
102static int tx_queues = BOND_DEFAULT_TX_QUEUES;
103static int num_peer_notif = 1;
104static int miimon;
105static int updelay;
106static int downdelay;
107static int use_carrier = 1;
108static char *mode;
109static char *primary;
110static char *primary_reselect;
111static char *lacp_rate;
112static int min_links;
113static char *ad_select;
114static char *xmit_hash_policy;
115static int arp_interval;
116static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
117static char *arp_validate;
118static char *arp_all_targets;
119static char *fail_over_mac;
120static int all_slaves_active;
121static struct bond_params bonding_defaults;
122static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
123static int packets_per_slave = 1;
124static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
125
126module_param(max_bonds, int, 0);
127MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
128module_param(tx_queues, int, 0);
129MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
130module_param_named(num_grat_arp, num_peer_notif, int, 0644);
131MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
132 "failover event (alias of num_unsol_na)");
133module_param_named(num_unsol_na, num_peer_notif, int, 0644);
134MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
135 "failover event (alias of num_grat_arp)");
136module_param(miimon, int, 0);
137MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
138module_param(updelay, int, 0);
139MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
140module_param(downdelay, int, 0);
141MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
142 "in milliseconds");
143module_param(use_carrier, int, 0);
144MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
145 "0 for off, 1 for on (default)");
146module_param(mode, charp, 0);
147MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
148 "1 for active-backup, 2 for balance-xor, "
149 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
150 "6 for balance-alb");
151module_param(primary, charp, 0);
152MODULE_PARM_DESC(primary, "Primary network device to use");
153module_param(primary_reselect, charp, 0);
154MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
155 "once it comes up; "
156 "0 for always (default), "
157 "1 for only if speed of primary is "
158 "better, "
159 "2 for only on active slave "
160 "failure");
161module_param(lacp_rate, charp, 0);
162MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163 "0 for slow, 1 for fast");
164module_param(ad_select, charp, 0);
165MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166 "0 for stable (default), 1 for bandwidth, "
167 "2 for count");
168module_param(min_links, int, 0);
169MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
170
171module_param(xmit_hash_policy, charp, 0);
172MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
173 "0 for layer 2 (default), 1 for layer 3+4, "
174 "2 for layer 2+3, 3 for encap layer 2+3, "
175 "4 for encap layer 3+4, 5 for vlan+srcmac");
176module_param(arp_interval, int, 0);
177MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178module_param_array(arp_ip_target, charp, NULL, 0);
179MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180module_param(arp_validate, charp, 0);
181MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
182 "0 for none (default), 1 for active, "
183 "2 for backup, 3 for all");
184module_param(arp_all_targets, charp, 0);
185MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186module_param(fail_over_mac, charp, 0);
187MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
188 "the same MAC; 0 for none (default), "
189 "1 for active, 2 for follow");
190module_param(all_slaves_active, int, 0);
191MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
192 "by setting active flag for all slaves; "
193 "0 for never (default), 1 for always.");
194module_param(resend_igmp, int, 0);
195MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
196 "link failure");
197module_param(packets_per_slave, int, 0);
198MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
199 "mode; 0 for a random slave, 1 packet per "
200 "slave (default), >1 packets per slave.");
201module_param(lp_interval, uint, 0);
202MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
203 "the bonding driver sends learning packets to "
204 "each slaves peer switch. The default is 1.");
205
206/*----------------------------- Global variables ----------------------------*/
207
208#ifdef CONFIG_NET_POLL_CONTROLLER
209atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210#endif
211
212unsigned int bond_net_id __read_mostly;
213
214static const struct flow_dissector_key flow_keys_bonding_keys[] = {
215 {
216 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
217 .offset = offsetof(struct flow_keys, control),
218 },
219 {
220 .key_id = FLOW_DISSECTOR_KEY_BASIC,
221 .offset = offsetof(struct flow_keys, basic),
222 },
223 {
224 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
225 .offset = offsetof(struct flow_keys, addrs.v4addrs),
226 },
227 {
228 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
229 .offset = offsetof(struct flow_keys, addrs.v6addrs),
230 },
231 {
232 .key_id = FLOW_DISSECTOR_KEY_TIPC,
233 .offset = offsetof(struct flow_keys, addrs.tipckey),
234 },
235 {
236 .key_id = FLOW_DISSECTOR_KEY_PORTS,
237 .offset = offsetof(struct flow_keys, ports),
238 },
239 {
240 .key_id = FLOW_DISSECTOR_KEY_ICMP,
241 .offset = offsetof(struct flow_keys, icmp),
242 },
243 {
244 .key_id = FLOW_DISSECTOR_KEY_VLAN,
245 .offset = offsetof(struct flow_keys, vlan),
246 },
247 {
248 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
249 .offset = offsetof(struct flow_keys, tags),
250 },
251 {
252 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
253 .offset = offsetof(struct flow_keys, keyid),
254 },
255};
256
257static struct flow_dissector flow_keys_bonding __read_mostly;
258
259/*-------------------------- Forward declarations ---------------------------*/
260
261static int bond_init(struct net_device *bond_dev);
262static void bond_uninit(struct net_device *bond_dev);
263static void bond_get_stats(struct net_device *bond_dev,
264 struct rtnl_link_stats64 *stats);
265static void bond_slave_arr_handler(struct work_struct *work);
266static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
267 int mod);
268static void bond_netdev_notify_work(struct work_struct *work);
269
270/*---------------------------- General routines -----------------------------*/
271
272const char *bond_mode_name(int mode)
273{
274 static const char *names[] = {
275 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
276 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
277 [BOND_MODE_XOR] = "load balancing (xor)",
278 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
279 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
280 [BOND_MODE_TLB] = "transmit load balancing",
281 [BOND_MODE_ALB] = "adaptive load balancing",
282 };
283
284 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
285 return "unknown";
286
287 return names[mode];
288}
289
290/**
291 * bond_dev_queue_xmit - Prepare skb for xmit.
292 *
293 * @bond: bond device that got this skb for tx.
294 * @skb: hw accel VLAN tagged skb to transmit
295 * @slave_dev: slave that is supposed to xmit this skbuff
296 */
297netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
298 struct net_device *slave_dev)
299{
300 skb->dev = slave_dev;
301
302 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
303 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
304 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
305
306 if (unlikely(netpoll_tx_running(bond->dev)))
307 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
308
309 return dev_queue_xmit(skb);
310}
311
312static bool bond_sk_check(struct bonding *bond)
313{
314 switch (BOND_MODE(bond)) {
315 case BOND_MODE_8023AD:
316 case BOND_MODE_XOR:
317 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
318 return true;
319 fallthrough;
320 default:
321 return false;
322 }
323}
324
325static bool bond_xdp_check(struct bonding *bond)
326{
327 switch (BOND_MODE(bond)) {
328 case BOND_MODE_ROUNDROBIN:
329 case BOND_MODE_ACTIVEBACKUP:
330 return true;
331 case BOND_MODE_8023AD:
332 case BOND_MODE_XOR:
333 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
334 * payload is not in the packet due to hardware offload.
335 */
336 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
337 return true;
338 fallthrough;
339 default:
340 return false;
341 }
342}
343
344/*---------------------------------- VLAN -----------------------------------*/
345
346/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
347 * We don't protect the slave list iteration with a lock because:
348 * a. This operation is performed in IOCTL context,
349 * b. The operation is protected by the RTNL semaphore in the 8021q code,
350 * c. Holding a lock with BH disabled while directly calling a base driver
351 * entry point is generally a BAD idea.
352 *
353 * The design of synchronization/protection for this operation in the 8021q
354 * module is good for one or more VLAN devices over a single physical device
355 * and cannot be extended for a teaming solution like bonding, so there is a
356 * potential race condition here where a net device from the vlan group might
357 * be referenced (either by a base driver or the 8021q code) while it is being
358 * removed from the system. However, it turns out we're not making matters
359 * worse, and if it works for regular VLAN usage it will work here too.
360*/
361
362/**
363 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
364 * @bond_dev: bonding net device that got called
365 * @proto: network protocol ID
366 * @vid: vlan id being added
367 */
368static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
369 __be16 proto, u16 vid)
370{
371 struct bonding *bond = netdev_priv(bond_dev);
372 struct slave *slave, *rollback_slave;
373 struct list_head *iter;
374 int res;
375
376 bond_for_each_slave(bond, slave, iter) {
377 res = vlan_vid_add(slave->dev, proto, vid);
378 if (res)
379 goto unwind;
380 }
381
382 return 0;
383
384unwind:
385 /* unwind to the slave that failed */
386 bond_for_each_slave(bond, rollback_slave, iter) {
387 if (rollback_slave == slave)
388 break;
389
390 vlan_vid_del(rollback_slave->dev, proto, vid);
391 }
392
393 return res;
394}
395
396/**
397 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
398 * @bond_dev: bonding net device that got called
399 * @proto: network protocol ID
400 * @vid: vlan id being removed
401 */
402static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
403 __be16 proto, u16 vid)
404{
405 struct bonding *bond = netdev_priv(bond_dev);
406 struct list_head *iter;
407 struct slave *slave;
408
409 bond_for_each_slave(bond, slave, iter)
410 vlan_vid_del(slave->dev, proto, vid);
411
412 if (bond_is_lb(bond))
413 bond_alb_clear_vlan(bond, vid);
414
415 return 0;
416}
417
418/*---------------------------------- XFRM -----------------------------------*/
419
420#ifdef CONFIG_XFRM_OFFLOAD
421/**
422 * bond_ipsec_dev - Get active device for IPsec offload
423 * @xs: pointer to transformer state struct
424 *
425 * Context: caller must hold rcu_read_lock.
426 *
427 * Return: the device for ipsec offload, or NULL if not exist.
428 **/
429static struct net_device *bond_ipsec_dev(struct xfrm_state *xs)
430{
431 struct net_device *bond_dev = xs->xso.dev;
432 struct bonding *bond;
433 struct slave *slave;
434
435 if (!bond_dev)
436 return NULL;
437
438 bond = netdev_priv(bond_dev);
439 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
440 return NULL;
441
442 slave = rcu_dereference(bond->curr_active_slave);
443 if (!slave)
444 return NULL;
445
446 if (!xs->xso.real_dev)
447 return NULL;
448
449 if (xs->xso.real_dev != slave->dev)
450 pr_warn_ratelimited("%s: (slave %s): not same with IPsec offload real dev %s\n",
451 bond_dev->name, slave->dev->name, xs->xso.real_dev->name);
452
453 return slave->dev;
454}
455
456/**
457 * bond_ipsec_add_sa - program device with a security association
458 * @xs: pointer to transformer state struct
459 * @extack: extack point to fill failure reason
460 **/
461static int bond_ipsec_add_sa(struct xfrm_state *xs,
462 struct netlink_ext_ack *extack)
463{
464 struct net_device *bond_dev = xs->xso.dev;
465 struct net_device *real_dev;
466 netdevice_tracker tracker;
467 struct bond_ipsec *ipsec;
468 struct bonding *bond;
469 struct slave *slave;
470 int err;
471
472 if (!bond_dev)
473 return -EINVAL;
474
475 rcu_read_lock();
476 bond = netdev_priv(bond_dev);
477 slave = rcu_dereference(bond->curr_active_slave);
478 real_dev = slave ? slave->dev : NULL;
479 netdev_hold(real_dev, &tracker, GFP_ATOMIC);
480 rcu_read_unlock();
481 if (!real_dev) {
482 err = -ENODEV;
483 goto out;
484 }
485
486 if (!real_dev->xfrmdev_ops ||
487 !real_dev->xfrmdev_ops->xdo_dev_state_add ||
488 netif_is_bond_master(real_dev)) {
489 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
490 err = -EINVAL;
491 goto out;
492 }
493
494 ipsec = kmalloc(sizeof(*ipsec), GFP_KERNEL);
495 if (!ipsec) {
496 err = -ENOMEM;
497 goto out;
498 }
499
500 xs->xso.real_dev = real_dev;
501 err = real_dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
502 if (!err) {
503 ipsec->xs = xs;
504 INIT_LIST_HEAD(&ipsec->list);
505 mutex_lock(&bond->ipsec_lock);
506 list_add(&ipsec->list, &bond->ipsec_list);
507 mutex_unlock(&bond->ipsec_lock);
508 } else {
509 kfree(ipsec);
510 }
511out:
512 netdev_put(real_dev, &tracker);
513 return err;
514}
515
516static void bond_ipsec_add_sa_all(struct bonding *bond)
517{
518 struct net_device *bond_dev = bond->dev;
519 struct net_device *real_dev;
520 struct bond_ipsec *ipsec;
521 struct slave *slave;
522
523 slave = rtnl_dereference(bond->curr_active_slave);
524 real_dev = slave ? slave->dev : NULL;
525 if (!real_dev)
526 return;
527
528 mutex_lock(&bond->ipsec_lock);
529 if (!real_dev->xfrmdev_ops ||
530 !real_dev->xfrmdev_ops->xdo_dev_state_add ||
531 netif_is_bond_master(real_dev)) {
532 if (!list_empty(&bond->ipsec_list))
533 slave_warn(bond_dev, real_dev,
534 "%s: no slave xdo_dev_state_add\n",
535 __func__);
536 goto out;
537 }
538
539 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
540 /* If new state is added before ipsec_lock acquired */
541 if (ipsec->xs->xso.real_dev == real_dev)
542 continue;
543
544 ipsec->xs->xso.real_dev = real_dev;
545 if (real_dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
546 slave_warn(bond_dev, real_dev, "%s: failed to add SA\n", __func__);
547 ipsec->xs->xso.real_dev = NULL;
548 }
549 }
550out:
551 mutex_unlock(&bond->ipsec_lock);
552}
553
554/**
555 * bond_ipsec_del_sa - clear out this specific SA
556 * @xs: pointer to transformer state struct
557 **/
558static void bond_ipsec_del_sa(struct xfrm_state *xs)
559{
560 struct net_device *bond_dev = xs->xso.dev;
561 struct net_device *real_dev;
562 netdevice_tracker tracker;
563 struct bond_ipsec *ipsec;
564 struct bonding *bond;
565 struct slave *slave;
566
567 if (!bond_dev)
568 return;
569
570 rcu_read_lock();
571 bond = netdev_priv(bond_dev);
572 slave = rcu_dereference(bond->curr_active_slave);
573 real_dev = slave ? slave->dev : NULL;
574 netdev_hold(real_dev, &tracker, GFP_ATOMIC);
575 rcu_read_unlock();
576
577 if (!slave)
578 goto out;
579
580 if (!xs->xso.real_dev)
581 goto out;
582
583 WARN_ON(xs->xso.real_dev != real_dev);
584
585 if (!real_dev->xfrmdev_ops ||
586 !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
587 netif_is_bond_master(real_dev)) {
588 slave_warn(bond_dev, real_dev, "%s: no slave xdo_dev_state_delete\n", __func__);
589 goto out;
590 }
591
592 real_dev->xfrmdev_ops->xdo_dev_state_delete(xs);
593out:
594 netdev_put(real_dev, &tracker);
595 mutex_lock(&bond->ipsec_lock);
596 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
597 if (ipsec->xs == xs) {
598 list_del(&ipsec->list);
599 kfree(ipsec);
600 break;
601 }
602 }
603 mutex_unlock(&bond->ipsec_lock);
604}
605
606static void bond_ipsec_del_sa_all(struct bonding *bond)
607{
608 struct net_device *bond_dev = bond->dev;
609 struct net_device *real_dev;
610 struct bond_ipsec *ipsec;
611 struct slave *slave;
612
613 slave = rtnl_dereference(bond->curr_active_slave);
614 real_dev = slave ? slave->dev : NULL;
615 if (!real_dev)
616 return;
617
618 mutex_lock(&bond->ipsec_lock);
619 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
620 if (!ipsec->xs->xso.real_dev)
621 continue;
622
623 if (!real_dev->xfrmdev_ops ||
624 !real_dev->xfrmdev_ops->xdo_dev_state_delete ||
625 netif_is_bond_master(real_dev)) {
626 slave_warn(bond_dev, real_dev,
627 "%s: no slave xdo_dev_state_delete\n",
628 __func__);
629 } else {
630 real_dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
631 if (real_dev->xfrmdev_ops->xdo_dev_state_free)
632 real_dev->xfrmdev_ops->xdo_dev_state_free(ipsec->xs);
633 }
634 }
635 mutex_unlock(&bond->ipsec_lock);
636}
637
638static void bond_ipsec_free_sa(struct xfrm_state *xs)
639{
640 struct net_device *bond_dev = xs->xso.dev;
641 struct net_device *real_dev;
642 netdevice_tracker tracker;
643 struct bonding *bond;
644 struct slave *slave;
645
646 if (!bond_dev)
647 return;
648
649 rcu_read_lock();
650 bond = netdev_priv(bond_dev);
651 slave = rcu_dereference(bond->curr_active_slave);
652 real_dev = slave ? slave->dev : NULL;
653 netdev_hold(real_dev, &tracker, GFP_ATOMIC);
654 rcu_read_unlock();
655
656 if (!slave)
657 goto out;
658
659 if (!xs->xso.real_dev)
660 goto out;
661
662 WARN_ON(xs->xso.real_dev != real_dev);
663
664 if (real_dev && real_dev->xfrmdev_ops &&
665 real_dev->xfrmdev_ops->xdo_dev_state_free)
666 real_dev->xfrmdev_ops->xdo_dev_state_free(xs);
667out:
668 netdev_put(real_dev, &tracker);
669}
670
671/**
672 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
673 * @skb: current data packet
674 * @xs: pointer to transformer state struct
675 **/
676static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
677{
678 struct net_device *real_dev;
679 bool ok = false;
680
681 rcu_read_lock();
682 real_dev = bond_ipsec_dev(xs);
683 if (!real_dev)
684 goto out;
685
686 if (!real_dev->xfrmdev_ops ||
687 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
688 netif_is_bond_master(real_dev))
689 goto out;
690
691 ok = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
692out:
693 rcu_read_unlock();
694 return ok;
695}
696
697/**
698 * bond_advance_esn_state - ESN support for IPSec HW offload
699 * @xs: pointer to transformer state struct
700 **/
701static void bond_advance_esn_state(struct xfrm_state *xs)
702{
703 struct net_device *real_dev;
704
705 rcu_read_lock();
706 real_dev = bond_ipsec_dev(xs);
707 if (!real_dev)
708 goto out;
709
710 if (!real_dev->xfrmdev_ops ||
711 !real_dev->xfrmdev_ops->xdo_dev_state_advance_esn) {
712 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_advance_esn\n", __func__, real_dev->name);
713 goto out;
714 }
715
716 real_dev->xfrmdev_ops->xdo_dev_state_advance_esn(xs);
717out:
718 rcu_read_unlock();
719}
720
721/**
722 * bond_xfrm_update_stats - Update xfrm state
723 * @xs: pointer to transformer state struct
724 **/
725static void bond_xfrm_update_stats(struct xfrm_state *xs)
726{
727 struct net_device *real_dev;
728
729 rcu_read_lock();
730 real_dev = bond_ipsec_dev(xs);
731 if (!real_dev)
732 goto out;
733
734 if (!real_dev->xfrmdev_ops ||
735 !real_dev->xfrmdev_ops->xdo_dev_state_update_stats) {
736 pr_warn_ratelimited("%s: %s doesn't support xdo_dev_state_update_stats\n", __func__, real_dev->name);
737 goto out;
738 }
739
740 real_dev->xfrmdev_ops->xdo_dev_state_update_stats(xs);
741out:
742 rcu_read_unlock();
743}
744
745static const struct xfrmdev_ops bond_xfrmdev_ops = {
746 .xdo_dev_state_add = bond_ipsec_add_sa,
747 .xdo_dev_state_delete = bond_ipsec_del_sa,
748 .xdo_dev_state_free = bond_ipsec_free_sa,
749 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
750 .xdo_dev_state_advance_esn = bond_advance_esn_state,
751 .xdo_dev_state_update_stats = bond_xfrm_update_stats,
752};
753#endif /* CONFIG_XFRM_OFFLOAD */
754
755/*------------------------------- Link status -------------------------------*/
756
757/* Set the carrier state for the master according to the state of its
758 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
759 * do special 802.3ad magic.
760 *
761 * Returns zero if carrier state does not change, nonzero if it does.
762 */
763int bond_set_carrier(struct bonding *bond)
764{
765 struct list_head *iter;
766 struct slave *slave;
767
768 if (!bond_has_slaves(bond))
769 goto down;
770
771 if (BOND_MODE(bond) == BOND_MODE_8023AD)
772 return bond_3ad_set_carrier(bond);
773
774 bond_for_each_slave(bond, slave, iter) {
775 if (slave->link == BOND_LINK_UP) {
776 if (!netif_carrier_ok(bond->dev)) {
777 netif_carrier_on(bond->dev);
778 return 1;
779 }
780 return 0;
781 }
782 }
783
784down:
785 if (netif_carrier_ok(bond->dev)) {
786 netif_carrier_off(bond->dev);
787 return 1;
788 }
789 return 0;
790}
791
792/* Get link speed and duplex from the slave's base driver
793 * using ethtool. If for some reason the call fails or the
794 * values are invalid, set speed and duplex to -1,
795 * and return. Return 1 if speed or duplex settings are
796 * UNKNOWN; 0 otherwise.
797 */
798static int bond_update_speed_duplex(struct slave *slave)
799{
800 struct net_device *slave_dev = slave->dev;
801 struct ethtool_link_ksettings ecmd;
802 int res;
803
804 slave->speed = SPEED_UNKNOWN;
805 slave->duplex = DUPLEX_UNKNOWN;
806
807 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
808 if (res < 0)
809 return 1;
810 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
811 return 1;
812 switch (ecmd.base.duplex) {
813 case DUPLEX_FULL:
814 case DUPLEX_HALF:
815 break;
816 default:
817 return 1;
818 }
819
820 slave->speed = ecmd.base.speed;
821 slave->duplex = ecmd.base.duplex;
822
823 return 0;
824}
825
826const char *bond_slave_link_status(s8 link)
827{
828 switch (link) {
829 case BOND_LINK_UP:
830 return "up";
831 case BOND_LINK_FAIL:
832 return "going down";
833 case BOND_LINK_DOWN:
834 return "down";
835 case BOND_LINK_BACK:
836 return "going back";
837 default:
838 return "unknown";
839 }
840}
841
842/* if <dev> supports MII link status reporting, check its link status.
843 *
844 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
845 * depending upon the setting of the use_carrier parameter.
846 *
847 * Return either BMSR_LSTATUS, meaning that the link is up (or we
848 * can't tell and just pretend it is), or 0, meaning that the link is
849 * down.
850 *
851 * If reporting is non-zero, instead of faking link up, return -1 if
852 * both ETHTOOL and MII ioctls fail (meaning the device does not
853 * support them). If use_carrier is set, return whatever it says.
854 * It'd be nice if there was a good way to tell if a driver supports
855 * netif_carrier, but there really isn't.
856 */
857static int bond_check_dev_link(struct bonding *bond,
858 struct net_device *slave_dev, int reporting)
859{
860 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
861 int (*ioctl)(struct net_device *, struct ifreq *, int);
862 struct ifreq ifr;
863 struct mii_ioctl_data *mii;
864
865 if (!reporting && !netif_running(slave_dev))
866 return 0;
867
868 if (bond->params.use_carrier)
869 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
870
871 /* Try to get link status using Ethtool first. */
872 if (slave_dev->ethtool_ops->get_link)
873 return slave_dev->ethtool_ops->get_link(slave_dev) ?
874 BMSR_LSTATUS : 0;
875
876 /* Ethtool can't be used, fallback to MII ioctls. */
877 ioctl = slave_ops->ndo_eth_ioctl;
878 if (ioctl) {
879 /* TODO: set pointer to correct ioctl on a per team member
880 * bases to make this more efficient. that is, once
881 * we determine the correct ioctl, we will always
882 * call it and not the others for that team
883 * member.
884 */
885
886 /* We cannot assume that SIOCGMIIPHY will also read a
887 * register; not all network drivers (e.g., e100)
888 * support that.
889 */
890
891 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
892 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
893 mii = if_mii(&ifr);
894 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
895 mii->reg_num = MII_BMSR;
896 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
897 return mii->val_out & BMSR_LSTATUS;
898 }
899 }
900
901 /* If reporting, report that either there's no ndo_eth_ioctl,
902 * or both SIOCGMIIREG and get_link failed (meaning that we
903 * cannot report link status). If not reporting, pretend
904 * we're ok.
905 */
906 return reporting ? -1 : BMSR_LSTATUS;
907}
908
909/*----------------------------- Multicast list ------------------------------*/
910
911/* Push the promiscuity flag down to appropriate slaves */
912static int bond_set_promiscuity(struct bonding *bond, int inc)
913{
914 struct list_head *iter;
915 int err = 0;
916
917 if (bond_uses_primary(bond)) {
918 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
919
920 if (curr_active)
921 err = dev_set_promiscuity(curr_active->dev, inc);
922 } else {
923 struct slave *slave;
924
925 bond_for_each_slave(bond, slave, iter) {
926 err = dev_set_promiscuity(slave->dev, inc);
927 if (err)
928 return err;
929 }
930 }
931 return err;
932}
933
934/* Push the allmulti flag down to all slaves */
935static int bond_set_allmulti(struct bonding *bond, int inc)
936{
937 struct list_head *iter;
938 int err = 0;
939
940 if (bond_uses_primary(bond)) {
941 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
942
943 if (curr_active)
944 err = dev_set_allmulti(curr_active->dev, inc);
945 } else {
946 struct slave *slave;
947
948 bond_for_each_slave(bond, slave, iter) {
949 err = dev_set_allmulti(slave->dev, inc);
950 if (err)
951 return err;
952 }
953 }
954 return err;
955}
956
957/* Retrieve the list of registered multicast addresses for the bonding
958 * device and retransmit an IGMP JOIN request to the current active
959 * slave.
960 */
961static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
962{
963 struct bonding *bond = container_of(work, struct bonding,
964 mcast_work.work);
965
966 if (!rtnl_trylock()) {
967 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
968 return;
969 }
970 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
971
972 if (bond->igmp_retrans > 1) {
973 bond->igmp_retrans--;
974 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
975 }
976 rtnl_unlock();
977}
978
979/* Flush bond's hardware addresses from slave */
980static void bond_hw_addr_flush(struct net_device *bond_dev,
981 struct net_device *slave_dev)
982{
983 struct bonding *bond = netdev_priv(bond_dev);
984
985 dev_uc_unsync(slave_dev, bond_dev);
986 dev_mc_unsync(slave_dev, bond_dev);
987
988 if (BOND_MODE(bond) == BOND_MODE_8023AD)
989 dev_mc_del(slave_dev, lacpdu_mcast_addr);
990}
991
992/*--------------------------- Active slave change ---------------------------*/
993
994/* Update the hardware address list and promisc/allmulti for the new and
995 * old active slaves (if any). Modes that are not using primary keep all
996 * slaves up date at all times; only the modes that use primary need to call
997 * this function to swap these settings during a failover.
998 */
999static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
1000 struct slave *old_active)
1001{
1002 if (old_active) {
1003 if (bond->dev->flags & IFF_PROMISC)
1004 dev_set_promiscuity(old_active->dev, -1);
1005
1006 if (bond->dev->flags & IFF_ALLMULTI)
1007 dev_set_allmulti(old_active->dev, -1);
1008
1009 if (bond->dev->flags & IFF_UP)
1010 bond_hw_addr_flush(bond->dev, old_active->dev);
1011
1012 bond_slave_ns_maddrs_add(bond, old_active);
1013 }
1014
1015 if (new_active) {
1016 /* FIXME: Signal errors upstream. */
1017 if (bond->dev->flags & IFF_PROMISC)
1018 dev_set_promiscuity(new_active->dev, 1);
1019
1020 if (bond->dev->flags & IFF_ALLMULTI)
1021 dev_set_allmulti(new_active->dev, 1);
1022
1023 if (bond->dev->flags & IFF_UP) {
1024 netif_addr_lock_bh(bond->dev);
1025 dev_uc_sync(new_active->dev, bond->dev);
1026 dev_mc_sync(new_active->dev, bond->dev);
1027 netif_addr_unlock_bh(bond->dev);
1028 }
1029
1030 bond_slave_ns_maddrs_del(bond, new_active);
1031 }
1032}
1033
1034/**
1035 * bond_set_dev_addr - clone slave's address to bond
1036 * @bond_dev: bond net device
1037 * @slave_dev: slave net device
1038 *
1039 * Should be called with RTNL held.
1040 */
1041static int bond_set_dev_addr(struct net_device *bond_dev,
1042 struct net_device *slave_dev)
1043{
1044 int err;
1045
1046 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
1047 bond_dev, slave_dev, slave_dev->addr_len);
1048 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
1049 if (err)
1050 return err;
1051
1052 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
1053 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
1054 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1055 return 0;
1056}
1057
1058static struct slave *bond_get_old_active(struct bonding *bond,
1059 struct slave *new_active)
1060{
1061 struct slave *slave;
1062 struct list_head *iter;
1063
1064 bond_for_each_slave(bond, slave, iter) {
1065 if (slave == new_active)
1066 continue;
1067
1068 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
1069 return slave;
1070 }
1071
1072 return NULL;
1073}
1074
1075/* bond_do_fail_over_mac
1076 *
1077 * Perform special MAC address swapping for fail_over_mac settings
1078 *
1079 * Called with RTNL
1080 */
1081static void bond_do_fail_over_mac(struct bonding *bond,
1082 struct slave *new_active,
1083 struct slave *old_active)
1084{
1085 u8 tmp_mac[MAX_ADDR_LEN];
1086 struct sockaddr_storage ss;
1087 int rv;
1088
1089 switch (bond->params.fail_over_mac) {
1090 case BOND_FOM_ACTIVE:
1091 if (new_active) {
1092 rv = bond_set_dev_addr(bond->dev, new_active->dev);
1093 if (rv)
1094 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
1095 -rv);
1096 }
1097 break;
1098 case BOND_FOM_FOLLOW:
1099 /* if new_active && old_active, swap them
1100 * if just old_active, do nothing (going to no active slave)
1101 * if just new_active, set new_active to bond's MAC
1102 */
1103 if (!new_active)
1104 return;
1105
1106 if (!old_active)
1107 old_active = bond_get_old_active(bond, new_active);
1108
1109 if (old_active) {
1110 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
1111 new_active->dev->addr_len);
1112 bond_hw_addr_copy(ss.__data,
1113 old_active->dev->dev_addr,
1114 old_active->dev->addr_len);
1115 ss.ss_family = new_active->dev->type;
1116 } else {
1117 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
1118 bond->dev->addr_len);
1119 ss.ss_family = bond->dev->type;
1120 }
1121
1122 rv = dev_set_mac_address(new_active->dev,
1123 (struct sockaddr *)&ss, NULL);
1124 if (rv) {
1125 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1126 -rv);
1127 goto out;
1128 }
1129
1130 if (!old_active)
1131 goto out;
1132
1133 bond_hw_addr_copy(ss.__data, tmp_mac,
1134 new_active->dev->addr_len);
1135 ss.ss_family = old_active->dev->type;
1136
1137 rv = dev_set_mac_address(old_active->dev,
1138 (struct sockaddr *)&ss, NULL);
1139 if (rv)
1140 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1141 -rv);
1142out:
1143 break;
1144 default:
1145 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1146 bond->params.fail_over_mac);
1147 break;
1148 }
1149
1150}
1151
1152/**
1153 * bond_choose_primary_or_current - select the primary or high priority slave
1154 * @bond: our bonding struct
1155 *
1156 * - Check if there is a primary link. If the primary link was set and is up,
1157 * go on and do link reselection.
1158 *
1159 * - If primary link is not set or down, find the highest priority link.
1160 * If the highest priority link is not current slave, set it as primary
1161 * link and do link reselection.
1162 */
1163static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1164{
1165 struct slave *prim = rtnl_dereference(bond->primary_slave);
1166 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1167 struct slave *slave, *hprio = NULL;
1168 struct list_head *iter;
1169
1170 if (!prim || prim->link != BOND_LINK_UP) {
1171 bond_for_each_slave(bond, slave, iter) {
1172 if (slave->link == BOND_LINK_UP) {
1173 hprio = hprio ?: slave;
1174 if (slave->prio > hprio->prio)
1175 hprio = slave;
1176 }
1177 }
1178
1179 if (hprio && hprio != curr) {
1180 prim = hprio;
1181 goto link_reselect;
1182 }
1183
1184 if (!curr || curr->link != BOND_LINK_UP)
1185 return NULL;
1186 return curr;
1187 }
1188
1189 if (bond->force_primary) {
1190 bond->force_primary = false;
1191 return prim;
1192 }
1193
1194link_reselect:
1195 if (!curr || curr->link != BOND_LINK_UP)
1196 return prim;
1197
1198 /* At this point, prim and curr are both up */
1199 switch (bond->params.primary_reselect) {
1200 case BOND_PRI_RESELECT_ALWAYS:
1201 return prim;
1202 case BOND_PRI_RESELECT_BETTER:
1203 if (prim->speed < curr->speed)
1204 return curr;
1205 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1206 return curr;
1207 return prim;
1208 case BOND_PRI_RESELECT_FAILURE:
1209 return curr;
1210 default:
1211 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1212 bond->params.primary_reselect);
1213 return curr;
1214 }
1215}
1216
1217/**
1218 * bond_find_best_slave - select the best available slave to be the active one
1219 * @bond: our bonding struct
1220 */
1221static struct slave *bond_find_best_slave(struct bonding *bond)
1222{
1223 struct slave *slave, *bestslave = NULL;
1224 struct list_head *iter;
1225 int mintime = bond->params.updelay;
1226
1227 slave = bond_choose_primary_or_current(bond);
1228 if (slave)
1229 return slave;
1230
1231 bond_for_each_slave(bond, slave, iter) {
1232 if (slave->link == BOND_LINK_UP)
1233 return slave;
1234 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1235 slave->delay < mintime) {
1236 mintime = slave->delay;
1237 bestslave = slave;
1238 }
1239 }
1240
1241 return bestslave;
1242}
1243
1244/* must be called in RCU critical section or with RTNL held */
1245static bool bond_should_notify_peers(struct bonding *bond)
1246{
1247 struct slave *slave = rcu_dereference_rtnl(bond->curr_active_slave);
1248
1249 if (!slave || !bond->send_peer_notif ||
1250 bond->send_peer_notif %
1251 max(1, bond->params.peer_notif_delay) != 0 ||
1252 !netif_carrier_ok(bond->dev) ||
1253 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1254 return false;
1255
1256 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1257 slave ? slave->dev->name : "NULL");
1258
1259 return true;
1260}
1261
1262/**
1263 * bond_change_active_slave - change the active slave into the specified one
1264 * @bond: our bonding struct
1265 * @new_active: the new slave to make the active one
1266 *
1267 * Set the new slave to the bond's settings and unset them on the old
1268 * curr_active_slave.
1269 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1270 *
1271 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1272 * because it is apparently the best available slave we have, even though its
1273 * updelay hasn't timed out yet.
1274 *
1275 * Caller must hold RTNL.
1276 */
1277void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1278{
1279 struct slave *old_active;
1280
1281 ASSERT_RTNL();
1282
1283 old_active = rtnl_dereference(bond->curr_active_slave);
1284
1285 if (old_active == new_active)
1286 return;
1287
1288#ifdef CONFIG_XFRM_OFFLOAD
1289 bond_ipsec_del_sa_all(bond);
1290#endif /* CONFIG_XFRM_OFFLOAD */
1291
1292 if (new_active) {
1293 new_active->last_link_up = jiffies;
1294
1295 if (new_active->link == BOND_LINK_BACK) {
1296 if (bond_uses_primary(bond)) {
1297 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1298 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1299 }
1300
1301 new_active->delay = 0;
1302 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1303 BOND_SLAVE_NOTIFY_NOW);
1304
1305 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1306 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1307
1308 if (bond_is_lb(bond))
1309 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1310 } else {
1311 if (bond_uses_primary(bond))
1312 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1313 }
1314 }
1315
1316 if (bond_uses_primary(bond))
1317 bond_hw_addr_swap(bond, new_active, old_active);
1318
1319 if (bond_is_lb(bond)) {
1320 bond_alb_handle_active_change(bond, new_active);
1321 if (old_active)
1322 bond_set_slave_inactive_flags(old_active,
1323 BOND_SLAVE_NOTIFY_NOW);
1324 if (new_active)
1325 bond_set_slave_active_flags(new_active,
1326 BOND_SLAVE_NOTIFY_NOW);
1327 } else {
1328 rcu_assign_pointer(bond->curr_active_slave, new_active);
1329 }
1330
1331 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1332 if (old_active)
1333 bond_set_slave_inactive_flags(old_active,
1334 BOND_SLAVE_NOTIFY_NOW);
1335
1336 if (new_active) {
1337 bool should_notify_peers = false;
1338
1339 bond_set_slave_active_flags(new_active,
1340 BOND_SLAVE_NOTIFY_NOW);
1341
1342 if (bond->params.fail_over_mac)
1343 bond_do_fail_over_mac(bond, new_active,
1344 old_active);
1345
1346 if (netif_running(bond->dev)) {
1347 bond->send_peer_notif =
1348 bond->params.num_peer_notif *
1349 max(1, bond->params.peer_notif_delay);
1350 should_notify_peers =
1351 bond_should_notify_peers(bond);
1352 }
1353
1354 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1355 if (should_notify_peers) {
1356 bond->send_peer_notif--;
1357 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1358 bond->dev);
1359 }
1360 }
1361 }
1362
1363#ifdef CONFIG_XFRM_OFFLOAD
1364 bond_ipsec_add_sa_all(bond);
1365#endif /* CONFIG_XFRM_OFFLOAD */
1366
1367 /* resend IGMP joins since active slave has changed or
1368 * all were sent on curr_active_slave.
1369 * resend only if bond is brought up with the affected
1370 * bonding modes and the retransmission is enabled
1371 */
1372 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1373 ((bond_uses_primary(bond) && new_active) ||
1374 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1375 bond->igmp_retrans = bond->params.resend_igmp;
1376 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1377 }
1378}
1379
1380/**
1381 * bond_select_active_slave - select a new active slave, if needed
1382 * @bond: our bonding struct
1383 *
1384 * This functions should be called when one of the following occurs:
1385 * - The old curr_active_slave has been released or lost its link.
1386 * - The primary_slave has got its link back.
1387 * - A slave has got its link back and there's no old curr_active_slave.
1388 *
1389 * Caller must hold RTNL.
1390 */
1391void bond_select_active_slave(struct bonding *bond)
1392{
1393 struct slave *best_slave;
1394 int rv;
1395
1396 ASSERT_RTNL();
1397
1398 best_slave = bond_find_best_slave(bond);
1399 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1400 bond_change_active_slave(bond, best_slave);
1401 rv = bond_set_carrier(bond);
1402 if (!rv)
1403 return;
1404
1405 if (netif_carrier_ok(bond->dev))
1406 netdev_info(bond->dev, "active interface up!\n");
1407 else
1408 netdev_info(bond->dev, "now running without any active interface!\n");
1409 }
1410}
1411
1412#ifdef CONFIG_NET_POLL_CONTROLLER
1413static inline int slave_enable_netpoll(struct slave *slave)
1414{
1415 struct netpoll *np;
1416 int err = 0;
1417
1418 np = kzalloc(sizeof(*np), GFP_KERNEL);
1419 err = -ENOMEM;
1420 if (!np)
1421 goto out;
1422
1423 err = __netpoll_setup(np, slave->dev);
1424 if (err) {
1425 kfree(np);
1426 goto out;
1427 }
1428 slave->np = np;
1429out:
1430 return err;
1431}
1432static inline void slave_disable_netpoll(struct slave *slave)
1433{
1434 struct netpoll *np = slave->np;
1435
1436 if (!np)
1437 return;
1438
1439 slave->np = NULL;
1440
1441 __netpoll_free(np);
1442}
1443
1444static void bond_poll_controller(struct net_device *bond_dev)
1445{
1446 struct bonding *bond = netdev_priv(bond_dev);
1447 struct slave *slave = NULL;
1448 struct list_head *iter;
1449 struct ad_info ad_info;
1450
1451 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1452 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1453 return;
1454
1455 bond_for_each_slave_rcu(bond, slave, iter) {
1456 if (!bond_slave_is_up(slave))
1457 continue;
1458
1459 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1460 struct aggregator *agg =
1461 SLAVE_AD_INFO(slave)->port.aggregator;
1462
1463 if (agg &&
1464 agg->aggregator_identifier != ad_info.aggregator_id)
1465 continue;
1466 }
1467
1468 netpoll_poll_dev(slave->dev);
1469 }
1470}
1471
1472static void bond_netpoll_cleanup(struct net_device *bond_dev)
1473{
1474 struct bonding *bond = netdev_priv(bond_dev);
1475 struct list_head *iter;
1476 struct slave *slave;
1477
1478 bond_for_each_slave(bond, slave, iter)
1479 if (bond_slave_is_up(slave))
1480 slave_disable_netpoll(slave);
1481}
1482
1483static int bond_netpoll_setup(struct net_device *dev)
1484{
1485 struct bonding *bond = netdev_priv(dev);
1486 struct list_head *iter;
1487 struct slave *slave;
1488 int err = 0;
1489
1490 bond_for_each_slave(bond, slave, iter) {
1491 err = slave_enable_netpoll(slave);
1492 if (err) {
1493 bond_netpoll_cleanup(dev);
1494 break;
1495 }
1496 }
1497 return err;
1498}
1499#else
1500static inline int slave_enable_netpoll(struct slave *slave)
1501{
1502 return 0;
1503}
1504static inline void slave_disable_netpoll(struct slave *slave)
1505{
1506}
1507static void bond_netpoll_cleanup(struct net_device *bond_dev)
1508{
1509}
1510#endif
1511
1512/*---------------------------------- IOCTL ----------------------------------*/
1513
1514static netdev_features_t bond_fix_features(struct net_device *dev,
1515 netdev_features_t features)
1516{
1517 struct bonding *bond = netdev_priv(dev);
1518 struct list_head *iter;
1519 netdev_features_t mask;
1520 struct slave *slave;
1521
1522 mask = features;
1523 features = netdev_base_features(features);
1524
1525 bond_for_each_slave(bond, slave, iter) {
1526 features = netdev_increment_features(features,
1527 slave->dev->features,
1528 mask);
1529 }
1530 features = netdev_add_tso_features(features, mask);
1531
1532 return features;
1533}
1534
1535#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1536 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1537 NETIF_F_GSO_ENCAP_ALL | \
1538 NETIF_F_HIGHDMA | NETIF_F_LRO)
1539
1540#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1541 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE | \
1542 NETIF_F_GSO_PARTIAL)
1543
1544#define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1545 NETIF_F_GSO_SOFTWARE)
1546
1547#define BOND_GSO_PARTIAL_FEATURES (NETIF_F_GSO_ESP)
1548
1549
1550static void bond_compute_features(struct bonding *bond)
1551{
1552 netdev_features_t gso_partial_features = BOND_GSO_PARTIAL_FEATURES;
1553 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1554 IFF_XMIT_DST_RELEASE_PERM;
1555 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1556 netdev_features_t enc_features = BOND_ENC_FEATURES;
1557#ifdef CONFIG_XFRM_OFFLOAD
1558 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1559#endif /* CONFIG_XFRM_OFFLOAD */
1560 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1561 struct net_device *bond_dev = bond->dev;
1562 struct list_head *iter;
1563 struct slave *slave;
1564 unsigned short max_hard_header_len = ETH_HLEN;
1565 unsigned int tso_max_size = TSO_MAX_SIZE;
1566 u16 tso_max_segs = TSO_MAX_SEGS;
1567
1568 if (!bond_has_slaves(bond))
1569 goto done;
1570
1571 vlan_features = netdev_base_features(vlan_features);
1572 mpls_features = netdev_base_features(mpls_features);
1573
1574 bond_for_each_slave(bond, slave, iter) {
1575 vlan_features = netdev_increment_features(vlan_features,
1576 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1577
1578 enc_features = netdev_increment_features(enc_features,
1579 slave->dev->hw_enc_features,
1580 BOND_ENC_FEATURES);
1581
1582#ifdef CONFIG_XFRM_OFFLOAD
1583 xfrm_features = netdev_increment_features(xfrm_features,
1584 slave->dev->hw_enc_features,
1585 BOND_XFRM_FEATURES);
1586#endif /* CONFIG_XFRM_OFFLOAD */
1587
1588 gso_partial_features = netdev_increment_features(gso_partial_features,
1589 slave->dev->gso_partial_features,
1590 BOND_GSO_PARTIAL_FEATURES);
1591
1592 mpls_features = netdev_increment_features(mpls_features,
1593 slave->dev->mpls_features,
1594 BOND_MPLS_FEATURES);
1595
1596 dst_release_flag &= slave->dev->priv_flags;
1597 if (slave->dev->hard_header_len > max_hard_header_len)
1598 max_hard_header_len = slave->dev->hard_header_len;
1599
1600 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1601 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1602 }
1603 bond_dev->hard_header_len = max_hard_header_len;
1604
1605done:
1606 bond_dev->gso_partial_features = gso_partial_features;
1607 bond_dev->vlan_features = vlan_features;
1608 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1609 NETIF_F_HW_VLAN_CTAG_TX |
1610 NETIF_F_HW_VLAN_STAG_TX;
1611#ifdef CONFIG_XFRM_OFFLOAD
1612 bond_dev->hw_enc_features |= xfrm_features;
1613#endif /* CONFIG_XFRM_OFFLOAD */
1614 bond_dev->mpls_features = mpls_features;
1615 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1616 netif_set_tso_max_size(bond_dev, tso_max_size);
1617
1618 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1619 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1620 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1621 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1622
1623 netdev_change_features(bond_dev);
1624}
1625
1626static void bond_setup_by_slave(struct net_device *bond_dev,
1627 struct net_device *slave_dev)
1628{
1629 bool was_up = !!(bond_dev->flags & IFF_UP);
1630
1631 dev_close(bond_dev);
1632
1633 bond_dev->header_ops = slave_dev->header_ops;
1634
1635 bond_dev->type = slave_dev->type;
1636 bond_dev->hard_header_len = slave_dev->hard_header_len;
1637 bond_dev->needed_headroom = slave_dev->needed_headroom;
1638 bond_dev->addr_len = slave_dev->addr_len;
1639
1640 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1641 slave_dev->addr_len);
1642
1643 if (slave_dev->flags & IFF_POINTOPOINT) {
1644 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1645 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1646 }
1647 if (was_up)
1648 dev_open(bond_dev, NULL);
1649}
1650
1651/* On bonding slaves other than the currently active slave, suppress
1652 * duplicates except for alb non-mcast/bcast.
1653 */
1654static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1655 struct slave *slave,
1656 struct bonding *bond)
1657{
1658 if (bond_is_slave_inactive(slave)) {
1659 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1660 skb->pkt_type != PACKET_BROADCAST &&
1661 skb->pkt_type != PACKET_MULTICAST)
1662 return false;
1663 return true;
1664 }
1665 return false;
1666}
1667
1668static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1669{
1670 struct sk_buff *skb = *pskb;
1671 struct slave *slave;
1672 struct bonding *bond;
1673 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1674 struct slave *);
1675 int ret = RX_HANDLER_ANOTHER;
1676
1677 skb = skb_share_check(skb, GFP_ATOMIC);
1678 if (unlikely(!skb))
1679 return RX_HANDLER_CONSUMED;
1680
1681 *pskb = skb;
1682
1683 slave = bond_slave_get_rcu(skb->dev);
1684 bond = slave->bond;
1685
1686 recv_probe = READ_ONCE(bond->recv_probe);
1687 if (recv_probe) {
1688 ret = recv_probe(skb, bond, slave);
1689 if (ret == RX_HANDLER_CONSUMED) {
1690 consume_skb(skb);
1691 return ret;
1692 }
1693 }
1694
1695 /*
1696 * For packets determined by bond_should_deliver_exact_match() call to
1697 * be suppressed we want to make an exception for link-local packets.
1698 * This is necessary for e.g. LLDP daemons to be able to monitor
1699 * inactive slave links without being forced to bind to them
1700 * explicitly.
1701 *
1702 * At the same time, packets that are passed to the bonding master
1703 * (including link-local ones) can have their originating interface
1704 * determined via PACKET_ORIGDEV socket option.
1705 */
1706 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1707 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1708 return RX_HANDLER_PASS;
1709 return RX_HANDLER_EXACT;
1710 }
1711
1712 skb->dev = bond->dev;
1713
1714 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1715 netif_is_bridge_port(bond->dev) &&
1716 skb->pkt_type == PACKET_HOST) {
1717
1718 if (unlikely(skb_cow_head(skb,
1719 skb->data - skb_mac_header(skb)))) {
1720 kfree_skb(skb);
1721 return RX_HANDLER_CONSUMED;
1722 }
1723 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1724 bond->dev->addr_len);
1725 }
1726
1727 return ret;
1728}
1729
1730static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1731{
1732 switch (BOND_MODE(bond)) {
1733 case BOND_MODE_ROUNDROBIN:
1734 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1735 case BOND_MODE_ACTIVEBACKUP:
1736 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1737 case BOND_MODE_BROADCAST:
1738 return NETDEV_LAG_TX_TYPE_BROADCAST;
1739 case BOND_MODE_XOR:
1740 case BOND_MODE_8023AD:
1741 return NETDEV_LAG_TX_TYPE_HASH;
1742 default:
1743 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1744 }
1745}
1746
1747static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1748 enum netdev_lag_tx_type type)
1749{
1750 if (type != NETDEV_LAG_TX_TYPE_HASH)
1751 return NETDEV_LAG_HASH_NONE;
1752
1753 switch (bond->params.xmit_policy) {
1754 case BOND_XMIT_POLICY_LAYER2:
1755 return NETDEV_LAG_HASH_L2;
1756 case BOND_XMIT_POLICY_LAYER34:
1757 return NETDEV_LAG_HASH_L34;
1758 case BOND_XMIT_POLICY_LAYER23:
1759 return NETDEV_LAG_HASH_L23;
1760 case BOND_XMIT_POLICY_ENCAP23:
1761 return NETDEV_LAG_HASH_E23;
1762 case BOND_XMIT_POLICY_ENCAP34:
1763 return NETDEV_LAG_HASH_E34;
1764 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1765 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1766 default:
1767 return NETDEV_LAG_HASH_UNKNOWN;
1768 }
1769}
1770
1771static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1772 struct netlink_ext_ack *extack)
1773{
1774 struct netdev_lag_upper_info lag_upper_info;
1775 enum netdev_lag_tx_type type;
1776 int err;
1777
1778 type = bond_lag_tx_type(bond);
1779 lag_upper_info.tx_type = type;
1780 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1781
1782 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1783 &lag_upper_info, extack);
1784 if (err)
1785 return err;
1786
1787 slave->dev->flags |= IFF_SLAVE;
1788 return 0;
1789}
1790
1791static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1792{
1793 netdev_upper_dev_unlink(slave->dev, bond->dev);
1794 slave->dev->flags &= ~IFF_SLAVE;
1795}
1796
1797static void slave_kobj_release(struct kobject *kobj)
1798{
1799 struct slave *slave = to_slave(kobj);
1800 struct bonding *bond = bond_get_bond_by_slave(slave);
1801
1802 cancel_delayed_work_sync(&slave->notify_work);
1803 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1804 kfree(SLAVE_AD_INFO(slave));
1805
1806 kfree(slave);
1807}
1808
1809static struct kobj_type slave_ktype = {
1810 .release = slave_kobj_release,
1811#ifdef CONFIG_SYSFS
1812 .sysfs_ops = &slave_sysfs_ops,
1813#endif
1814};
1815
1816static int bond_kobj_init(struct slave *slave)
1817{
1818 int err;
1819
1820 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1821 &(slave->dev->dev.kobj), "bonding_slave");
1822 if (err)
1823 kobject_put(&slave->kobj);
1824
1825 return err;
1826}
1827
1828static struct slave *bond_alloc_slave(struct bonding *bond,
1829 struct net_device *slave_dev)
1830{
1831 struct slave *slave = NULL;
1832
1833 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1834 if (!slave)
1835 return NULL;
1836
1837 slave->bond = bond;
1838 slave->dev = slave_dev;
1839 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1840
1841 if (bond_kobj_init(slave))
1842 return NULL;
1843
1844 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1845 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1846 GFP_KERNEL);
1847 if (!SLAVE_AD_INFO(slave)) {
1848 kobject_put(&slave->kobj);
1849 return NULL;
1850 }
1851 }
1852
1853 return slave;
1854}
1855
1856static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1857{
1858 info->bond_mode = BOND_MODE(bond);
1859 info->miimon = bond->params.miimon;
1860 info->num_slaves = bond->slave_cnt;
1861}
1862
1863static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1864{
1865 strcpy(info->slave_name, slave->dev->name);
1866 info->link = slave->link;
1867 info->state = bond_slave_state(slave);
1868 info->link_failure_count = slave->link_failure_count;
1869}
1870
1871static void bond_netdev_notify_work(struct work_struct *_work)
1872{
1873 struct slave *slave = container_of(_work, struct slave,
1874 notify_work.work);
1875
1876 if (rtnl_trylock()) {
1877 struct netdev_bonding_info binfo;
1878
1879 bond_fill_ifslave(slave, &binfo.slave);
1880 bond_fill_ifbond(slave->bond, &binfo.master);
1881 netdev_bonding_info_change(slave->dev, &binfo);
1882 rtnl_unlock();
1883 } else {
1884 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1885 }
1886}
1887
1888void bond_queue_slave_event(struct slave *slave)
1889{
1890 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1891}
1892
1893void bond_lower_state_changed(struct slave *slave)
1894{
1895 struct netdev_lag_lower_state_info info;
1896
1897 info.link_up = slave->link == BOND_LINK_UP ||
1898 slave->link == BOND_LINK_FAIL;
1899 info.tx_enabled = bond_is_active_slave(slave);
1900 netdev_lower_state_changed(slave->dev, &info);
1901}
1902
1903#define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1904 if (extack) \
1905 NL_SET_ERR_MSG(extack, errmsg); \
1906 else \
1907 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1908} while (0)
1909
1910#define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1911 if (extack) \
1912 NL_SET_ERR_MSG(extack, errmsg); \
1913 else \
1914 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1915} while (0)
1916
1917/* The bonding driver uses ether_setup() to convert a master bond device
1918 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1919 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1920 * if they were set
1921 */
1922static void bond_ether_setup(struct net_device *bond_dev)
1923{
1924 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1925
1926 ether_setup(bond_dev);
1927 bond_dev->flags |= IFF_MASTER | flags;
1928 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1929}
1930
1931void bond_xdp_set_features(struct net_device *bond_dev)
1932{
1933 struct bonding *bond = netdev_priv(bond_dev);
1934 xdp_features_t val = NETDEV_XDP_ACT_MASK;
1935 struct list_head *iter;
1936 struct slave *slave;
1937
1938 ASSERT_RTNL();
1939
1940 if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) {
1941 xdp_clear_features_flag(bond_dev);
1942 return;
1943 }
1944
1945 bond_for_each_slave(bond, slave, iter)
1946 val &= slave->dev->xdp_features;
1947
1948 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1949
1950 xdp_set_features_flag(bond_dev, val);
1951}
1952
1953/* enslave device <slave> to bond device <master> */
1954int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1955 struct netlink_ext_ack *extack)
1956{
1957 struct bonding *bond = netdev_priv(bond_dev);
1958 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1959 struct slave *new_slave = NULL, *prev_slave;
1960 struct sockaddr_storage ss;
1961 int link_reporting;
1962 int res = 0, i;
1963
1964 if (slave_dev->flags & IFF_MASTER &&
1965 !netif_is_bond_master(slave_dev)) {
1966 BOND_NL_ERR(bond_dev, extack,
1967 "Device type (master device) cannot be enslaved");
1968 return -EPERM;
1969 }
1970
1971 if (!bond->params.use_carrier &&
1972 slave_dev->ethtool_ops->get_link == NULL &&
1973 slave_ops->ndo_eth_ioctl == NULL) {
1974 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1975 }
1976
1977 /* already in-use? */
1978 if (netdev_is_rx_handler_busy(slave_dev)) {
1979 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1980 "Device is in use and cannot be enslaved");
1981 return -EBUSY;
1982 }
1983
1984 if (bond_dev == slave_dev) {
1985 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1986 return -EPERM;
1987 }
1988
1989 /* vlan challenged mutual exclusion */
1990 /* no need to lock since we're protected by rtnl_lock */
1991 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1992 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1993 if (vlan_uses_dev(bond_dev)) {
1994 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1995 "Can not enslave VLAN challenged device to VLAN enabled bond");
1996 return -EPERM;
1997 } else {
1998 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1999 }
2000 } else {
2001 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
2002 }
2003
2004 if (slave_dev->features & NETIF_F_HW_ESP)
2005 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
2006
2007 /* Old ifenslave binaries are no longer supported. These can
2008 * be identified with moderate accuracy by the state of the slave:
2009 * the current ifenslave will set the interface down prior to
2010 * enslaving it; the old ifenslave will not.
2011 */
2012 if (slave_dev->flags & IFF_UP) {
2013 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2014 "Device can not be enslaved while up");
2015 return -EPERM;
2016 }
2017
2018 /* set bonding device ether type by slave - bonding netdevices are
2019 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
2020 * there is a need to override some of the type dependent attribs/funcs.
2021 *
2022 * bond ether type mutual exclusion - don't allow slaves of dissimilar
2023 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
2024 */
2025 if (!bond_has_slaves(bond)) {
2026 if (bond_dev->type != slave_dev->type) {
2027 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
2028 bond_dev->type, slave_dev->type);
2029
2030 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
2031 bond_dev);
2032 res = notifier_to_errno(res);
2033 if (res) {
2034 slave_err(bond_dev, slave_dev, "refused to change device type\n");
2035 return -EBUSY;
2036 }
2037
2038 /* Flush unicast and multicast addresses */
2039 dev_uc_flush(bond_dev);
2040 dev_mc_flush(bond_dev);
2041
2042 if (slave_dev->type != ARPHRD_ETHER)
2043 bond_setup_by_slave(bond_dev, slave_dev);
2044 else
2045 bond_ether_setup(bond_dev);
2046
2047 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
2048 bond_dev);
2049 }
2050 } else if (bond_dev->type != slave_dev->type) {
2051 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2052 "Device type is different from other slaves");
2053 return -EINVAL;
2054 }
2055
2056 if (slave_dev->type == ARPHRD_INFINIBAND &&
2057 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2058 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2059 "Only active-backup mode is supported for infiniband slaves");
2060 res = -EOPNOTSUPP;
2061 goto err_undo_flags;
2062 }
2063
2064 if (!slave_ops->ndo_set_mac_address ||
2065 slave_dev->type == ARPHRD_INFINIBAND) {
2066 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
2067 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
2068 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2069 if (!bond_has_slaves(bond)) {
2070 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
2071 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
2072 } else {
2073 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2074 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
2075 res = -EOPNOTSUPP;
2076 goto err_undo_flags;
2077 }
2078 }
2079 }
2080
2081 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
2082
2083 /* If this is the first slave, then we need to set the master's hardware
2084 * address to be the same as the slave's.
2085 */
2086 if (!bond_has_slaves(bond) &&
2087 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
2088 res = bond_set_dev_addr(bond->dev, slave_dev);
2089 if (res)
2090 goto err_undo_flags;
2091 }
2092
2093 new_slave = bond_alloc_slave(bond, slave_dev);
2094 if (!new_slave) {
2095 res = -ENOMEM;
2096 goto err_undo_flags;
2097 }
2098
2099 /* Set the new_slave's queue_id to be zero. Queue ID mapping
2100 * is set via sysfs or module option if desired.
2101 */
2102 new_slave->queue_id = 0;
2103
2104 /* Save slave's original mtu and then set it to match the bond */
2105 new_slave->original_mtu = slave_dev->mtu;
2106 res = dev_set_mtu(slave_dev, bond->dev->mtu);
2107 if (res) {
2108 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
2109 goto err_free;
2110 }
2111
2112 /* Save slave's original ("permanent") mac address for modes
2113 * that need it, and for restoring it upon release, and then
2114 * set it to the master's address
2115 */
2116 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
2117 slave_dev->addr_len);
2118
2119 if (!bond->params.fail_over_mac ||
2120 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2121 /* Set slave to master's mac address. The application already
2122 * set the master's mac address to that of the first slave
2123 */
2124 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
2125 ss.ss_family = slave_dev->type;
2126 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
2127 extack);
2128 if (res) {
2129 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
2130 goto err_restore_mtu;
2131 }
2132 }
2133
2134 /* set no_addrconf flag before open to prevent IPv6 addrconf */
2135 slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2136
2137 /* open the slave since the application closed it */
2138 res = dev_open(slave_dev, extack);
2139 if (res) {
2140 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2141 goto err_restore_mac;
2142 }
2143
2144 slave_dev->priv_flags |= IFF_BONDING;
2145 /* initialize slave stats */
2146 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2147
2148 if (bond_is_lb(bond)) {
2149 /* bond_alb_init_slave() must be called before all other stages since
2150 * it might fail and we do not want to have to undo everything
2151 */
2152 res = bond_alb_init_slave(bond, new_slave);
2153 if (res)
2154 goto err_close;
2155 }
2156
2157 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2158 if (res) {
2159 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2160 goto err_close;
2161 }
2162
2163 prev_slave = bond_last_slave(bond);
2164
2165 new_slave->delay = 0;
2166 new_slave->link_failure_count = 0;
2167
2168 if (bond_update_speed_duplex(new_slave) &&
2169 bond_needs_speed_duplex(bond))
2170 new_slave->link = BOND_LINK_DOWN;
2171
2172 new_slave->last_rx = jiffies -
2173 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2174 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2175 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2176
2177 new_slave->last_tx = new_slave->last_rx;
2178
2179 if (bond->params.miimon && !bond->params.use_carrier) {
2180 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2181
2182 if ((link_reporting == -1) && !bond->params.arp_interval) {
2183 /* miimon is set but a bonded network driver
2184 * does not support ETHTOOL/MII and
2185 * arp_interval is not set. Note: if
2186 * use_carrier is enabled, we will never go
2187 * here (because netif_carrier is always
2188 * supported); thus, we don't need to change
2189 * the messages for netif_carrier.
2190 */
2191 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");
2192 } else if (link_reporting == -1) {
2193 /* unable get link status using mii/ethtool */
2194 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");
2195 }
2196 }
2197
2198 /* check for initial state */
2199 new_slave->link = BOND_LINK_NOCHANGE;
2200 if (bond->params.miimon) {
2201 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2202 if (bond->params.updelay) {
2203 bond_set_slave_link_state(new_slave,
2204 BOND_LINK_BACK,
2205 BOND_SLAVE_NOTIFY_NOW);
2206 new_slave->delay = bond->params.updelay;
2207 } else {
2208 bond_set_slave_link_state(new_slave,
2209 BOND_LINK_UP,
2210 BOND_SLAVE_NOTIFY_NOW);
2211 }
2212 } else {
2213 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2214 BOND_SLAVE_NOTIFY_NOW);
2215 }
2216 } else if (bond->params.arp_interval) {
2217 bond_set_slave_link_state(new_slave,
2218 (netif_carrier_ok(slave_dev) ?
2219 BOND_LINK_UP : BOND_LINK_DOWN),
2220 BOND_SLAVE_NOTIFY_NOW);
2221 } else {
2222 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2223 BOND_SLAVE_NOTIFY_NOW);
2224 }
2225
2226 if (new_slave->link != BOND_LINK_DOWN)
2227 new_slave->last_link_up = jiffies;
2228 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2229 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2230 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2231
2232 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2233 /* if there is a primary slave, remember it */
2234 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2235 rcu_assign_pointer(bond->primary_slave, new_slave);
2236 bond->force_primary = true;
2237 }
2238 }
2239
2240 switch (BOND_MODE(bond)) {
2241 case BOND_MODE_ACTIVEBACKUP:
2242 bond_set_slave_inactive_flags(new_slave,
2243 BOND_SLAVE_NOTIFY_NOW);
2244 break;
2245 case BOND_MODE_8023AD:
2246 /* in 802.3ad mode, the internal mechanism
2247 * will activate the slaves in the selected
2248 * aggregator
2249 */
2250 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2251 /* if this is the first slave */
2252 if (!prev_slave) {
2253 SLAVE_AD_INFO(new_slave)->id = 1;
2254 /* Initialize AD with the number of times that the AD timer is called in 1 second
2255 * can be called only after the mac address of the bond is set
2256 */
2257 bond_3ad_initialize(bond);
2258 } else {
2259 SLAVE_AD_INFO(new_slave)->id =
2260 SLAVE_AD_INFO(prev_slave)->id + 1;
2261 }
2262
2263 bond_3ad_bind_slave(new_slave);
2264 break;
2265 case BOND_MODE_TLB:
2266 case BOND_MODE_ALB:
2267 bond_set_active_slave(new_slave);
2268 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2269 break;
2270 default:
2271 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2272
2273 /* always active in trunk mode */
2274 bond_set_active_slave(new_slave);
2275
2276 /* In trunking mode there is little meaning to curr_active_slave
2277 * anyway (it holds no special properties of the bond device),
2278 * so we can change it without calling change_active_interface()
2279 */
2280 if (!rcu_access_pointer(bond->curr_active_slave) &&
2281 new_slave->link == BOND_LINK_UP)
2282 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2283
2284 break;
2285 } /* switch(bond_mode) */
2286
2287#ifdef CONFIG_NET_POLL_CONTROLLER
2288 if (bond->dev->npinfo) {
2289 if (slave_enable_netpoll(new_slave)) {
2290 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2291 res = -EBUSY;
2292 goto err_detach;
2293 }
2294 }
2295#endif
2296
2297 if (!(bond_dev->features & NETIF_F_LRO))
2298 dev_disable_lro(slave_dev);
2299
2300 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2301 new_slave);
2302 if (res) {
2303 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2304 goto err_detach;
2305 }
2306
2307 res = bond_master_upper_dev_link(bond, new_slave, extack);
2308 if (res) {
2309 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2310 goto err_unregister;
2311 }
2312
2313 bond_lower_state_changed(new_slave);
2314
2315 res = bond_sysfs_slave_add(new_slave);
2316 if (res) {
2317 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2318 goto err_upper_unlink;
2319 }
2320
2321 /* If the mode uses primary, then the following is handled by
2322 * bond_change_active_slave().
2323 */
2324 if (!bond_uses_primary(bond)) {
2325 /* set promiscuity level to new slave */
2326 if (bond_dev->flags & IFF_PROMISC) {
2327 res = dev_set_promiscuity(slave_dev, 1);
2328 if (res)
2329 goto err_sysfs_del;
2330 }
2331
2332 /* set allmulti level to new slave */
2333 if (bond_dev->flags & IFF_ALLMULTI) {
2334 res = dev_set_allmulti(slave_dev, 1);
2335 if (res) {
2336 if (bond_dev->flags & IFF_PROMISC)
2337 dev_set_promiscuity(slave_dev, -1);
2338 goto err_sysfs_del;
2339 }
2340 }
2341
2342 if (bond_dev->flags & IFF_UP) {
2343 netif_addr_lock_bh(bond_dev);
2344 dev_mc_sync_multiple(slave_dev, bond_dev);
2345 dev_uc_sync_multiple(slave_dev, bond_dev);
2346 netif_addr_unlock_bh(bond_dev);
2347
2348 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2349 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2350 }
2351 }
2352
2353 bond->slave_cnt++;
2354 bond_compute_features(bond);
2355 bond_set_carrier(bond);
2356
2357 /* Needs to be called before bond_select_active_slave(), which will
2358 * remove the maddrs if the slave is selected as active slave.
2359 */
2360 bond_slave_ns_maddrs_add(bond, new_slave);
2361
2362 if (bond_uses_primary(bond)) {
2363 block_netpoll_tx();
2364 bond_select_active_slave(bond);
2365 unblock_netpoll_tx();
2366 }
2367
2368 if (bond_mode_can_use_xmit_hash(bond))
2369 bond_update_slave_arr(bond, NULL);
2370
2371 if (!slave_dev->netdev_ops->ndo_bpf ||
2372 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2373 if (bond->xdp_prog) {
2374 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2375 "Slave does not support XDP");
2376 res = -EOPNOTSUPP;
2377 goto err_sysfs_del;
2378 }
2379 } else if (bond->xdp_prog) {
2380 struct netdev_bpf xdp = {
2381 .command = XDP_SETUP_PROG,
2382 .flags = 0,
2383 .prog = bond->xdp_prog,
2384 .extack = extack,
2385 };
2386
2387 if (dev_xdp_prog_count(slave_dev) > 0) {
2388 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2389 "Slave has XDP program loaded, please unload before enslaving");
2390 res = -EOPNOTSUPP;
2391 goto err_sysfs_del;
2392 }
2393
2394 res = dev_xdp_propagate(slave_dev, &xdp);
2395 if (res < 0) {
2396 /* ndo_bpf() sets extack error message */
2397 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2398 goto err_sysfs_del;
2399 }
2400 if (bond->xdp_prog)
2401 bpf_prog_inc(bond->xdp_prog);
2402 }
2403
2404 bond_xdp_set_features(bond_dev);
2405
2406 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2407 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2408 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2409
2410 /* enslave is successful */
2411 bond_queue_slave_event(new_slave);
2412 return 0;
2413
2414/* Undo stages on error */
2415err_sysfs_del:
2416 bond_sysfs_slave_del(new_slave);
2417
2418err_upper_unlink:
2419 bond_upper_dev_unlink(bond, new_slave);
2420
2421err_unregister:
2422 netdev_rx_handler_unregister(slave_dev);
2423
2424err_detach:
2425 vlan_vids_del_by_dev(slave_dev, bond_dev);
2426 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2427 RCU_INIT_POINTER(bond->primary_slave, NULL);
2428 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2429 block_netpoll_tx();
2430 bond_change_active_slave(bond, NULL);
2431 bond_select_active_slave(bond);
2432 unblock_netpoll_tx();
2433 }
2434 /* either primary_slave or curr_active_slave might've changed */
2435 synchronize_rcu();
2436 slave_disable_netpoll(new_slave);
2437
2438err_close:
2439 if (!netif_is_bond_master(slave_dev))
2440 slave_dev->priv_flags &= ~IFF_BONDING;
2441 dev_close(slave_dev);
2442
2443err_restore_mac:
2444 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2445 if (!bond->params.fail_over_mac ||
2446 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2447 /* XXX TODO - fom follow mode needs to change master's
2448 * MAC if this slave's MAC is in use by the bond, or at
2449 * least print a warning.
2450 */
2451 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2452 new_slave->dev->addr_len);
2453 ss.ss_family = slave_dev->type;
2454 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2455 }
2456
2457err_restore_mtu:
2458 dev_set_mtu(slave_dev, new_slave->original_mtu);
2459
2460err_free:
2461 kobject_put(&new_slave->kobj);
2462
2463err_undo_flags:
2464 /* Enslave of first slave has failed and we need to fix master's mac */
2465 if (!bond_has_slaves(bond)) {
2466 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2467 slave_dev->dev_addr))
2468 eth_hw_addr_random(bond_dev);
2469 if (bond_dev->type != ARPHRD_ETHER) {
2470 dev_close(bond_dev);
2471 bond_ether_setup(bond_dev);
2472 }
2473 }
2474
2475 return res;
2476}
2477
2478/* Try to release the slave device <slave> from the bond device <master>
2479 * It is legal to access curr_active_slave without a lock because all the function
2480 * is RTNL-locked. If "all" is true it means that the function is being called
2481 * while destroying a bond interface and all slaves are being released.
2482 *
2483 * The rules for slave state should be:
2484 * for Active/Backup:
2485 * Active stays on all backups go down
2486 * for Bonded connections:
2487 * The first up interface should be left on and all others downed.
2488 */
2489static int __bond_release_one(struct net_device *bond_dev,
2490 struct net_device *slave_dev,
2491 bool all, bool unregister)
2492{
2493 struct bonding *bond = netdev_priv(bond_dev);
2494 struct slave *slave, *oldcurrent;
2495 struct sockaddr_storage ss;
2496 int old_flags = bond_dev->flags;
2497 netdev_features_t old_features = bond_dev->features;
2498
2499 /* slave is not a slave or master is not master of this slave */
2500 if (!(slave_dev->flags & IFF_SLAVE) ||
2501 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2502 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2503 return -EINVAL;
2504 }
2505
2506 block_netpoll_tx();
2507
2508 slave = bond_get_slave_by_dev(bond, slave_dev);
2509 if (!slave) {
2510 /* not a slave of this bond */
2511 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2512 unblock_netpoll_tx();
2513 return -EINVAL;
2514 }
2515
2516 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2517
2518 bond_sysfs_slave_del(slave);
2519
2520 /* recompute stats just before removing the slave */
2521 bond_get_stats(bond->dev, &bond->bond_stats);
2522
2523 if (bond->xdp_prog) {
2524 struct netdev_bpf xdp = {
2525 .command = XDP_SETUP_PROG,
2526 .flags = 0,
2527 .prog = NULL,
2528 .extack = NULL,
2529 };
2530 if (dev_xdp_propagate(slave_dev, &xdp))
2531 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2532 }
2533
2534 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2535 * for this slave anymore.
2536 */
2537 netdev_rx_handler_unregister(slave_dev);
2538
2539 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2540 bond_3ad_unbind_slave(slave);
2541
2542 bond_upper_dev_unlink(bond, slave);
2543
2544 if (bond_mode_can_use_xmit_hash(bond))
2545 bond_update_slave_arr(bond, slave);
2546
2547 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2548 bond_is_active_slave(slave) ? "active" : "backup");
2549
2550 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2551
2552 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2553
2554 if (!all && (!bond->params.fail_over_mac ||
2555 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2556 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2557 bond_has_slaves(bond))
2558 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",
2559 slave->perm_hwaddr);
2560 }
2561
2562 if (rtnl_dereference(bond->primary_slave) == slave)
2563 RCU_INIT_POINTER(bond->primary_slave, NULL);
2564
2565 if (oldcurrent == slave)
2566 bond_change_active_slave(bond, NULL);
2567
2568 /* Must be called after bond_change_active_slave () as the slave
2569 * might change from an active slave to a backup slave. Then it is
2570 * necessary to clear the maddrs on the backup slave.
2571 */
2572 bond_slave_ns_maddrs_del(bond, slave);
2573
2574 if (bond_is_lb(bond)) {
2575 /* Must be called only after the slave has been
2576 * detached from the list and the curr_active_slave
2577 * has been cleared (if our_slave == old_current),
2578 * but before a new active slave is selected.
2579 */
2580 bond_alb_deinit_slave(bond, slave);
2581 }
2582
2583 if (all) {
2584 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2585 } else if (oldcurrent == slave) {
2586 /* Note that we hold RTNL over this sequence, so there
2587 * is no concern that another slave add/remove event
2588 * will interfere.
2589 */
2590 bond_select_active_slave(bond);
2591 }
2592
2593 bond_set_carrier(bond);
2594 if (!bond_has_slaves(bond))
2595 eth_hw_addr_random(bond_dev);
2596
2597 unblock_netpoll_tx();
2598 synchronize_rcu();
2599 bond->slave_cnt--;
2600
2601 if (!bond_has_slaves(bond)) {
2602 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2603 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2604 }
2605
2606 bond_compute_features(bond);
2607 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2608 (old_features & NETIF_F_VLAN_CHALLENGED))
2609 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2610
2611 vlan_vids_del_by_dev(slave_dev, bond_dev);
2612
2613 /* If the mode uses primary, then this case was handled above by
2614 * bond_change_active_slave(..., NULL)
2615 */
2616 if (!bond_uses_primary(bond)) {
2617 /* unset promiscuity level from slave
2618 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2619 * of the IFF_PROMISC flag in the bond_dev, but we need the
2620 * value of that flag before that change, as that was the value
2621 * when this slave was attached, so we cache at the start of the
2622 * function and use it here. Same goes for ALLMULTI below
2623 */
2624 if (old_flags & IFF_PROMISC)
2625 dev_set_promiscuity(slave_dev, -1);
2626
2627 /* unset allmulti level from slave */
2628 if (old_flags & IFF_ALLMULTI)
2629 dev_set_allmulti(slave_dev, -1);
2630
2631 if (old_flags & IFF_UP)
2632 bond_hw_addr_flush(bond_dev, slave_dev);
2633 }
2634
2635 slave_disable_netpoll(slave);
2636
2637 /* close slave before restoring its mac address */
2638 dev_close(slave_dev);
2639
2640 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2641
2642 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2643 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2644 /* restore original ("permanent") mac address */
2645 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2646 slave->dev->addr_len);
2647 ss.ss_family = slave_dev->type;
2648 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2649 }
2650
2651 if (unregister)
2652 __dev_set_mtu(slave_dev, slave->original_mtu);
2653 else
2654 dev_set_mtu(slave_dev, slave->original_mtu);
2655
2656 if (!netif_is_bond_master(slave_dev))
2657 slave_dev->priv_flags &= ~IFF_BONDING;
2658
2659 bond_xdp_set_features(bond_dev);
2660 kobject_put(&slave->kobj);
2661
2662 return 0;
2663}
2664
2665/* A wrapper used because of ndo_del_link */
2666int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2667{
2668 return __bond_release_one(bond_dev, slave_dev, false, false);
2669}
2670
2671/* First release a slave and then destroy the bond if no more slaves are left.
2672 * Must be under rtnl_lock when this function is called.
2673 */
2674static int bond_release_and_destroy(struct net_device *bond_dev,
2675 struct net_device *slave_dev)
2676{
2677 struct bonding *bond = netdev_priv(bond_dev);
2678 int ret;
2679
2680 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2681 if (ret == 0 && !bond_has_slaves(bond) &&
2682 bond_dev->reg_state != NETREG_UNREGISTERING) {
2683 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2684 netdev_info(bond_dev, "Destroying bond\n");
2685 bond_remove_proc_entry(bond);
2686 unregister_netdevice(bond_dev);
2687 }
2688 return ret;
2689}
2690
2691static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2692{
2693 struct bonding *bond = netdev_priv(bond_dev);
2694
2695 bond_fill_ifbond(bond, info);
2696}
2697
2698static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2699{
2700 struct bonding *bond = netdev_priv(bond_dev);
2701 struct list_head *iter;
2702 int i = 0, res = -ENODEV;
2703 struct slave *slave;
2704
2705 bond_for_each_slave(bond, slave, iter) {
2706 if (i++ == (int)info->slave_id) {
2707 res = 0;
2708 bond_fill_ifslave(slave, info);
2709 break;
2710 }
2711 }
2712
2713 return res;
2714}
2715
2716/*-------------------------------- Monitoring -------------------------------*/
2717
2718/* called with rcu_read_lock() */
2719static int bond_miimon_inspect(struct bonding *bond)
2720{
2721 bool ignore_updelay = false;
2722 int link_state, commit = 0;
2723 struct list_head *iter;
2724 struct slave *slave;
2725
2726 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2727 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2728 } else {
2729 struct bond_up_slave *usable_slaves;
2730
2731 usable_slaves = rcu_dereference(bond->usable_slaves);
2732
2733 if (usable_slaves && usable_slaves->count == 0)
2734 ignore_updelay = true;
2735 }
2736
2737 bond_for_each_slave_rcu(bond, slave, iter) {
2738 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2739
2740 link_state = bond_check_dev_link(bond, slave->dev, 0);
2741
2742 switch (slave->link) {
2743 case BOND_LINK_UP:
2744 if (link_state)
2745 continue;
2746
2747 bond_propose_link_state(slave, BOND_LINK_FAIL);
2748 commit++;
2749 slave->delay = bond->params.downdelay;
2750 if (slave->delay && net_ratelimit()) {
2751 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2752 (BOND_MODE(bond) ==
2753 BOND_MODE_ACTIVEBACKUP) ?
2754 (bond_is_active_slave(slave) ?
2755 "active " : "backup ") : "",
2756 bond->params.downdelay * bond->params.miimon);
2757 }
2758 fallthrough;
2759 case BOND_LINK_FAIL:
2760 if (link_state) {
2761 /* recovered before downdelay expired */
2762 bond_propose_link_state(slave, BOND_LINK_UP);
2763 slave->last_link_up = jiffies;
2764 if (net_ratelimit())
2765 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2766 (bond->params.downdelay - slave->delay) *
2767 bond->params.miimon);
2768 commit++;
2769 continue;
2770 }
2771
2772 if (slave->delay <= 0) {
2773 bond_propose_link_state(slave, BOND_LINK_DOWN);
2774 commit++;
2775 continue;
2776 }
2777
2778 slave->delay--;
2779 break;
2780
2781 case BOND_LINK_DOWN:
2782 if (!link_state)
2783 continue;
2784
2785 bond_propose_link_state(slave, BOND_LINK_BACK);
2786 commit++;
2787 slave->delay = bond->params.updelay;
2788
2789 if (slave->delay && net_ratelimit()) {
2790 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2791 ignore_updelay ? 0 :
2792 bond->params.updelay *
2793 bond->params.miimon);
2794 }
2795 fallthrough;
2796 case BOND_LINK_BACK:
2797 if (!link_state) {
2798 bond_propose_link_state(slave, BOND_LINK_DOWN);
2799 if (net_ratelimit())
2800 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2801 (bond->params.updelay - slave->delay) *
2802 bond->params.miimon);
2803 commit++;
2804 continue;
2805 }
2806
2807 if (ignore_updelay)
2808 slave->delay = 0;
2809
2810 if (slave->delay <= 0) {
2811 bond_propose_link_state(slave, BOND_LINK_UP);
2812 commit++;
2813 ignore_updelay = false;
2814 continue;
2815 }
2816
2817 slave->delay--;
2818 break;
2819 }
2820 }
2821
2822 return commit;
2823}
2824
2825static void bond_miimon_link_change(struct bonding *bond,
2826 struct slave *slave,
2827 char link)
2828{
2829 switch (BOND_MODE(bond)) {
2830 case BOND_MODE_8023AD:
2831 bond_3ad_handle_link_change(slave, link);
2832 break;
2833 case BOND_MODE_TLB:
2834 case BOND_MODE_ALB:
2835 bond_alb_handle_link_change(bond, slave, link);
2836 break;
2837 case BOND_MODE_XOR:
2838 bond_update_slave_arr(bond, NULL);
2839 break;
2840 }
2841}
2842
2843static void bond_miimon_commit(struct bonding *bond)
2844{
2845 struct slave *slave, *primary, *active;
2846 bool do_failover = false;
2847 struct list_head *iter;
2848
2849 ASSERT_RTNL();
2850
2851 bond_for_each_slave(bond, slave, iter) {
2852 switch (slave->link_new_state) {
2853 case BOND_LINK_NOCHANGE:
2854 /* For 802.3ad mode, check current slave speed and
2855 * duplex again in case its port was disabled after
2856 * invalid speed/duplex reporting but recovered before
2857 * link monitoring could make a decision on the actual
2858 * link status
2859 */
2860 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2861 slave->link == BOND_LINK_UP)
2862 bond_3ad_adapter_speed_duplex_changed(slave);
2863 continue;
2864
2865 case BOND_LINK_UP:
2866 if (bond_update_speed_duplex(slave) &&
2867 bond_needs_speed_duplex(bond)) {
2868 slave->link = BOND_LINK_DOWN;
2869 if (net_ratelimit())
2870 slave_warn(bond->dev, slave->dev,
2871 "failed to get link speed/duplex\n");
2872 continue;
2873 }
2874 bond_set_slave_link_state(slave, BOND_LINK_UP,
2875 BOND_SLAVE_NOTIFY_NOW);
2876 slave->last_link_up = jiffies;
2877
2878 primary = rtnl_dereference(bond->primary_slave);
2879 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2880 /* prevent it from being the active one */
2881 bond_set_backup_slave(slave);
2882 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2883 /* make it immediately active */
2884 bond_set_active_slave(slave);
2885 }
2886
2887 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2888 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2889 slave->duplex ? "full" : "half");
2890
2891 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2892
2893 active = rtnl_dereference(bond->curr_active_slave);
2894 if (!active || slave == primary || slave->prio > active->prio)
2895 do_failover = true;
2896
2897 continue;
2898
2899 case BOND_LINK_DOWN:
2900 if (slave->link_failure_count < UINT_MAX)
2901 slave->link_failure_count++;
2902
2903 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2904 BOND_SLAVE_NOTIFY_NOW);
2905
2906 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2907 BOND_MODE(bond) == BOND_MODE_8023AD)
2908 bond_set_slave_inactive_flags(slave,
2909 BOND_SLAVE_NOTIFY_NOW);
2910
2911 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2912
2913 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2914
2915 if (slave == rcu_access_pointer(bond->curr_active_slave))
2916 do_failover = true;
2917
2918 continue;
2919
2920 default:
2921 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2922 slave->link_new_state);
2923 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2924
2925 continue;
2926 }
2927 }
2928
2929 if (do_failover) {
2930 block_netpoll_tx();
2931 bond_select_active_slave(bond);
2932 unblock_netpoll_tx();
2933 }
2934
2935 bond_set_carrier(bond);
2936}
2937
2938/* bond_mii_monitor
2939 *
2940 * Really a wrapper that splits the mii monitor into two phases: an
2941 * inspection, then (if inspection indicates something needs to be done)
2942 * an acquisition of appropriate locks followed by a commit phase to
2943 * implement whatever link state changes are indicated.
2944 */
2945static void bond_mii_monitor(struct work_struct *work)
2946{
2947 struct bonding *bond = container_of(work, struct bonding,
2948 mii_work.work);
2949 bool should_notify_peers = false;
2950 bool commit;
2951 unsigned long delay;
2952 struct slave *slave;
2953 struct list_head *iter;
2954
2955 delay = msecs_to_jiffies(bond->params.miimon);
2956
2957 if (!bond_has_slaves(bond))
2958 goto re_arm;
2959
2960 rcu_read_lock();
2961 should_notify_peers = bond_should_notify_peers(bond);
2962 commit = !!bond_miimon_inspect(bond);
2963 if (bond->send_peer_notif) {
2964 rcu_read_unlock();
2965 if (rtnl_trylock()) {
2966 bond->send_peer_notif--;
2967 rtnl_unlock();
2968 }
2969 } else {
2970 rcu_read_unlock();
2971 }
2972
2973 if (commit) {
2974 /* Race avoidance with bond_close cancel of workqueue */
2975 if (!rtnl_trylock()) {
2976 delay = 1;
2977 should_notify_peers = false;
2978 goto re_arm;
2979 }
2980
2981 bond_for_each_slave(bond, slave, iter) {
2982 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2983 }
2984 bond_miimon_commit(bond);
2985
2986 rtnl_unlock(); /* might sleep, hold no other locks */
2987 }
2988
2989re_arm:
2990 if (bond->params.miimon)
2991 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2992
2993 if (should_notify_peers) {
2994 if (!rtnl_trylock())
2995 return;
2996 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2997 rtnl_unlock();
2998 }
2999}
3000
3001static int bond_upper_dev_walk(struct net_device *upper,
3002 struct netdev_nested_priv *priv)
3003{
3004 __be32 ip = *(__be32 *)priv->data;
3005
3006 return ip == bond_confirm_addr(upper, 0, ip);
3007}
3008
3009static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
3010{
3011 struct netdev_nested_priv priv = {
3012 .data = (void *)&ip,
3013 };
3014 bool ret = false;
3015
3016 if (ip == bond_confirm_addr(bond->dev, 0, ip))
3017 return true;
3018
3019 rcu_read_lock();
3020 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
3021 ret = true;
3022 rcu_read_unlock();
3023
3024 return ret;
3025}
3026
3027#define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
3028
3029static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
3030 struct sk_buff *skb)
3031{
3032 struct net_device *bond_dev = slave->bond->dev;
3033 struct net_device *slave_dev = slave->dev;
3034 struct bond_vlan_tag *outer_tag = tags;
3035
3036 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
3037 return true;
3038
3039 tags++;
3040
3041 /* Go through all the tags backwards and add them to the packet */
3042 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
3043 if (!tags->vlan_id) {
3044 tags++;
3045 continue;
3046 }
3047
3048 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
3049 ntohs(outer_tag->vlan_proto), tags->vlan_id);
3050 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
3051 tags->vlan_id);
3052 if (!skb) {
3053 net_err_ratelimited("failed to insert inner VLAN tag\n");
3054 return false;
3055 }
3056
3057 tags++;
3058 }
3059 /* Set the outer tag */
3060 if (outer_tag->vlan_id) {
3061 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
3062 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
3063 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
3064 outer_tag->vlan_id);
3065 }
3066
3067 return true;
3068}
3069
3070/* We go to the (large) trouble of VLAN tagging ARP frames because
3071 * switches in VLAN mode (especially if ports are configured as
3072 * "native" to a VLAN) might not pass non-tagged frames.
3073 */
3074static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
3075 __be32 src_ip, struct bond_vlan_tag *tags)
3076{
3077 struct net_device *bond_dev = slave->bond->dev;
3078 struct net_device *slave_dev = slave->dev;
3079 struct sk_buff *skb;
3080
3081 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
3082 arp_op, &dest_ip, &src_ip);
3083
3084 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
3085 NULL, slave_dev->dev_addr, NULL);
3086
3087 if (!skb) {
3088 net_err_ratelimited("ARP packet allocation failed\n");
3089 return;
3090 }
3091
3092 if (bond_handle_vlan(slave, tags, skb)) {
3093 slave_update_last_tx(slave);
3094 arp_xmit(skb);
3095 }
3096
3097 return;
3098}
3099
3100/* Validate the device path between the @start_dev and the @end_dev.
3101 * The path is valid if the @end_dev is reachable through device
3102 * stacking.
3103 * When the path is validated, collect any vlan information in the
3104 * path.
3105 */
3106struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
3107 struct net_device *end_dev,
3108 int level)
3109{
3110 struct bond_vlan_tag *tags;
3111 struct net_device *upper;
3112 struct list_head *iter;
3113
3114 if (start_dev == end_dev) {
3115 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
3116 if (!tags)
3117 return ERR_PTR(-ENOMEM);
3118 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
3119 return tags;
3120 }
3121
3122 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
3123 tags = bond_verify_device_path(upper, end_dev, level + 1);
3124 if (IS_ERR_OR_NULL(tags)) {
3125 if (IS_ERR(tags))
3126 return tags;
3127 continue;
3128 }
3129 if (is_vlan_dev(upper)) {
3130 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
3131 tags[level].vlan_id = vlan_dev_vlan_id(upper);
3132 }
3133
3134 return tags;
3135 }
3136
3137 return NULL;
3138}
3139
3140static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
3141{
3142 struct rtable *rt;
3143 struct bond_vlan_tag *tags;
3144 __be32 *targets = bond->params.arp_targets, addr;
3145 int i;
3146
3147 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3148 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3149 __func__, &targets[i]);
3150 tags = NULL;
3151
3152 /* Find out through which dev should the packet go */
3153 rt = ip_route_output(dev_net(bond->dev), targets[i], 0, 0, 0,
3154 RT_SCOPE_LINK);
3155 if (IS_ERR(rt)) {
3156 /* there's no route to target - try to send arp
3157 * probe to generate any traffic (arp_validate=0)
3158 */
3159 if (bond->params.arp_validate)
3160 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3161 bond->dev->name,
3162 &targets[i]);
3163 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3164 0, tags);
3165 continue;
3166 }
3167
3168 /* bond device itself */
3169 if (rt->dst.dev == bond->dev)
3170 goto found;
3171
3172 rcu_read_lock();
3173 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3174 rcu_read_unlock();
3175
3176 if (!IS_ERR_OR_NULL(tags))
3177 goto found;
3178
3179 /* Not our device - skip */
3180 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3181 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3182
3183 ip_rt_put(rt);
3184 continue;
3185
3186found:
3187 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3188 ip_rt_put(rt);
3189 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3190 kfree(tags);
3191 }
3192}
3193
3194static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3195{
3196 int i;
3197
3198 if (!sip || !bond_has_this_ip(bond, tip)) {
3199 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3200 __func__, &sip, &tip);
3201 return;
3202 }
3203
3204 i = bond_get_targets_ip(bond->params.arp_targets, sip);
3205 if (i == -1) {
3206 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3207 __func__, &sip);
3208 return;
3209 }
3210 slave->last_rx = jiffies;
3211 slave->target_last_arp_rx[i] = jiffies;
3212}
3213
3214static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3215 struct slave *slave)
3216{
3217 struct arphdr *arp = (struct arphdr *)skb->data;
3218 struct slave *curr_active_slave, *curr_arp_slave;
3219 unsigned char *arp_ptr;
3220 __be32 sip, tip;
3221 unsigned int alen;
3222
3223 alen = arp_hdr_len(bond->dev);
3224
3225 if (alen > skb_headlen(skb)) {
3226 arp = kmalloc(alen, GFP_ATOMIC);
3227 if (!arp)
3228 goto out_unlock;
3229 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3230 goto out_unlock;
3231 }
3232
3233 if (arp->ar_hln != bond->dev->addr_len ||
3234 skb->pkt_type == PACKET_OTHERHOST ||
3235 skb->pkt_type == PACKET_LOOPBACK ||
3236 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3237 arp->ar_pro != htons(ETH_P_IP) ||
3238 arp->ar_pln != 4)
3239 goto out_unlock;
3240
3241 arp_ptr = (unsigned char *)(arp + 1);
3242 arp_ptr += bond->dev->addr_len;
3243 memcpy(&sip, arp_ptr, 4);
3244 arp_ptr += 4 + bond->dev->addr_len;
3245 memcpy(&tip, arp_ptr, 4);
3246
3247 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3248 __func__, slave->dev->name, bond_slave_state(slave),
3249 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3250 &sip, &tip);
3251
3252 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3253 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3254
3255 /* We 'trust' the received ARP enough to validate it if:
3256 *
3257 * (a) the slave receiving the ARP is active (which includes the
3258 * current ARP slave, if any), or
3259 *
3260 * (b) the receiving slave isn't active, but there is a currently
3261 * active slave and it received valid arp reply(s) after it became
3262 * the currently active slave, or
3263 *
3264 * (c) there is an ARP slave that sent an ARP during the prior ARP
3265 * interval, and we receive an ARP reply on any slave. We accept
3266 * these because switch FDB update delays may deliver the ARP
3267 * reply to a slave other than the sender of the ARP request.
3268 *
3269 * Note: for (b), backup slaves are receiving the broadcast ARP
3270 * request, not a reply. This request passes from the sending
3271 * slave through the L2 switch(es) to the receiving slave. Since
3272 * this is checking the request, sip/tip are swapped for
3273 * validation.
3274 *
3275 * This is done to avoid endless looping when we can't reach the
3276 * arp_ip_target and fool ourselves with our own arp requests.
3277 */
3278 if (bond_is_active_slave(slave))
3279 bond_validate_arp(bond, slave, sip, tip);
3280 else if (curr_active_slave &&
3281 time_after(slave_last_rx(bond, curr_active_slave),
3282 curr_active_slave->last_link_up))
3283 bond_validate_arp(bond, slave, tip, sip);
3284 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3285 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3286 bond_validate_arp(bond, slave, sip, tip);
3287
3288out_unlock:
3289 if (arp != (struct arphdr *)skb->data)
3290 kfree(arp);
3291 return RX_HANDLER_ANOTHER;
3292}
3293
3294#if IS_ENABLED(CONFIG_IPV6)
3295static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3296 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3297{
3298 struct net_device *bond_dev = slave->bond->dev;
3299 struct net_device *slave_dev = slave->dev;
3300 struct in6_addr mcaddr;
3301 struct sk_buff *skb;
3302
3303 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3304 daddr, saddr);
3305
3306 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3307 if (!skb) {
3308 net_err_ratelimited("NS packet allocation failed\n");
3309 return;
3310 }
3311
3312 addrconf_addr_solict_mult(daddr, &mcaddr);
3313 if (bond_handle_vlan(slave, tags, skb)) {
3314 slave_update_last_tx(slave);
3315 ndisc_send_skb(skb, &mcaddr, saddr);
3316 }
3317}
3318
3319static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3320{
3321 struct in6_addr *targets = bond->params.ns_targets;
3322 struct bond_vlan_tag *tags;
3323 struct dst_entry *dst;
3324 struct in6_addr saddr;
3325 struct flowi6 fl6;
3326 int i;
3327
3328 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3329 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3330 __func__, &targets[i]);
3331 tags = NULL;
3332
3333 /* Find out through which dev should the packet go */
3334 memset(&fl6, 0, sizeof(struct flowi6));
3335 fl6.daddr = targets[i];
3336 fl6.flowi6_oif = bond->dev->ifindex;
3337
3338 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3339 if (dst->error) {
3340 dst_release(dst);
3341 /* there's no route to target - try to send arp
3342 * probe to generate any traffic (arp_validate=0)
3343 */
3344 if (bond->params.arp_validate)
3345 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3346 bond->dev->name,
3347 &targets[i]);
3348 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3349 continue;
3350 }
3351
3352 /* bond device itself */
3353 if (dst->dev == bond->dev)
3354 goto found;
3355
3356 rcu_read_lock();
3357 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3358 rcu_read_unlock();
3359
3360 if (!IS_ERR_OR_NULL(tags))
3361 goto found;
3362
3363 /* Not our device - skip */
3364 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3365 &targets[i], dst->dev ? dst->dev->name : "NULL");
3366
3367 dst_release(dst);
3368 continue;
3369
3370found:
3371 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3372 bond_ns_send(slave, &targets[i], &saddr, tags);
3373 else
3374 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3375
3376 dst_release(dst);
3377 kfree(tags);
3378 }
3379}
3380
3381static int bond_confirm_addr6(struct net_device *dev,
3382 struct netdev_nested_priv *priv)
3383{
3384 struct in6_addr *addr = (struct in6_addr *)priv->data;
3385
3386 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3387}
3388
3389static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3390{
3391 struct netdev_nested_priv priv = {
3392 .data = addr,
3393 };
3394 int ret = false;
3395
3396 if (bond_confirm_addr6(bond->dev, &priv))
3397 return true;
3398
3399 rcu_read_lock();
3400 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3401 ret = true;
3402 rcu_read_unlock();
3403
3404 return ret;
3405}
3406
3407static void bond_validate_na(struct bonding *bond, struct slave *slave,
3408 struct in6_addr *saddr, struct in6_addr *daddr)
3409{
3410 int i;
3411
3412 /* Ignore NAs that:
3413 * 1. Source address is unspecified address.
3414 * 2. Dest address is neither all-nodes multicast address nor
3415 * exist on bond interface.
3416 */
3417 if (ipv6_addr_any(saddr) ||
3418 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3419 !bond_has_this_ip6(bond, daddr))) {
3420 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3421 __func__, saddr, daddr);
3422 return;
3423 }
3424
3425 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3426 if (i == -1) {
3427 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3428 __func__, saddr);
3429 return;
3430 }
3431 slave->last_rx = jiffies;
3432 slave->target_last_arp_rx[i] = jiffies;
3433}
3434
3435static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3436 struct slave *slave)
3437{
3438 struct slave *curr_active_slave, *curr_arp_slave;
3439 struct in6_addr *saddr, *daddr;
3440 struct {
3441 struct ipv6hdr ip6;
3442 struct icmp6hdr icmp6;
3443 } *combined, _combined;
3444
3445 if (skb->pkt_type == PACKET_OTHERHOST ||
3446 skb->pkt_type == PACKET_LOOPBACK)
3447 goto out;
3448
3449 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3450 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3451 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3452 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3453 goto out;
3454
3455 saddr = &combined->ip6.saddr;
3456 daddr = &combined->ip6.daddr;
3457
3458 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3459 __func__, slave->dev->name, bond_slave_state(slave),
3460 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3461 saddr, daddr);
3462
3463 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3464 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3465
3466 /* We 'trust' the received ARP enough to validate it if:
3467 * see bond_arp_rcv().
3468 */
3469 if (bond_is_active_slave(slave))
3470 bond_validate_na(bond, slave, saddr, daddr);
3471 else if (curr_active_slave &&
3472 time_after(slave_last_rx(bond, curr_active_slave),
3473 curr_active_slave->last_link_up))
3474 bond_validate_na(bond, slave, daddr, saddr);
3475 else if (curr_arp_slave &&
3476 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3477 bond_validate_na(bond, slave, saddr, daddr);
3478
3479out:
3480 return RX_HANDLER_ANOTHER;
3481}
3482#endif
3483
3484int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3485 struct slave *slave)
3486{
3487#if IS_ENABLED(CONFIG_IPV6)
3488 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3489#endif
3490 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3491
3492 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3493 __func__, skb->dev->name);
3494
3495 /* Use arp validate logic for both ARP and NS */
3496 if (!slave_do_arp_validate(bond, slave)) {
3497 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3498#if IS_ENABLED(CONFIG_IPV6)
3499 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3500#endif
3501 !slave_do_arp_validate_only(bond))
3502 slave->last_rx = jiffies;
3503 return RX_HANDLER_ANOTHER;
3504 } else if (is_arp) {
3505 return bond_arp_rcv(skb, bond, slave);
3506#if IS_ENABLED(CONFIG_IPV6)
3507 } else if (is_ipv6) {
3508 return bond_na_rcv(skb, bond, slave);
3509#endif
3510 } else {
3511 return RX_HANDLER_ANOTHER;
3512 }
3513}
3514
3515static void bond_send_validate(struct bonding *bond, struct slave *slave)
3516{
3517 bond_arp_send_all(bond, slave);
3518#if IS_ENABLED(CONFIG_IPV6)
3519 bond_ns_send_all(bond, slave);
3520#endif
3521}
3522
3523/* function to verify if we're in the arp_interval timeslice, returns true if
3524 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3525 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3526 */
3527static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3528 int mod)
3529{
3530 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3531
3532 return time_in_range(jiffies,
3533 last_act - delta_in_ticks,
3534 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3535}
3536
3537/* This function is called regularly to monitor each slave's link
3538 * ensuring that traffic is being sent and received when arp monitoring
3539 * is used in load-balancing mode. if the adapter has been dormant, then an
3540 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3541 * arp monitoring in active backup mode.
3542 */
3543static void bond_loadbalance_arp_mon(struct bonding *bond)
3544{
3545 struct slave *slave, *oldcurrent;
3546 struct list_head *iter;
3547 int do_failover = 0, slave_state_changed = 0;
3548
3549 if (!bond_has_slaves(bond))
3550 goto re_arm;
3551
3552 rcu_read_lock();
3553
3554 oldcurrent = rcu_dereference(bond->curr_active_slave);
3555 /* see if any of the previous devices are up now (i.e. they have
3556 * xmt and rcv traffic). the curr_active_slave does not come into
3557 * the picture unless it is null. also, slave->last_link_up is not
3558 * needed here because we send an arp on each slave and give a slave
3559 * as long as it needs to get the tx/rx within the delta.
3560 * TODO: what about up/down delay in arp mode? it wasn't here before
3561 * so it can wait
3562 */
3563 bond_for_each_slave_rcu(bond, slave, iter) {
3564 unsigned long last_tx = slave_last_tx(slave);
3565
3566 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3567
3568 if (slave->link != BOND_LINK_UP) {
3569 if (bond_time_in_interval(bond, last_tx, 1) &&
3570 bond_time_in_interval(bond, slave->last_rx, 1)) {
3571
3572 bond_propose_link_state(slave, BOND_LINK_UP);
3573 slave_state_changed = 1;
3574
3575 /* primary_slave has no meaning in round-robin
3576 * mode. the window of a slave being up and
3577 * curr_active_slave being null after enslaving
3578 * is closed.
3579 */
3580 if (!oldcurrent) {
3581 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3582 do_failover = 1;
3583 } else {
3584 slave_info(bond->dev, slave->dev, "interface is now up\n");
3585 }
3586 }
3587 } else {
3588 /* slave->link == BOND_LINK_UP */
3589
3590 /* not all switches will respond to an arp request
3591 * when the source ip is 0, so don't take the link down
3592 * if we don't know our ip yet
3593 */
3594 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3595 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3596
3597 bond_propose_link_state(slave, BOND_LINK_DOWN);
3598 slave_state_changed = 1;
3599
3600 if (slave->link_failure_count < UINT_MAX)
3601 slave->link_failure_count++;
3602
3603 slave_info(bond->dev, slave->dev, "interface is now down\n");
3604
3605 if (slave == oldcurrent)
3606 do_failover = 1;
3607 }
3608 }
3609
3610 /* note: if switch is in round-robin mode, all links
3611 * must tx arp to ensure all links rx an arp - otherwise
3612 * links may oscillate or not come up at all; if switch is
3613 * in something like xor mode, there is nothing we can
3614 * do - all replies will be rx'ed on same link causing slaves
3615 * to be unstable during low/no traffic periods
3616 */
3617 if (bond_slave_is_up(slave))
3618 bond_send_validate(bond, slave);
3619 }
3620
3621 rcu_read_unlock();
3622
3623 if (do_failover || slave_state_changed) {
3624 if (!rtnl_trylock())
3625 goto re_arm;
3626
3627 bond_for_each_slave(bond, slave, iter) {
3628 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3629 slave->link = slave->link_new_state;
3630 }
3631
3632 if (slave_state_changed) {
3633 bond_slave_state_change(bond);
3634 if (BOND_MODE(bond) == BOND_MODE_XOR)
3635 bond_update_slave_arr(bond, NULL);
3636 }
3637 if (do_failover) {
3638 block_netpoll_tx();
3639 bond_select_active_slave(bond);
3640 unblock_netpoll_tx();
3641 }
3642 rtnl_unlock();
3643 }
3644
3645re_arm:
3646 if (bond->params.arp_interval)
3647 queue_delayed_work(bond->wq, &bond->arp_work,
3648 msecs_to_jiffies(bond->params.arp_interval));
3649}
3650
3651/* Called to inspect slaves for active-backup mode ARP monitor link state
3652 * changes. Sets proposed link state in slaves to specify what action
3653 * should take place for the slave. Returns 0 if no changes are found, >0
3654 * if changes to link states must be committed.
3655 *
3656 * Called with rcu_read_lock held.
3657 */
3658static int bond_ab_arp_inspect(struct bonding *bond)
3659{
3660 unsigned long last_tx, last_rx;
3661 struct list_head *iter;
3662 struct slave *slave;
3663 int commit = 0;
3664
3665 bond_for_each_slave_rcu(bond, slave, iter) {
3666 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3667 last_rx = slave_last_rx(bond, slave);
3668
3669 if (slave->link != BOND_LINK_UP) {
3670 if (bond_time_in_interval(bond, last_rx, 1)) {
3671 bond_propose_link_state(slave, BOND_LINK_UP);
3672 commit++;
3673 } else if (slave->link == BOND_LINK_BACK) {
3674 bond_propose_link_state(slave, BOND_LINK_FAIL);
3675 commit++;
3676 }
3677 continue;
3678 }
3679
3680 /* Give slaves 2*delta after being enslaved or made
3681 * active. This avoids bouncing, as the last receive
3682 * times need a full ARP monitor cycle to be updated.
3683 */
3684 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3685 continue;
3686
3687 /* Backup slave is down if:
3688 * - No current_arp_slave AND
3689 * - more than (missed_max+1)*delta since last receive AND
3690 * - the bond has an IP address
3691 *
3692 * Note: a non-null current_arp_slave indicates
3693 * the curr_active_slave went down and we are
3694 * searching for a new one; under this condition
3695 * we only take the curr_active_slave down - this
3696 * gives each slave a chance to tx/rx traffic
3697 * before being taken out
3698 */
3699 if (!bond_is_active_slave(slave) &&
3700 !rcu_access_pointer(bond->current_arp_slave) &&
3701 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3702 bond_propose_link_state(slave, BOND_LINK_DOWN);
3703 commit++;
3704 }
3705
3706 /* Active slave is down if:
3707 * - more than missed_max*delta since transmitting OR
3708 * - (more than missed_max*delta since receive AND
3709 * the bond has an IP address)
3710 */
3711 last_tx = slave_last_tx(slave);
3712 if (bond_is_active_slave(slave) &&
3713 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3714 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3715 bond_propose_link_state(slave, BOND_LINK_DOWN);
3716 commit++;
3717 }
3718 }
3719
3720 return commit;
3721}
3722
3723/* Called to commit link state changes noted by inspection step of
3724 * active-backup mode ARP monitor.
3725 *
3726 * Called with RTNL hold.
3727 */
3728static void bond_ab_arp_commit(struct bonding *bond)
3729{
3730 bool do_failover = false;
3731 struct list_head *iter;
3732 unsigned long last_tx;
3733 struct slave *slave;
3734
3735 bond_for_each_slave(bond, slave, iter) {
3736 switch (slave->link_new_state) {
3737 case BOND_LINK_NOCHANGE:
3738 continue;
3739
3740 case BOND_LINK_UP:
3741 last_tx = slave_last_tx(slave);
3742 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3743 (!rtnl_dereference(bond->curr_active_slave) &&
3744 bond_time_in_interval(bond, last_tx, 1))) {
3745 struct slave *current_arp_slave;
3746
3747 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3748 bond_set_slave_link_state(slave, BOND_LINK_UP,
3749 BOND_SLAVE_NOTIFY_NOW);
3750 if (current_arp_slave) {
3751 bond_set_slave_inactive_flags(
3752 current_arp_slave,
3753 BOND_SLAVE_NOTIFY_NOW);
3754 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3755 }
3756
3757 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3758
3759 if (!rtnl_dereference(bond->curr_active_slave) ||
3760 slave == rtnl_dereference(bond->primary_slave) ||
3761 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3762 do_failover = true;
3763
3764 }
3765
3766 continue;
3767
3768 case BOND_LINK_DOWN:
3769 if (slave->link_failure_count < UINT_MAX)
3770 slave->link_failure_count++;
3771
3772 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3773 BOND_SLAVE_NOTIFY_NOW);
3774 bond_set_slave_inactive_flags(slave,
3775 BOND_SLAVE_NOTIFY_NOW);
3776
3777 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3778
3779 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3780 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3781 do_failover = true;
3782 }
3783
3784 continue;
3785
3786 case BOND_LINK_FAIL:
3787 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3788 BOND_SLAVE_NOTIFY_NOW);
3789 bond_set_slave_inactive_flags(slave,
3790 BOND_SLAVE_NOTIFY_NOW);
3791
3792 /* A slave has just been enslaved and has become
3793 * the current active slave.
3794 */
3795 if (rtnl_dereference(bond->curr_active_slave))
3796 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3797 continue;
3798
3799 default:
3800 slave_err(bond->dev, slave->dev,
3801 "impossible: link_new_state %d on slave\n",
3802 slave->link_new_state);
3803 continue;
3804 }
3805 }
3806
3807 if (do_failover) {
3808 block_netpoll_tx();
3809 bond_select_active_slave(bond);
3810 unblock_netpoll_tx();
3811 }
3812
3813 bond_set_carrier(bond);
3814}
3815
3816/* Send ARP probes for active-backup mode ARP monitor.
3817 *
3818 * Called with rcu_read_lock held.
3819 */
3820static bool bond_ab_arp_probe(struct bonding *bond)
3821{
3822 struct slave *slave, *before = NULL, *new_slave = NULL,
3823 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3824 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3825 struct list_head *iter;
3826 bool found = false;
3827 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3828
3829 if (curr_arp_slave && curr_active_slave)
3830 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3831 curr_arp_slave->dev->name,
3832 curr_active_slave->dev->name);
3833
3834 if (curr_active_slave) {
3835 bond_send_validate(bond, curr_active_slave);
3836 return should_notify_rtnl;
3837 }
3838
3839 /* if we don't have a curr_active_slave, search for the next available
3840 * backup slave from the current_arp_slave and make it the candidate
3841 * for becoming the curr_active_slave
3842 */
3843
3844 if (!curr_arp_slave) {
3845 curr_arp_slave = bond_first_slave_rcu(bond);
3846 if (!curr_arp_slave)
3847 return should_notify_rtnl;
3848 }
3849
3850 bond_for_each_slave_rcu(bond, slave, iter) {
3851 if (!found && !before && bond_slave_is_up(slave))
3852 before = slave;
3853
3854 if (found && !new_slave && bond_slave_is_up(slave))
3855 new_slave = slave;
3856 /* if the link state is up at this point, we
3857 * mark it down - this can happen if we have
3858 * simultaneous link failures and
3859 * reselect_active_interface doesn't make this
3860 * one the current slave so it is still marked
3861 * up when it is actually down
3862 */
3863 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3864 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3865 BOND_SLAVE_NOTIFY_LATER);
3866 if (slave->link_failure_count < UINT_MAX)
3867 slave->link_failure_count++;
3868
3869 bond_set_slave_inactive_flags(slave,
3870 BOND_SLAVE_NOTIFY_LATER);
3871
3872 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3873 }
3874 if (slave == curr_arp_slave)
3875 found = true;
3876 }
3877
3878 if (!new_slave && before)
3879 new_slave = before;
3880
3881 if (!new_slave)
3882 goto check_state;
3883
3884 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3885 BOND_SLAVE_NOTIFY_LATER);
3886 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3887 bond_send_validate(bond, new_slave);
3888 new_slave->last_link_up = jiffies;
3889 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3890
3891check_state:
3892 bond_for_each_slave_rcu(bond, slave, iter) {
3893 if (slave->should_notify || slave->should_notify_link) {
3894 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3895 break;
3896 }
3897 }
3898 return should_notify_rtnl;
3899}
3900
3901static void bond_activebackup_arp_mon(struct bonding *bond)
3902{
3903 bool should_notify_peers = false;
3904 bool should_notify_rtnl = false;
3905 int delta_in_ticks;
3906
3907 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3908
3909 if (!bond_has_slaves(bond))
3910 goto re_arm;
3911
3912 rcu_read_lock();
3913
3914 should_notify_peers = bond_should_notify_peers(bond);
3915
3916 if (bond_ab_arp_inspect(bond)) {
3917 rcu_read_unlock();
3918
3919 /* Race avoidance with bond_close flush of workqueue */
3920 if (!rtnl_trylock()) {
3921 delta_in_ticks = 1;
3922 should_notify_peers = false;
3923 goto re_arm;
3924 }
3925
3926 bond_ab_arp_commit(bond);
3927
3928 rtnl_unlock();
3929 rcu_read_lock();
3930 }
3931
3932 should_notify_rtnl = bond_ab_arp_probe(bond);
3933 rcu_read_unlock();
3934
3935re_arm:
3936 if (bond->params.arp_interval)
3937 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3938
3939 if (should_notify_peers || should_notify_rtnl) {
3940 if (!rtnl_trylock())
3941 return;
3942
3943 if (should_notify_peers) {
3944 bond->send_peer_notif--;
3945 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3946 bond->dev);
3947 }
3948 if (should_notify_rtnl) {
3949 bond_slave_state_notify(bond);
3950 bond_slave_link_notify(bond);
3951 }
3952
3953 rtnl_unlock();
3954 }
3955}
3956
3957static void bond_arp_monitor(struct work_struct *work)
3958{
3959 struct bonding *bond = container_of(work, struct bonding,
3960 arp_work.work);
3961
3962 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3963 bond_activebackup_arp_mon(bond);
3964 else
3965 bond_loadbalance_arp_mon(bond);
3966}
3967
3968/*-------------------------- netdev event handling --------------------------*/
3969
3970/* Change device name */
3971static int bond_event_changename(struct bonding *bond)
3972{
3973 bond_remove_proc_entry(bond);
3974 bond_create_proc_entry(bond);
3975
3976 bond_debug_reregister(bond);
3977
3978 return NOTIFY_DONE;
3979}
3980
3981static int bond_master_netdev_event(unsigned long event,
3982 struct net_device *bond_dev)
3983{
3984 struct bonding *event_bond = netdev_priv(bond_dev);
3985
3986 netdev_dbg(bond_dev, "%s called\n", __func__);
3987
3988 switch (event) {
3989 case NETDEV_CHANGENAME:
3990 return bond_event_changename(event_bond);
3991 case NETDEV_UNREGISTER:
3992 bond_remove_proc_entry(event_bond);
3993#ifdef CONFIG_XFRM_OFFLOAD
3994 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3995#endif /* CONFIG_XFRM_OFFLOAD */
3996 break;
3997 case NETDEV_REGISTER:
3998 bond_create_proc_entry(event_bond);
3999 break;
4000 default:
4001 break;
4002 }
4003
4004 return NOTIFY_DONE;
4005}
4006
4007static int bond_slave_netdev_event(unsigned long event,
4008 struct net_device *slave_dev)
4009{
4010 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
4011 struct bonding *bond;
4012 struct net_device *bond_dev;
4013
4014 /* A netdev event can be generated while enslaving a device
4015 * before netdev_rx_handler_register is called in which case
4016 * slave will be NULL
4017 */
4018 if (!slave) {
4019 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
4020 return NOTIFY_DONE;
4021 }
4022
4023 bond_dev = slave->bond->dev;
4024 bond = slave->bond;
4025 primary = rtnl_dereference(bond->primary_slave);
4026
4027 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
4028
4029 switch (event) {
4030 case NETDEV_UNREGISTER:
4031 if (bond_dev->type != ARPHRD_ETHER)
4032 bond_release_and_destroy(bond_dev, slave_dev);
4033 else
4034 __bond_release_one(bond_dev, slave_dev, false, true);
4035 break;
4036 case NETDEV_UP:
4037 case NETDEV_CHANGE:
4038 /* For 802.3ad mode only:
4039 * Getting invalid Speed/Duplex values here will put slave
4040 * in weird state. Mark it as link-fail if the link was
4041 * previously up or link-down if it hasn't yet come up, and
4042 * let link-monitoring (miimon) set it right when correct
4043 * speeds/duplex are available.
4044 */
4045 if (bond_update_speed_duplex(slave) &&
4046 BOND_MODE(bond) == BOND_MODE_8023AD) {
4047 if (slave->last_link_up)
4048 slave->link = BOND_LINK_FAIL;
4049 else
4050 slave->link = BOND_LINK_DOWN;
4051 }
4052
4053 if (BOND_MODE(bond) == BOND_MODE_8023AD)
4054 bond_3ad_adapter_speed_duplex_changed(slave);
4055 fallthrough;
4056 case NETDEV_DOWN:
4057 /* Refresh slave-array if applicable!
4058 * If the setup does not use miimon or arpmon (mode-specific!),
4059 * then these events will not cause the slave-array to be
4060 * refreshed. This will cause xmit to use a slave that is not
4061 * usable. Avoid such situation by refeshing the array at these
4062 * events. If these (miimon/arpmon) parameters are configured
4063 * then array gets refreshed twice and that should be fine!
4064 */
4065 if (bond_mode_can_use_xmit_hash(bond))
4066 bond_update_slave_arr(bond, NULL);
4067 break;
4068 case NETDEV_CHANGEMTU:
4069 /* TODO: Should slaves be allowed to
4070 * independently alter their MTU? For
4071 * an active-backup bond, slaves need
4072 * not be the same type of device, so
4073 * MTUs may vary. For other modes,
4074 * slaves arguably should have the
4075 * same MTUs. To do this, we'd need to
4076 * take over the slave's change_mtu
4077 * function for the duration of their
4078 * servitude.
4079 */
4080 break;
4081 case NETDEV_CHANGENAME:
4082 /* we don't care if we don't have primary set */
4083 if (!bond_uses_primary(bond) ||
4084 !bond->params.primary[0])
4085 break;
4086
4087 if (slave == primary) {
4088 /* slave's name changed - he's no longer primary */
4089 RCU_INIT_POINTER(bond->primary_slave, NULL);
4090 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
4091 /* we have a new primary slave */
4092 rcu_assign_pointer(bond->primary_slave, slave);
4093 } else { /* we didn't change primary - exit */
4094 break;
4095 }
4096
4097 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
4098 primary ? slave_dev->name : "none");
4099
4100 block_netpoll_tx();
4101 bond_select_active_slave(bond);
4102 unblock_netpoll_tx();
4103 break;
4104 case NETDEV_FEAT_CHANGE:
4105 if (!bond->notifier_ctx) {
4106 bond->notifier_ctx = true;
4107 bond_compute_features(bond);
4108 bond->notifier_ctx = false;
4109 }
4110 break;
4111 case NETDEV_RESEND_IGMP:
4112 /* Propagate to master device */
4113 call_netdevice_notifiers(event, slave->bond->dev);
4114 break;
4115 case NETDEV_XDP_FEAT_CHANGE:
4116 bond_xdp_set_features(bond_dev);
4117 break;
4118 default:
4119 break;
4120 }
4121
4122 return NOTIFY_DONE;
4123}
4124
4125/* bond_netdev_event: handle netdev notifier chain events.
4126 *
4127 * This function receives events for the netdev chain. The caller (an
4128 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
4129 * locks for us to safely manipulate the slave devices (RTNL lock,
4130 * dev_probe_lock).
4131 */
4132static int bond_netdev_event(struct notifier_block *this,
4133 unsigned long event, void *ptr)
4134{
4135 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
4136
4137 netdev_dbg(event_dev, "%s received %s\n",
4138 __func__, netdev_cmd_to_name(event));
4139
4140 if (!(event_dev->priv_flags & IFF_BONDING))
4141 return NOTIFY_DONE;
4142
4143 if (event_dev->flags & IFF_MASTER) {
4144 int ret;
4145
4146 ret = bond_master_netdev_event(event, event_dev);
4147 if (ret != NOTIFY_DONE)
4148 return ret;
4149 }
4150
4151 if (event_dev->flags & IFF_SLAVE)
4152 return bond_slave_netdev_event(event, event_dev);
4153
4154 return NOTIFY_DONE;
4155}
4156
4157static struct notifier_block bond_netdev_notifier = {
4158 .notifier_call = bond_netdev_event,
4159};
4160
4161/*---------------------------- Hashing Policies -----------------------------*/
4162
4163/* Helper to access data in a packet, with or without a backing skb.
4164 * If skb is given the data is linearized if necessary via pskb_may_pull.
4165 */
4166static inline const void *bond_pull_data(struct sk_buff *skb,
4167 const void *data, int hlen, int n)
4168{
4169 if (likely(n <= hlen))
4170 return data;
4171 else if (skb && likely(pskb_may_pull(skb, n)))
4172 return skb->data;
4173
4174 return NULL;
4175}
4176
4177/* L2 hash helper */
4178static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4179{
4180 struct ethhdr *ep;
4181
4182 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4183 if (!data)
4184 return 0;
4185
4186 ep = (struct ethhdr *)(data + mhoff);
4187 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4188}
4189
4190static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4191 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4192{
4193 const struct ipv6hdr *iph6;
4194 const struct iphdr *iph;
4195
4196 if (l2_proto == htons(ETH_P_IP)) {
4197 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4198 if (!data)
4199 return false;
4200
4201 iph = (const struct iphdr *)(data + *nhoff);
4202 iph_to_flow_copy_v4addrs(fk, iph);
4203 *nhoff += iph->ihl << 2;
4204 if (!ip_is_fragment(iph))
4205 *ip_proto = iph->protocol;
4206 } else if (l2_proto == htons(ETH_P_IPV6)) {
4207 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4208 if (!data)
4209 return false;
4210
4211 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4212 iph_to_flow_copy_v6addrs(fk, iph6);
4213 *nhoff += sizeof(*iph6);
4214 *ip_proto = iph6->nexthdr;
4215 } else {
4216 return false;
4217 }
4218
4219 if (l34 && *ip_proto >= 0)
4220 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4221
4222 return true;
4223}
4224
4225static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4226{
4227 u32 srcmac_vendor = 0, srcmac_dev = 0;
4228 struct ethhdr *mac_hdr;
4229 u16 vlan = 0;
4230 int i;
4231
4232 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4233 if (!data)
4234 return 0;
4235 mac_hdr = (struct ethhdr *)(data + mhoff);
4236
4237 for (i = 0; i < 3; i++)
4238 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4239
4240 for (i = 3; i < ETH_ALEN; i++)
4241 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4242
4243 if (skb && skb_vlan_tag_present(skb))
4244 vlan = skb_vlan_tag_get(skb);
4245
4246 return vlan ^ srcmac_vendor ^ srcmac_dev;
4247}
4248
4249/* Extract the appropriate headers based on bond's xmit policy */
4250static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4251 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4252{
4253 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4254 int ip_proto = -1;
4255
4256 switch (bond->params.xmit_policy) {
4257 case BOND_XMIT_POLICY_ENCAP23:
4258 case BOND_XMIT_POLICY_ENCAP34:
4259 memset(fk, 0, sizeof(*fk));
4260 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4261 fk, data, l2_proto, nhoff, hlen, 0);
4262 default:
4263 break;
4264 }
4265
4266 fk->ports.ports = 0;
4267 memset(&fk->icmp, 0, sizeof(fk->icmp));
4268 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4269 return false;
4270
4271 /* ICMP error packets contains at least 8 bytes of the header
4272 * of the packet which generated the error. Use this information
4273 * to correlate ICMP error packets within the same flow which
4274 * generated the error.
4275 */
4276 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4277 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4278 if (ip_proto == IPPROTO_ICMP) {
4279 if (!icmp_is_err(fk->icmp.type))
4280 return true;
4281
4282 nhoff += sizeof(struct icmphdr);
4283 } else if (ip_proto == IPPROTO_ICMPV6) {
4284 if (!icmpv6_is_err(fk->icmp.type))
4285 return true;
4286
4287 nhoff += sizeof(struct icmp6hdr);
4288 }
4289 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4290 }
4291
4292 return true;
4293}
4294
4295static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4296{
4297 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4298 (__force u32)flow_get_u32_src(flow);
4299 hash ^= (hash >> 16);
4300 hash ^= (hash >> 8);
4301
4302 /* discard lowest hash bit to deal with the common even ports pattern */
4303 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4304 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4305 return hash >> 1;
4306
4307 return hash;
4308}
4309
4310/* Generate hash based on xmit policy. If @skb is given it is used to linearize
4311 * the data as required, but this function can be used without it if the data is
4312 * known to be linear (e.g. with xdp_buff).
4313 */
4314static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4315 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4316{
4317 struct flow_keys flow;
4318 u32 hash;
4319
4320 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4321 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4322
4323 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4324 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4325 return bond_eth_hash(skb, data, mhoff, hlen);
4326
4327 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4328 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4329 hash = bond_eth_hash(skb, data, mhoff, hlen);
4330 } else {
4331 if (flow.icmp.id)
4332 memcpy(&hash, &flow.icmp, sizeof(hash));
4333 else
4334 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4335 }
4336
4337 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4338}
4339
4340/**
4341 * bond_xmit_hash - generate a hash value based on the xmit policy
4342 * @bond: bonding device
4343 * @skb: buffer to use for headers
4344 *
4345 * This function will extract the necessary headers from the skb buffer and use
4346 * them to generate a hash based on the xmit_policy set in the bonding device
4347 */
4348u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4349{
4350 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4351 skb->l4_hash)
4352 return skb->hash;
4353
4354 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4355 0, skb_network_offset(skb),
4356 skb_headlen(skb));
4357}
4358
4359/**
4360 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4361 * @bond: bonding device
4362 * @xdp: buffer to use for headers
4363 *
4364 * The XDP variant of bond_xmit_hash.
4365 */
4366static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4367{
4368 struct ethhdr *eth;
4369
4370 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4371 return 0;
4372
4373 eth = (struct ethhdr *)xdp->data;
4374
4375 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4376 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4377}
4378
4379/*-------------------------- Device entry points ----------------------------*/
4380
4381void bond_work_init_all(struct bonding *bond)
4382{
4383 INIT_DELAYED_WORK(&bond->mcast_work,
4384 bond_resend_igmp_join_requests_delayed);
4385 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4386 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4387 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4388 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4389 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4390}
4391
4392static void bond_work_cancel_all(struct bonding *bond)
4393{
4394 cancel_delayed_work_sync(&bond->mii_work);
4395 cancel_delayed_work_sync(&bond->arp_work);
4396 cancel_delayed_work_sync(&bond->alb_work);
4397 cancel_delayed_work_sync(&bond->ad_work);
4398 cancel_delayed_work_sync(&bond->mcast_work);
4399 cancel_delayed_work_sync(&bond->slave_arr_work);
4400}
4401
4402static int bond_open(struct net_device *bond_dev)
4403{
4404 struct bonding *bond = netdev_priv(bond_dev);
4405 struct list_head *iter;
4406 struct slave *slave;
4407
4408 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4409 bond->rr_tx_counter = alloc_percpu(u32);
4410 if (!bond->rr_tx_counter)
4411 return -ENOMEM;
4412 }
4413
4414 /* reset slave->backup and slave->inactive */
4415 if (bond_has_slaves(bond)) {
4416 bond_for_each_slave(bond, slave, iter) {
4417 if (bond_uses_primary(bond) &&
4418 slave != rcu_access_pointer(bond->curr_active_slave)) {
4419 bond_set_slave_inactive_flags(slave,
4420 BOND_SLAVE_NOTIFY_NOW);
4421 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4422 bond_set_slave_active_flags(slave,
4423 BOND_SLAVE_NOTIFY_NOW);
4424 }
4425 }
4426 }
4427
4428 if (bond_is_lb(bond)) {
4429 /* bond_alb_initialize must be called before the timer
4430 * is started.
4431 */
4432 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4433 return -ENOMEM;
4434 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4435 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4436 }
4437
4438 if (bond->params.miimon) /* link check interval, in milliseconds. */
4439 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4440
4441 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4442 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4443 bond->recv_probe = bond_rcv_validate;
4444 }
4445
4446 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4447 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4448 /* register to receive LACPDUs */
4449 bond->recv_probe = bond_3ad_lacpdu_recv;
4450 bond_3ad_initiate_agg_selection(bond, 1);
4451
4452 bond_for_each_slave(bond, slave, iter)
4453 dev_mc_add(slave->dev, lacpdu_mcast_addr);
4454 }
4455
4456 if (bond_mode_can_use_xmit_hash(bond))
4457 bond_update_slave_arr(bond, NULL);
4458
4459 return 0;
4460}
4461
4462static int bond_close(struct net_device *bond_dev)
4463{
4464 struct bonding *bond = netdev_priv(bond_dev);
4465 struct slave *slave;
4466
4467 bond_work_cancel_all(bond);
4468 bond->send_peer_notif = 0;
4469 if (bond_is_lb(bond))
4470 bond_alb_deinitialize(bond);
4471 bond->recv_probe = NULL;
4472
4473 if (bond_uses_primary(bond)) {
4474 rcu_read_lock();
4475 slave = rcu_dereference(bond->curr_active_slave);
4476 if (slave)
4477 bond_hw_addr_flush(bond_dev, slave->dev);
4478 rcu_read_unlock();
4479 } else {
4480 struct list_head *iter;
4481
4482 bond_for_each_slave(bond, slave, iter)
4483 bond_hw_addr_flush(bond_dev, slave->dev);
4484 }
4485
4486 return 0;
4487}
4488
4489/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4490 * that some drivers can provide 32bit values only.
4491 */
4492static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4493 const struct rtnl_link_stats64 *_new,
4494 const struct rtnl_link_stats64 *_old)
4495{
4496 const u64 *new = (const u64 *)_new;
4497 const u64 *old = (const u64 *)_old;
4498 u64 *res = (u64 *)_res;
4499 int i;
4500
4501 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4502 u64 nv = new[i];
4503 u64 ov = old[i];
4504 s64 delta = nv - ov;
4505
4506 /* detects if this particular field is 32bit only */
4507 if (((nv | ov) >> 32) == 0)
4508 delta = (s64)(s32)((u32)nv - (u32)ov);
4509
4510 /* filter anomalies, some drivers reset their stats
4511 * at down/up events.
4512 */
4513 if (delta > 0)
4514 res[i] += delta;
4515 }
4516}
4517
4518#ifdef CONFIG_LOCKDEP
4519static int bond_get_lowest_level_rcu(struct net_device *dev)
4520{
4521 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4522 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4523 int cur = 0, max = 0;
4524
4525 now = dev;
4526 iter = &dev->adj_list.lower;
4527
4528 while (1) {
4529 next = NULL;
4530 while (1) {
4531 ldev = netdev_next_lower_dev_rcu(now, &iter);
4532 if (!ldev)
4533 break;
4534
4535 next = ldev;
4536 niter = &ldev->adj_list.lower;
4537 dev_stack[cur] = now;
4538 iter_stack[cur++] = iter;
4539 if (max <= cur)
4540 max = cur;
4541 break;
4542 }
4543
4544 if (!next) {
4545 if (!cur)
4546 return max;
4547 next = dev_stack[--cur];
4548 niter = iter_stack[cur];
4549 }
4550
4551 now = next;
4552 iter = niter;
4553 }
4554
4555 return max;
4556}
4557#endif
4558
4559static void bond_get_stats(struct net_device *bond_dev,
4560 struct rtnl_link_stats64 *stats)
4561{
4562 struct bonding *bond = netdev_priv(bond_dev);
4563 struct rtnl_link_stats64 temp;
4564 struct list_head *iter;
4565 struct slave *slave;
4566 int nest_level = 0;
4567
4568
4569 rcu_read_lock();
4570#ifdef CONFIG_LOCKDEP
4571 nest_level = bond_get_lowest_level_rcu(bond_dev);
4572#endif
4573
4574 spin_lock_nested(&bond->stats_lock, nest_level);
4575 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4576
4577 bond_for_each_slave_rcu(bond, slave, iter) {
4578 const struct rtnl_link_stats64 *new =
4579 dev_get_stats(slave->dev, &temp);
4580
4581 bond_fold_stats(stats, new, &slave->slave_stats);
4582
4583 /* save off the slave stats for the next run */
4584 memcpy(&slave->slave_stats, new, sizeof(*new));
4585 }
4586
4587 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4588 spin_unlock(&bond->stats_lock);
4589 rcu_read_unlock();
4590}
4591
4592static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4593{
4594 struct bonding *bond = netdev_priv(bond_dev);
4595 struct mii_ioctl_data *mii = NULL;
4596
4597 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4598
4599 switch (cmd) {
4600 case SIOCGMIIPHY:
4601 mii = if_mii(ifr);
4602 if (!mii)
4603 return -EINVAL;
4604
4605 mii->phy_id = 0;
4606 fallthrough;
4607 case SIOCGMIIREG:
4608 /* We do this again just in case we were called by SIOCGMIIREG
4609 * instead of SIOCGMIIPHY.
4610 */
4611 mii = if_mii(ifr);
4612 if (!mii)
4613 return -EINVAL;
4614
4615 if (mii->reg_num == 1) {
4616 mii->val_out = 0;
4617 if (netif_carrier_ok(bond->dev))
4618 mii->val_out = BMSR_LSTATUS;
4619 }
4620
4621 break;
4622 default:
4623 return -EOPNOTSUPP;
4624 }
4625
4626 return 0;
4627}
4628
4629static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4630{
4631 struct bonding *bond = netdev_priv(bond_dev);
4632 struct net_device *slave_dev = NULL;
4633 struct ifbond k_binfo;
4634 struct ifbond __user *u_binfo = NULL;
4635 struct ifslave k_sinfo;
4636 struct ifslave __user *u_sinfo = NULL;
4637 struct bond_opt_value newval;
4638 struct net *net;
4639 int res = 0;
4640
4641 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4642
4643 switch (cmd) {
4644 case SIOCBONDINFOQUERY:
4645 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4646
4647 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4648 return -EFAULT;
4649
4650 bond_info_query(bond_dev, &k_binfo);
4651 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4652 return -EFAULT;
4653
4654 return 0;
4655 case SIOCBONDSLAVEINFOQUERY:
4656 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4657
4658 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4659 return -EFAULT;
4660
4661 res = bond_slave_info_query(bond_dev, &k_sinfo);
4662 if (res == 0 &&
4663 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4664 return -EFAULT;
4665
4666 return res;
4667 default:
4668 break;
4669 }
4670
4671 net = dev_net(bond_dev);
4672
4673 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4674 return -EPERM;
4675
4676 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4677
4678 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4679
4680 if (!slave_dev)
4681 return -ENODEV;
4682
4683 switch (cmd) {
4684 case SIOCBONDENSLAVE:
4685 res = bond_enslave(bond_dev, slave_dev, NULL);
4686 break;
4687 case SIOCBONDRELEASE:
4688 res = bond_release(bond_dev, slave_dev);
4689 break;
4690 case SIOCBONDSETHWADDR:
4691 res = bond_set_dev_addr(bond_dev, slave_dev);
4692 break;
4693 case SIOCBONDCHANGEACTIVE:
4694 bond_opt_initstr(&newval, slave_dev->name);
4695 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4696 &newval);
4697 break;
4698 default:
4699 res = -EOPNOTSUPP;
4700 }
4701
4702 return res;
4703}
4704
4705static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4706 void __user *data, int cmd)
4707{
4708 struct ifreq ifrdata = { .ifr_data = data };
4709
4710 switch (cmd) {
4711 case BOND_INFO_QUERY_OLD:
4712 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4713 case BOND_SLAVE_INFO_QUERY_OLD:
4714 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4715 case BOND_ENSLAVE_OLD:
4716 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4717 case BOND_RELEASE_OLD:
4718 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4719 case BOND_SETHWADDR_OLD:
4720 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4721 case BOND_CHANGE_ACTIVE_OLD:
4722 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4723 }
4724
4725 return -EOPNOTSUPP;
4726}
4727
4728static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4729{
4730 struct bonding *bond = netdev_priv(bond_dev);
4731
4732 if (change & IFF_PROMISC)
4733 bond_set_promiscuity(bond,
4734 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4735
4736 if (change & IFF_ALLMULTI)
4737 bond_set_allmulti(bond,
4738 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4739}
4740
4741static void bond_set_rx_mode(struct net_device *bond_dev)
4742{
4743 struct bonding *bond = netdev_priv(bond_dev);
4744 struct list_head *iter;
4745 struct slave *slave;
4746
4747 rcu_read_lock();
4748 if (bond_uses_primary(bond)) {
4749 slave = rcu_dereference(bond->curr_active_slave);
4750 if (slave) {
4751 dev_uc_sync(slave->dev, bond_dev);
4752 dev_mc_sync(slave->dev, bond_dev);
4753 }
4754 } else {
4755 bond_for_each_slave_rcu(bond, slave, iter) {
4756 dev_uc_sync_multiple(slave->dev, bond_dev);
4757 dev_mc_sync_multiple(slave->dev, bond_dev);
4758 }
4759 }
4760 rcu_read_unlock();
4761}
4762
4763static int bond_neigh_init(struct neighbour *n)
4764{
4765 struct bonding *bond = netdev_priv(n->dev);
4766 const struct net_device_ops *slave_ops;
4767 struct neigh_parms parms;
4768 struct slave *slave;
4769 int ret = 0;
4770
4771 rcu_read_lock();
4772 slave = bond_first_slave_rcu(bond);
4773 if (!slave)
4774 goto out;
4775 slave_ops = slave->dev->netdev_ops;
4776 if (!slave_ops->ndo_neigh_setup)
4777 goto out;
4778
4779 /* TODO: find another way [1] to implement this.
4780 * Passing a zeroed structure is fragile,
4781 * but at least we do not pass garbage.
4782 *
4783 * [1] One way would be that ndo_neigh_setup() never touch
4784 * struct neigh_parms, but propagate the new neigh_setup()
4785 * back to ___neigh_create() / neigh_parms_alloc()
4786 */
4787 memset(&parms, 0, sizeof(parms));
4788 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4789
4790 if (ret)
4791 goto out;
4792
4793 if (parms.neigh_setup)
4794 ret = parms.neigh_setup(n);
4795out:
4796 rcu_read_unlock();
4797 return ret;
4798}
4799
4800/* The bonding ndo_neigh_setup is called at init time beofre any
4801 * slave exists. So we must declare proxy setup function which will
4802 * be used at run time to resolve the actual slave neigh param setup.
4803 *
4804 * It's also called by master devices (such as vlans) to setup their
4805 * underlying devices. In that case - do nothing, we're already set up from
4806 * our init.
4807 */
4808static int bond_neigh_setup(struct net_device *dev,
4809 struct neigh_parms *parms)
4810{
4811 /* modify only our neigh_parms */
4812 if (parms->dev == dev)
4813 parms->neigh_setup = bond_neigh_init;
4814
4815 return 0;
4816}
4817
4818/* Change the MTU of all of a master's slaves to match the master */
4819static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4820{
4821 struct bonding *bond = netdev_priv(bond_dev);
4822 struct slave *slave, *rollback_slave;
4823 struct list_head *iter;
4824 int res = 0;
4825
4826 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4827
4828 bond_for_each_slave(bond, slave, iter) {
4829 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4830 slave, slave->dev->netdev_ops->ndo_change_mtu);
4831
4832 res = dev_set_mtu(slave->dev, new_mtu);
4833
4834 if (res) {
4835 /* If we failed to set the slave's mtu to the new value
4836 * we must abort the operation even in ACTIVE_BACKUP
4837 * mode, because if we allow the backup slaves to have
4838 * different mtu values than the active slave we'll
4839 * need to change their mtu when doing a failover. That
4840 * means changing their mtu from timer context, which
4841 * is probably not a good idea.
4842 */
4843 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4844 res, new_mtu);
4845 goto unwind;
4846 }
4847 }
4848
4849 WRITE_ONCE(bond_dev->mtu, new_mtu);
4850
4851 return 0;
4852
4853unwind:
4854 /* unwind from head to the slave that failed */
4855 bond_for_each_slave(bond, rollback_slave, iter) {
4856 int tmp_res;
4857
4858 if (rollback_slave == slave)
4859 break;
4860
4861 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4862 if (tmp_res)
4863 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4864 tmp_res);
4865 }
4866
4867 return res;
4868}
4869
4870/* Change HW address
4871 *
4872 * Note that many devices must be down to change the HW address, and
4873 * downing the master releases all slaves. We can make bonds full of
4874 * bonding devices to test this, however.
4875 */
4876static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4877{
4878 struct bonding *bond = netdev_priv(bond_dev);
4879 struct slave *slave, *rollback_slave;
4880 struct sockaddr_storage *ss = addr, tmp_ss;
4881 struct list_head *iter;
4882 int res = 0;
4883
4884 if (BOND_MODE(bond) == BOND_MODE_ALB)
4885 return bond_alb_set_mac_address(bond_dev, addr);
4886
4887
4888 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4889
4890 /* If fail_over_mac is enabled, do nothing and return success.
4891 * Returning an error causes ifenslave to fail.
4892 */
4893 if (bond->params.fail_over_mac &&
4894 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4895 return 0;
4896
4897 if (!is_valid_ether_addr(ss->__data))
4898 return -EADDRNOTAVAIL;
4899
4900 bond_for_each_slave(bond, slave, iter) {
4901 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4902 __func__, slave);
4903 res = dev_set_mac_address(slave->dev, addr, NULL);
4904 if (res) {
4905 /* TODO: consider downing the slave
4906 * and retry ?
4907 * User should expect communications
4908 * breakage anyway until ARP finish
4909 * updating, so...
4910 */
4911 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4912 __func__, res);
4913 goto unwind;
4914 }
4915 }
4916
4917 /* success */
4918 dev_addr_set(bond_dev, ss->__data);
4919 return 0;
4920
4921unwind:
4922 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4923 tmp_ss.ss_family = bond_dev->type;
4924
4925 /* unwind from head to the slave that failed */
4926 bond_for_each_slave(bond, rollback_slave, iter) {
4927 int tmp_res;
4928
4929 if (rollback_slave == slave)
4930 break;
4931
4932 tmp_res = dev_set_mac_address(rollback_slave->dev,
4933 (struct sockaddr *)&tmp_ss, NULL);
4934 if (tmp_res) {
4935 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4936 __func__, tmp_res);
4937 }
4938 }
4939
4940 return res;
4941}
4942
4943/**
4944 * bond_get_slave_by_id - get xmit slave with slave_id
4945 * @bond: bonding device that is transmitting
4946 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4947 *
4948 * This function tries to get slave with slave_id but in case
4949 * it fails, it tries to find the first available slave for transmission.
4950 */
4951static struct slave *bond_get_slave_by_id(struct bonding *bond,
4952 int slave_id)
4953{
4954 struct list_head *iter;
4955 struct slave *slave;
4956 int i = slave_id;
4957
4958 /* Here we start from the slave with slave_id */
4959 bond_for_each_slave_rcu(bond, slave, iter) {
4960 if (--i < 0) {
4961 if (bond_slave_can_tx(slave))
4962 return slave;
4963 }
4964 }
4965
4966 /* Here we start from the first slave up to slave_id */
4967 i = slave_id;
4968 bond_for_each_slave_rcu(bond, slave, iter) {
4969 if (--i < 0)
4970 break;
4971 if (bond_slave_can_tx(slave))
4972 return slave;
4973 }
4974 /* no slave that can tx has been found */
4975 return NULL;
4976}
4977
4978/**
4979 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4980 * @bond: bonding device to use
4981 *
4982 * Based on the value of the bonding device's packets_per_slave parameter
4983 * this function generates a slave id, which is usually used as the next
4984 * slave to transmit through.
4985 */
4986static u32 bond_rr_gen_slave_id(struct bonding *bond)
4987{
4988 u32 slave_id;
4989 struct reciprocal_value reciprocal_packets_per_slave;
4990 int packets_per_slave = bond->params.packets_per_slave;
4991
4992 switch (packets_per_slave) {
4993 case 0:
4994 slave_id = get_random_u32();
4995 break;
4996 case 1:
4997 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4998 break;
4999 default:
5000 reciprocal_packets_per_slave =
5001 bond->params.reciprocal_packets_per_slave;
5002 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
5003 slave_id = reciprocal_divide(slave_id,
5004 reciprocal_packets_per_slave);
5005 break;
5006 }
5007
5008 return slave_id;
5009}
5010
5011static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
5012 struct sk_buff *skb)
5013{
5014 struct slave *slave;
5015 int slave_cnt;
5016 u32 slave_id;
5017
5018 /* Start with the curr_active_slave that joined the bond as the
5019 * default for sending IGMP traffic. For failover purposes one
5020 * needs to maintain some consistency for the interface that will
5021 * send the join/membership reports. The curr_active_slave found
5022 * will send all of this type of traffic.
5023 */
5024 if (skb->protocol == htons(ETH_P_IP)) {
5025 int noff = skb_network_offset(skb);
5026 struct iphdr *iph;
5027
5028 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
5029 goto non_igmp;
5030
5031 iph = ip_hdr(skb);
5032 if (iph->protocol == IPPROTO_IGMP) {
5033 slave = rcu_dereference(bond->curr_active_slave);
5034 if (slave)
5035 return slave;
5036 return bond_get_slave_by_id(bond, 0);
5037 }
5038 }
5039
5040non_igmp:
5041 slave_cnt = READ_ONCE(bond->slave_cnt);
5042 if (likely(slave_cnt)) {
5043 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
5044 return bond_get_slave_by_id(bond, slave_id);
5045 }
5046 return NULL;
5047}
5048
5049static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
5050 struct xdp_buff *xdp)
5051{
5052 struct slave *slave;
5053 int slave_cnt;
5054 u32 slave_id;
5055 const struct ethhdr *eth;
5056 void *data = xdp->data;
5057
5058 if (data + sizeof(struct ethhdr) > xdp->data_end)
5059 goto non_igmp;
5060
5061 eth = (struct ethhdr *)data;
5062 data += sizeof(struct ethhdr);
5063
5064 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
5065 if (eth->h_proto == htons(ETH_P_IP)) {
5066 const struct iphdr *iph;
5067
5068 if (data + sizeof(struct iphdr) > xdp->data_end)
5069 goto non_igmp;
5070
5071 iph = (struct iphdr *)data;
5072
5073 if (iph->protocol == IPPROTO_IGMP) {
5074 slave = rcu_dereference(bond->curr_active_slave);
5075 if (slave)
5076 return slave;
5077 return bond_get_slave_by_id(bond, 0);
5078 }
5079 }
5080
5081non_igmp:
5082 slave_cnt = READ_ONCE(bond->slave_cnt);
5083 if (likely(slave_cnt)) {
5084 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
5085 return bond_get_slave_by_id(bond, slave_id);
5086 }
5087 return NULL;
5088}
5089
5090static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
5091 struct net_device *bond_dev)
5092{
5093 struct bonding *bond = netdev_priv(bond_dev);
5094 struct slave *slave;
5095
5096 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5097 if (likely(slave))
5098 return bond_dev_queue_xmit(bond, skb, slave->dev);
5099
5100 return bond_tx_drop(bond_dev, skb);
5101}
5102
5103static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
5104{
5105 return rcu_dereference(bond->curr_active_slave);
5106}
5107
5108/* In active-backup mode, we know that bond->curr_active_slave is always valid if
5109 * the bond has a usable interface.
5110 */
5111static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
5112 struct net_device *bond_dev)
5113{
5114 struct bonding *bond = netdev_priv(bond_dev);
5115 struct slave *slave;
5116
5117 slave = bond_xmit_activebackup_slave_get(bond);
5118 if (slave)
5119 return bond_dev_queue_xmit(bond, skb, slave->dev);
5120
5121 return bond_tx_drop(bond_dev, skb);
5122}
5123
5124/* Use this to update slave_array when (a) it's not appropriate to update
5125 * slave_array right away (note that update_slave_array() may sleep)
5126 * and / or (b) RTNL is not held.
5127 */
5128void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
5129{
5130 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
5131}
5132
5133/* Slave array work handler. Holds only RTNL */
5134static void bond_slave_arr_handler(struct work_struct *work)
5135{
5136 struct bonding *bond = container_of(work, struct bonding,
5137 slave_arr_work.work);
5138 int ret;
5139
5140 if (!rtnl_trylock())
5141 goto err;
5142
5143 ret = bond_update_slave_arr(bond, NULL);
5144 rtnl_unlock();
5145 if (ret) {
5146 pr_warn_ratelimited("Failed to update slave array from WT\n");
5147 goto err;
5148 }
5149 return;
5150
5151err:
5152 bond_slave_arr_work_rearm(bond, 1);
5153}
5154
5155static void bond_skip_slave(struct bond_up_slave *slaves,
5156 struct slave *skipslave)
5157{
5158 int idx;
5159
5160 /* Rare situation where caller has asked to skip a specific
5161 * slave but allocation failed (most likely!). BTW this is
5162 * only possible when the call is initiated from
5163 * __bond_release_one(). In this situation; overwrite the
5164 * skipslave entry in the array with the last entry from the
5165 * array to avoid a situation where the xmit path may choose
5166 * this to-be-skipped slave to send a packet out.
5167 */
5168 for (idx = 0; slaves && idx < slaves->count; idx++) {
5169 if (skipslave == slaves->arr[idx]) {
5170 slaves->arr[idx] =
5171 slaves->arr[slaves->count - 1];
5172 slaves->count--;
5173 break;
5174 }
5175 }
5176}
5177
5178static void bond_set_slave_arr(struct bonding *bond,
5179 struct bond_up_slave *usable_slaves,
5180 struct bond_up_slave *all_slaves)
5181{
5182 struct bond_up_slave *usable, *all;
5183
5184 usable = rtnl_dereference(bond->usable_slaves);
5185 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5186 kfree_rcu(usable, rcu);
5187
5188 all = rtnl_dereference(bond->all_slaves);
5189 rcu_assign_pointer(bond->all_slaves, all_slaves);
5190 kfree_rcu(all, rcu);
5191}
5192
5193static void bond_reset_slave_arr(struct bonding *bond)
5194{
5195 bond_set_slave_arr(bond, NULL, NULL);
5196}
5197
5198/* Build the usable slaves array in control path for modes that use xmit-hash
5199 * to determine the slave interface -
5200 * (a) BOND_MODE_8023AD
5201 * (b) BOND_MODE_XOR
5202 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5203 *
5204 * The caller is expected to hold RTNL only and NO other lock!
5205 */
5206int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5207{
5208 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5209 struct slave *slave;
5210 struct list_head *iter;
5211 int agg_id = 0;
5212 int ret = 0;
5213
5214 might_sleep();
5215
5216 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5217 bond->slave_cnt), GFP_KERNEL);
5218 all_slaves = kzalloc(struct_size(all_slaves, arr,
5219 bond->slave_cnt), GFP_KERNEL);
5220 if (!usable_slaves || !all_slaves) {
5221 ret = -ENOMEM;
5222 goto out;
5223 }
5224 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5225 struct ad_info ad_info;
5226
5227 spin_lock_bh(&bond->mode_lock);
5228 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5229 spin_unlock_bh(&bond->mode_lock);
5230 pr_debug("bond_3ad_get_active_agg_info failed\n");
5231 /* No active aggragator means it's not safe to use
5232 * the previous array.
5233 */
5234 bond_reset_slave_arr(bond);
5235 goto out;
5236 }
5237 spin_unlock_bh(&bond->mode_lock);
5238 agg_id = ad_info.aggregator_id;
5239 }
5240 bond_for_each_slave(bond, slave, iter) {
5241 if (skipslave == slave)
5242 continue;
5243
5244 all_slaves->arr[all_slaves->count++] = slave;
5245 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5246 struct aggregator *agg;
5247
5248 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5249 if (!agg || agg->aggregator_identifier != agg_id)
5250 continue;
5251 }
5252 if (!bond_slave_can_tx(slave))
5253 continue;
5254
5255 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5256 usable_slaves->count);
5257
5258 usable_slaves->arr[usable_slaves->count++] = slave;
5259 }
5260
5261 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5262 return ret;
5263out:
5264 if (ret != 0 && skipslave) {
5265 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5266 skipslave);
5267 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5268 skipslave);
5269 }
5270 kfree_rcu(all_slaves, rcu);
5271 kfree_rcu(usable_slaves, rcu);
5272
5273 return ret;
5274}
5275
5276static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5277 struct sk_buff *skb,
5278 struct bond_up_slave *slaves)
5279{
5280 struct slave *slave;
5281 unsigned int count;
5282 u32 hash;
5283
5284 hash = bond_xmit_hash(bond, skb);
5285 count = slaves ? READ_ONCE(slaves->count) : 0;
5286 if (unlikely(!count))
5287 return NULL;
5288
5289 slave = slaves->arr[hash % count];
5290 return slave;
5291}
5292
5293static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5294 struct xdp_buff *xdp)
5295{
5296 struct bond_up_slave *slaves;
5297 unsigned int count;
5298 u32 hash;
5299
5300 hash = bond_xmit_hash_xdp(bond, xdp);
5301 slaves = rcu_dereference(bond->usable_slaves);
5302 count = slaves ? READ_ONCE(slaves->count) : 0;
5303 if (unlikely(!count))
5304 return NULL;
5305
5306 return slaves->arr[hash % count];
5307}
5308
5309/* Use this Xmit function for 3AD as well as XOR modes. The current
5310 * usable slave array is formed in the control path. The xmit function
5311 * just calculates hash and sends the packet out.
5312 */
5313static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5314 struct net_device *dev)
5315{
5316 struct bonding *bond = netdev_priv(dev);
5317 struct bond_up_slave *slaves;
5318 struct slave *slave;
5319
5320 slaves = rcu_dereference(bond->usable_slaves);
5321 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5322 if (likely(slave))
5323 return bond_dev_queue_xmit(bond, skb, slave->dev);
5324
5325 return bond_tx_drop(dev, skb);
5326}
5327
5328/* in broadcast mode, we send everything to all usable interfaces. */
5329static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5330 struct net_device *bond_dev)
5331{
5332 struct bonding *bond = netdev_priv(bond_dev);
5333 struct slave *slave = NULL;
5334 struct list_head *iter;
5335 bool xmit_suc = false;
5336 bool skb_used = false;
5337
5338 bond_for_each_slave_rcu(bond, slave, iter) {
5339 struct sk_buff *skb2;
5340
5341 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5342 continue;
5343
5344 if (bond_is_last_slave(bond, slave)) {
5345 skb2 = skb;
5346 skb_used = true;
5347 } else {
5348 skb2 = skb_clone(skb, GFP_ATOMIC);
5349 if (!skb2) {
5350 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5351 bond_dev->name, __func__);
5352 continue;
5353 }
5354 }
5355
5356 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5357 xmit_suc = true;
5358 }
5359
5360 if (!skb_used)
5361 dev_kfree_skb_any(skb);
5362
5363 if (xmit_suc)
5364 return NETDEV_TX_OK;
5365
5366 dev_core_stats_tx_dropped_inc(bond_dev);
5367 return NET_XMIT_DROP;
5368}
5369
5370/*------------------------- Device initialization ---------------------------*/
5371
5372/* Lookup the slave that corresponds to a qid */
5373static inline int bond_slave_override(struct bonding *bond,
5374 struct sk_buff *skb)
5375{
5376 struct slave *slave = NULL;
5377 struct list_head *iter;
5378
5379 if (!skb_rx_queue_recorded(skb))
5380 return 1;
5381
5382 /* Find out if any slaves have the same mapping as this skb. */
5383 bond_for_each_slave_rcu(bond, slave, iter) {
5384 if (READ_ONCE(slave->queue_id) == skb_get_queue_mapping(skb)) {
5385 if (bond_slave_is_up(slave) &&
5386 slave->link == BOND_LINK_UP) {
5387 bond_dev_queue_xmit(bond, skb, slave->dev);
5388 return 0;
5389 }
5390 /* If the slave isn't UP, use default transmit policy. */
5391 break;
5392 }
5393 }
5394
5395 return 1;
5396}
5397
5398
5399static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5400 struct net_device *sb_dev)
5401{
5402 /* This helper function exists to help dev_pick_tx get the correct
5403 * destination queue. Using a helper function skips a call to
5404 * skb_tx_hash and will put the skbs in the queue we expect on their
5405 * way down to the bonding driver.
5406 */
5407 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5408
5409 /* Save the original txq to restore before passing to the driver */
5410 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5411
5412 if (unlikely(txq >= dev->real_num_tx_queues)) {
5413 do {
5414 txq -= dev->real_num_tx_queues;
5415 } while (txq >= dev->real_num_tx_queues);
5416 }
5417 return txq;
5418}
5419
5420static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5421 struct sk_buff *skb,
5422 bool all_slaves)
5423{
5424 struct bonding *bond = netdev_priv(master_dev);
5425 struct bond_up_slave *slaves;
5426 struct slave *slave = NULL;
5427
5428 switch (BOND_MODE(bond)) {
5429 case BOND_MODE_ROUNDROBIN:
5430 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5431 break;
5432 case BOND_MODE_ACTIVEBACKUP:
5433 slave = bond_xmit_activebackup_slave_get(bond);
5434 break;
5435 case BOND_MODE_8023AD:
5436 case BOND_MODE_XOR:
5437 if (all_slaves)
5438 slaves = rcu_dereference(bond->all_slaves);
5439 else
5440 slaves = rcu_dereference(bond->usable_slaves);
5441 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5442 break;
5443 case BOND_MODE_BROADCAST:
5444 break;
5445 case BOND_MODE_ALB:
5446 slave = bond_xmit_alb_slave_get(bond, skb);
5447 break;
5448 case BOND_MODE_TLB:
5449 slave = bond_xmit_tlb_slave_get(bond, skb);
5450 break;
5451 default:
5452 /* Should never happen, mode already checked */
5453 WARN_ONCE(true, "Unknown bonding mode");
5454 break;
5455 }
5456
5457 if (slave)
5458 return slave->dev;
5459 return NULL;
5460}
5461
5462static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5463{
5464 switch (sk->sk_family) {
5465#if IS_ENABLED(CONFIG_IPV6)
5466 case AF_INET6:
5467 if (ipv6_only_sock(sk) ||
5468 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5469 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5470 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5471 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5472 break;
5473 }
5474 fallthrough;
5475#endif
5476 default: /* AF_INET */
5477 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5478 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5479 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5480 break;
5481 }
5482
5483 flow->ports.src = inet_sk(sk)->inet_sport;
5484 flow->ports.dst = inet_sk(sk)->inet_dport;
5485}
5486
5487/**
5488 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5489 * @sk: socket to use for headers
5490 *
5491 * This function will extract the necessary field from the socket and use
5492 * them to generate a hash based on the LAYER34 xmit_policy.
5493 * Assumes that sk is a TCP or UDP socket.
5494 */
5495static u32 bond_sk_hash_l34(struct sock *sk)
5496{
5497 struct flow_keys flow;
5498 u32 hash;
5499
5500 bond_sk_to_flow(sk, &flow);
5501
5502 /* L4 */
5503 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5504 /* L3 */
5505 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5506}
5507
5508static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5509 struct sock *sk)
5510{
5511 struct bond_up_slave *slaves;
5512 struct slave *slave;
5513 unsigned int count;
5514 u32 hash;
5515
5516 slaves = rcu_dereference(bond->usable_slaves);
5517 count = slaves ? READ_ONCE(slaves->count) : 0;
5518 if (unlikely(!count))
5519 return NULL;
5520
5521 hash = bond_sk_hash_l34(sk);
5522 slave = slaves->arr[hash % count];
5523
5524 return slave->dev;
5525}
5526
5527static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5528 struct sock *sk)
5529{
5530 struct bonding *bond = netdev_priv(dev);
5531 struct net_device *lower = NULL;
5532
5533 rcu_read_lock();
5534 if (bond_sk_check(bond))
5535 lower = __bond_sk_get_lower_dev(bond, sk);
5536 rcu_read_unlock();
5537
5538 return lower;
5539}
5540
5541#if IS_ENABLED(CONFIG_TLS_DEVICE)
5542static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5543 struct net_device *dev)
5544{
5545 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5546
5547 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5548 * was true, if tls_device_down is running in parallel, but it's OK,
5549 * because bond_get_slave_by_dev has a NULL check.
5550 */
5551 if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5552 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5553 return bond_tx_drop(dev, skb);
5554}
5555#endif
5556
5557static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5558{
5559 struct bonding *bond = netdev_priv(dev);
5560
5561 if (bond_should_override_tx_queue(bond) &&
5562 !bond_slave_override(bond, skb))
5563 return NETDEV_TX_OK;
5564
5565#if IS_ENABLED(CONFIG_TLS_DEVICE)
5566 if (tls_is_skb_tx_device_offloaded(skb))
5567 return bond_tls_device_xmit(bond, skb, dev);
5568#endif
5569
5570 switch (BOND_MODE(bond)) {
5571 case BOND_MODE_ROUNDROBIN:
5572 return bond_xmit_roundrobin(skb, dev);
5573 case BOND_MODE_ACTIVEBACKUP:
5574 return bond_xmit_activebackup(skb, dev);
5575 case BOND_MODE_8023AD:
5576 case BOND_MODE_XOR:
5577 return bond_3ad_xor_xmit(skb, dev);
5578 case BOND_MODE_BROADCAST:
5579 return bond_xmit_broadcast(skb, dev);
5580 case BOND_MODE_ALB:
5581 return bond_alb_xmit(skb, dev);
5582 case BOND_MODE_TLB:
5583 return bond_tlb_xmit(skb, dev);
5584 default:
5585 /* Should never happen, mode already checked */
5586 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5587 WARN_ON_ONCE(1);
5588 return bond_tx_drop(dev, skb);
5589 }
5590}
5591
5592static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5593{
5594 struct bonding *bond = netdev_priv(dev);
5595 netdev_tx_t ret = NETDEV_TX_OK;
5596
5597 /* If we risk deadlock from transmitting this in the
5598 * netpoll path, tell netpoll to queue the frame for later tx
5599 */
5600 if (unlikely(is_netpoll_tx_blocked(dev)))
5601 return NETDEV_TX_BUSY;
5602
5603 rcu_read_lock();
5604 if (bond_has_slaves(bond))
5605 ret = __bond_start_xmit(skb, dev);
5606 else
5607 ret = bond_tx_drop(dev, skb);
5608 rcu_read_unlock();
5609
5610 return ret;
5611}
5612
5613static struct net_device *
5614bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5615{
5616 struct bonding *bond = netdev_priv(bond_dev);
5617 struct slave *slave;
5618
5619 /* Caller needs to hold rcu_read_lock() */
5620
5621 switch (BOND_MODE(bond)) {
5622 case BOND_MODE_ROUNDROBIN:
5623 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5624 break;
5625
5626 case BOND_MODE_ACTIVEBACKUP:
5627 slave = bond_xmit_activebackup_slave_get(bond);
5628 break;
5629
5630 case BOND_MODE_8023AD:
5631 case BOND_MODE_XOR:
5632 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5633 break;
5634
5635 default:
5636 if (net_ratelimit())
5637 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n",
5638 BOND_MODE(bond));
5639 return NULL;
5640 }
5641
5642 if (slave)
5643 return slave->dev;
5644
5645 return NULL;
5646}
5647
5648static int bond_xdp_xmit(struct net_device *bond_dev,
5649 int n, struct xdp_frame **frames, u32 flags)
5650{
5651 int nxmit, err = -ENXIO;
5652
5653 rcu_read_lock();
5654
5655 for (nxmit = 0; nxmit < n; nxmit++) {
5656 struct xdp_frame *frame = frames[nxmit];
5657 struct xdp_frame *frames1[] = {frame};
5658 struct net_device *slave_dev;
5659 struct xdp_buff xdp;
5660
5661 xdp_convert_frame_to_buff(frame, &xdp);
5662
5663 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5664 if (!slave_dev) {
5665 err = -ENXIO;
5666 break;
5667 }
5668
5669 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5670 if (err < 1)
5671 break;
5672 }
5673
5674 rcu_read_unlock();
5675
5676 /* If error happened on the first frame then we can pass the error up, otherwise
5677 * report the number of frames that were xmitted.
5678 */
5679 if (err < 0)
5680 return (nxmit == 0 ? err : nxmit);
5681
5682 return nxmit;
5683}
5684
5685static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5686 struct netlink_ext_ack *extack)
5687{
5688 struct bonding *bond = netdev_priv(dev);
5689 struct list_head *iter;
5690 struct slave *slave, *rollback_slave;
5691 struct bpf_prog *old_prog;
5692 struct netdev_bpf xdp = {
5693 .command = XDP_SETUP_PROG,
5694 .flags = 0,
5695 .prog = prog,
5696 .extack = extack,
5697 };
5698 int err;
5699
5700 ASSERT_RTNL();
5701
5702 if (!bond_xdp_check(bond)) {
5703 BOND_NL_ERR(dev, extack,
5704 "No native XDP support for the current bonding mode");
5705 return -EOPNOTSUPP;
5706 }
5707
5708 old_prog = bond->xdp_prog;
5709 bond->xdp_prog = prog;
5710
5711 bond_for_each_slave(bond, slave, iter) {
5712 struct net_device *slave_dev = slave->dev;
5713
5714 if (!slave_dev->netdev_ops->ndo_bpf ||
5715 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5716 SLAVE_NL_ERR(dev, slave_dev, extack,
5717 "Slave device does not support XDP");
5718 err = -EOPNOTSUPP;
5719 goto err;
5720 }
5721
5722 if (dev_xdp_prog_count(slave_dev) > 0) {
5723 SLAVE_NL_ERR(dev, slave_dev, extack,
5724 "Slave has XDP program loaded, please unload before enslaving");
5725 err = -EOPNOTSUPP;
5726 goto err;
5727 }
5728
5729 err = dev_xdp_propagate(slave_dev, &xdp);
5730 if (err < 0) {
5731 /* ndo_bpf() sets extack error message */
5732 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5733 goto err;
5734 }
5735 if (prog)
5736 bpf_prog_inc(prog);
5737 }
5738
5739 if (prog) {
5740 static_branch_inc(&bpf_master_redirect_enabled_key);
5741 } else if (old_prog) {
5742 bpf_prog_put(old_prog);
5743 static_branch_dec(&bpf_master_redirect_enabled_key);
5744 }
5745
5746 return 0;
5747
5748err:
5749 /* unwind the program changes */
5750 bond->xdp_prog = old_prog;
5751 xdp.prog = old_prog;
5752 xdp.extack = NULL; /* do not overwrite original error */
5753
5754 bond_for_each_slave(bond, rollback_slave, iter) {
5755 struct net_device *slave_dev = rollback_slave->dev;
5756 int err_unwind;
5757
5758 if (slave == rollback_slave)
5759 break;
5760
5761 err_unwind = dev_xdp_propagate(slave_dev, &xdp);
5762 if (err_unwind < 0)
5763 slave_err(dev, slave_dev,
5764 "Error %d when unwinding XDP program change\n", err_unwind);
5765 else if (xdp.prog)
5766 bpf_prog_inc(xdp.prog);
5767 }
5768 return err;
5769}
5770
5771static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5772{
5773 switch (xdp->command) {
5774 case XDP_SETUP_PROG:
5775 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5776 default:
5777 return -EINVAL;
5778 }
5779}
5780
5781static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5782{
5783 if (speed == 0 || speed == SPEED_UNKNOWN)
5784 speed = slave->speed;
5785 else
5786 speed = min(speed, slave->speed);
5787
5788 return speed;
5789}
5790
5791/* Set the BOND_PHC_INDEX flag to notify user space */
5792static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5793{
5794 struct ifreq *ifr = kernel_cfg->ifr;
5795 struct hwtstamp_config cfg;
5796
5797 if (kernel_cfg->copied_to_user) {
5798 /* Lower device has a legacy implementation */
5799 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
5800 return -EFAULT;
5801
5802 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5803 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)))
5804 return -EFAULT;
5805 } else {
5806 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5807 }
5808
5809 return 0;
5810}
5811
5812static int bond_hwtstamp_get(struct net_device *dev,
5813 struct kernel_hwtstamp_config *cfg)
5814{
5815 struct bonding *bond = netdev_priv(dev);
5816 struct net_device *real_dev;
5817 int err;
5818
5819 real_dev = bond_option_active_slave_get_rcu(bond);
5820 if (!real_dev)
5821 return -EOPNOTSUPP;
5822
5823 err = generic_hwtstamp_get_lower(real_dev, cfg);
5824 if (err)
5825 return err;
5826
5827 return bond_set_phc_index_flag(cfg);
5828}
5829
5830static int bond_hwtstamp_set(struct net_device *dev,
5831 struct kernel_hwtstamp_config *cfg,
5832 struct netlink_ext_ack *extack)
5833{
5834 struct bonding *bond = netdev_priv(dev);
5835 struct net_device *real_dev;
5836 int err;
5837
5838 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5839 return -EOPNOTSUPP;
5840
5841 real_dev = bond_option_active_slave_get_rcu(bond);
5842 if (!real_dev)
5843 return -EOPNOTSUPP;
5844
5845 err = generic_hwtstamp_set_lower(real_dev, cfg, extack);
5846 if (err)
5847 return err;
5848
5849 return bond_set_phc_index_flag(cfg);
5850}
5851
5852static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5853 struct ethtool_link_ksettings *cmd)
5854{
5855 struct bonding *bond = netdev_priv(bond_dev);
5856 struct list_head *iter;
5857 struct slave *slave;
5858 u32 speed = 0;
5859
5860 cmd->base.duplex = DUPLEX_UNKNOWN;
5861 cmd->base.port = PORT_OTHER;
5862
5863 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5864 * do not need to check mode. Though link speed might not represent
5865 * the true receive or transmit bandwidth (not all modes are symmetric)
5866 * this is an accurate maximum.
5867 */
5868 bond_for_each_slave(bond, slave, iter) {
5869 if (bond_slave_can_tx(slave)) {
5870 bond_update_speed_duplex(slave);
5871 if (slave->speed != SPEED_UNKNOWN) {
5872 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5873 speed = bond_mode_bcast_speed(slave,
5874 speed);
5875 else
5876 speed += slave->speed;
5877 }
5878 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5879 slave->duplex != DUPLEX_UNKNOWN)
5880 cmd->base.duplex = slave->duplex;
5881 }
5882 }
5883 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5884
5885 return 0;
5886}
5887
5888static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5889 struct ethtool_drvinfo *drvinfo)
5890{
5891 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5892 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5893 BOND_ABI_VERSION);
5894}
5895
5896static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5897 struct kernel_ethtool_ts_info *info)
5898{
5899 struct bonding *bond = netdev_priv(bond_dev);
5900 struct kernel_ethtool_ts_info ts_info;
5901 struct net_device *real_dev;
5902 bool sw_tx_support = false;
5903 struct list_head *iter;
5904 struct slave *slave;
5905 int ret = 0;
5906
5907 rcu_read_lock();
5908 real_dev = bond_option_active_slave_get_rcu(bond);
5909 dev_hold(real_dev);
5910 rcu_read_unlock();
5911
5912 if (real_dev) {
5913 ret = ethtool_get_ts_info_by_layer(real_dev, info);
5914 } else {
5915 /* Check if all slaves support software tx timestamping */
5916 rcu_read_lock();
5917 bond_for_each_slave_rcu(bond, slave, iter) {
5918 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info);
5919 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5920 sw_tx_support = true;
5921 continue;
5922 }
5923
5924 sw_tx_support = false;
5925 break;
5926 }
5927 rcu_read_unlock();
5928 }
5929
5930 if (sw_tx_support)
5931 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5932
5933 dev_put(real_dev);
5934 return ret;
5935}
5936
5937static const struct ethtool_ops bond_ethtool_ops = {
5938 .get_drvinfo = bond_ethtool_get_drvinfo,
5939 .get_link = ethtool_op_get_link,
5940 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5941 .get_ts_info = bond_ethtool_get_ts_info,
5942};
5943
5944static const struct net_device_ops bond_netdev_ops = {
5945 .ndo_init = bond_init,
5946 .ndo_uninit = bond_uninit,
5947 .ndo_open = bond_open,
5948 .ndo_stop = bond_close,
5949 .ndo_start_xmit = bond_start_xmit,
5950 .ndo_select_queue = bond_select_queue,
5951 .ndo_get_stats64 = bond_get_stats,
5952 .ndo_eth_ioctl = bond_eth_ioctl,
5953 .ndo_siocbond = bond_do_ioctl,
5954 .ndo_siocdevprivate = bond_siocdevprivate,
5955 .ndo_change_rx_flags = bond_change_rx_flags,
5956 .ndo_set_rx_mode = bond_set_rx_mode,
5957 .ndo_change_mtu = bond_change_mtu,
5958 .ndo_set_mac_address = bond_set_mac_address,
5959 .ndo_neigh_setup = bond_neigh_setup,
5960 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5961 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5962#ifdef CONFIG_NET_POLL_CONTROLLER
5963 .ndo_netpoll_setup = bond_netpoll_setup,
5964 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5965 .ndo_poll_controller = bond_poll_controller,
5966#endif
5967 .ndo_add_slave = bond_enslave,
5968 .ndo_del_slave = bond_release,
5969 .ndo_fix_features = bond_fix_features,
5970 .ndo_features_check = passthru_features_check,
5971 .ndo_get_xmit_slave = bond_xmit_get_slave,
5972 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5973 .ndo_bpf = bond_xdp,
5974 .ndo_xdp_xmit = bond_xdp_xmit,
5975 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5976 .ndo_hwtstamp_get = bond_hwtstamp_get,
5977 .ndo_hwtstamp_set = bond_hwtstamp_set,
5978};
5979
5980static const struct device_type bond_type = {
5981 .name = "bond",
5982};
5983
5984static void bond_destructor(struct net_device *bond_dev)
5985{
5986 struct bonding *bond = netdev_priv(bond_dev);
5987
5988 if (bond->wq)
5989 destroy_workqueue(bond->wq);
5990
5991 free_percpu(bond->rr_tx_counter);
5992}
5993
5994void bond_setup(struct net_device *bond_dev)
5995{
5996 struct bonding *bond = netdev_priv(bond_dev);
5997
5998 spin_lock_init(&bond->mode_lock);
5999 bond->params = bonding_defaults;
6000
6001 /* Initialize pointers */
6002 bond->dev = bond_dev;
6003
6004 /* Initialize the device entry points */
6005 ether_setup(bond_dev);
6006 bond_dev->max_mtu = ETH_MAX_MTU;
6007 bond_dev->netdev_ops = &bond_netdev_ops;
6008 bond_dev->ethtool_ops = &bond_ethtool_ops;
6009
6010 bond_dev->needs_free_netdev = true;
6011 bond_dev->priv_destructor = bond_destructor;
6012
6013 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
6014
6015 /* Initialize the device options */
6016 bond_dev->flags |= IFF_MASTER;
6017 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
6018 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
6019
6020#ifdef CONFIG_XFRM_OFFLOAD
6021 /* set up xfrm device ops (only supported in active-backup right now) */
6022 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
6023 INIT_LIST_HEAD(&bond->ipsec_list);
6024 mutex_init(&bond->ipsec_lock);
6025#endif /* CONFIG_XFRM_OFFLOAD */
6026
6027 /* don't acquire bond device's netif_tx_lock when transmitting */
6028 bond_dev->lltx = true;
6029
6030 /* Don't allow bond devices to change network namespaces. */
6031 bond_dev->netns_local = true;
6032
6033 /* By default, we declare the bond to be fully
6034 * VLAN hardware accelerated capable. Special
6035 * care is taken in the various xmit functions
6036 * when there are slaves that are not hw accel
6037 * capable
6038 */
6039
6040 bond_dev->hw_features = BOND_VLAN_FEATURES |
6041 NETIF_F_HW_VLAN_CTAG_RX |
6042 NETIF_F_HW_VLAN_CTAG_FILTER |
6043 NETIF_F_HW_VLAN_STAG_RX |
6044 NETIF_F_HW_VLAN_STAG_FILTER;
6045
6046 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
6047 bond_dev->features |= bond_dev->hw_features;
6048 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
6049 bond_dev->features |= NETIF_F_GSO_PARTIAL;
6050#ifdef CONFIG_XFRM_OFFLOAD
6051 bond_dev->hw_features |= BOND_XFRM_FEATURES;
6052 /* Only enable XFRM features if this is an active-backup config */
6053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
6054 bond_dev->features |= BOND_XFRM_FEATURES;
6055#endif /* CONFIG_XFRM_OFFLOAD */
6056}
6057
6058/* Destroy a bonding device.
6059 * Must be under rtnl_lock when this function is called.
6060 */
6061static void bond_uninit(struct net_device *bond_dev)
6062{
6063 struct bonding *bond = netdev_priv(bond_dev);
6064 struct list_head *iter;
6065 struct slave *slave;
6066
6067 bond_netpoll_cleanup(bond_dev);
6068
6069 /* Release the bonded slaves */
6070 bond_for_each_slave(bond, slave, iter)
6071 __bond_release_one(bond_dev, slave->dev, true, true);
6072 netdev_info(bond_dev, "Released all slaves\n");
6073
6074#ifdef CONFIG_XFRM_OFFLOAD
6075 mutex_destroy(&bond->ipsec_lock);
6076#endif /* CONFIG_XFRM_OFFLOAD */
6077
6078 bond_set_slave_arr(bond, NULL, NULL);
6079
6080 list_del_rcu(&bond->bond_list);
6081
6082 bond_debug_unregister(bond);
6083}
6084
6085/*------------------------- Module initialization ---------------------------*/
6086
6087static int __init bond_check_params(struct bond_params *params)
6088{
6089 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
6090 struct bond_opt_value newval;
6091 const struct bond_opt_value *valptr;
6092 int arp_all_targets_value = 0;
6093 u16 ad_actor_sys_prio = 0;
6094 u16 ad_user_port_key = 0;
6095 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
6096 int arp_ip_count;
6097 int bond_mode = BOND_MODE_ROUNDROBIN;
6098 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
6099 int lacp_fast = 0;
6100 int tlb_dynamic_lb;
6101
6102 /* Convert string parameters. */
6103 if (mode) {
6104 bond_opt_initstr(&newval, mode);
6105 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
6106 if (!valptr) {
6107 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
6108 return -EINVAL;
6109 }
6110 bond_mode = valptr->value;
6111 }
6112
6113 if (xmit_hash_policy) {
6114 if (bond_mode == BOND_MODE_ROUNDROBIN ||
6115 bond_mode == BOND_MODE_ACTIVEBACKUP ||
6116 bond_mode == BOND_MODE_BROADCAST) {
6117 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
6118 bond_mode_name(bond_mode));
6119 } else {
6120 bond_opt_initstr(&newval, xmit_hash_policy);
6121 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
6122 &newval);
6123 if (!valptr) {
6124 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
6125 xmit_hash_policy);
6126 return -EINVAL;
6127 }
6128 xmit_hashtype = valptr->value;
6129 }
6130 }
6131
6132 if (lacp_rate) {
6133 if (bond_mode != BOND_MODE_8023AD) {
6134 pr_info("lacp_rate param is irrelevant in mode %s\n",
6135 bond_mode_name(bond_mode));
6136 } else {
6137 bond_opt_initstr(&newval, lacp_rate);
6138 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
6139 &newval);
6140 if (!valptr) {
6141 pr_err("Error: Invalid lacp rate \"%s\"\n",
6142 lacp_rate);
6143 return -EINVAL;
6144 }
6145 lacp_fast = valptr->value;
6146 }
6147 }
6148
6149 if (ad_select) {
6150 bond_opt_initstr(&newval, ad_select);
6151 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6152 &newval);
6153 if (!valptr) {
6154 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6155 return -EINVAL;
6156 }
6157 params->ad_select = valptr->value;
6158 if (bond_mode != BOND_MODE_8023AD)
6159 pr_warn("ad_select param only affects 802.3ad mode\n");
6160 } else {
6161 params->ad_select = BOND_AD_STABLE;
6162 }
6163
6164 if (max_bonds < 0) {
6165 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6166 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6167 max_bonds = BOND_DEFAULT_MAX_BONDS;
6168 }
6169
6170 if (miimon < 0) {
6171 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6172 miimon, INT_MAX);
6173 miimon = 0;
6174 }
6175
6176 if (updelay < 0) {
6177 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6178 updelay, INT_MAX);
6179 updelay = 0;
6180 }
6181
6182 if (downdelay < 0) {
6183 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6184 downdelay, INT_MAX);
6185 downdelay = 0;
6186 }
6187
6188 if ((use_carrier != 0) && (use_carrier != 1)) {
6189 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6190 use_carrier);
6191 use_carrier = 1;
6192 }
6193
6194 if (num_peer_notif < 0 || num_peer_notif > 255) {
6195 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6196 num_peer_notif);
6197 num_peer_notif = 1;
6198 }
6199
6200 /* reset values for 802.3ad/TLB/ALB */
6201 if (!bond_mode_uses_arp(bond_mode)) {
6202 if (!miimon) {
6203 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");
6204 pr_warn("Forcing miimon to 100msec\n");
6205 miimon = BOND_DEFAULT_MIIMON;
6206 }
6207 }
6208
6209 if (tx_queues < 1 || tx_queues > 255) {
6210 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6211 tx_queues, BOND_DEFAULT_TX_QUEUES);
6212 tx_queues = BOND_DEFAULT_TX_QUEUES;
6213 }
6214
6215 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6216 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6217 all_slaves_active);
6218 all_slaves_active = 0;
6219 }
6220
6221 if (resend_igmp < 0 || resend_igmp > 255) {
6222 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6223 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6224 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6225 }
6226
6227 bond_opt_initval(&newval, packets_per_slave);
6228 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6229 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6230 packets_per_slave, USHRT_MAX);
6231 packets_per_slave = 1;
6232 }
6233
6234 if (bond_mode == BOND_MODE_ALB) {
6235 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",
6236 updelay);
6237 }
6238
6239 if (!miimon) {
6240 if (updelay || downdelay) {
6241 /* just warn the user the up/down delay will have
6242 * no effect since miimon is zero...
6243 */
6244 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",
6245 updelay, downdelay);
6246 }
6247 } else {
6248 /* don't allow arp monitoring */
6249 if (arp_interval) {
6250 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6251 miimon, arp_interval);
6252 arp_interval = 0;
6253 }
6254
6255 if ((updelay % miimon) != 0) {
6256 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6257 updelay, miimon, (updelay / miimon) * miimon);
6258 }
6259
6260 updelay /= miimon;
6261
6262 if ((downdelay % miimon) != 0) {
6263 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6264 downdelay, miimon,
6265 (downdelay / miimon) * miimon);
6266 }
6267
6268 downdelay /= miimon;
6269 }
6270
6271 if (arp_interval < 0) {
6272 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6273 arp_interval, INT_MAX);
6274 arp_interval = 0;
6275 }
6276
6277 for (arp_ip_count = 0, i = 0;
6278 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6279 __be32 ip;
6280
6281 /* not a complete check, but good enough to catch mistakes */
6282 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6283 !bond_is_ip_target_ok(ip)) {
6284 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6285 arp_ip_target[i]);
6286 arp_interval = 0;
6287 } else {
6288 if (bond_get_targets_ip(arp_target, ip) == -1)
6289 arp_target[arp_ip_count++] = ip;
6290 else
6291 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6292 &ip);
6293 }
6294 }
6295
6296 if (arp_interval && !arp_ip_count) {
6297 /* don't allow arping if no arp_ip_target given... */
6298 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6299 arp_interval);
6300 arp_interval = 0;
6301 }
6302
6303 if (arp_validate) {
6304 if (!arp_interval) {
6305 pr_err("arp_validate requires arp_interval\n");
6306 return -EINVAL;
6307 }
6308
6309 bond_opt_initstr(&newval, arp_validate);
6310 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6311 &newval);
6312 if (!valptr) {
6313 pr_err("Error: invalid arp_validate \"%s\"\n",
6314 arp_validate);
6315 return -EINVAL;
6316 }
6317 arp_validate_value = valptr->value;
6318 } else {
6319 arp_validate_value = 0;
6320 }
6321
6322 if (arp_all_targets) {
6323 bond_opt_initstr(&newval, arp_all_targets);
6324 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6325 &newval);
6326 if (!valptr) {
6327 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6328 arp_all_targets);
6329 arp_all_targets_value = 0;
6330 } else {
6331 arp_all_targets_value = valptr->value;
6332 }
6333 }
6334
6335 if (miimon) {
6336 pr_info("MII link monitoring set to %d ms\n", miimon);
6337 } else if (arp_interval) {
6338 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6339 arp_validate_value);
6340 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6341 arp_interval, valptr->string, arp_ip_count);
6342
6343 for (i = 0; i < arp_ip_count; i++)
6344 pr_cont(" %s", arp_ip_target[i]);
6345
6346 pr_cont("\n");
6347
6348 } else if (max_bonds) {
6349 /* miimon and arp_interval not set, we need one so things
6350 * work as expected, see bonding.txt for details
6351 */
6352 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");
6353 }
6354
6355 if (primary && !bond_mode_uses_primary(bond_mode)) {
6356 /* currently, using a primary only makes sense
6357 * in active backup, TLB or ALB modes
6358 */
6359 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6360 primary, bond_mode_name(bond_mode));
6361 primary = NULL;
6362 }
6363
6364 if (primary && primary_reselect) {
6365 bond_opt_initstr(&newval, primary_reselect);
6366 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6367 &newval);
6368 if (!valptr) {
6369 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6370 primary_reselect);
6371 return -EINVAL;
6372 }
6373 primary_reselect_value = valptr->value;
6374 } else {
6375 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6376 }
6377
6378 if (fail_over_mac) {
6379 bond_opt_initstr(&newval, fail_over_mac);
6380 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6381 &newval);
6382 if (!valptr) {
6383 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6384 fail_over_mac);
6385 return -EINVAL;
6386 }
6387 fail_over_mac_value = valptr->value;
6388 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6389 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6390 } else {
6391 fail_over_mac_value = BOND_FOM_NONE;
6392 }
6393
6394 bond_opt_initstr(&newval, "default");
6395 valptr = bond_opt_parse(
6396 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6397 &newval);
6398 if (!valptr) {
6399 pr_err("Error: No ad_actor_sys_prio default value");
6400 return -EINVAL;
6401 }
6402 ad_actor_sys_prio = valptr->value;
6403
6404 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6405 &newval);
6406 if (!valptr) {
6407 pr_err("Error: No ad_user_port_key default value");
6408 return -EINVAL;
6409 }
6410 ad_user_port_key = valptr->value;
6411
6412 bond_opt_initstr(&newval, "default");
6413 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6414 if (!valptr) {
6415 pr_err("Error: No tlb_dynamic_lb default value");
6416 return -EINVAL;
6417 }
6418 tlb_dynamic_lb = valptr->value;
6419
6420 if (lp_interval == 0) {
6421 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6422 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6423 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6424 }
6425
6426 /* fill params struct with the proper values */
6427 params->mode = bond_mode;
6428 params->xmit_policy = xmit_hashtype;
6429 params->miimon = miimon;
6430 params->num_peer_notif = num_peer_notif;
6431 params->arp_interval = arp_interval;
6432 params->arp_validate = arp_validate_value;
6433 params->arp_all_targets = arp_all_targets_value;
6434 params->missed_max = 2;
6435 params->updelay = updelay;
6436 params->downdelay = downdelay;
6437 params->peer_notif_delay = 0;
6438 params->use_carrier = use_carrier;
6439 params->lacp_active = 1;
6440 params->lacp_fast = lacp_fast;
6441 params->primary[0] = 0;
6442 params->primary_reselect = primary_reselect_value;
6443 params->fail_over_mac = fail_over_mac_value;
6444 params->tx_queues = tx_queues;
6445 params->all_slaves_active = all_slaves_active;
6446 params->resend_igmp = resend_igmp;
6447 params->min_links = min_links;
6448 params->lp_interval = lp_interval;
6449 params->packets_per_slave = packets_per_slave;
6450 params->tlb_dynamic_lb = tlb_dynamic_lb;
6451 params->ad_actor_sys_prio = ad_actor_sys_prio;
6452 eth_zero_addr(params->ad_actor_system);
6453 params->ad_user_port_key = ad_user_port_key;
6454 params->coupled_control = 1;
6455 if (packets_per_slave > 0) {
6456 params->reciprocal_packets_per_slave =
6457 reciprocal_value(packets_per_slave);
6458 } else {
6459 /* reciprocal_packets_per_slave is unused if
6460 * packets_per_slave is 0 or 1, just initialize it
6461 */
6462 params->reciprocal_packets_per_slave =
6463 (struct reciprocal_value) { 0 };
6464 }
6465
6466 if (primary)
6467 strscpy_pad(params->primary, primary, sizeof(params->primary));
6468
6469 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6470#if IS_ENABLED(CONFIG_IPV6)
6471 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6472#endif
6473
6474 return 0;
6475}
6476
6477/* Called from registration process */
6478static int bond_init(struct net_device *bond_dev)
6479{
6480 struct bonding *bond = netdev_priv(bond_dev);
6481 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6482
6483 netdev_dbg(bond_dev, "Begin bond_init\n");
6484
6485 bond->wq = alloc_ordered_workqueue("%s", WQ_MEM_RECLAIM,
6486 bond_dev->name);
6487 if (!bond->wq)
6488 return -ENOMEM;
6489
6490 bond->notifier_ctx = false;
6491
6492 spin_lock_init(&bond->stats_lock);
6493 netdev_lockdep_set_classes(bond_dev);
6494
6495 list_add_tail_rcu(&bond->bond_list, &bn->dev_list);
6496
6497 bond_prepare_sysfs_group(bond);
6498
6499 bond_debug_register(bond);
6500
6501 /* Ensure valid dev_addr */
6502 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6503 bond_dev->addr_assign_type == NET_ADDR_PERM)
6504 eth_hw_addr_random(bond_dev);
6505
6506 return 0;
6507}
6508
6509unsigned int bond_get_num_tx_queues(void)
6510{
6511 return tx_queues;
6512}
6513
6514/* Create a new bond based on the specified name and bonding parameters.
6515 * If name is NULL, obtain a suitable "bond%d" name for us.
6516 * Caller must NOT hold rtnl_lock; we need to release it here before we
6517 * set up our sysfs entries.
6518 */
6519int bond_create(struct net *net, const char *name)
6520{
6521 struct net_device *bond_dev;
6522 struct bonding *bond;
6523 int res = -ENOMEM;
6524
6525 rtnl_lock();
6526
6527 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6528 name ? name : "bond%d", NET_NAME_UNKNOWN,
6529 bond_setup, tx_queues);
6530 if (!bond_dev)
6531 goto out;
6532
6533 bond = netdev_priv(bond_dev);
6534 dev_net_set(bond_dev, net);
6535 bond_dev->rtnl_link_ops = &bond_link_ops;
6536
6537 res = register_netdevice(bond_dev);
6538 if (res < 0) {
6539 free_netdev(bond_dev);
6540 goto out;
6541 }
6542
6543 netif_carrier_off(bond_dev);
6544
6545 bond_work_init_all(bond);
6546
6547out:
6548 rtnl_unlock();
6549 return res;
6550}
6551
6552static int __net_init bond_net_init(struct net *net)
6553{
6554 struct bond_net *bn = net_generic(net, bond_net_id);
6555
6556 bn->net = net;
6557 INIT_LIST_HEAD(&bn->dev_list);
6558
6559 bond_create_proc_dir(bn);
6560 bond_create_sysfs(bn);
6561
6562 return 0;
6563}
6564
6565/* According to commit 69b0216ac255 ("bonding: fix bonding_masters
6566 * race condition in bond unloading") we need to remove sysfs files
6567 * before we remove our devices (done later in bond_net_exit_batch_rtnl())
6568 */
6569static void __net_exit bond_net_pre_exit(struct net *net)
6570{
6571 struct bond_net *bn = net_generic(net, bond_net_id);
6572
6573 bond_destroy_sysfs(bn);
6574}
6575
6576static void __net_exit bond_net_exit_batch_rtnl(struct list_head *net_list,
6577 struct list_head *dev_kill_list)
6578{
6579 struct bond_net *bn;
6580 struct net *net;
6581
6582 /* Kill off any bonds created after unregistering bond rtnl ops */
6583 list_for_each_entry(net, net_list, exit_list) {
6584 struct bonding *bond, *tmp_bond;
6585
6586 bn = net_generic(net, bond_net_id);
6587 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6588 unregister_netdevice_queue(bond->dev, dev_kill_list);
6589 }
6590}
6591
6592/* According to commit 23fa5c2caae0 ("bonding: destroy proc directory
6593 * only after all bonds are gone") bond_destroy_proc_dir() is called
6594 * after bond_net_exit_batch_rtnl() has completed.
6595 */
6596static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6597{
6598 struct bond_net *bn;
6599 struct net *net;
6600
6601 list_for_each_entry(net, net_list, exit_list) {
6602 bn = net_generic(net, bond_net_id);
6603 bond_destroy_proc_dir(bn);
6604 }
6605}
6606
6607static struct pernet_operations bond_net_ops = {
6608 .init = bond_net_init,
6609 .pre_exit = bond_net_pre_exit,
6610 .exit_batch_rtnl = bond_net_exit_batch_rtnl,
6611 .exit_batch = bond_net_exit_batch,
6612 .id = &bond_net_id,
6613 .size = sizeof(struct bond_net),
6614};
6615
6616static int __init bonding_init(void)
6617{
6618 int i;
6619 int res;
6620
6621 res = bond_check_params(&bonding_defaults);
6622 if (res)
6623 goto out;
6624
6625 bond_create_debugfs();
6626
6627 res = register_pernet_subsys(&bond_net_ops);
6628 if (res)
6629 goto err_net_ops;
6630
6631 res = bond_netlink_init();
6632 if (res)
6633 goto err_link;
6634
6635 for (i = 0; i < max_bonds; i++) {
6636 res = bond_create(&init_net, NULL);
6637 if (res)
6638 goto err;
6639 }
6640
6641 skb_flow_dissector_init(&flow_keys_bonding,
6642 flow_keys_bonding_keys,
6643 ARRAY_SIZE(flow_keys_bonding_keys));
6644
6645 register_netdevice_notifier(&bond_netdev_notifier);
6646out:
6647 return res;
6648err:
6649 bond_netlink_fini();
6650err_link:
6651 unregister_pernet_subsys(&bond_net_ops);
6652err_net_ops:
6653 bond_destroy_debugfs();
6654 goto out;
6655
6656}
6657
6658static void __exit bonding_exit(void)
6659{
6660 unregister_netdevice_notifier(&bond_netdev_notifier);
6661
6662 bond_netlink_fini();
6663 unregister_pernet_subsys(&bond_net_ops);
6664
6665 bond_destroy_debugfs();
6666
6667#ifdef CONFIG_NET_POLL_CONTROLLER
6668 /* Make sure we don't have an imbalance on our netpoll blocking */
6669 WARN_ON(atomic_read(&netpoll_block_tx));
6670#endif
6671}
6672
6673module_init(bonding_init);
6674module_exit(bonding_exit);
6675MODULE_LICENSE("GPL");
6676MODULE_DESCRIPTION(DRV_DESCRIPTION);
6677MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");