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/bonding.h>
81#include <net/bond_3ad.h>
82#include <net/bond_alb.h>
83
84#include "bonding_priv.h"
85
86/*---------------------------- Module parameters ----------------------------*/
87
88/* monitor all links that often (in milliseconds). <=0 disables monitoring */
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92static int num_peer_notif = 1;
93static int miimon;
94static int updelay;
95static int downdelay;
96static int use_carrier = 1;
97static char *mode;
98static char *primary;
99static char *primary_reselect;
100static char *lacp_rate;
101static int min_links;
102static char *ad_select;
103static char *xmit_hash_policy;
104static int arp_interval;
105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *arp_all_targets;
108static char *fail_over_mac;
109static int all_slaves_active;
110static struct bond_params bonding_defaults;
111static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112static int packets_per_slave = 1;
113static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
114
115module_param(max_bonds, int, 0);
116MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117module_param(tx_queues, int, 0);
118MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125module_param(miimon, int, 0);
126MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127module_param(updelay, int, 0);
128MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129module_param(downdelay, int, 0);
130MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 "in milliseconds");
132module_param(use_carrier, int, 0);
133MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135module_param(mode, charp, 0);
136MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140module_param(primary, charp, 0);
141MODULE_PARM_DESC(primary, "Primary network device to use");
142module_param(primary_reselect, charp, 0);
143MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "once it comes up; "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
147 "better, "
148 "2 for only on active slave "
149 "failure");
150module_param(lacp_rate, charp, 0);
151MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153module_param(ad_select, charp, 0);
154MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
156 "2 for count");
157module_param(min_links, int, 0);
158MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159
160module_param(xmit_hash_policy, charp, 0);
161MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165module_param(arp_interval, int, 0);
166MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167module_param_array(arp_ip_target, charp, NULL, 0);
168MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169module_param(arp_validate, charp, 0);
170MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173module_param(arp_all_targets, charp, 0);
174MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175module_param(fail_over_mac, charp, 0);
176MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179module_param(all_slaves_active, int, 0);
180MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183module_param(resend_igmp, int, 0);
184MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 "link failure");
186module_param(packets_per_slave, int, 0);
187MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190module_param(lp_interval, uint, 0);
191MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
194
195/*----------------------------- Global variables ----------------------------*/
196
197#ifdef CONFIG_NET_POLL_CONTROLLER
198atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199#endif
200
201unsigned int bond_net_id __read_mostly;
202
203/*-------------------------- Forward declarations ---------------------------*/
204
205static int bond_init(struct net_device *bond_dev);
206static void bond_uninit(struct net_device *bond_dev);
207static void bond_get_stats(struct net_device *bond_dev,
208 struct rtnl_link_stats64 *stats);
209static void bond_slave_arr_handler(struct work_struct *work);
210static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
211 int mod);
212static void bond_netdev_notify_work(struct work_struct *work);
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_set_queue_mapping(skb, 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 int bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
613{
614 int err;
615
616 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
617 bond_dev, slave_dev, slave_dev->addr_len);
618 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
619 if (err)
620 return err;
621
622 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
623 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
624 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
625 return 0;
626}
627
628static struct slave *bond_get_old_active(struct bonding *bond,
629 struct slave *new_active)
630{
631 struct slave *slave;
632 struct list_head *iter;
633
634 bond_for_each_slave(bond, slave, iter) {
635 if (slave == new_active)
636 continue;
637
638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
639 return slave;
640 }
641
642 return NULL;
643}
644
645/* bond_do_fail_over_mac
646 *
647 * Perform special MAC address swapping for fail_over_mac settings
648 *
649 * Called with RTNL
650 */
651static void bond_do_fail_over_mac(struct bonding *bond,
652 struct slave *new_active,
653 struct slave *old_active)
654{
655 u8 tmp_mac[MAX_ADDR_LEN];
656 struct sockaddr_storage ss;
657 int rv;
658
659 switch (bond->params.fail_over_mac) {
660 case BOND_FOM_ACTIVE:
661 if (new_active) {
662 rv = bond_set_dev_addr(bond->dev, new_active->dev);
663 if (rv)
664 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
665 -rv);
666 }
667 break;
668 case BOND_FOM_FOLLOW:
669 /* if new_active && old_active, swap them
670 * if just old_active, do nothing (going to no active slave)
671 * if just new_active, set new_active to bond's MAC
672 */
673 if (!new_active)
674 return;
675
676 if (!old_active)
677 old_active = bond_get_old_active(bond, new_active);
678
679 if (old_active) {
680 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
681 new_active->dev->addr_len);
682 bond_hw_addr_copy(ss.__data,
683 old_active->dev->dev_addr,
684 old_active->dev->addr_len);
685 ss.ss_family = new_active->dev->type;
686 } else {
687 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
688 bond->dev->addr_len);
689 ss.ss_family = bond->dev->type;
690 }
691
692 rv = dev_set_mac_address(new_active->dev,
693 (struct sockaddr *)&ss, NULL);
694 if (rv) {
695 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
696 -rv);
697 goto out;
698 }
699
700 if (!old_active)
701 goto out;
702
703 bond_hw_addr_copy(ss.__data, tmp_mac,
704 new_active->dev->addr_len);
705 ss.ss_family = old_active->dev->type;
706
707 rv = dev_set_mac_address(old_active->dev,
708 (struct sockaddr *)&ss, NULL);
709 if (rv)
710 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
711 -rv);
712out:
713 break;
714 default:
715 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
716 bond->params.fail_over_mac);
717 break;
718 }
719
720}
721
722static struct slave *bond_choose_primary_or_current(struct bonding *bond)
723{
724 struct slave *prim = rtnl_dereference(bond->primary_slave);
725 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
726
727 if (!prim || prim->link != BOND_LINK_UP) {
728 if (!curr || curr->link != BOND_LINK_UP)
729 return NULL;
730 return curr;
731 }
732
733 if (bond->force_primary) {
734 bond->force_primary = false;
735 return prim;
736 }
737
738 if (!curr || curr->link != BOND_LINK_UP)
739 return prim;
740
741 /* At this point, prim and curr are both up */
742 switch (bond->params.primary_reselect) {
743 case BOND_PRI_RESELECT_ALWAYS:
744 return prim;
745 case BOND_PRI_RESELECT_BETTER:
746 if (prim->speed < curr->speed)
747 return curr;
748 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
749 return curr;
750 return prim;
751 case BOND_PRI_RESELECT_FAILURE:
752 return curr;
753 default:
754 netdev_err(bond->dev, "impossible primary_reselect %d\n",
755 bond->params.primary_reselect);
756 return curr;
757 }
758}
759
760/**
761 * bond_find_best_slave - select the best available slave to be the active one
762 * @bond: our bonding struct
763 */
764static struct slave *bond_find_best_slave(struct bonding *bond)
765{
766 struct slave *slave, *bestslave = NULL;
767 struct list_head *iter;
768 int mintime = bond->params.updelay;
769
770 slave = bond_choose_primary_or_current(bond);
771 if (slave)
772 return slave;
773
774 bond_for_each_slave(bond, slave, iter) {
775 if (slave->link == BOND_LINK_UP)
776 return slave;
777 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
778 slave->delay < mintime) {
779 mintime = slave->delay;
780 bestslave = slave;
781 }
782 }
783
784 return bestslave;
785}
786
787static bool bond_should_notify_peers(struct bonding *bond)
788{
789 struct slave *slave;
790
791 rcu_read_lock();
792 slave = rcu_dereference(bond->curr_active_slave);
793 rcu_read_unlock();
794
795 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
796 slave ? slave->dev->name : "NULL");
797
798 if (!slave || !bond->send_peer_notif ||
799 bond->send_peer_notif %
800 max(1, bond->params.peer_notif_delay) != 0 ||
801 !netif_carrier_ok(bond->dev) ||
802 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
803 return false;
804
805 return true;
806}
807
808/**
809 * change_active_interface - change the active slave into the specified one
810 * @bond: our bonding struct
811 * @new: the new slave to make the active one
812 *
813 * Set the new slave to the bond's settings and unset them on the old
814 * curr_active_slave.
815 * Setting include flags, mc-list, promiscuity, allmulti, etc.
816 *
817 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
818 * because it is apparently the best available slave we have, even though its
819 * updelay hasn't timed out yet.
820 *
821 * Caller must hold RTNL.
822 */
823void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
824{
825 struct slave *old_active;
826
827 ASSERT_RTNL();
828
829 old_active = rtnl_dereference(bond->curr_active_slave);
830
831 if (old_active == new_active)
832 return;
833
834 if (new_active) {
835 new_active->last_link_up = jiffies;
836
837 if (new_active->link == BOND_LINK_BACK) {
838 if (bond_uses_primary(bond)) {
839 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
840 (bond->params.updelay - new_active->delay) * bond->params.miimon);
841 }
842
843 new_active->delay = 0;
844 bond_set_slave_link_state(new_active, BOND_LINK_UP,
845 BOND_SLAVE_NOTIFY_NOW);
846
847 if (BOND_MODE(bond) == BOND_MODE_8023AD)
848 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
849
850 if (bond_is_lb(bond))
851 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
852 } else {
853 if (bond_uses_primary(bond)) {
854 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
855 }
856 }
857 }
858
859 if (bond_uses_primary(bond))
860 bond_hw_addr_swap(bond, new_active, old_active);
861
862 if (bond_is_lb(bond)) {
863 bond_alb_handle_active_change(bond, new_active);
864 if (old_active)
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
867 if (new_active)
868 bond_set_slave_active_flags(new_active,
869 BOND_SLAVE_NOTIFY_NOW);
870 } else {
871 rcu_assign_pointer(bond->curr_active_slave, new_active);
872 }
873
874 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
875 if (old_active)
876 bond_set_slave_inactive_flags(old_active,
877 BOND_SLAVE_NOTIFY_NOW);
878
879 if (new_active) {
880 bool should_notify_peers = false;
881
882 bond_set_slave_active_flags(new_active,
883 BOND_SLAVE_NOTIFY_NOW);
884
885 if (bond->params.fail_over_mac)
886 bond_do_fail_over_mac(bond, new_active,
887 old_active);
888
889 if (netif_running(bond->dev)) {
890 bond->send_peer_notif =
891 bond->params.num_peer_notif *
892 max(1, bond->params.peer_notif_delay);
893 should_notify_peers =
894 bond_should_notify_peers(bond);
895 }
896
897 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
898 if (should_notify_peers) {
899 bond->send_peer_notif--;
900 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
901 bond->dev);
902 }
903 }
904 }
905
906 /* resend IGMP joins since active slave has changed or
907 * all were sent on curr_active_slave.
908 * resend only if bond is brought up with the affected
909 * bonding modes and the retransmission is enabled
910 */
911 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
912 ((bond_uses_primary(bond) && new_active) ||
913 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
914 bond->igmp_retrans = bond->params.resend_igmp;
915 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
916 }
917}
918
919/**
920 * bond_select_active_slave - select a new active slave, if needed
921 * @bond: our bonding struct
922 *
923 * This functions should be called when one of the following occurs:
924 * - The old curr_active_slave has been released or lost its link.
925 * - The primary_slave has got its link back.
926 * - A slave has got its link back and there's no old curr_active_slave.
927 *
928 * Caller must hold RTNL.
929 */
930void bond_select_active_slave(struct bonding *bond)
931{
932 struct slave *best_slave;
933 int rv;
934
935 ASSERT_RTNL();
936
937 best_slave = bond_find_best_slave(bond);
938 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
939 bond_change_active_slave(bond, best_slave);
940 rv = bond_set_carrier(bond);
941 if (!rv)
942 return;
943
944 if (netif_carrier_ok(bond->dev))
945 netdev_info(bond->dev, "active interface up!\n");
946 else
947 netdev_info(bond->dev, "now running without any active interface!\n");
948 }
949}
950
951#ifdef CONFIG_NET_POLL_CONTROLLER
952static inline int slave_enable_netpoll(struct slave *slave)
953{
954 struct netpoll *np;
955 int err = 0;
956
957 np = kzalloc(sizeof(*np), GFP_KERNEL);
958 err = -ENOMEM;
959 if (!np)
960 goto out;
961
962 err = __netpoll_setup(np, slave->dev);
963 if (err) {
964 kfree(np);
965 goto out;
966 }
967 slave->np = np;
968out:
969 return err;
970}
971static inline void slave_disable_netpoll(struct slave *slave)
972{
973 struct netpoll *np = slave->np;
974
975 if (!np)
976 return;
977
978 slave->np = NULL;
979
980 __netpoll_free(np);
981}
982
983static void bond_poll_controller(struct net_device *bond_dev)
984{
985 struct bonding *bond = netdev_priv(bond_dev);
986 struct slave *slave = NULL;
987 struct list_head *iter;
988 struct ad_info ad_info;
989
990 if (BOND_MODE(bond) == BOND_MODE_8023AD)
991 if (bond_3ad_get_active_agg_info(bond, &ad_info))
992 return;
993
994 bond_for_each_slave_rcu(bond, slave, iter) {
995 if (!bond_slave_is_up(slave))
996 continue;
997
998 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
999 struct aggregator *agg =
1000 SLAVE_AD_INFO(slave)->port.aggregator;
1001
1002 if (agg &&
1003 agg->aggregator_identifier != ad_info.aggregator_id)
1004 continue;
1005 }
1006
1007 netpoll_poll_dev(slave->dev);
1008 }
1009}
1010
1011static void bond_netpoll_cleanup(struct net_device *bond_dev)
1012{
1013 struct bonding *bond = netdev_priv(bond_dev);
1014 struct list_head *iter;
1015 struct slave *slave;
1016
1017 bond_for_each_slave(bond, slave, iter)
1018 if (bond_slave_is_up(slave))
1019 slave_disable_netpoll(slave);
1020}
1021
1022static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1023{
1024 struct bonding *bond = netdev_priv(dev);
1025 struct list_head *iter;
1026 struct slave *slave;
1027 int err = 0;
1028
1029 bond_for_each_slave(bond, slave, iter) {
1030 err = slave_enable_netpoll(slave);
1031 if (err) {
1032 bond_netpoll_cleanup(dev);
1033 break;
1034 }
1035 }
1036 return err;
1037}
1038#else
1039static inline int slave_enable_netpoll(struct slave *slave)
1040{
1041 return 0;
1042}
1043static inline void slave_disable_netpoll(struct slave *slave)
1044{
1045}
1046static void bond_netpoll_cleanup(struct net_device *bond_dev)
1047{
1048}
1049#endif
1050
1051/*---------------------------------- IOCTL ----------------------------------*/
1052
1053static netdev_features_t bond_fix_features(struct net_device *dev,
1054 netdev_features_t features)
1055{
1056 struct bonding *bond = netdev_priv(dev);
1057 struct list_head *iter;
1058 netdev_features_t mask;
1059 struct slave *slave;
1060
1061 mask = features;
1062
1063 features &= ~NETIF_F_ONE_FOR_ALL;
1064 features |= NETIF_F_ALL_FOR_ALL;
1065
1066 bond_for_each_slave(bond, slave, iter) {
1067 features = netdev_increment_features(features,
1068 slave->dev->features,
1069 mask);
1070 }
1071 features = netdev_add_tso_features(features, mask);
1072
1073 return features;
1074}
1075
1076#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1077 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1078 NETIF_F_HIGHDMA | NETIF_F_LRO)
1079
1080#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1081 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1082
1083#define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1084 NETIF_F_ALL_TSO)
1085
1086static void bond_compute_features(struct bonding *bond)
1087{
1088 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1089 IFF_XMIT_DST_RELEASE_PERM;
1090 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1091 netdev_features_t enc_features = BOND_ENC_FEATURES;
1092 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1093 struct net_device *bond_dev = bond->dev;
1094 struct list_head *iter;
1095 struct slave *slave;
1096 unsigned short max_hard_header_len = ETH_HLEN;
1097 unsigned int gso_max_size = GSO_MAX_SIZE;
1098 u16 gso_max_segs = GSO_MAX_SEGS;
1099
1100 if (!bond_has_slaves(bond))
1101 goto done;
1102 vlan_features &= NETIF_F_ALL_FOR_ALL;
1103 mpls_features &= NETIF_F_ALL_FOR_ALL;
1104
1105 bond_for_each_slave(bond, slave, iter) {
1106 vlan_features = netdev_increment_features(vlan_features,
1107 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1108
1109 enc_features = netdev_increment_features(enc_features,
1110 slave->dev->hw_enc_features,
1111 BOND_ENC_FEATURES);
1112
1113 mpls_features = netdev_increment_features(mpls_features,
1114 slave->dev->mpls_features,
1115 BOND_MPLS_FEATURES);
1116
1117 dst_release_flag &= slave->dev->priv_flags;
1118 if (slave->dev->hard_header_len > max_hard_header_len)
1119 max_hard_header_len = slave->dev->hard_header_len;
1120
1121 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1122 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1123 }
1124 bond_dev->hard_header_len = max_hard_header_len;
1125
1126done:
1127 bond_dev->vlan_features = vlan_features;
1128 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1129 NETIF_F_HW_VLAN_CTAG_TX |
1130 NETIF_F_HW_VLAN_STAG_TX |
1131 NETIF_F_GSO_UDP_L4;
1132 bond_dev->mpls_features = mpls_features;
1133 bond_dev->gso_max_segs = gso_max_segs;
1134 netif_set_gso_max_size(bond_dev, gso_max_size);
1135
1136 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1137 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1138 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1139 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1140
1141 netdev_change_features(bond_dev);
1142}
1143
1144static void bond_setup_by_slave(struct net_device *bond_dev,
1145 struct net_device *slave_dev)
1146{
1147 bond_dev->header_ops = slave_dev->header_ops;
1148
1149 bond_dev->type = slave_dev->type;
1150 bond_dev->hard_header_len = slave_dev->hard_header_len;
1151 bond_dev->addr_len = slave_dev->addr_len;
1152
1153 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1154 slave_dev->addr_len);
1155}
1156
1157/* On bonding slaves other than the currently active slave, suppress
1158 * duplicates except for alb non-mcast/bcast.
1159 */
1160static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1161 struct slave *slave,
1162 struct bonding *bond)
1163{
1164 if (bond_is_slave_inactive(slave)) {
1165 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1166 skb->pkt_type != PACKET_BROADCAST &&
1167 skb->pkt_type != PACKET_MULTICAST)
1168 return false;
1169 return true;
1170 }
1171 return false;
1172}
1173
1174static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1175{
1176 struct sk_buff *skb = *pskb;
1177 struct slave *slave;
1178 struct bonding *bond;
1179 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1180 struct slave *);
1181 int ret = RX_HANDLER_ANOTHER;
1182
1183 skb = skb_share_check(skb, GFP_ATOMIC);
1184 if (unlikely(!skb))
1185 return RX_HANDLER_CONSUMED;
1186
1187 *pskb = skb;
1188
1189 slave = bond_slave_get_rcu(skb->dev);
1190 bond = slave->bond;
1191
1192 recv_probe = READ_ONCE(bond->recv_probe);
1193 if (recv_probe) {
1194 ret = recv_probe(skb, bond, slave);
1195 if (ret == RX_HANDLER_CONSUMED) {
1196 consume_skb(skb);
1197 return ret;
1198 }
1199 }
1200
1201 /*
1202 * For packets determined by bond_should_deliver_exact_match() call to
1203 * be suppressed we want to make an exception for link-local packets.
1204 * This is necessary for e.g. LLDP daemons to be able to monitor
1205 * inactive slave links without being forced to bind to them
1206 * explicitly.
1207 *
1208 * At the same time, packets that are passed to the bonding master
1209 * (including link-local ones) can have their originating interface
1210 * determined via PACKET_ORIGDEV socket option.
1211 */
1212 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1213 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1214 return RX_HANDLER_PASS;
1215 return RX_HANDLER_EXACT;
1216 }
1217
1218 skb->dev = bond->dev;
1219
1220 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1221 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1222 skb->pkt_type == PACKET_HOST) {
1223
1224 if (unlikely(skb_cow_head(skb,
1225 skb->data - skb_mac_header(skb)))) {
1226 kfree_skb(skb);
1227 return RX_HANDLER_CONSUMED;
1228 }
1229 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1230 bond->dev->addr_len);
1231 }
1232
1233 return ret;
1234}
1235
1236static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1237{
1238 switch (BOND_MODE(bond)) {
1239 case BOND_MODE_ROUNDROBIN:
1240 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1241 case BOND_MODE_ACTIVEBACKUP:
1242 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1243 case BOND_MODE_BROADCAST:
1244 return NETDEV_LAG_TX_TYPE_BROADCAST;
1245 case BOND_MODE_XOR:
1246 case BOND_MODE_8023AD:
1247 return NETDEV_LAG_TX_TYPE_HASH;
1248 default:
1249 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1250 }
1251}
1252
1253static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1254 enum netdev_lag_tx_type type)
1255{
1256 if (type != NETDEV_LAG_TX_TYPE_HASH)
1257 return NETDEV_LAG_HASH_NONE;
1258
1259 switch (bond->params.xmit_policy) {
1260 case BOND_XMIT_POLICY_LAYER2:
1261 return NETDEV_LAG_HASH_L2;
1262 case BOND_XMIT_POLICY_LAYER34:
1263 return NETDEV_LAG_HASH_L34;
1264 case BOND_XMIT_POLICY_LAYER23:
1265 return NETDEV_LAG_HASH_L23;
1266 case BOND_XMIT_POLICY_ENCAP23:
1267 return NETDEV_LAG_HASH_E23;
1268 case BOND_XMIT_POLICY_ENCAP34:
1269 return NETDEV_LAG_HASH_E34;
1270 default:
1271 return NETDEV_LAG_HASH_UNKNOWN;
1272 }
1273}
1274
1275static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1276 struct netlink_ext_ack *extack)
1277{
1278 struct netdev_lag_upper_info lag_upper_info;
1279 enum netdev_lag_tx_type type;
1280
1281 type = bond_lag_tx_type(bond);
1282 lag_upper_info.tx_type = type;
1283 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1284
1285 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1286 &lag_upper_info, extack);
1287}
1288
1289static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1290{
1291 netdev_upper_dev_unlink(slave->dev, bond->dev);
1292 slave->dev->flags &= ~IFF_SLAVE;
1293}
1294
1295static struct slave *bond_alloc_slave(struct bonding *bond)
1296{
1297 struct slave *slave = NULL;
1298
1299 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1300 if (!slave)
1301 return NULL;
1302
1303 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1304 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1305 GFP_KERNEL);
1306 if (!SLAVE_AD_INFO(slave)) {
1307 kfree(slave);
1308 return NULL;
1309 }
1310 }
1311 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1312
1313 return slave;
1314}
1315
1316static void bond_free_slave(struct slave *slave)
1317{
1318 struct bonding *bond = bond_get_bond_by_slave(slave);
1319
1320 cancel_delayed_work_sync(&slave->notify_work);
1321 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1322 kfree(SLAVE_AD_INFO(slave));
1323
1324 kfree(slave);
1325}
1326
1327static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1328{
1329 info->bond_mode = BOND_MODE(bond);
1330 info->miimon = bond->params.miimon;
1331 info->num_slaves = bond->slave_cnt;
1332}
1333
1334static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1335{
1336 strcpy(info->slave_name, slave->dev->name);
1337 info->link = slave->link;
1338 info->state = bond_slave_state(slave);
1339 info->link_failure_count = slave->link_failure_count;
1340}
1341
1342static void bond_netdev_notify_work(struct work_struct *_work)
1343{
1344 struct slave *slave = container_of(_work, struct slave,
1345 notify_work.work);
1346
1347 if (rtnl_trylock()) {
1348 struct netdev_bonding_info binfo;
1349
1350 bond_fill_ifslave(slave, &binfo.slave);
1351 bond_fill_ifbond(slave->bond, &binfo.master);
1352 netdev_bonding_info_change(slave->dev, &binfo);
1353 rtnl_unlock();
1354 } else {
1355 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1356 }
1357}
1358
1359void bond_queue_slave_event(struct slave *slave)
1360{
1361 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1362}
1363
1364void bond_lower_state_changed(struct slave *slave)
1365{
1366 struct netdev_lag_lower_state_info info;
1367
1368 info.link_up = slave->link == BOND_LINK_UP ||
1369 slave->link == BOND_LINK_FAIL;
1370 info.tx_enabled = bond_is_active_slave(slave);
1371 netdev_lower_state_changed(slave->dev, &info);
1372}
1373
1374/* enslave device <slave> to bond device <master> */
1375int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1376 struct netlink_ext_ack *extack)
1377{
1378 struct bonding *bond = netdev_priv(bond_dev);
1379 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1380 struct slave *new_slave = NULL, *prev_slave;
1381 struct sockaddr_storage ss;
1382 int link_reporting;
1383 int res = 0, i;
1384
1385 if (!bond->params.use_carrier &&
1386 slave_dev->ethtool_ops->get_link == NULL &&
1387 slave_ops->ndo_do_ioctl == NULL) {
1388 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1389 }
1390
1391 /* already in-use? */
1392 if (netdev_is_rx_handler_busy(slave_dev)) {
1393 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1394 slave_err(bond_dev, slave_dev,
1395 "Error: Device is in use and cannot be enslaved\n");
1396 return -EBUSY;
1397 }
1398
1399 if (bond_dev == slave_dev) {
1400 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1401 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1402 return -EPERM;
1403 }
1404
1405 /* vlan challenged mutual exclusion */
1406 /* no need to lock since we're protected by rtnl_lock */
1407 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1408 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1409 if (vlan_uses_dev(bond_dev)) {
1410 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1411 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1412 return -EPERM;
1413 } else {
1414 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1415 }
1416 } else {
1417 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1418 }
1419
1420 /* Old ifenslave binaries are no longer supported. These can
1421 * be identified with moderate accuracy by the state of the slave:
1422 * the current ifenslave will set the interface down prior to
1423 * enslaving it; the old ifenslave will not.
1424 */
1425 if (slave_dev->flags & IFF_UP) {
1426 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1427 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1428 return -EPERM;
1429 }
1430
1431 /* set bonding device ether type by slave - bonding netdevices are
1432 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1433 * there is a need to override some of the type dependent attribs/funcs.
1434 *
1435 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1436 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1437 */
1438 if (!bond_has_slaves(bond)) {
1439 if (bond_dev->type != slave_dev->type) {
1440 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1441 bond_dev->type, slave_dev->type);
1442
1443 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1444 bond_dev);
1445 res = notifier_to_errno(res);
1446 if (res) {
1447 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1448 return -EBUSY;
1449 }
1450
1451 /* Flush unicast and multicast addresses */
1452 dev_uc_flush(bond_dev);
1453 dev_mc_flush(bond_dev);
1454
1455 if (slave_dev->type != ARPHRD_ETHER)
1456 bond_setup_by_slave(bond_dev, slave_dev);
1457 else {
1458 ether_setup(bond_dev);
1459 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1460 }
1461
1462 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1463 bond_dev);
1464 }
1465 } else if (bond_dev->type != slave_dev->type) {
1466 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1467 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1468 slave_dev->type, bond_dev->type);
1469 return -EINVAL;
1470 }
1471
1472 if (slave_dev->type == ARPHRD_INFINIBAND &&
1473 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1474 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1475 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1476 slave_dev->type);
1477 res = -EOPNOTSUPP;
1478 goto err_undo_flags;
1479 }
1480
1481 if (!slave_ops->ndo_set_mac_address ||
1482 slave_dev->type == ARPHRD_INFINIBAND) {
1483 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1484 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1485 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1486 if (!bond_has_slaves(bond)) {
1487 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1488 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1489 } else {
1490 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1491 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1492 res = -EOPNOTSUPP;
1493 goto err_undo_flags;
1494 }
1495 }
1496 }
1497
1498 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1499
1500 /* If this is the first slave, then we need to set the master's hardware
1501 * address to be the same as the slave's.
1502 */
1503 if (!bond_has_slaves(bond) &&
1504 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1505 res = bond_set_dev_addr(bond->dev, slave_dev);
1506 if (res)
1507 goto err_undo_flags;
1508 }
1509
1510 new_slave = bond_alloc_slave(bond);
1511 if (!new_slave) {
1512 res = -ENOMEM;
1513 goto err_undo_flags;
1514 }
1515
1516 new_slave->bond = bond;
1517 new_slave->dev = slave_dev;
1518 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1519 * is set via sysfs or module option if desired.
1520 */
1521 new_slave->queue_id = 0;
1522
1523 /* Save slave's original mtu and then set it to match the bond */
1524 new_slave->original_mtu = slave_dev->mtu;
1525 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1526 if (res) {
1527 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1528 goto err_free;
1529 }
1530
1531 /* Save slave's original ("permanent") mac address for modes
1532 * that need it, and for restoring it upon release, and then
1533 * set it to the master's address
1534 */
1535 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1536 slave_dev->addr_len);
1537
1538 if (!bond->params.fail_over_mac ||
1539 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1540 /* Set slave to master's mac address. The application already
1541 * set the master's mac address to that of the first slave
1542 */
1543 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1544 ss.ss_family = slave_dev->type;
1545 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1546 extack);
1547 if (res) {
1548 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1549 goto err_restore_mtu;
1550 }
1551 }
1552
1553 /* set slave flag before open to prevent IPv6 addrconf */
1554 slave_dev->flags |= IFF_SLAVE;
1555
1556 /* open the slave since the application closed it */
1557 res = dev_open(slave_dev, extack);
1558 if (res) {
1559 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1560 goto err_restore_mac;
1561 }
1562
1563 slave_dev->priv_flags |= IFF_BONDING;
1564 /* initialize slave stats */
1565 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1566
1567 if (bond_is_lb(bond)) {
1568 /* bond_alb_init_slave() must be called before all other stages since
1569 * it might fail and we do not want to have to undo everything
1570 */
1571 res = bond_alb_init_slave(bond, new_slave);
1572 if (res)
1573 goto err_close;
1574 }
1575
1576 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1577 if (res) {
1578 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1579 goto err_close;
1580 }
1581
1582 prev_slave = bond_last_slave(bond);
1583
1584 new_slave->delay = 0;
1585 new_slave->link_failure_count = 0;
1586
1587 if (bond_update_speed_duplex(new_slave) &&
1588 bond_needs_speed_duplex(bond))
1589 new_slave->link = BOND_LINK_DOWN;
1590
1591 new_slave->last_rx = jiffies -
1592 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1593 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1594 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1595
1596 if (bond->params.miimon && !bond->params.use_carrier) {
1597 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1598
1599 if ((link_reporting == -1) && !bond->params.arp_interval) {
1600 /* miimon is set but a bonded network driver
1601 * does not support ETHTOOL/MII and
1602 * arp_interval is not set. Note: if
1603 * use_carrier is enabled, we will never go
1604 * here (because netif_carrier is always
1605 * supported); thus, we don't need to change
1606 * the messages for netif_carrier.
1607 */
1608 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");
1609 } else if (link_reporting == -1) {
1610 /* unable get link status using mii/ethtool */
1611 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");
1612 }
1613 }
1614
1615 /* check for initial state */
1616 new_slave->link = BOND_LINK_NOCHANGE;
1617 if (bond->params.miimon) {
1618 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1619 if (bond->params.updelay) {
1620 bond_set_slave_link_state(new_slave,
1621 BOND_LINK_BACK,
1622 BOND_SLAVE_NOTIFY_NOW);
1623 new_slave->delay = bond->params.updelay;
1624 } else {
1625 bond_set_slave_link_state(new_slave,
1626 BOND_LINK_UP,
1627 BOND_SLAVE_NOTIFY_NOW);
1628 }
1629 } else {
1630 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1631 BOND_SLAVE_NOTIFY_NOW);
1632 }
1633 } else if (bond->params.arp_interval) {
1634 bond_set_slave_link_state(new_slave,
1635 (netif_carrier_ok(slave_dev) ?
1636 BOND_LINK_UP : BOND_LINK_DOWN),
1637 BOND_SLAVE_NOTIFY_NOW);
1638 } else {
1639 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1640 BOND_SLAVE_NOTIFY_NOW);
1641 }
1642
1643 if (new_slave->link != BOND_LINK_DOWN)
1644 new_slave->last_link_up = jiffies;
1645 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1646 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1647 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1648
1649 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1650 /* if there is a primary slave, remember it */
1651 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1652 rcu_assign_pointer(bond->primary_slave, new_slave);
1653 bond->force_primary = true;
1654 }
1655 }
1656
1657 switch (BOND_MODE(bond)) {
1658 case BOND_MODE_ACTIVEBACKUP:
1659 bond_set_slave_inactive_flags(new_slave,
1660 BOND_SLAVE_NOTIFY_NOW);
1661 break;
1662 case BOND_MODE_8023AD:
1663 /* in 802.3ad mode, the internal mechanism
1664 * will activate the slaves in the selected
1665 * aggregator
1666 */
1667 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1668 /* if this is the first slave */
1669 if (!prev_slave) {
1670 SLAVE_AD_INFO(new_slave)->id = 1;
1671 /* Initialize AD with the number of times that the AD timer is called in 1 second
1672 * can be called only after the mac address of the bond is set
1673 */
1674 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1675 } else {
1676 SLAVE_AD_INFO(new_slave)->id =
1677 SLAVE_AD_INFO(prev_slave)->id + 1;
1678 }
1679
1680 bond_3ad_bind_slave(new_slave);
1681 break;
1682 case BOND_MODE_TLB:
1683 case BOND_MODE_ALB:
1684 bond_set_active_slave(new_slave);
1685 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1686 break;
1687 default:
1688 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1689
1690 /* always active in trunk mode */
1691 bond_set_active_slave(new_slave);
1692
1693 /* In trunking mode there is little meaning to curr_active_slave
1694 * anyway (it holds no special properties of the bond device),
1695 * so we can change it without calling change_active_interface()
1696 */
1697 if (!rcu_access_pointer(bond->curr_active_slave) &&
1698 new_slave->link == BOND_LINK_UP)
1699 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1700
1701 break;
1702 } /* switch(bond_mode) */
1703
1704#ifdef CONFIG_NET_POLL_CONTROLLER
1705 if (bond->dev->npinfo) {
1706 if (slave_enable_netpoll(new_slave)) {
1707 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1708 res = -EBUSY;
1709 goto err_detach;
1710 }
1711 }
1712#endif
1713
1714 if (!(bond_dev->features & NETIF_F_LRO))
1715 dev_disable_lro(slave_dev);
1716
1717 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1718 new_slave);
1719 if (res) {
1720 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1721 goto err_detach;
1722 }
1723
1724 res = bond_master_upper_dev_link(bond, new_slave, extack);
1725 if (res) {
1726 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1727 goto err_unregister;
1728 }
1729
1730 res = bond_sysfs_slave_add(new_slave);
1731 if (res) {
1732 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1733 goto err_upper_unlink;
1734 }
1735
1736 /* If the mode uses primary, then the following is handled by
1737 * bond_change_active_slave().
1738 */
1739 if (!bond_uses_primary(bond)) {
1740 /* set promiscuity level to new slave */
1741 if (bond_dev->flags & IFF_PROMISC) {
1742 res = dev_set_promiscuity(slave_dev, 1);
1743 if (res)
1744 goto err_sysfs_del;
1745 }
1746
1747 /* set allmulti level to new slave */
1748 if (bond_dev->flags & IFF_ALLMULTI) {
1749 res = dev_set_allmulti(slave_dev, 1);
1750 if (res) {
1751 if (bond_dev->flags & IFF_PROMISC)
1752 dev_set_promiscuity(slave_dev, -1);
1753 goto err_sysfs_del;
1754 }
1755 }
1756
1757 netif_addr_lock_bh(bond_dev);
1758 dev_mc_sync_multiple(slave_dev, bond_dev);
1759 dev_uc_sync_multiple(slave_dev, bond_dev);
1760 netif_addr_unlock_bh(bond_dev);
1761
1762 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1763 /* add lacpdu mc addr to mc list */
1764 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1765
1766 dev_mc_add(slave_dev, lacpdu_multicast);
1767 }
1768 }
1769
1770 bond->slave_cnt++;
1771 bond_compute_features(bond);
1772 bond_set_carrier(bond);
1773
1774 if (bond_uses_primary(bond)) {
1775 block_netpoll_tx();
1776 bond_select_active_slave(bond);
1777 unblock_netpoll_tx();
1778 }
1779
1780 if (bond_mode_can_use_xmit_hash(bond))
1781 bond_update_slave_arr(bond, NULL);
1782
1783
1784 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1785 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1786 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1787
1788 /* enslave is successful */
1789 bond_queue_slave_event(new_slave);
1790 return 0;
1791
1792/* Undo stages on error */
1793err_sysfs_del:
1794 bond_sysfs_slave_del(new_slave);
1795
1796err_upper_unlink:
1797 bond_upper_dev_unlink(bond, new_slave);
1798
1799err_unregister:
1800 netdev_rx_handler_unregister(slave_dev);
1801
1802err_detach:
1803 vlan_vids_del_by_dev(slave_dev, bond_dev);
1804 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1805 RCU_INIT_POINTER(bond->primary_slave, NULL);
1806 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1807 block_netpoll_tx();
1808 bond_change_active_slave(bond, NULL);
1809 bond_select_active_slave(bond);
1810 unblock_netpoll_tx();
1811 }
1812 /* either primary_slave or curr_active_slave might've changed */
1813 synchronize_rcu();
1814 slave_disable_netpoll(new_slave);
1815
1816err_close:
1817 if (!netif_is_bond_master(slave_dev))
1818 slave_dev->priv_flags &= ~IFF_BONDING;
1819 dev_close(slave_dev);
1820
1821err_restore_mac:
1822 slave_dev->flags &= ~IFF_SLAVE;
1823 if (!bond->params.fail_over_mac ||
1824 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1825 /* XXX TODO - fom follow mode needs to change master's
1826 * MAC if this slave's MAC is in use by the bond, or at
1827 * least print a warning.
1828 */
1829 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1830 new_slave->dev->addr_len);
1831 ss.ss_family = slave_dev->type;
1832 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1833 }
1834
1835err_restore_mtu:
1836 dev_set_mtu(slave_dev, new_slave->original_mtu);
1837
1838err_free:
1839 bond_free_slave(new_slave);
1840
1841err_undo_flags:
1842 /* Enslave of first slave has failed and we need to fix master's mac */
1843 if (!bond_has_slaves(bond)) {
1844 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1845 slave_dev->dev_addr))
1846 eth_hw_addr_random(bond_dev);
1847 if (bond_dev->type != ARPHRD_ETHER) {
1848 dev_close(bond_dev);
1849 ether_setup(bond_dev);
1850 bond_dev->flags |= IFF_MASTER;
1851 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1852 }
1853 }
1854
1855 return res;
1856}
1857
1858/* Try to release the slave device <slave> from the bond device <master>
1859 * It is legal to access curr_active_slave without a lock because all the function
1860 * is RTNL-locked. If "all" is true it means that the function is being called
1861 * while destroying a bond interface and all slaves are being released.
1862 *
1863 * The rules for slave state should be:
1864 * for Active/Backup:
1865 * Active stays on all backups go down
1866 * for Bonded connections:
1867 * The first up interface should be left on and all others downed.
1868 */
1869static int __bond_release_one(struct net_device *bond_dev,
1870 struct net_device *slave_dev,
1871 bool all, bool unregister)
1872{
1873 struct bonding *bond = netdev_priv(bond_dev);
1874 struct slave *slave, *oldcurrent;
1875 struct sockaddr_storage ss;
1876 int old_flags = bond_dev->flags;
1877 netdev_features_t old_features = bond_dev->features;
1878
1879 /* slave is not a slave or master is not master of this slave */
1880 if (!(slave_dev->flags & IFF_SLAVE) ||
1881 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1882 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
1883 return -EINVAL;
1884 }
1885
1886 block_netpoll_tx();
1887
1888 slave = bond_get_slave_by_dev(bond, slave_dev);
1889 if (!slave) {
1890 /* not a slave of this bond */
1891 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
1892 unblock_netpoll_tx();
1893 return -EINVAL;
1894 }
1895
1896 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1897
1898 bond_sysfs_slave_del(slave);
1899
1900 /* recompute stats just before removing the slave */
1901 bond_get_stats(bond->dev, &bond->bond_stats);
1902
1903 bond_upper_dev_unlink(bond, slave);
1904 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1905 * for this slave anymore.
1906 */
1907 netdev_rx_handler_unregister(slave_dev);
1908
1909 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1910 bond_3ad_unbind_slave(slave);
1911
1912 if (bond_mode_can_use_xmit_hash(bond))
1913 bond_update_slave_arr(bond, slave);
1914
1915 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
1916 bond_is_active_slave(slave) ? "active" : "backup");
1917
1918 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1919
1920 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1921
1922 if (!all && (!bond->params.fail_over_mac ||
1923 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1924 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1925 bond_has_slaves(bond))
1926 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",
1927 slave->perm_hwaddr);
1928 }
1929
1930 if (rtnl_dereference(bond->primary_slave) == slave)
1931 RCU_INIT_POINTER(bond->primary_slave, NULL);
1932
1933 if (oldcurrent == slave)
1934 bond_change_active_slave(bond, NULL);
1935
1936 if (bond_is_lb(bond)) {
1937 /* Must be called only after the slave has been
1938 * detached from the list and the curr_active_slave
1939 * has been cleared (if our_slave == old_current),
1940 * but before a new active slave is selected.
1941 */
1942 bond_alb_deinit_slave(bond, slave);
1943 }
1944
1945 if (all) {
1946 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1947 } else if (oldcurrent == slave) {
1948 /* Note that we hold RTNL over this sequence, so there
1949 * is no concern that another slave add/remove event
1950 * will interfere.
1951 */
1952 bond_select_active_slave(bond);
1953 }
1954
1955 if (!bond_has_slaves(bond)) {
1956 bond_set_carrier(bond);
1957 eth_hw_addr_random(bond_dev);
1958 }
1959
1960 unblock_netpoll_tx();
1961 synchronize_rcu();
1962 bond->slave_cnt--;
1963
1964 if (!bond_has_slaves(bond)) {
1965 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1966 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1967 }
1968
1969 bond_compute_features(bond);
1970 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1971 (old_features & NETIF_F_VLAN_CHALLENGED))
1972 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
1973
1974 vlan_vids_del_by_dev(slave_dev, bond_dev);
1975
1976 /* If the mode uses primary, then this case was handled above by
1977 * bond_change_active_slave(..., NULL)
1978 */
1979 if (!bond_uses_primary(bond)) {
1980 /* unset promiscuity level from slave
1981 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1982 * of the IFF_PROMISC flag in the bond_dev, but we need the
1983 * value of that flag before that change, as that was the value
1984 * when this slave was attached, so we cache at the start of the
1985 * function and use it here. Same goes for ALLMULTI below
1986 */
1987 if (old_flags & IFF_PROMISC)
1988 dev_set_promiscuity(slave_dev, -1);
1989
1990 /* unset allmulti level from slave */
1991 if (old_flags & IFF_ALLMULTI)
1992 dev_set_allmulti(slave_dev, -1);
1993
1994 bond_hw_addr_flush(bond_dev, slave_dev);
1995 }
1996
1997 slave_disable_netpoll(slave);
1998
1999 /* close slave before restoring its mac address */
2000 dev_close(slave_dev);
2001
2002 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2003 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2004 /* restore original ("permanent") mac address */
2005 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2006 slave->dev->addr_len);
2007 ss.ss_family = slave_dev->type;
2008 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2009 }
2010
2011 if (unregister)
2012 __dev_set_mtu(slave_dev, slave->original_mtu);
2013 else
2014 dev_set_mtu(slave_dev, slave->original_mtu);
2015
2016 if (!netif_is_bond_master(slave_dev))
2017 slave_dev->priv_flags &= ~IFF_BONDING;
2018
2019 bond_free_slave(slave);
2020
2021 return 0;
2022}
2023
2024/* A wrapper used because of ndo_del_link */
2025int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2026{
2027 return __bond_release_one(bond_dev, slave_dev, false, false);
2028}
2029
2030/* First release a slave and then destroy the bond if no more slaves are left.
2031 * Must be under rtnl_lock when this function is called.
2032 */
2033static int bond_release_and_destroy(struct net_device *bond_dev,
2034 struct net_device *slave_dev)
2035{
2036 struct bonding *bond = netdev_priv(bond_dev);
2037 int ret;
2038
2039 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2040 if (ret == 0 && !bond_has_slaves(bond)) {
2041 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2042 netdev_info(bond_dev, "Destroying bond\n");
2043 bond_remove_proc_entry(bond);
2044 unregister_netdevice(bond_dev);
2045 }
2046 return ret;
2047}
2048
2049static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2050{
2051 struct bonding *bond = netdev_priv(bond_dev);
2052 bond_fill_ifbond(bond, info);
2053}
2054
2055static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2056{
2057 struct bonding *bond = netdev_priv(bond_dev);
2058 struct list_head *iter;
2059 int i = 0, res = -ENODEV;
2060 struct slave *slave;
2061
2062 bond_for_each_slave(bond, slave, iter) {
2063 if (i++ == (int)info->slave_id) {
2064 res = 0;
2065 bond_fill_ifslave(slave, info);
2066 break;
2067 }
2068 }
2069
2070 return res;
2071}
2072
2073/*-------------------------------- Monitoring -------------------------------*/
2074
2075/* called with rcu_read_lock() */
2076static int bond_miimon_inspect(struct bonding *bond)
2077{
2078 int link_state, commit = 0;
2079 struct list_head *iter;
2080 struct slave *slave;
2081 bool ignore_updelay;
2082
2083 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2084
2085 bond_for_each_slave_rcu(bond, slave, iter) {
2086 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2087
2088 link_state = bond_check_dev_link(bond, slave->dev, 0);
2089
2090 switch (slave->link) {
2091 case BOND_LINK_UP:
2092 if (link_state)
2093 continue;
2094
2095 bond_propose_link_state(slave, BOND_LINK_FAIL);
2096 commit++;
2097 slave->delay = bond->params.downdelay;
2098 if (slave->delay) {
2099 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2100 (BOND_MODE(bond) ==
2101 BOND_MODE_ACTIVEBACKUP) ?
2102 (bond_is_active_slave(slave) ?
2103 "active " : "backup ") : "",
2104 bond->params.downdelay * bond->params.miimon);
2105 }
2106 /*FALLTHRU*/
2107 case BOND_LINK_FAIL:
2108 if (link_state) {
2109 /* recovered before downdelay expired */
2110 bond_propose_link_state(slave, BOND_LINK_UP);
2111 slave->last_link_up = jiffies;
2112 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2113 (bond->params.downdelay - slave->delay) *
2114 bond->params.miimon);
2115 commit++;
2116 continue;
2117 }
2118
2119 if (slave->delay <= 0) {
2120 bond_propose_link_state(slave, BOND_LINK_DOWN);
2121 commit++;
2122 continue;
2123 }
2124
2125 slave->delay--;
2126 break;
2127
2128 case BOND_LINK_DOWN:
2129 if (!link_state)
2130 continue;
2131
2132 bond_propose_link_state(slave, BOND_LINK_BACK);
2133 commit++;
2134 slave->delay = bond->params.updelay;
2135
2136 if (slave->delay) {
2137 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2138 ignore_updelay ? 0 :
2139 bond->params.updelay *
2140 bond->params.miimon);
2141 }
2142 /*FALLTHRU*/
2143 case BOND_LINK_BACK:
2144 if (!link_state) {
2145 bond_propose_link_state(slave, BOND_LINK_DOWN);
2146 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2147 (bond->params.updelay - slave->delay) *
2148 bond->params.miimon);
2149 commit++;
2150 continue;
2151 }
2152
2153 if (ignore_updelay)
2154 slave->delay = 0;
2155
2156 if (slave->delay <= 0) {
2157 bond_propose_link_state(slave, BOND_LINK_UP);
2158 commit++;
2159 ignore_updelay = false;
2160 continue;
2161 }
2162
2163 slave->delay--;
2164 break;
2165 }
2166 }
2167
2168 return commit;
2169}
2170
2171static void bond_miimon_link_change(struct bonding *bond,
2172 struct slave *slave,
2173 char link)
2174{
2175 switch (BOND_MODE(bond)) {
2176 case BOND_MODE_8023AD:
2177 bond_3ad_handle_link_change(slave, link);
2178 break;
2179 case BOND_MODE_TLB:
2180 case BOND_MODE_ALB:
2181 bond_alb_handle_link_change(bond, slave, link);
2182 break;
2183 case BOND_MODE_XOR:
2184 bond_update_slave_arr(bond, NULL);
2185 break;
2186 }
2187}
2188
2189static void bond_miimon_commit(struct bonding *bond)
2190{
2191 struct list_head *iter;
2192 struct slave *slave, *primary;
2193
2194 bond_for_each_slave(bond, slave, iter) {
2195 switch (slave->link_new_state) {
2196 case BOND_LINK_NOCHANGE:
2197 /* For 802.3ad mode, check current slave speed and
2198 * duplex again in case its port was disabled after
2199 * invalid speed/duplex reporting but recovered before
2200 * link monitoring could make a decision on the actual
2201 * link status
2202 */
2203 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2204 slave->link == BOND_LINK_UP)
2205 bond_3ad_adapter_speed_duplex_changed(slave);
2206 continue;
2207
2208 case BOND_LINK_UP:
2209 if (bond_update_speed_duplex(slave) &&
2210 bond_needs_speed_duplex(bond)) {
2211 slave->link = BOND_LINK_DOWN;
2212 if (net_ratelimit())
2213 slave_warn(bond->dev, slave->dev,
2214 "failed to get link speed/duplex\n");
2215 continue;
2216 }
2217 bond_set_slave_link_state(slave, BOND_LINK_UP,
2218 BOND_SLAVE_NOTIFY_NOW);
2219 slave->last_link_up = jiffies;
2220
2221 primary = rtnl_dereference(bond->primary_slave);
2222 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2223 /* prevent it from being the active one */
2224 bond_set_backup_slave(slave);
2225 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2226 /* make it immediately active */
2227 bond_set_active_slave(slave);
2228 } else if (slave != primary) {
2229 /* prevent it from being the active one */
2230 bond_set_backup_slave(slave);
2231 }
2232
2233 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2234 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2235 slave->duplex ? "full" : "half");
2236
2237 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2238
2239 if (!bond->curr_active_slave || slave == primary)
2240 goto do_failover;
2241
2242 continue;
2243
2244 case BOND_LINK_DOWN:
2245 if (slave->link_failure_count < UINT_MAX)
2246 slave->link_failure_count++;
2247
2248 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2249 BOND_SLAVE_NOTIFY_NOW);
2250
2251 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2252 BOND_MODE(bond) == BOND_MODE_8023AD)
2253 bond_set_slave_inactive_flags(slave,
2254 BOND_SLAVE_NOTIFY_NOW);
2255
2256 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2257
2258 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2259
2260 if (slave == rcu_access_pointer(bond->curr_active_slave))
2261 goto do_failover;
2262
2263 continue;
2264
2265 default:
2266 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2267 slave->link_new_state);
2268 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2269
2270 continue;
2271 }
2272
2273do_failover:
2274 block_netpoll_tx();
2275 bond_select_active_slave(bond);
2276 unblock_netpoll_tx();
2277 }
2278
2279 bond_set_carrier(bond);
2280}
2281
2282/* bond_mii_monitor
2283 *
2284 * Really a wrapper that splits the mii monitor into two phases: an
2285 * inspection, then (if inspection indicates something needs to be done)
2286 * an acquisition of appropriate locks followed by a commit phase to
2287 * implement whatever link state changes are indicated.
2288 */
2289static void bond_mii_monitor(struct work_struct *work)
2290{
2291 struct bonding *bond = container_of(work, struct bonding,
2292 mii_work.work);
2293 bool should_notify_peers = false;
2294 bool commit;
2295 unsigned long delay;
2296 struct slave *slave;
2297 struct list_head *iter;
2298
2299 delay = msecs_to_jiffies(bond->params.miimon);
2300
2301 if (!bond_has_slaves(bond))
2302 goto re_arm;
2303
2304 rcu_read_lock();
2305 should_notify_peers = bond_should_notify_peers(bond);
2306 commit = !!bond_miimon_inspect(bond);
2307 if (bond->send_peer_notif) {
2308 rcu_read_unlock();
2309 if (rtnl_trylock()) {
2310 bond->send_peer_notif--;
2311 rtnl_unlock();
2312 }
2313 } else {
2314 rcu_read_unlock();
2315 }
2316
2317 if (commit) {
2318 /* Race avoidance with bond_close cancel of workqueue */
2319 if (!rtnl_trylock()) {
2320 delay = 1;
2321 should_notify_peers = false;
2322 goto re_arm;
2323 }
2324
2325 bond_for_each_slave(bond, slave, iter) {
2326 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2327 }
2328 bond_miimon_commit(bond);
2329
2330 rtnl_unlock(); /* might sleep, hold no other locks */
2331 }
2332
2333re_arm:
2334 if (bond->params.miimon)
2335 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2336
2337 if (should_notify_peers) {
2338 if (!rtnl_trylock())
2339 return;
2340 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2341 rtnl_unlock();
2342 }
2343}
2344
2345static int bond_upper_dev_walk(struct net_device *upper, void *data)
2346{
2347 __be32 ip = *((__be32 *)data);
2348
2349 return ip == bond_confirm_addr(upper, 0, ip);
2350}
2351
2352static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2353{
2354 bool ret = false;
2355
2356 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2357 return true;
2358
2359 rcu_read_lock();
2360 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2361 ret = true;
2362 rcu_read_unlock();
2363
2364 return ret;
2365}
2366
2367/* We go to the (large) trouble of VLAN tagging ARP frames because
2368 * switches in VLAN mode (especially if ports are configured as
2369 * "native" to a VLAN) might not pass non-tagged frames.
2370 */
2371static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2372 __be32 src_ip, struct bond_vlan_tag *tags)
2373{
2374 struct sk_buff *skb;
2375 struct bond_vlan_tag *outer_tag = tags;
2376 struct net_device *slave_dev = slave->dev;
2377 struct net_device *bond_dev = slave->bond->dev;
2378
2379 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2380 arp_op, &dest_ip, &src_ip);
2381
2382 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2383 NULL, slave_dev->dev_addr, NULL);
2384
2385 if (!skb) {
2386 net_err_ratelimited("ARP packet allocation failed\n");
2387 return;
2388 }
2389
2390 if (!tags || tags->vlan_proto == VLAN_N_VID)
2391 goto xmit;
2392
2393 tags++;
2394
2395 /* Go through all the tags backwards and add them to the packet */
2396 while (tags->vlan_proto != VLAN_N_VID) {
2397 if (!tags->vlan_id) {
2398 tags++;
2399 continue;
2400 }
2401
2402 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2403 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2404 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2405 tags->vlan_id);
2406 if (!skb) {
2407 net_err_ratelimited("failed to insert inner VLAN tag\n");
2408 return;
2409 }
2410
2411 tags++;
2412 }
2413 /* Set the outer tag */
2414 if (outer_tag->vlan_id) {
2415 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2416 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2417 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2418 outer_tag->vlan_id);
2419 }
2420
2421xmit:
2422 arp_xmit(skb);
2423}
2424
2425/* Validate the device path between the @start_dev and the @end_dev.
2426 * The path is valid if the @end_dev is reachable through device
2427 * stacking.
2428 * When the path is validated, collect any vlan information in the
2429 * path.
2430 */
2431struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2432 struct net_device *end_dev,
2433 int level)
2434{
2435 struct bond_vlan_tag *tags;
2436 struct net_device *upper;
2437 struct list_head *iter;
2438
2439 if (start_dev == end_dev) {
2440 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2441 if (!tags)
2442 return ERR_PTR(-ENOMEM);
2443 tags[level].vlan_proto = VLAN_N_VID;
2444 return tags;
2445 }
2446
2447 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2448 tags = bond_verify_device_path(upper, end_dev, level + 1);
2449 if (IS_ERR_OR_NULL(tags)) {
2450 if (IS_ERR(tags))
2451 return tags;
2452 continue;
2453 }
2454 if (is_vlan_dev(upper)) {
2455 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2456 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2457 }
2458
2459 return tags;
2460 }
2461
2462 return NULL;
2463}
2464
2465static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2466{
2467 struct rtable *rt;
2468 struct bond_vlan_tag *tags;
2469 __be32 *targets = bond->params.arp_targets, addr;
2470 int i;
2471
2472 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2473 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2474 __func__, &targets[i]);
2475 tags = NULL;
2476
2477 /* Find out through which dev should the packet go */
2478 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2479 RTO_ONLINK, 0);
2480 if (IS_ERR(rt)) {
2481 /* there's no route to target - try to send arp
2482 * probe to generate any traffic (arp_validate=0)
2483 */
2484 if (bond->params.arp_validate)
2485 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2486 bond->dev->name,
2487 &targets[i]);
2488 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2489 0, tags);
2490 continue;
2491 }
2492
2493 /* bond device itself */
2494 if (rt->dst.dev == bond->dev)
2495 goto found;
2496
2497 rcu_read_lock();
2498 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2499 rcu_read_unlock();
2500
2501 if (!IS_ERR_OR_NULL(tags))
2502 goto found;
2503
2504 /* Not our device - skip */
2505 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2506 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2507
2508 ip_rt_put(rt);
2509 continue;
2510
2511found:
2512 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2513 ip_rt_put(rt);
2514 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2515 kfree(tags);
2516 }
2517}
2518
2519static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2520{
2521 int i;
2522
2523 if (!sip || !bond_has_this_ip(bond, tip)) {
2524 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2525 __func__, &sip, &tip);
2526 return;
2527 }
2528
2529 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2530 if (i == -1) {
2531 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2532 __func__, &sip);
2533 return;
2534 }
2535 slave->last_rx = jiffies;
2536 slave->target_last_arp_rx[i] = jiffies;
2537}
2538
2539int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2540 struct slave *slave)
2541{
2542 struct arphdr *arp = (struct arphdr *)skb->data;
2543 struct slave *curr_active_slave, *curr_arp_slave;
2544 unsigned char *arp_ptr;
2545 __be32 sip, tip;
2546 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2547 unsigned int alen;
2548
2549 if (!slave_do_arp_validate(bond, slave)) {
2550 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2551 !slave_do_arp_validate_only(bond))
2552 slave->last_rx = jiffies;
2553 return RX_HANDLER_ANOTHER;
2554 } else if (!is_arp) {
2555 return RX_HANDLER_ANOTHER;
2556 }
2557
2558 alen = arp_hdr_len(bond->dev);
2559
2560 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2561 __func__, skb->dev->name);
2562
2563 if (alen > skb_headlen(skb)) {
2564 arp = kmalloc(alen, GFP_ATOMIC);
2565 if (!arp)
2566 goto out_unlock;
2567 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2568 goto out_unlock;
2569 }
2570
2571 if (arp->ar_hln != bond->dev->addr_len ||
2572 skb->pkt_type == PACKET_OTHERHOST ||
2573 skb->pkt_type == PACKET_LOOPBACK ||
2574 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2575 arp->ar_pro != htons(ETH_P_IP) ||
2576 arp->ar_pln != 4)
2577 goto out_unlock;
2578
2579 arp_ptr = (unsigned char *)(arp + 1);
2580 arp_ptr += bond->dev->addr_len;
2581 memcpy(&sip, arp_ptr, 4);
2582 arp_ptr += 4 + bond->dev->addr_len;
2583 memcpy(&tip, arp_ptr, 4);
2584
2585 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2586 __func__, slave->dev->name, bond_slave_state(slave),
2587 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2588 &sip, &tip);
2589
2590 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2591 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2592
2593 /* We 'trust' the received ARP enough to validate it if:
2594 *
2595 * (a) the slave receiving the ARP is active (which includes the
2596 * current ARP slave, if any), or
2597 *
2598 * (b) the receiving slave isn't active, but there is a currently
2599 * active slave and it received valid arp reply(s) after it became
2600 * the currently active slave, or
2601 *
2602 * (c) there is an ARP slave that sent an ARP during the prior ARP
2603 * interval, and we receive an ARP reply on any slave. We accept
2604 * these because switch FDB update delays may deliver the ARP
2605 * reply to a slave other than the sender of the ARP request.
2606 *
2607 * Note: for (b), backup slaves are receiving the broadcast ARP
2608 * request, not a reply. This request passes from the sending
2609 * slave through the L2 switch(es) to the receiving slave. Since
2610 * this is checking the request, sip/tip are swapped for
2611 * validation.
2612 *
2613 * This is done to avoid endless looping when we can't reach the
2614 * arp_ip_target and fool ourselves with our own arp requests.
2615 */
2616 if (bond_is_active_slave(slave))
2617 bond_validate_arp(bond, slave, sip, tip);
2618 else if (curr_active_slave &&
2619 time_after(slave_last_rx(bond, curr_active_slave),
2620 curr_active_slave->last_link_up))
2621 bond_validate_arp(bond, slave, tip, sip);
2622 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2623 bond_time_in_interval(bond,
2624 dev_trans_start(curr_arp_slave->dev), 1))
2625 bond_validate_arp(bond, slave, sip, tip);
2626
2627out_unlock:
2628 if (arp != (struct arphdr *)skb->data)
2629 kfree(arp);
2630 return RX_HANDLER_ANOTHER;
2631}
2632
2633/* function to verify if we're in the arp_interval timeslice, returns true if
2634 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2635 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2636 */
2637static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2638 int mod)
2639{
2640 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2641
2642 return time_in_range(jiffies,
2643 last_act - delta_in_ticks,
2644 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2645}
2646
2647/* This function is called regularly to monitor each slave's link
2648 * ensuring that traffic is being sent and received when arp monitoring
2649 * is used in load-balancing mode. if the adapter has been dormant, then an
2650 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2651 * arp monitoring in active backup mode.
2652 */
2653static void bond_loadbalance_arp_mon(struct bonding *bond)
2654{
2655 struct slave *slave, *oldcurrent;
2656 struct list_head *iter;
2657 int do_failover = 0, slave_state_changed = 0;
2658
2659 if (!bond_has_slaves(bond))
2660 goto re_arm;
2661
2662 rcu_read_lock();
2663
2664 oldcurrent = rcu_dereference(bond->curr_active_slave);
2665 /* see if any of the previous devices are up now (i.e. they have
2666 * xmt and rcv traffic). the curr_active_slave does not come into
2667 * the picture unless it is null. also, slave->last_link_up is not
2668 * needed here because we send an arp on each slave and give a slave
2669 * as long as it needs to get the tx/rx within the delta.
2670 * TODO: what about up/down delay in arp mode? it wasn't here before
2671 * so it can wait
2672 */
2673 bond_for_each_slave_rcu(bond, slave, iter) {
2674 unsigned long trans_start = dev_trans_start(slave->dev);
2675
2676 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2677
2678 if (slave->link != BOND_LINK_UP) {
2679 if (bond_time_in_interval(bond, trans_start, 1) &&
2680 bond_time_in_interval(bond, slave->last_rx, 1)) {
2681
2682 bond_propose_link_state(slave, BOND_LINK_UP);
2683 slave_state_changed = 1;
2684
2685 /* primary_slave has no meaning in round-robin
2686 * mode. the window of a slave being up and
2687 * curr_active_slave being null after enslaving
2688 * is closed.
2689 */
2690 if (!oldcurrent) {
2691 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2692 do_failover = 1;
2693 } else {
2694 slave_info(bond->dev, slave->dev, "interface is now up\n");
2695 }
2696 }
2697 } else {
2698 /* slave->link == BOND_LINK_UP */
2699
2700 /* not all switches will respond to an arp request
2701 * when the source ip is 0, so don't take the link down
2702 * if we don't know our ip yet
2703 */
2704 if (!bond_time_in_interval(bond, trans_start, 2) ||
2705 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2706
2707 bond_propose_link_state(slave, BOND_LINK_DOWN);
2708 slave_state_changed = 1;
2709
2710 if (slave->link_failure_count < UINT_MAX)
2711 slave->link_failure_count++;
2712
2713 slave_info(bond->dev, slave->dev, "interface is now down\n");
2714
2715 if (slave == oldcurrent)
2716 do_failover = 1;
2717 }
2718 }
2719
2720 /* note: if switch is in round-robin mode, all links
2721 * must tx arp to ensure all links rx an arp - otherwise
2722 * links may oscillate or not come up at all; if switch is
2723 * in something like xor mode, there is nothing we can
2724 * do - all replies will be rx'ed on same link causing slaves
2725 * to be unstable during low/no traffic periods
2726 */
2727 if (bond_slave_is_up(slave))
2728 bond_arp_send_all(bond, slave);
2729 }
2730
2731 rcu_read_unlock();
2732
2733 if (do_failover || slave_state_changed) {
2734 if (!rtnl_trylock())
2735 goto re_arm;
2736
2737 bond_for_each_slave(bond, slave, iter) {
2738 if (slave->link_new_state != BOND_LINK_NOCHANGE)
2739 slave->link = slave->link_new_state;
2740 }
2741
2742 if (slave_state_changed) {
2743 bond_slave_state_change(bond);
2744 if (BOND_MODE(bond) == BOND_MODE_XOR)
2745 bond_update_slave_arr(bond, NULL);
2746 }
2747 if (do_failover) {
2748 block_netpoll_tx();
2749 bond_select_active_slave(bond);
2750 unblock_netpoll_tx();
2751 }
2752 rtnl_unlock();
2753 }
2754
2755re_arm:
2756 if (bond->params.arp_interval)
2757 queue_delayed_work(bond->wq, &bond->arp_work,
2758 msecs_to_jiffies(bond->params.arp_interval));
2759}
2760
2761/* Called to inspect slaves for active-backup mode ARP monitor link state
2762 * changes. Sets proposed link state in slaves to specify what action
2763 * should take place for the slave. Returns 0 if no changes are found, >0
2764 * if changes to link states must be committed.
2765 *
2766 * Called with rcu_read_lock held.
2767 */
2768static int bond_ab_arp_inspect(struct bonding *bond)
2769{
2770 unsigned long trans_start, last_rx;
2771 struct list_head *iter;
2772 struct slave *slave;
2773 int commit = 0;
2774
2775 bond_for_each_slave_rcu(bond, slave, iter) {
2776 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2777 last_rx = slave_last_rx(bond, slave);
2778
2779 if (slave->link != BOND_LINK_UP) {
2780 if (bond_time_in_interval(bond, last_rx, 1)) {
2781 bond_propose_link_state(slave, BOND_LINK_UP);
2782 commit++;
2783 }
2784 continue;
2785 }
2786
2787 /* Give slaves 2*delta after being enslaved or made
2788 * active. This avoids bouncing, as the last receive
2789 * times need a full ARP monitor cycle to be updated.
2790 */
2791 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2792 continue;
2793
2794 /* Backup slave is down if:
2795 * - No current_arp_slave AND
2796 * - more than 3*delta since last receive AND
2797 * - the bond has an IP address
2798 *
2799 * Note: a non-null current_arp_slave indicates
2800 * the curr_active_slave went down and we are
2801 * searching for a new one; under this condition
2802 * we only take the curr_active_slave down - this
2803 * gives each slave a chance to tx/rx traffic
2804 * before being taken out
2805 */
2806 if (!bond_is_active_slave(slave) &&
2807 !rcu_access_pointer(bond->current_arp_slave) &&
2808 !bond_time_in_interval(bond, last_rx, 3)) {
2809 bond_propose_link_state(slave, BOND_LINK_DOWN);
2810 commit++;
2811 }
2812
2813 /* Active slave is down if:
2814 * - more than 2*delta since transmitting OR
2815 * - (more than 2*delta since receive AND
2816 * the bond has an IP address)
2817 */
2818 trans_start = dev_trans_start(slave->dev);
2819 if (bond_is_active_slave(slave) &&
2820 (!bond_time_in_interval(bond, trans_start, 2) ||
2821 !bond_time_in_interval(bond, last_rx, 2))) {
2822 bond_propose_link_state(slave, BOND_LINK_DOWN);
2823 commit++;
2824 }
2825 }
2826
2827 return commit;
2828}
2829
2830/* Called to commit link state changes noted by inspection step of
2831 * active-backup mode ARP monitor.
2832 *
2833 * Called with RTNL hold.
2834 */
2835static void bond_ab_arp_commit(struct bonding *bond)
2836{
2837 unsigned long trans_start;
2838 struct list_head *iter;
2839 struct slave *slave;
2840
2841 bond_for_each_slave(bond, slave, iter) {
2842 switch (slave->link_new_state) {
2843 case BOND_LINK_NOCHANGE:
2844 continue;
2845
2846 case BOND_LINK_UP:
2847 trans_start = dev_trans_start(slave->dev);
2848 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2849 (!rtnl_dereference(bond->curr_active_slave) &&
2850 bond_time_in_interval(bond, trans_start, 1))) {
2851 struct slave *current_arp_slave;
2852
2853 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2854 bond_set_slave_link_state(slave, BOND_LINK_UP,
2855 BOND_SLAVE_NOTIFY_NOW);
2856 if (current_arp_slave) {
2857 bond_set_slave_inactive_flags(
2858 current_arp_slave,
2859 BOND_SLAVE_NOTIFY_NOW);
2860 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2861 }
2862
2863 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2864
2865 if (!rtnl_dereference(bond->curr_active_slave) ||
2866 slave == rtnl_dereference(bond->primary_slave))
2867 goto do_failover;
2868
2869 }
2870
2871 continue;
2872
2873 case BOND_LINK_DOWN:
2874 if (slave->link_failure_count < UINT_MAX)
2875 slave->link_failure_count++;
2876
2877 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2878 BOND_SLAVE_NOTIFY_NOW);
2879 bond_set_slave_inactive_flags(slave,
2880 BOND_SLAVE_NOTIFY_NOW);
2881
2882 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2883
2884 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2885 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2886 goto do_failover;
2887 }
2888
2889 continue;
2890
2891 default:
2892 slave_err(bond->dev, slave->dev,
2893 "impossible: link_new_state %d on slave\n",
2894 slave->link_new_state);
2895 continue;
2896 }
2897
2898do_failover:
2899 block_netpoll_tx();
2900 bond_select_active_slave(bond);
2901 unblock_netpoll_tx();
2902 }
2903
2904 bond_set_carrier(bond);
2905}
2906
2907/* Send ARP probes for active-backup mode ARP monitor.
2908 *
2909 * Called with rcu_read_lock held.
2910 */
2911static bool bond_ab_arp_probe(struct bonding *bond)
2912{
2913 struct slave *slave, *before = NULL, *new_slave = NULL,
2914 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2915 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2916 struct list_head *iter;
2917 bool found = false;
2918 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2919
2920 if (curr_arp_slave && curr_active_slave)
2921 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2922 curr_arp_slave->dev->name,
2923 curr_active_slave->dev->name);
2924
2925 if (curr_active_slave) {
2926 bond_arp_send_all(bond, curr_active_slave);
2927 return should_notify_rtnl;
2928 }
2929
2930 /* if we don't have a curr_active_slave, search for the next available
2931 * backup slave from the current_arp_slave and make it the candidate
2932 * for becoming the curr_active_slave
2933 */
2934
2935 if (!curr_arp_slave) {
2936 curr_arp_slave = bond_first_slave_rcu(bond);
2937 if (!curr_arp_slave)
2938 return should_notify_rtnl;
2939 }
2940
2941 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2942
2943 bond_for_each_slave_rcu(bond, slave, iter) {
2944 if (!found && !before && bond_slave_is_up(slave))
2945 before = slave;
2946
2947 if (found && !new_slave && bond_slave_is_up(slave))
2948 new_slave = slave;
2949 /* if the link state is up at this point, we
2950 * mark it down - this can happen if we have
2951 * simultaneous link failures and
2952 * reselect_active_interface doesn't make this
2953 * one the current slave so it is still marked
2954 * up when it is actually down
2955 */
2956 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2957 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2958 BOND_SLAVE_NOTIFY_LATER);
2959 if (slave->link_failure_count < UINT_MAX)
2960 slave->link_failure_count++;
2961
2962 bond_set_slave_inactive_flags(slave,
2963 BOND_SLAVE_NOTIFY_LATER);
2964
2965 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
2966 }
2967 if (slave == curr_arp_slave)
2968 found = true;
2969 }
2970
2971 if (!new_slave && before)
2972 new_slave = before;
2973
2974 if (!new_slave)
2975 goto check_state;
2976
2977 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2978 BOND_SLAVE_NOTIFY_LATER);
2979 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2980 bond_arp_send_all(bond, new_slave);
2981 new_slave->last_link_up = jiffies;
2982 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2983
2984check_state:
2985 bond_for_each_slave_rcu(bond, slave, iter) {
2986 if (slave->should_notify || slave->should_notify_link) {
2987 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2988 break;
2989 }
2990 }
2991 return should_notify_rtnl;
2992}
2993
2994static void bond_activebackup_arp_mon(struct bonding *bond)
2995{
2996 bool should_notify_peers = false;
2997 bool should_notify_rtnl = false;
2998 int delta_in_ticks;
2999
3000 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3001
3002 if (!bond_has_slaves(bond))
3003 goto re_arm;
3004
3005 rcu_read_lock();
3006
3007 should_notify_peers = bond_should_notify_peers(bond);
3008
3009 if (bond_ab_arp_inspect(bond)) {
3010 rcu_read_unlock();
3011
3012 /* Race avoidance with bond_close flush of workqueue */
3013 if (!rtnl_trylock()) {
3014 delta_in_ticks = 1;
3015 should_notify_peers = false;
3016 goto re_arm;
3017 }
3018
3019 bond_ab_arp_commit(bond);
3020
3021 rtnl_unlock();
3022 rcu_read_lock();
3023 }
3024
3025 should_notify_rtnl = bond_ab_arp_probe(bond);
3026 rcu_read_unlock();
3027
3028re_arm:
3029 if (bond->params.arp_interval)
3030 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3031
3032 if (should_notify_peers || should_notify_rtnl) {
3033 if (!rtnl_trylock())
3034 return;
3035
3036 if (should_notify_peers)
3037 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3038 bond->dev);
3039 if (should_notify_rtnl) {
3040 bond_slave_state_notify(bond);
3041 bond_slave_link_notify(bond);
3042 }
3043
3044 rtnl_unlock();
3045 }
3046}
3047
3048static void bond_arp_monitor(struct work_struct *work)
3049{
3050 struct bonding *bond = container_of(work, struct bonding,
3051 arp_work.work);
3052
3053 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3054 bond_activebackup_arp_mon(bond);
3055 else
3056 bond_loadbalance_arp_mon(bond);
3057}
3058
3059/*-------------------------- netdev event handling --------------------------*/
3060
3061/* Change device name */
3062static int bond_event_changename(struct bonding *bond)
3063{
3064 bond_remove_proc_entry(bond);
3065 bond_create_proc_entry(bond);
3066
3067 bond_debug_reregister(bond);
3068
3069 return NOTIFY_DONE;
3070}
3071
3072static int bond_master_netdev_event(unsigned long event,
3073 struct net_device *bond_dev)
3074{
3075 struct bonding *event_bond = netdev_priv(bond_dev);
3076
3077 netdev_dbg(bond_dev, "%s called\n", __func__);
3078
3079 switch (event) {
3080 case NETDEV_CHANGENAME:
3081 return bond_event_changename(event_bond);
3082 case NETDEV_UNREGISTER:
3083 bond_remove_proc_entry(event_bond);
3084 break;
3085 case NETDEV_REGISTER:
3086 bond_create_proc_entry(event_bond);
3087 break;
3088 default:
3089 break;
3090 }
3091
3092 return NOTIFY_DONE;
3093}
3094
3095static int bond_slave_netdev_event(unsigned long event,
3096 struct net_device *slave_dev)
3097{
3098 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3099 struct bonding *bond;
3100 struct net_device *bond_dev;
3101
3102 /* A netdev event can be generated while enslaving a device
3103 * before netdev_rx_handler_register is called in which case
3104 * slave will be NULL
3105 */
3106 if (!slave) {
3107 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3108 return NOTIFY_DONE;
3109 }
3110
3111 bond_dev = slave->bond->dev;
3112 bond = slave->bond;
3113 primary = rtnl_dereference(bond->primary_slave);
3114
3115 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3116
3117 switch (event) {
3118 case NETDEV_UNREGISTER:
3119 if (bond_dev->type != ARPHRD_ETHER)
3120 bond_release_and_destroy(bond_dev, slave_dev);
3121 else
3122 __bond_release_one(bond_dev, slave_dev, false, true);
3123 break;
3124 case NETDEV_UP:
3125 case NETDEV_CHANGE:
3126 /* For 802.3ad mode only:
3127 * Getting invalid Speed/Duplex values here will put slave
3128 * in weird state. Mark it as link-fail if the link was
3129 * previously up or link-down if it hasn't yet come up, and
3130 * let link-monitoring (miimon) set it right when correct
3131 * speeds/duplex are available.
3132 */
3133 if (bond_update_speed_duplex(slave) &&
3134 BOND_MODE(bond) == BOND_MODE_8023AD) {
3135 if (slave->last_link_up)
3136 slave->link = BOND_LINK_FAIL;
3137 else
3138 slave->link = BOND_LINK_DOWN;
3139 }
3140
3141 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3142 bond_3ad_adapter_speed_duplex_changed(slave);
3143 /* Fallthrough */
3144 case NETDEV_DOWN:
3145 /* Refresh slave-array if applicable!
3146 * If the setup does not use miimon or arpmon (mode-specific!),
3147 * then these events will not cause the slave-array to be
3148 * refreshed. This will cause xmit to use a slave that is not
3149 * usable. Avoid such situation by refeshing the array at these
3150 * events. If these (miimon/arpmon) parameters are configured
3151 * then array gets refreshed twice and that should be fine!
3152 */
3153 if (bond_mode_can_use_xmit_hash(bond))
3154 bond_update_slave_arr(bond, NULL);
3155 break;
3156 case NETDEV_CHANGEMTU:
3157 /* TODO: Should slaves be allowed to
3158 * independently alter their MTU? For
3159 * an active-backup bond, slaves need
3160 * not be the same type of device, so
3161 * MTUs may vary. For other modes,
3162 * slaves arguably should have the
3163 * same MTUs. To do this, we'd need to
3164 * take over the slave's change_mtu
3165 * function for the duration of their
3166 * servitude.
3167 */
3168 break;
3169 case NETDEV_CHANGENAME:
3170 /* we don't care if we don't have primary set */
3171 if (!bond_uses_primary(bond) ||
3172 !bond->params.primary[0])
3173 break;
3174
3175 if (slave == primary) {
3176 /* slave's name changed - he's no longer primary */
3177 RCU_INIT_POINTER(bond->primary_slave, NULL);
3178 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3179 /* we have a new primary slave */
3180 rcu_assign_pointer(bond->primary_slave, slave);
3181 } else { /* we didn't change primary - exit */
3182 break;
3183 }
3184
3185 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3186 primary ? slave_dev->name : "none");
3187
3188 block_netpoll_tx();
3189 bond_select_active_slave(bond);
3190 unblock_netpoll_tx();
3191 break;
3192 case NETDEV_FEAT_CHANGE:
3193 bond_compute_features(bond);
3194 break;
3195 case NETDEV_RESEND_IGMP:
3196 /* Propagate to master device */
3197 call_netdevice_notifiers(event, slave->bond->dev);
3198 break;
3199 default:
3200 break;
3201 }
3202
3203 return NOTIFY_DONE;
3204}
3205
3206/* bond_netdev_event: handle netdev notifier chain events.
3207 *
3208 * This function receives events for the netdev chain. The caller (an
3209 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3210 * locks for us to safely manipulate the slave devices (RTNL lock,
3211 * dev_probe_lock).
3212 */
3213static int bond_netdev_event(struct notifier_block *this,
3214 unsigned long event, void *ptr)
3215{
3216 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3217
3218 netdev_dbg(event_dev, "%s received %s\n",
3219 __func__, netdev_cmd_to_name(event));
3220
3221 if (!(event_dev->priv_flags & IFF_BONDING))
3222 return NOTIFY_DONE;
3223
3224 if (event_dev->flags & IFF_MASTER) {
3225 int ret;
3226
3227 ret = bond_master_netdev_event(event, event_dev);
3228 if (ret != NOTIFY_DONE)
3229 return ret;
3230 }
3231
3232 if (event_dev->flags & IFF_SLAVE)
3233 return bond_slave_netdev_event(event, event_dev);
3234
3235 return NOTIFY_DONE;
3236}
3237
3238static struct notifier_block bond_netdev_notifier = {
3239 .notifier_call = bond_netdev_event,
3240};
3241
3242/*---------------------------- Hashing Policies -----------------------------*/
3243
3244/* L2 hash helper */
3245static inline u32 bond_eth_hash(struct sk_buff *skb)
3246{
3247 struct ethhdr *ep, hdr_tmp;
3248
3249 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3250 if (ep)
3251 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3252 return 0;
3253}
3254
3255/* Extract the appropriate headers based on bond's xmit policy */
3256static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3257 struct flow_keys *fk)
3258{
3259 const struct ipv6hdr *iph6;
3260 const struct iphdr *iph;
3261 int noff, proto = -1;
3262
3263 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3264 return skb_flow_dissect_flow_keys(skb, fk, 0);
3265
3266 fk->ports.ports = 0;
3267 noff = skb_network_offset(skb);
3268 if (skb->protocol == htons(ETH_P_IP)) {
3269 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3270 return false;
3271 iph = ip_hdr(skb);
3272 iph_to_flow_copy_v4addrs(fk, iph);
3273 noff += iph->ihl << 2;
3274 if (!ip_is_fragment(iph))
3275 proto = iph->protocol;
3276 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3277 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3278 return false;
3279 iph6 = ipv6_hdr(skb);
3280 iph_to_flow_copy_v6addrs(fk, iph6);
3281 noff += sizeof(*iph6);
3282 proto = iph6->nexthdr;
3283 } else {
3284 return false;
3285 }
3286 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3287 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3288
3289 return true;
3290}
3291
3292/**
3293 * bond_xmit_hash - generate a hash value based on the xmit policy
3294 * @bond: bonding device
3295 * @skb: buffer to use for headers
3296 *
3297 * This function will extract the necessary headers from the skb buffer and use
3298 * them to generate a hash based on the xmit_policy set in the bonding device
3299 */
3300u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3301{
3302 struct flow_keys flow;
3303 u32 hash;
3304
3305 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3306 skb->l4_hash)
3307 return skb->hash;
3308
3309 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3310 !bond_flow_dissect(bond, skb, &flow))
3311 return bond_eth_hash(skb);
3312
3313 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3314 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3315 hash = bond_eth_hash(skb);
3316 else
3317 hash = (__force u32)flow.ports.ports;
3318 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3319 (__force u32)flow_get_u32_src(&flow);
3320 hash ^= (hash >> 16);
3321 hash ^= (hash >> 8);
3322
3323 return hash >> 1;
3324}
3325
3326/*-------------------------- Device entry points ----------------------------*/
3327
3328void bond_work_init_all(struct bonding *bond)
3329{
3330 INIT_DELAYED_WORK(&bond->mcast_work,
3331 bond_resend_igmp_join_requests_delayed);
3332 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3333 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3334 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3335 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3336 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3337}
3338
3339static void bond_work_cancel_all(struct bonding *bond)
3340{
3341 cancel_delayed_work_sync(&bond->mii_work);
3342 cancel_delayed_work_sync(&bond->arp_work);
3343 cancel_delayed_work_sync(&bond->alb_work);
3344 cancel_delayed_work_sync(&bond->ad_work);
3345 cancel_delayed_work_sync(&bond->mcast_work);
3346 cancel_delayed_work_sync(&bond->slave_arr_work);
3347}
3348
3349static int bond_open(struct net_device *bond_dev)
3350{
3351 struct bonding *bond = netdev_priv(bond_dev);
3352 struct list_head *iter;
3353 struct slave *slave;
3354
3355 /* reset slave->backup and slave->inactive */
3356 if (bond_has_slaves(bond)) {
3357 bond_for_each_slave(bond, slave, iter) {
3358 if (bond_uses_primary(bond) &&
3359 slave != rcu_access_pointer(bond->curr_active_slave)) {
3360 bond_set_slave_inactive_flags(slave,
3361 BOND_SLAVE_NOTIFY_NOW);
3362 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3363 bond_set_slave_active_flags(slave,
3364 BOND_SLAVE_NOTIFY_NOW);
3365 }
3366 }
3367 }
3368
3369 if (bond_is_lb(bond)) {
3370 /* bond_alb_initialize must be called before the timer
3371 * is started.
3372 */
3373 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3374 return -ENOMEM;
3375 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3376 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3377 }
3378
3379 if (bond->params.miimon) /* link check interval, in milliseconds. */
3380 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3381
3382 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3383 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3384 bond->recv_probe = bond_arp_rcv;
3385 }
3386
3387 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3388 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3389 /* register to receive LACPDUs */
3390 bond->recv_probe = bond_3ad_lacpdu_recv;
3391 bond_3ad_initiate_agg_selection(bond, 1);
3392 }
3393
3394 if (bond_mode_can_use_xmit_hash(bond))
3395 bond_update_slave_arr(bond, NULL);
3396
3397 return 0;
3398}
3399
3400static int bond_close(struct net_device *bond_dev)
3401{
3402 struct bonding *bond = netdev_priv(bond_dev);
3403
3404 bond_work_cancel_all(bond);
3405 bond->send_peer_notif = 0;
3406 if (bond_is_lb(bond))
3407 bond_alb_deinitialize(bond);
3408 bond->recv_probe = NULL;
3409
3410 return 0;
3411}
3412
3413/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3414 * that some drivers can provide 32bit values only.
3415 */
3416static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3417 const struct rtnl_link_stats64 *_new,
3418 const struct rtnl_link_stats64 *_old)
3419{
3420 const u64 *new = (const u64 *)_new;
3421 const u64 *old = (const u64 *)_old;
3422 u64 *res = (u64 *)_res;
3423 int i;
3424
3425 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3426 u64 nv = new[i];
3427 u64 ov = old[i];
3428 s64 delta = nv - ov;
3429
3430 /* detects if this particular field is 32bit only */
3431 if (((nv | ov) >> 32) == 0)
3432 delta = (s64)(s32)((u32)nv - (u32)ov);
3433
3434 /* filter anomalies, some drivers reset their stats
3435 * at down/up events.
3436 */
3437 if (delta > 0)
3438 res[i] += delta;
3439 }
3440}
3441
3442static void bond_get_stats(struct net_device *bond_dev,
3443 struct rtnl_link_stats64 *stats)
3444{
3445 struct bonding *bond = netdev_priv(bond_dev);
3446 struct rtnl_link_stats64 temp;
3447 struct list_head *iter;
3448 struct slave *slave;
3449
3450 spin_lock(&bond->stats_lock);
3451 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3452
3453 rcu_read_lock();
3454 bond_for_each_slave_rcu(bond, slave, iter) {
3455 const struct rtnl_link_stats64 *new =
3456 dev_get_stats(slave->dev, &temp);
3457
3458 bond_fold_stats(stats, new, &slave->slave_stats);
3459
3460 /* save off the slave stats for the next run */
3461 memcpy(&slave->slave_stats, new, sizeof(*new));
3462 }
3463 rcu_read_unlock();
3464
3465 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3466 spin_unlock(&bond->stats_lock);
3467}
3468
3469static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3470{
3471 struct bonding *bond = netdev_priv(bond_dev);
3472 struct net_device *slave_dev = NULL;
3473 struct ifbond k_binfo;
3474 struct ifbond __user *u_binfo = NULL;
3475 struct ifslave k_sinfo;
3476 struct ifslave __user *u_sinfo = NULL;
3477 struct mii_ioctl_data *mii = NULL;
3478 struct bond_opt_value newval;
3479 struct net *net;
3480 int res = 0;
3481
3482 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3483
3484 switch (cmd) {
3485 case SIOCGMIIPHY:
3486 mii = if_mii(ifr);
3487 if (!mii)
3488 return -EINVAL;
3489
3490 mii->phy_id = 0;
3491 /* Fall Through */
3492 case SIOCGMIIREG:
3493 /* We do this again just in case we were called by SIOCGMIIREG
3494 * instead of SIOCGMIIPHY.
3495 */
3496 mii = if_mii(ifr);
3497 if (!mii)
3498 return -EINVAL;
3499
3500 if (mii->reg_num == 1) {
3501 mii->val_out = 0;
3502 if (netif_carrier_ok(bond->dev))
3503 mii->val_out = BMSR_LSTATUS;
3504 }
3505
3506 return 0;
3507 case BOND_INFO_QUERY_OLD:
3508 case SIOCBONDINFOQUERY:
3509 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3510
3511 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3512 return -EFAULT;
3513
3514 bond_info_query(bond_dev, &k_binfo);
3515 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3516 return -EFAULT;
3517
3518 return 0;
3519 case BOND_SLAVE_INFO_QUERY_OLD:
3520 case SIOCBONDSLAVEINFOQUERY:
3521 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3522
3523 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3524 return -EFAULT;
3525
3526 res = bond_slave_info_query(bond_dev, &k_sinfo);
3527 if (res == 0 &&
3528 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3529 return -EFAULT;
3530
3531 return res;
3532 default:
3533 break;
3534 }
3535
3536 net = dev_net(bond_dev);
3537
3538 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3539 return -EPERM;
3540
3541 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3542
3543 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3544
3545 if (!slave_dev)
3546 return -ENODEV;
3547
3548 switch (cmd) {
3549 case BOND_ENSLAVE_OLD:
3550 case SIOCBONDENSLAVE:
3551 res = bond_enslave(bond_dev, slave_dev, NULL);
3552 break;
3553 case BOND_RELEASE_OLD:
3554 case SIOCBONDRELEASE:
3555 res = bond_release(bond_dev, slave_dev);
3556 break;
3557 case BOND_SETHWADDR_OLD:
3558 case SIOCBONDSETHWADDR:
3559 res = bond_set_dev_addr(bond_dev, slave_dev);
3560 break;
3561 case BOND_CHANGE_ACTIVE_OLD:
3562 case SIOCBONDCHANGEACTIVE:
3563 bond_opt_initstr(&newval, slave_dev->name);
3564 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3565 &newval);
3566 break;
3567 default:
3568 res = -EOPNOTSUPP;
3569 }
3570
3571 return res;
3572}
3573
3574static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3575{
3576 struct bonding *bond = netdev_priv(bond_dev);
3577
3578 if (change & IFF_PROMISC)
3579 bond_set_promiscuity(bond,
3580 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3581
3582 if (change & IFF_ALLMULTI)
3583 bond_set_allmulti(bond,
3584 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3585}
3586
3587static void bond_set_rx_mode(struct net_device *bond_dev)
3588{
3589 struct bonding *bond = netdev_priv(bond_dev);
3590 struct list_head *iter;
3591 struct slave *slave;
3592
3593 rcu_read_lock();
3594 if (bond_uses_primary(bond)) {
3595 slave = rcu_dereference(bond->curr_active_slave);
3596 if (slave) {
3597 dev_uc_sync(slave->dev, bond_dev);
3598 dev_mc_sync(slave->dev, bond_dev);
3599 }
3600 } else {
3601 bond_for_each_slave_rcu(bond, slave, iter) {
3602 dev_uc_sync_multiple(slave->dev, bond_dev);
3603 dev_mc_sync_multiple(slave->dev, bond_dev);
3604 }
3605 }
3606 rcu_read_unlock();
3607}
3608
3609static int bond_neigh_init(struct neighbour *n)
3610{
3611 struct bonding *bond = netdev_priv(n->dev);
3612 const struct net_device_ops *slave_ops;
3613 struct neigh_parms parms;
3614 struct slave *slave;
3615 int ret;
3616
3617 slave = bond_first_slave(bond);
3618 if (!slave)
3619 return 0;
3620 slave_ops = slave->dev->netdev_ops;
3621 if (!slave_ops->ndo_neigh_setup)
3622 return 0;
3623
3624 parms.neigh_setup = NULL;
3625 parms.neigh_cleanup = NULL;
3626 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3627 if (ret)
3628 return ret;
3629
3630 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3631 * after the last slave has been detached. Assumes that all slaves
3632 * utilize the same neigh_cleanup (true at this writing as only user
3633 * is ipoib).
3634 */
3635 n->parms->neigh_cleanup = parms.neigh_cleanup;
3636
3637 if (!parms.neigh_setup)
3638 return 0;
3639
3640 return parms.neigh_setup(n);
3641}
3642
3643/* The bonding ndo_neigh_setup is called at init time beofre any
3644 * slave exists. So we must declare proxy setup function which will
3645 * be used at run time to resolve the actual slave neigh param setup.
3646 *
3647 * It's also called by master devices (such as vlans) to setup their
3648 * underlying devices. In that case - do nothing, we're already set up from
3649 * our init.
3650 */
3651static int bond_neigh_setup(struct net_device *dev,
3652 struct neigh_parms *parms)
3653{
3654 /* modify only our neigh_parms */
3655 if (parms->dev == dev)
3656 parms->neigh_setup = bond_neigh_init;
3657
3658 return 0;
3659}
3660
3661/* Change the MTU of all of a master's slaves to match the master */
3662static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3663{
3664 struct bonding *bond = netdev_priv(bond_dev);
3665 struct slave *slave, *rollback_slave;
3666 struct list_head *iter;
3667 int res = 0;
3668
3669 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3670
3671 bond_for_each_slave(bond, slave, iter) {
3672 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3673 slave, slave->dev->netdev_ops->ndo_change_mtu);
3674
3675 res = dev_set_mtu(slave->dev, new_mtu);
3676
3677 if (res) {
3678 /* If we failed to set the slave's mtu to the new value
3679 * we must abort the operation even in ACTIVE_BACKUP
3680 * mode, because if we allow the backup slaves to have
3681 * different mtu values than the active slave we'll
3682 * need to change their mtu when doing a failover. That
3683 * means changing their mtu from timer context, which
3684 * is probably not a good idea.
3685 */
3686 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3687 res, new_mtu);
3688 goto unwind;
3689 }
3690 }
3691
3692 bond_dev->mtu = new_mtu;
3693
3694 return 0;
3695
3696unwind:
3697 /* unwind from head to the slave that failed */
3698 bond_for_each_slave(bond, rollback_slave, iter) {
3699 int tmp_res;
3700
3701 if (rollback_slave == slave)
3702 break;
3703
3704 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3705 if (tmp_res)
3706 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3707 tmp_res);
3708 }
3709
3710 return res;
3711}
3712
3713/* Change HW address
3714 *
3715 * Note that many devices must be down to change the HW address, and
3716 * downing the master releases all slaves. We can make bonds full of
3717 * bonding devices to test this, however.
3718 */
3719static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3720{
3721 struct bonding *bond = netdev_priv(bond_dev);
3722 struct slave *slave, *rollback_slave;
3723 struct sockaddr_storage *ss = addr, tmp_ss;
3724 struct list_head *iter;
3725 int res = 0;
3726
3727 if (BOND_MODE(bond) == BOND_MODE_ALB)
3728 return bond_alb_set_mac_address(bond_dev, addr);
3729
3730
3731 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
3732
3733 /* If fail_over_mac is enabled, do nothing and return success.
3734 * Returning an error causes ifenslave to fail.
3735 */
3736 if (bond->params.fail_over_mac &&
3737 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3738 return 0;
3739
3740 if (!is_valid_ether_addr(ss->__data))
3741 return -EADDRNOTAVAIL;
3742
3743 bond_for_each_slave(bond, slave, iter) {
3744 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
3745 __func__, slave);
3746 res = dev_set_mac_address(slave->dev, addr, NULL);
3747 if (res) {
3748 /* TODO: consider downing the slave
3749 * and retry ?
3750 * User should expect communications
3751 * breakage anyway until ARP finish
3752 * updating, so...
3753 */
3754 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
3755 __func__, res);
3756 goto unwind;
3757 }
3758 }
3759
3760 /* success */
3761 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3762 return 0;
3763
3764unwind:
3765 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3766 tmp_ss.ss_family = bond_dev->type;
3767
3768 /* unwind from head to the slave that failed */
3769 bond_for_each_slave(bond, rollback_slave, iter) {
3770 int tmp_res;
3771
3772 if (rollback_slave == slave)
3773 break;
3774
3775 tmp_res = dev_set_mac_address(rollback_slave->dev,
3776 (struct sockaddr *)&tmp_ss, NULL);
3777 if (tmp_res) {
3778 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
3779 __func__, tmp_res);
3780 }
3781 }
3782
3783 return res;
3784}
3785
3786/**
3787 * bond_xmit_slave_id - transmit skb through slave with slave_id
3788 * @bond: bonding device that is transmitting
3789 * @skb: buffer to transmit
3790 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3791 *
3792 * This function tries to transmit through slave with slave_id but in case
3793 * it fails, it tries to find the first available slave for transmission.
3794 * The skb is consumed in all cases, thus the function is void.
3795 */
3796static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3797{
3798 struct list_head *iter;
3799 struct slave *slave;
3800 int i = slave_id;
3801
3802 /* Here we start from the slave with slave_id */
3803 bond_for_each_slave_rcu(bond, slave, iter) {
3804 if (--i < 0) {
3805 if (bond_slave_can_tx(slave)) {
3806 bond_dev_queue_xmit(bond, skb, slave->dev);
3807 return;
3808 }
3809 }
3810 }
3811
3812 /* Here we start from the first slave up to slave_id */
3813 i = slave_id;
3814 bond_for_each_slave_rcu(bond, slave, iter) {
3815 if (--i < 0)
3816 break;
3817 if (bond_slave_can_tx(slave)) {
3818 bond_dev_queue_xmit(bond, skb, slave->dev);
3819 return;
3820 }
3821 }
3822 /* no slave that can tx has been found */
3823 bond_tx_drop(bond->dev, skb);
3824}
3825
3826/**
3827 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3828 * @bond: bonding device to use
3829 *
3830 * Based on the value of the bonding device's packets_per_slave parameter
3831 * this function generates a slave id, which is usually used as the next
3832 * slave to transmit through.
3833 */
3834static u32 bond_rr_gen_slave_id(struct bonding *bond)
3835{
3836 u32 slave_id;
3837 struct reciprocal_value reciprocal_packets_per_slave;
3838 int packets_per_slave = bond->params.packets_per_slave;
3839
3840 switch (packets_per_slave) {
3841 case 0:
3842 slave_id = prandom_u32();
3843 break;
3844 case 1:
3845 slave_id = bond->rr_tx_counter;
3846 break;
3847 default:
3848 reciprocal_packets_per_slave =
3849 bond->params.reciprocal_packets_per_slave;
3850 slave_id = reciprocal_divide(bond->rr_tx_counter,
3851 reciprocal_packets_per_slave);
3852 break;
3853 }
3854 bond->rr_tx_counter++;
3855
3856 return slave_id;
3857}
3858
3859static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3860 struct net_device *bond_dev)
3861{
3862 struct bonding *bond = netdev_priv(bond_dev);
3863 struct slave *slave;
3864 int slave_cnt;
3865 u32 slave_id;
3866
3867 /* Start with the curr_active_slave that joined the bond as the
3868 * default for sending IGMP traffic. For failover purposes one
3869 * needs to maintain some consistency for the interface that will
3870 * send the join/membership reports. The curr_active_slave found
3871 * will send all of this type of traffic.
3872 */
3873 if (skb->protocol == htons(ETH_P_IP)) {
3874 int noff = skb_network_offset(skb);
3875 struct iphdr *iph;
3876
3877 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3878 goto non_igmp;
3879
3880 iph = ip_hdr(skb);
3881 if (iph->protocol == IPPROTO_IGMP) {
3882 slave = rcu_dereference(bond->curr_active_slave);
3883 if (slave)
3884 bond_dev_queue_xmit(bond, skb, slave->dev);
3885 else
3886 bond_xmit_slave_id(bond, skb, 0);
3887 return NETDEV_TX_OK;
3888 }
3889 }
3890
3891non_igmp:
3892 slave_cnt = READ_ONCE(bond->slave_cnt);
3893 if (likely(slave_cnt)) {
3894 slave_id = bond_rr_gen_slave_id(bond);
3895 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3896 } else {
3897 bond_tx_drop(bond_dev, skb);
3898 }
3899 return NETDEV_TX_OK;
3900}
3901
3902/* In active-backup mode, we know that bond->curr_active_slave is always valid if
3903 * the bond has a usable interface.
3904 */
3905static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3906 struct net_device *bond_dev)
3907{
3908 struct bonding *bond = netdev_priv(bond_dev);
3909 struct slave *slave;
3910
3911 slave = rcu_dereference(bond->curr_active_slave);
3912 if (slave)
3913 bond_dev_queue_xmit(bond, skb, slave->dev);
3914 else
3915 bond_tx_drop(bond_dev, skb);
3916
3917 return NETDEV_TX_OK;
3918}
3919
3920/* Use this to update slave_array when (a) it's not appropriate to update
3921 * slave_array right away (note that update_slave_array() may sleep)
3922 * and / or (b) RTNL is not held.
3923 */
3924void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3925{
3926 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3927}
3928
3929/* Slave array work handler. Holds only RTNL */
3930static void bond_slave_arr_handler(struct work_struct *work)
3931{
3932 struct bonding *bond = container_of(work, struct bonding,
3933 slave_arr_work.work);
3934 int ret;
3935
3936 if (!rtnl_trylock())
3937 goto err;
3938
3939 ret = bond_update_slave_arr(bond, NULL);
3940 rtnl_unlock();
3941 if (ret) {
3942 pr_warn_ratelimited("Failed to update slave array from WT\n");
3943 goto err;
3944 }
3945 return;
3946
3947err:
3948 bond_slave_arr_work_rearm(bond, 1);
3949}
3950
3951/* Build the usable slaves array in control path for modes that use xmit-hash
3952 * to determine the slave interface -
3953 * (a) BOND_MODE_8023AD
3954 * (b) BOND_MODE_XOR
3955 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3956 *
3957 * The caller is expected to hold RTNL only and NO other lock!
3958 */
3959int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3960{
3961 struct slave *slave;
3962 struct list_head *iter;
3963 struct bond_up_slave *new_arr, *old_arr;
3964 int agg_id = 0;
3965 int ret = 0;
3966
3967#ifdef CONFIG_LOCKDEP
3968 WARN_ON(lockdep_is_held(&bond->mode_lock));
3969#endif
3970
3971 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3972 GFP_KERNEL);
3973 if (!new_arr) {
3974 ret = -ENOMEM;
3975 pr_err("Failed to build slave-array.\n");
3976 goto out;
3977 }
3978 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3979 struct ad_info ad_info;
3980
3981 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3982 pr_debug("bond_3ad_get_active_agg_info failed\n");
3983 kfree_rcu(new_arr, rcu);
3984 /* No active aggragator means it's not safe to use
3985 * the previous array.
3986 */
3987 old_arr = rtnl_dereference(bond->slave_arr);
3988 if (old_arr) {
3989 RCU_INIT_POINTER(bond->slave_arr, NULL);
3990 kfree_rcu(old_arr, rcu);
3991 }
3992 goto out;
3993 }
3994 agg_id = ad_info.aggregator_id;
3995 }
3996 bond_for_each_slave(bond, slave, iter) {
3997 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3998 struct aggregator *agg;
3999
4000 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4001 if (!agg || agg->aggregator_identifier != agg_id)
4002 continue;
4003 }
4004 if (!bond_slave_can_tx(slave))
4005 continue;
4006 if (skipslave == slave)
4007 continue;
4008
4009 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4010 new_arr->count);
4011
4012 new_arr->arr[new_arr->count++] = slave;
4013 }
4014
4015 old_arr = rtnl_dereference(bond->slave_arr);
4016 rcu_assign_pointer(bond->slave_arr, new_arr);
4017 if (old_arr)
4018 kfree_rcu(old_arr, rcu);
4019out:
4020 if (ret != 0 && skipslave) {
4021 int idx;
4022
4023 /* Rare situation where caller has asked to skip a specific
4024 * slave but allocation failed (most likely!). BTW this is
4025 * only possible when the call is initiated from
4026 * __bond_release_one(). In this situation; overwrite the
4027 * skipslave entry in the array with the last entry from the
4028 * array to avoid a situation where the xmit path may choose
4029 * this to-be-skipped slave to send a packet out.
4030 */
4031 old_arr = rtnl_dereference(bond->slave_arr);
4032 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) {
4033 if (skipslave == old_arr->arr[idx]) {
4034 old_arr->arr[idx] =
4035 old_arr->arr[old_arr->count-1];
4036 old_arr->count--;
4037 break;
4038 }
4039 }
4040 }
4041 return ret;
4042}
4043
4044/* Use this Xmit function for 3AD as well as XOR modes. The current
4045 * usable slave array is formed in the control path. The xmit function
4046 * just calculates hash and sends the packet out.
4047 */
4048static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4049 struct net_device *dev)
4050{
4051 struct bonding *bond = netdev_priv(dev);
4052 struct slave *slave;
4053 struct bond_up_slave *slaves;
4054 unsigned int count;
4055
4056 slaves = rcu_dereference(bond->slave_arr);
4057 count = slaves ? READ_ONCE(slaves->count) : 0;
4058 if (likely(count)) {
4059 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4060 bond_dev_queue_xmit(bond, skb, slave->dev);
4061 } else {
4062 bond_tx_drop(dev, skb);
4063 }
4064
4065 return NETDEV_TX_OK;
4066}
4067
4068/* in broadcast mode, we send everything to all usable interfaces. */
4069static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4070 struct net_device *bond_dev)
4071{
4072 struct bonding *bond = netdev_priv(bond_dev);
4073 struct slave *slave = NULL;
4074 struct list_head *iter;
4075
4076 bond_for_each_slave_rcu(bond, slave, iter) {
4077 if (bond_is_last_slave(bond, slave))
4078 break;
4079 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4080 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4081
4082 if (!skb2) {
4083 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4084 bond_dev->name, __func__);
4085 continue;
4086 }
4087 bond_dev_queue_xmit(bond, skb2, slave->dev);
4088 }
4089 }
4090 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4091 bond_dev_queue_xmit(bond, skb, slave->dev);
4092 else
4093 bond_tx_drop(bond_dev, skb);
4094
4095 return NETDEV_TX_OK;
4096}
4097
4098/*------------------------- Device initialization ---------------------------*/
4099
4100/* Lookup the slave that corresponds to a qid */
4101static inline int bond_slave_override(struct bonding *bond,
4102 struct sk_buff *skb)
4103{
4104 struct slave *slave = NULL;
4105 struct list_head *iter;
4106
4107 if (!skb_rx_queue_recorded(skb))
4108 return 1;
4109
4110 /* Find out if any slaves have the same mapping as this skb. */
4111 bond_for_each_slave_rcu(bond, slave, iter) {
4112 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4113 if (bond_slave_is_up(slave) &&
4114 slave->link == BOND_LINK_UP) {
4115 bond_dev_queue_xmit(bond, skb, slave->dev);
4116 return 0;
4117 }
4118 /* If the slave isn't UP, use default transmit policy. */
4119 break;
4120 }
4121 }
4122
4123 return 1;
4124}
4125
4126
4127static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4128 struct net_device *sb_dev)
4129{
4130 /* This helper function exists to help dev_pick_tx get the correct
4131 * destination queue. Using a helper function skips a call to
4132 * skb_tx_hash and will put the skbs in the queue we expect on their
4133 * way down to the bonding driver.
4134 */
4135 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4136
4137 /* Save the original txq to restore before passing to the driver */
4138 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4139
4140 if (unlikely(txq >= dev->real_num_tx_queues)) {
4141 do {
4142 txq -= dev->real_num_tx_queues;
4143 } while (txq >= dev->real_num_tx_queues);
4144 }
4145 return txq;
4146}
4147
4148static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4149{
4150 struct bonding *bond = netdev_priv(dev);
4151
4152 if (bond_should_override_tx_queue(bond) &&
4153 !bond_slave_override(bond, skb))
4154 return NETDEV_TX_OK;
4155
4156 switch (BOND_MODE(bond)) {
4157 case BOND_MODE_ROUNDROBIN:
4158 return bond_xmit_roundrobin(skb, dev);
4159 case BOND_MODE_ACTIVEBACKUP:
4160 return bond_xmit_activebackup(skb, dev);
4161 case BOND_MODE_8023AD:
4162 case BOND_MODE_XOR:
4163 return bond_3ad_xor_xmit(skb, dev);
4164 case BOND_MODE_BROADCAST:
4165 return bond_xmit_broadcast(skb, dev);
4166 case BOND_MODE_ALB:
4167 return bond_alb_xmit(skb, dev);
4168 case BOND_MODE_TLB:
4169 return bond_tlb_xmit(skb, dev);
4170 default:
4171 /* Should never happen, mode already checked */
4172 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4173 WARN_ON_ONCE(1);
4174 bond_tx_drop(dev, skb);
4175 return NETDEV_TX_OK;
4176 }
4177}
4178
4179static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4180{
4181 struct bonding *bond = netdev_priv(dev);
4182 netdev_tx_t ret = NETDEV_TX_OK;
4183
4184 /* If we risk deadlock from transmitting this in the
4185 * netpoll path, tell netpoll to queue the frame for later tx
4186 */
4187 if (unlikely(is_netpoll_tx_blocked(dev)))
4188 return NETDEV_TX_BUSY;
4189
4190 rcu_read_lock();
4191 if (bond_has_slaves(bond))
4192 ret = __bond_start_xmit(skb, dev);
4193 else
4194 bond_tx_drop(dev, skb);
4195 rcu_read_unlock();
4196
4197 return ret;
4198}
4199
4200static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4201 struct ethtool_link_ksettings *cmd)
4202{
4203 struct bonding *bond = netdev_priv(bond_dev);
4204 unsigned long speed = 0;
4205 struct list_head *iter;
4206 struct slave *slave;
4207
4208 cmd->base.duplex = DUPLEX_UNKNOWN;
4209 cmd->base.port = PORT_OTHER;
4210
4211 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4212 * do not need to check mode. Though link speed might not represent
4213 * the true receive or transmit bandwidth (not all modes are symmetric)
4214 * this is an accurate maximum.
4215 */
4216 bond_for_each_slave(bond, slave, iter) {
4217 if (bond_slave_can_tx(slave)) {
4218 if (slave->speed != SPEED_UNKNOWN)
4219 speed += slave->speed;
4220 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4221 slave->duplex != DUPLEX_UNKNOWN)
4222 cmd->base.duplex = slave->duplex;
4223 }
4224 }
4225 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4226
4227 return 0;
4228}
4229
4230static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4231 struct ethtool_drvinfo *drvinfo)
4232{
4233 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4234 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4235 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4236 BOND_ABI_VERSION);
4237}
4238
4239static const struct ethtool_ops bond_ethtool_ops = {
4240 .get_drvinfo = bond_ethtool_get_drvinfo,
4241 .get_link = ethtool_op_get_link,
4242 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4243};
4244
4245static const struct net_device_ops bond_netdev_ops = {
4246 .ndo_init = bond_init,
4247 .ndo_uninit = bond_uninit,
4248 .ndo_open = bond_open,
4249 .ndo_stop = bond_close,
4250 .ndo_start_xmit = bond_start_xmit,
4251 .ndo_select_queue = bond_select_queue,
4252 .ndo_get_stats64 = bond_get_stats,
4253 .ndo_do_ioctl = bond_do_ioctl,
4254 .ndo_change_rx_flags = bond_change_rx_flags,
4255 .ndo_set_rx_mode = bond_set_rx_mode,
4256 .ndo_change_mtu = bond_change_mtu,
4257 .ndo_set_mac_address = bond_set_mac_address,
4258 .ndo_neigh_setup = bond_neigh_setup,
4259 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4260 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4261#ifdef CONFIG_NET_POLL_CONTROLLER
4262 .ndo_netpoll_setup = bond_netpoll_setup,
4263 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4264 .ndo_poll_controller = bond_poll_controller,
4265#endif
4266 .ndo_add_slave = bond_enslave,
4267 .ndo_del_slave = bond_release,
4268 .ndo_fix_features = bond_fix_features,
4269 .ndo_features_check = passthru_features_check,
4270};
4271
4272static const struct device_type bond_type = {
4273 .name = "bond",
4274};
4275
4276static void bond_destructor(struct net_device *bond_dev)
4277{
4278 struct bonding *bond = netdev_priv(bond_dev);
4279 if (bond->wq)
4280 destroy_workqueue(bond->wq);
4281}
4282
4283void bond_setup(struct net_device *bond_dev)
4284{
4285 struct bonding *bond = netdev_priv(bond_dev);
4286
4287 spin_lock_init(&bond->mode_lock);
4288 bond->params = bonding_defaults;
4289
4290 /* Initialize pointers */
4291 bond->dev = bond_dev;
4292
4293 /* Initialize the device entry points */
4294 ether_setup(bond_dev);
4295 bond_dev->max_mtu = ETH_MAX_MTU;
4296 bond_dev->netdev_ops = &bond_netdev_ops;
4297 bond_dev->ethtool_ops = &bond_ethtool_ops;
4298
4299 bond_dev->needs_free_netdev = true;
4300 bond_dev->priv_destructor = bond_destructor;
4301
4302 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4303
4304 /* Initialize the device options */
4305 bond_dev->flags |= IFF_MASTER;
4306 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4307 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4308
4309 /* don't acquire bond device's netif_tx_lock when transmitting */
4310 bond_dev->features |= NETIF_F_LLTX;
4311
4312 /* By default, we declare the bond to be fully
4313 * VLAN hardware accelerated capable. Special
4314 * care is taken in the various xmit functions
4315 * when there are slaves that are not hw accel
4316 * capable
4317 */
4318
4319 /* Don't allow bond devices to change network namespaces. */
4320 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4321
4322 bond_dev->hw_features = BOND_VLAN_FEATURES |
4323 NETIF_F_HW_VLAN_CTAG_RX |
4324 NETIF_F_HW_VLAN_CTAG_FILTER;
4325
4326 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4327 bond_dev->features |= bond_dev->hw_features;
4328 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4329}
4330
4331/* Destroy a bonding device.
4332 * Must be under rtnl_lock when this function is called.
4333 */
4334static void bond_uninit(struct net_device *bond_dev)
4335{
4336 struct bonding *bond = netdev_priv(bond_dev);
4337 struct list_head *iter;
4338 struct slave *slave;
4339 struct bond_up_slave *arr;
4340
4341 bond_netpoll_cleanup(bond_dev);
4342
4343 /* Release the bonded slaves */
4344 bond_for_each_slave(bond, slave, iter)
4345 __bond_release_one(bond_dev, slave->dev, true, true);
4346 netdev_info(bond_dev, "Released all slaves\n");
4347
4348 arr = rtnl_dereference(bond->slave_arr);
4349 if (arr) {
4350 RCU_INIT_POINTER(bond->slave_arr, NULL);
4351 kfree_rcu(arr, rcu);
4352 }
4353
4354 list_del(&bond->bond_list);
4355
4356 lockdep_unregister_key(&bond->stats_lock_key);
4357 bond_debug_unregister(bond);
4358}
4359
4360/*------------------------- Module initialization ---------------------------*/
4361
4362static int bond_check_params(struct bond_params *params)
4363{
4364 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4365 struct bond_opt_value newval;
4366 const struct bond_opt_value *valptr;
4367 int arp_all_targets_value = 0;
4368 u16 ad_actor_sys_prio = 0;
4369 u16 ad_user_port_key = 0;
4370 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4371 int arp_ip_count;
4372 int bond_mode = BOND_MODE_ROUNDROBIN;
4373 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4374 int lacp_fast = 0;
4375 int tlb_dynamic_lb;
4376
4377 /* Convert string parameters. */
4378 if (mode) {
4379 bond_opt_initstr(&newval, mode);
4380 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4381 if (!valptr) {
4382 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4383 return -EINVAL;
4384 }
4385 bond_mode = valptr->value;
4386 }
4387
4388 if (xmit_hash_policy) {
4389 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4390 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4391 bond_mode == BOND_MODE_BROADCAST) {
4392 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4393 bond_mode_name(bond_mode));
4394 } else {
4395 bond_opt_initstr(&newval, xmit_hash_policy);
4396 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4397 &newval);
4398 if (!valptr) {
4399 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4400 xmit_hash_policy);
4401 return -EINVAL;
4402 }
4403 xmit_hashtype = valptr->value;
4404 }
4405 }
4406
4407 if (lacp_rate) {
4408 if (bond_mode != BOND_MODE_8023AD) {
4409 pr_info("lacp_rate param is irrelevant in mode %s\n",
4410 bond_mode_name(bond_mode));
4411 } else {
4412 bond_opt_initstr(&newval, lacp_rate);
4413 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4414 &newval);
4415 if (!valptr) {
4416 pr_err("Error: Invalid lacp rate \"%s\"\n",
4417 lacp_rate);
4418 return -EINVAL;
4419 }
4420 lacp_fast = valptr->value;
4421 }
4422 }
4423
4424 if (ad_select) {
4425 bond_opt_initstr(&newval, ad_select);
4426 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4427 &newval);
4428 if (!valptr) {
4429 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4430 return -EINVAL;
4431 }
4432 params->ad_select = valptr->value;
4433 if (bond_mode != BOND_MODE_8023AD)
4434 pr_warn("ad_select param only affects 802.3ad mode\n");
4435 } else {
4436 params->ad_select = BOND_AD_STABLE;
4437 }
4438
4439 if (max_bonds < 0) {
4440 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4441 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4442 max_bonds = BOND_DEFAULT_MAX_BONDS;
4443 }
4444
4445 if (miimon < 0) {
4446 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4447 miimon, INT_MAX);
4448 miimon = 0;
4449 }
4450
4451 if (updelay < 0) {
4452 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4453 updelay, INT_MAX);
4454 updelay = 0;
4455 }
4456
4457 if (downdelay < 0) {
4458 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4459 downdelay, INT_MAX);
4460 downdelay = 0;
4461 }
4462
4463 if ((use_carrier != 0) && (use_carrier != 1)) {
4464 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4465 use_carrier);
4466 use_carrier = 1;
4467 }
4468
4469 if (num_peer_notif < 0 || num_peer_notif > 255) {
4470 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4471 num_peer_notif);
4472 num_peer_notif = 1;
4473 }
4474
4475 /* reset values for 802.3ad/TLB/ALB */
4476 if (!bond_mode_uses_arp(bond_mode)) {
4477 if (!miimon) {
4478 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");
4479 pr_warn("Forcing miimon to 100msec\n");
4480 miimon = BOND_DEFAULT_MIIMON;
4481 }
4482 }
4483
4484 if (tx_queues < 1 || tx_queues > 255) {
4485 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4486 tx_queues, BOND_DEFAULT_TX_QUEUES);
4487 tx_queues = BOND_DEFAULT_TX_QUEUES;
4488 }
4489
4490 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4491 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4492 all_slaves_active);
4493 all_slaves_active = 0;
4494 }
4495
4496 if (resend_igmp < 0 || resend_igmp > 255) {
4497 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4498 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4499 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4500 }
4501
4502 bond_opt_initval(&newval, packets_per_slave);
4503 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4504 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4505 packets_per_slave, USHRT_MAX);
4506 packets_per_slave = 1;
4507 }
4508
4509 if (bond_mode == BOND_MODE_ALB) {
4510 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",
4511 updelay);
4512 }
4513
4514 if (!miimon) {
4515 if (updelay || downdelay) {
4516 /* just warn the user the up/down delay will have
4517 * no effect since miimon is zero...
4518 */
4519 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",
4520 updelay, downdelay);
4521 }
4522 } else {
4523 /* don't allow arp monitoring */
4524 if (arp_interval) {
4525 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4526 miimon, arp_interval);
4527 arp_interval = 0;
4528 }
4529
4530 if ((updelay % miimon) != 0) {
4531 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4532 updelay, miimon, (updelay / miimon) * miimon);
4533 }
4534
4535 updelay /= miimon;
4536
4537 if ((downdelay % miimon) != 0) {
4538 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4539 downdelay, miimon,
4540 (downdelay / miimon) * miimon);
4541 }
4542
4543 downdelay /= miimon;
4544 }
4545
4546 if (arp_interval < 0) {
4547 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4548 arp_interval, INT_MAX);
4549 arp_interval = 0;
4550 }
4551
4552 for (arp_ip_count = 0, i = 0;
4553 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4554 __be32 ip;
4555
4556 /* not a complete check, but good enough to catch mistakes */
4557 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4558 !bond_is_ip_target_ok(ip)) {
4559 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4560 arp_ip_target[i]);
4561 arp_interval = 0;
4562 } else {
4563 if (bond_get_targets_ip(arp_target, ip) == -1)
4564 arp_target[arp_ip_count++] = ip;
4565 else
4566 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4567 &ip);
4568 }
4569 }
4570
4571 if (arp_interval && !arp_ip_count) {
4572 /* don't allow arping if no arp_ip_target given... */
4573 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4574 arp_interval);
4575 arp_interval = 0;
4576 }
4577
4578 if (arp_validate) {
4579 if (!arp_interval) {
4580 pr_err("arp_validate requires arp_interval\n");
4581 return -EINVAL;
4582 }
4583
4584 bond_opt_initstr(&newval, arp_validate);
4585 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4586 &newval);
4587 if (!valptr) {
4588 pr_err("Error: invalid arp_validate \"%s\"\n",
4589 arp_validate);
4590 return -EINVAL;
4591 }
4592 arp_validate_value = valptr->value;
4593 } else {
4594 arp_validate_value = 0;
4595 }
4596
4597 if (arp_all_targets) {
4598 bond_opt_initstr(&newval, arp_all_targets);
4599 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4600 &newval);
4601 if (!valptr) {
4602 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4603 arp_all_targets);
4604 arp_all_targets_value = 0;
4605 } else {
4606 arp_all_targets_value = valptr->value;
4607 }
4608 }
4609
4610 if (miimon) {
4611 pr_info("MII link monitoring set to %d ms\n", miimon);
4612 } else if (arp_interval) {
4613 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4614 arp_validate_value);
4615 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4616 arp_interval, valptr->string, arp_ip_count);
4617
4618 for (i = 0; i < arp_ip_count; i++)
4619 pr_cont(" %s", arp_ip_target[i]);
4620
4621 pr_cont("\n");
4622
4623 } else if (max_bonds) {
4624 /* miimon and arp_interval not set, we need one so things
4625 * work as expected, see bonding.txt for details
4626 */
4627 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");
4628 }
4629
4630 if (primary && !bond_mode_uses_primary(bond_mode)) {
4631 /* currently, using a primary only makes sense
4632 * in active backup, TLB or ALB modes
4633 */
4634 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4635 primary, bond_mode_name(bond_mode));
4636 primary = NULL;
4637 }
4638
4639 if (primary && primary_reselect) {
4640 bond_opt_initstr(&newval, primary_reselect);
4641 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4642 &newval);
4643 if (!valptr) {
4644 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4645 primary_reselect);
4646 return -EINVAL;
4647 }
4648 primary_reselect_value = valptr->value;
4649 } else {
4650 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4651 }
4652
4653 if (fail_over_mac) {
4654 bond_opt_initstr(&newval, fail_over_mac);
4655 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4656 &newval);
4657 if (!valptr) {
4658 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4659 fail_over_mac);
4660 return -EINVAL;
4661 }
4662 fail_over_mac_value = valptr->value;
4663 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4664 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4665 } else {
4666 fail_over_mac_value = BOND_FOM_NONE;
4667 }
4668
4669 bond_opt_initstr(&newval, "default");
4670 valptr = bond_opt_parse(
4671 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4672 &newval);
4673 if (!valptr) {
4674 pr_err("Error: No ad_actor_sys_prio default value");
4675 return -EINVAL;
4676 }
4677 ad_actor_sys_prio = valptr->value;
4678
4679 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4680 &newval);
4681 if (!valptr) {
4682 pr_err("Error: No ad_user_port_key default value");
4683 return -EINVAL;
4684 }
4685 ad_user_port_key = valptr->value;
4686
4687 bond_opt_initstr(&newval, "default");
4688 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4689 if (!valptr) {
4690 pr_err("Error: No tlb_dynamic_lb default value");
4691 return -EINVAL;
4692 }
4693 tlb_dynamic_lb = valptr->value;
4694
4695 if (lp_interval == 0) {
4696 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4697 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4698 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4699 }
4700
4701 /* fill params struct with the proper values */
4702 params->mode = bond_mode;
4703 params->xmit_policy = xmit_hashtype;
4704 params->miimon = miimon;
4705 params->num_peer_notif = num_peer_notif;
4706 params->arp_interval = arp_interval;
4707 params->arp_validate = arp_validate_value;
4708 params->arp_all_targets = arp_all_targets_value;
4709 params->updelay = updelay;
4710 params->downdelay = downdelay;
4711 params->peer_notif_delay = 0;
4712 params->use_carrier = use_carrier;
4713 params->lacp_fast = lacp_fast;
4714 params->primary[0] = 0;
4715 params->primary_reselect = primary_reselect_value;
4716 params->fail_over_mac = fail_over_mac_value;
4717 params->tx_queues = tx_queues;
4718 params->all_slaves_active = all_slaves_active;
4719 params->resend_igmp = resend_igmp;
4720 params->min_links = min_links;
4721 params->lp_interval = lp_interval;
4722 params->packets_per_slave = packets_per_slave;
4723 params->tlb_dynamic_lb = tlb_dynamic_lb;
4724 params->ad_actor_sys_prio = ad_actor_sys_prio;
4725 eth_zero_addr(params->ad_actor_system);
4726 params->ad_user_port_key = ad_user_port_key;
4727 if (packets_per_slave > 0) {
4728 params->reciprocal_packets_per_slave =
4729 reciprocal_value(packets_per_slave);
4730 } else {
4731 /* reciprocal_packets_per_slave is unused if
4732 * packets_per_slave is 0 or 1, just initialize it
4733 */
4734 params->reciprocal_packets_per_slave =
4735 (struct reciprocal_value) { 0 };
4736 }
4737
4738 if (primary) {
4739 strncpy(params->primary, primary, IFNAMSIZ);
4740 params->primary[IFNAMSIZ - 1] = 0;
4741 }
4742
4743 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4744
4745 return 0;
4746}
4747
4748/* Called from registration process */
4749static int bond_init(struct net_device *bond_dev)
4750{
4751 struct bonding *bond = netdev_priv(bond_dev);
4752 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4753
4754 netdev_dbg(bond_dev, "Begin bond_init\n");
4755
4756 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4757 if (!bond->wq)
4758 return -ENOMEM;
4759
4760 spin_lock_init(&bond->stats_lock);
4761 lockdep_register_key(&bond->stats_lock_key);
4762 lockdep_set_class(&bond->stats_lock, &bond->stats_lock_key);
4763
4764 list_add_tail(&bond->bond_list, &bn->dev_list);
4765
4766 bond_prepare_sysfs_group(bond);
4767
4768 bond_debug_register(bond);
4769
4770 /* Ensure valid dev_addr */
4771 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4772 bond_dev->addr_assign_type == NET_ADDR_PERM)
4773 eth_hw_addr_random(bond_dev);
4774
4775 return 0;
4776}
4777
4778unsigned int bond_get_num_tx_queues(void)
4779{
4780 return tx_queues;
4781}
4782
4783/* Create a new bond based on the specified name and bonding parameters.
4784 * If name is NULL, obtain a suitable "bond%d" name for us.
4785 * Caller must NOT hold rtnl_lock; we need to release it here before we
4786 * set up our sysfs entries.
4787 */
4788int bond_create(struct net *net, const char *name)
4789{
4790 struct net_device *bond_dev;
4791 struct bonding *bond;
4792 struct alb_bond_info *bond_info;
4793 int res;
4794
4795 rtnl_lock();
4796
4797 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4798 name ? name : "bond%d", NET_NAME_UNKNOWN,
4799 bond_setup, tx_queues);
4800 if (!bond_dev) {
4801 pr_err("%s: eek! can't alloc netdev!\n", name);
4802 rtnl_unlock();
4803 return -ENOMEM;
4804 }
4805
4806 /*
4807 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4808 * It is set to 0 by default which is wrong.
4809 */
4810 bond = netdev_priv(bond_dev);
4811 bond_info = &(BOND_ALB_INFO(bond));
4812 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4813
4814 dev_net_set(bond_dev, net);
4815 bond_dev->rtnl_link_ops = &bond_link_ops;
4816
4817 res = register_netdevice(bond_dev);
4818
4819 netif_carrier_off(bond_dev);
4820
4821 bond_work_init_all(bond);
4822
4823 rtnl_unlock();
4824 if (res < 0)
4825 free_netdev(bond_dev);
4826 return res;
4827}
4828
4829static int __net_init bond_net_init(struct net *net)
4830{
4831 struct bond_net *bn = net_generic(net, bond_net_id);
4832
4833 bn->net = net;
4834 INIT_LIST_HEAD(&bn->dev_list);
4835
4836 bond_create_proc_dir(bn);
4837 bond_create_sysfs(bn);
4838
4839 return 0;
4840}
4841
4842static void __net_exit bond_net_exit(struct net *net)
4843{
4844 struct bond_net *bn = net_generic(net, bond_net_id);
4845 struct bonding *bond, *tmp_bond;
4846 LIST_HEAD(list);
4847
4848 bond_destroy_sysfs(bn);
4849
4850 /* Kill off any bonds created after unregistering bond rtnl ops */
4851 rtnl_lock();
4852 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4853 unregister_netdevice_queue(bond->dev, &list);
4854 unregister_netdevice_many(&list);
4855 rtnl_unlock();
4856
4857 bond_destroy_proc_dir(bn);
4858}
4859
4860static struct pernet_operations bond_net_ops = {
4861 .init = bond_net_init,
4862 .exit = bond_net_exit,
4863 .id = &bond_net_id,
4864 .size = sizeof(struct bond_net),
4865};
4866
4867static int __init bonding_init(void)
4868{
4869 int i;
4870 int res;
4871
4872 pr_info("%s", bond_version);
4873
4874 res = bond_check_params(&bonding_defaults);
4875 if (res)
4876 goto out;
4877
4878 res = register_pernet_subsys(&bond_net_ops);
4879 if (res)
4880 goto out;
4881
4882 res = bond_netlink_init();
4883 if (res)
4884 goto err_link;
4885
4886 bond_create_debugfs();
4887
4888 for (i = 0; i < max_bonds; i++) {
4889 res = bond_create(&init_net, NULL);
4890 if (res)
4891 goto err;
4892 }
4893
4894 register_netdevice_notifier(&bond_netdev_notifier);
4895out:
4896 return res;
4897err:
4898 bond_destroy_debugfs();
4899 bond_netlink_fini();
4900err_link:
4901 unregister_pernet_subsys(&bond_net_ops);
4902 goto out;
4903
4904}
4905
4906static void __exit bonding_exit(void)
4907{
4908 unregister_netdevice_notifier(&bond_netdev_notifier);
4909
4910 bond_destroy_debugfs();
4911
4912 bond_netlink_fini();
4913 unregister_pernet_subsys(&bond_net_ops);
4914
4915#ifdef CONFIG_NET_POLL_CONTROLLER
4916 /* Make sure we don't have an imbalance on our netpoll blocking */
4917 WARN_ON(atomic_read(&netpoll_block_tx));
4918#endif
4919}
4920
4921module_init(bonding_init);
4922module_exit(bonding_exit);
4923MODULE_LICENSE("GPL");
4924MODULE_VERSION(DRV_VERSION);
4925MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4926MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
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/icmp.h>
45#include <linux/icmpv6.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <linux/io.h>
57#include <asm/dma.h>
58#include <linux/uaccess.h>
59#include <linux/errno.h>
60#include <linux/netdevice.h>
61#include <linux/inetdevice.h>
62#include <linux/igmp.h>
63#include <linux/etherdevice.h>
64#include <linux/skbuff.h>
65#include <net/sock.h>
66#include <linux/rtnetlink.h>
67#include <linux/smp.h>
68#include <linux/if_ether.h>
69#include <net/arp.h>
70#include <linux/mii.h>
71#include <linux/ethtool.h>
72#include <linux/if_vlan.h>
73#include <linux/if_bonding.h>
74#include <linux/jiffies.h>
75#include <linux/preempt.h>
76#include <net/route.h>
77#include <net/net_namespace.h>
78#include <net/netns/generic.h>
79#include <net/pkt_sched.h>
80#include <linux/rculist.h>
81#include <net/flow_dissector.h>
82#include <net/xfrm.h>
83#include <net/bonding.h>
84#include <net/bond_3ad.h>
85#include <net/bond_alb.h>
86#if IS_ENABLED(CONFIG_TLS_DEVICE)
87#include <net/tls.h>
88#endif
89
90#include "bonding_priv.h"
91
92/*---------------------------- Module parameters ----------------------------*/
93
94/* monitor all links that often (in milliseconds). <=0 disables monitoring */
95
96static int max_bonds = BOND_DEFAULT_MAX_BONDS;
97static int tx_queues = BOND_DEFAULT_TX_QUEUES;
98static int num_peer_notif = 1;
99static int miimon;
100static int updelay;
101static int downdelay;
102static int use_carrier = 1;
103static char *mode;
104static char *primary;
105static char *primary_reselect;
106static char *lacp_rate;
107static int min_links;
108static char *ad_select;
109static char *xmit_hash_policy;
110static int arp_interval;
111static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
112static char *arp_validate;
113static char *arp_all_targets;
114static char *fail_over_mac;
115static int all_slaves_active;
116static struct bond_params bonding_defaults;
117static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
118static int packets_per_slave = 1;
119static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
120
121module_param(max_bonds, int, 0);
122MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
123module_param(tx_queues, int, 0);
124MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
125module_param_named(num_grat_arp, num_peer_notif, int, 0644);
126MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
127 "failover event (alias of num_unsol_na)");
128module_param_named(num_unsol_na, num_peer_notif, int, 0644);
129MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
130 "failover event (alias of num_grat_arp)");
131module_param(miimon, int, 0);
132MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
133module_param(updelay, int, 0);
134MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
135module_param(downdelay, int, 0);
136MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
137 "in milliseconds");
138module_param(use_carrier, int, 0);
139MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
140 "0 for off, 1 for on (default)");
141module_param(mode, charp, 0);
142MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
143 "1 for active-backup, 2 for balance-xor, "
144 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
145 "6 for balance-alb");
146module_param(primary, charp, 0);
147MODULE_PARM_DESC(primary, "Primary network device to use");
148module_param(primary_reselect, charp, 0);
149MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
150 "once it comes up; "
151 "0 for always (default), "
152 "1 for only if speed of primary is "
153 "better, "
154 "2 for only on active slave "
155 "failure");
156module_param(lacp_rate, charp, 0);
157MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
158 "0 for slow, 1 for fast");
159module_param(ad_select, charp, 0);
160MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
161 "0 for stable (default), 1 for bandwidth, "
162 "2 for count");
163module_param(min_links, int, 0);
164MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
165
166module_param(xmit_hash_policy, charp, 0);
167MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
168 "0 for layer 2 (default), 1 for layer 3+4, "
169 "2 for layer 2+3, 3 for encap layer 2+3, "
170 "4 for encap layer 3+4, 5 for vlan+srcmac");
171module_param(arp_interval, int, 0);
172MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
173module_param_array(arp_ip_target, charp, NULL, 0);
174MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
175module_param(arp_validate, charp, 0);
176MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
177 "0 for none (default), 1 for active, "
178 "2 for backup, 3 for all");
179module_param(arp_all_targets, charp, 0);
180MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
181module_param(fail_over_mac, charp, 0);
182MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
183 "the same MAC; 0 for none (default), "
184 "1 for active, 2 for follow");
185module_param(all_slaves_active, int, 0);
186MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
187 "by setting active flag for all slaves; "
188 "0 for never (default), 1 for always.");
189module_param(resend_igmp, int, 0);
190MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
191 "link failure");
192module_param(packets_per_slave, int, 0);
193MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
194 "mode; 0 for a random slave, 1 packet per "
195 "slave (default), >1 packets per slave.");
196module_param(lp_interval, uint, 0);
197MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
198 "the bonding driver sends learning packets to "
199 "each slaves peer switch. The default is 1.");
200
201/*----------------------------- Global variables ----------------------------*/
202
203#ifdef CONFIG_NET_POLL_CONTROLLER
204atomic_t netpoll_block_tx = ATOMIC_INIT(0);
205#endif
206
207unsigned int bond_net_id __read_mostly;
208
209static const struct flow_dissector_key flow_keys_bonding_keys[] = {
210 {
211 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
212 .offset = offsetof(struct flow_keys, control),
213 },
214 {
215 .key_id = FLOW_DISSECTOR_KEY_BASIC,
216 .offset = offsetof(struct flow_keys, basic),
217 },
218 {
219 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
220 .offset = offsetof(struct flow_keys, addrs.v4addrs),
221 },
222 {
223 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
224 .offset = offsetof(struct flow_keys, addrs.v6addrs),
225 },
226 {
227 .key_id = FLOW_DISSECTOR_KEY_TIPC,
228 .offset = offsetof(struct flow_keys, addrs.tipckey),
229 },
230 {
231 .key_id = FLOW_DISSECTOR_KEY_PORTS,
232 .offset = offsetof(struct flow_keys, ports),
233 },
234 {
235 .key_id = FLOW_DISSECTOR_KEY_ICMP,
236 .offset = offsetof(struct flow_keys, icmp),
237 },
238 {
239 .key_id = FLOW_DISSECTOR_KEY_VLAN,
240 .offset = offsetof(struct flow_keys, vlan),
241 },
242 {
243 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
244 .offset = offsetof(struct flow_keys, tags),
245 },
246 {
247 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
248 .offset = offsetof(struct flow_keys, keyid),
249 },
250};
251
252static struct flow_dissector flow_keys_bonding __read_mostly;
253
254/*-------------------------- Forward declarations ---------------------------*/
255
256static int bond_init(struct net_device *bond_dev);
257static void bond_uninit(struct net_device *bond_dev);
258static void bond_get_stats(struct net_device *bond_dev,
259 struct rtnl_link_stats64 *stats);
260static void bond_slave_arr_handler(struct work_struct *work);
261static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
262 int mod);
263static void bond_netdev_notify_work(struct work_struct *work);
264
265/*---------------------------- General routines -----------------------------*/
266
267const char *bond_mode_name(int mode)
268{
269 static const char *names[] = {
270 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
271 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
272 [BOND_MODE_XOR] = "load balancing (xor)",
273 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
274 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
275 [BOND_MODE_TLB] = "transmit load balancing",
276 [BOND_MODE_ALB] = "adaptive load balancing",
277 };
278
279 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
280 return "unknown";
281
282 return names[mode];
283}
284
285/**
286 * bond_dev_queue_xmit - Prepare skb for xmit.
287 *
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
291 */
292netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
294{
295 skb->dev = slave_dev;
296
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
300
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
303
304 return dev_queue_xmit(skb);
305}
306
307bool bond_sk_check(struct bonding *bond)
308{
309 switch (BOND_MODE(bond)) {
310 case BOND_MODE_8023AD:
311 case BOND_MODE_XOR:
312 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
313 return true;
314 fallthrough;
315 default:
316 return false;
317 }
318}
319
320/*---------------------------------- VLAN -----------------------------------*/
321
322/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
323 * We don't protect the slave list iteration with a lock because:
324 * a. This operation is performed in IOCTL context,
325 * b. The operation is protected by the RTNL semaphore in the 8021q code,
326 * c. Holding a lock with BH disabled while directly calling a base driver
327 * entry point is generally a BAD idea.
328 *
329 * The design of synchronization/protection for this operation in the 8021q
330 * module is good for one or more VLAN devices over a single physical device
331 * and cannot be extended for a teaming solution like bonding, so there is a
332 * potential race condition here where a net device from the vlan group might
333 * be referenced (either by a base driver or the 8021q code) while it is being
334 * removed from the system. However, it turns out we're not making matters
335 * worse, and if it works for regular VLAN usage it will work here too.
336*/
337
338/**
339 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
340 * @bond_dev: bonding net device that got called
341 * @proto: network protocol ID
342 * @vid: vlan id being added
343 */
344static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
345 __be16 proto, u16 vid)
346{
347 struct bonding *bond = netdev_priv(bond_dev);
348 struct slave *slave, *rollback_slave;
349 struct list_head *iter;
350 int res;
351
352 bond_for_each_slave(bond, slave, iter) {
353 res = vlan_vid_add(slave->dev, proto, vid);
354 if (res)
355 goto unwind;
356 }
357
358 return 0;
359
360unwind:
361 /* unwind to the slave that failed */
362 bond_for_each_slave(bond, rollback_slave, iter) {
363 if (rollback_slave == slave)
364 break;
365
366 vlan_vid_del(rollback_slave->dev, proto, vid);
367 }
368
369 return res;
370}
371
372/**
373 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
374 * @bond_dev: bonding net device that got called
375 * @proto: network protocol ID
376 * @vid: vlan id being removed
377 */
378static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
379 __be16 proto, u16 vid)
380{
381 struct bonding *bond = netdev_priv(bond_dev);
382 struct list_head *iter;
383 struct slave *slave;
384
385 bond_for_each_slave(bond, slave, iter)
386 vlan_vid_del(slave->dev, proto, vid);
387
388 if (bond_is_lb(bond))
389 bond_alb_clear_vlan(bond, vid);
390
391 return 0;
392}
393
394/*---------------------------------- XFRM -----------------------------------*/
395
396#ifdef CONFIG_XFRM_OFFLOAD
397/**
398 * bond_ipsec_add_sa - program device with a security association
399 * @xs: pointer to transformer state struct
400 **/
401static int bond_ipsec_add_sa(struct xfrm_state *xs)
402{
403 struct net_device *bond_dev = xs->xso.dev;
404 struct bond_ipsec *ipsec;
405 struct bonding *bond;
406 struct slave *slave;
407 int err;
408
409 if (!bond_dev)
410 return -EINVAL;
411
412 rcu_read_lock();
413 bond = netdev_priv(bond_dev);
414 slave = rcu_dereference(bond->curr_active_slave);
415 if (!slave) {
416 rcu_read_unlock();
417 return -ENODEV;
418 }
419
420 if (!slave->dev->xfrmdev_ops ||
421 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
422 netif_is_bond_master(slave->dev)) {
423 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
424 rcu_read_unlock();
425 return -EINVAL;
426 }
427
428 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
429 if (!ipsec) {
430 rcu_read_unlock();
431 return -ENOMEM;
432 }
433 xs->xso.real_dev = slave->dev;
434
435 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
436 if (!err) {
437 ipsec->xs = xs;
438 INIT_LIST_HEAD(&ipsec->list);
439 spin_lock_bh(&bond->ipsec_lock);
440 list_add(&ipsec->list, &bond->ipsec_list);
441 spin_unlock_bh(&bond->ipsec_lock);
442 } else {
443 kfree(ipsec);
444 }
445 rcu_read_unlock();
446 return err;
447}
448
449static void bond_ipsec_add_sa_all(struct bonding *bond)
450{
451 struct net_device *bond_dev = bond->dev;
452 struct bond_ipsec *ipsec;
453 struct slave *slave;
454
455 rcu_read_lock();
456 slave = rcu_dereference(bond->curr_active_slave);
457 if (!slave)
458 goto out;
459
460 if (!slave->dev->xfrmdev_ops ||
461 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
462 netif_is_bond_master(slave->dev)) {
463 spin_lock_bh(&bond->ipsec_lock);
464 if (!list_empty(&bond->ipsec_list))
465 slave_warn(bond_dev, slave->dev,
466 "%s: no slave xdo_dev_state_add\n",
467 __func__);
468 spin_unlock_bh(&bond->ipsec_lock);
469 goto out;
470 }
471
472 spin_lock_bh(&bond->ipsec_lock);
473 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
474 ipsec->xs->xso.real_dev = slave->dev;
475 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
476 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
477 ipsec->xs->xso.real_dev = NULL;
478 }
479 }
480 spin_unlock_bh(&bond->ipsec_lock);
481out:
482 rcu_read_unlock();
483}
484
485/**
486 * bond_ipsec_del_sa - clear out this specific SA
487 * @xs: pointer to transformer state struct
488 **/
489static void bond_ipsec_del_sa(struct xfrm_state *xs)
490{
491 struct net_device *bond_dev = xs->xso.dev;
492 struct bond_ipsec *ipsec;
493 struct bonding *bond;
494 struct slave *slave;
495
496 if (!bond_dev)
497 return;
498
499 rcu_read_lock();
500 bond = netdev_priv(bond_dev);
501 slave = rcu_dereference(bond->curr_active_slave);
502
503 if (!slave)
504 goto out;
505
506 if (!xs->xso.real_dev)
507 goto out;
508
509 WARN_ON(xs->xso.real_dev != slave->dev);
510
511 if (!slave->dev->xfrmdev_ops ||
512 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
513 netif_is_bond_master(slave->dev)) {
514 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
515 goto out;
516 }
517
518 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
519out:
520 spin_lock_bh(&bond->ipsec_lock);
521 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
522 if (ipsec->xs == xs) {
523 list_del(&ipsec->list);
524 kfree(ipsec);
525 break;
526 }
527 }
528 spin_unlock_bh(&bond->ipsec_lock);
529 rcu_read_unlock();
530}
531
532static void bond_ipsec_del_sa_all(struct bonding *bond)
533{
534 struct net_device *bond_dev = bond->dev;
535 struct bond_ipsec *ipsec;
536 struct slave *slave;
537
538 rcu_read_lock();
539 slave = rcu_dereference(bond->curr_active_slave);
540 if (!slave) {
541 rcu_read_unlock();
542 return;
543 }
544
545 spin_lock_bh(&bond->ipsec_lock);
546 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
547 if (!ipsec->xs->xso.real_dev)
548 continue;
549
550 if (!slave->dev->xfrmdev_ops ||
551 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
552 netif_is_bond_master(slave->dev)) {
553 slave_warn(bond_dev, slave->dev,
554 "%s: no slave xdo_dev_state_delete\n",
555 __func__);
556 } else {
557 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
558 }
559 ipsec->xs->xso.real_dev = NULL;
560 }
561 spin_unlock_bh(&bond->ipsec_lock);
562 rcu_read_unlock();
563}
564
565/**
566 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
567 * @skb: current data packet
568 * @xs: pointer to transformer state struct
569 **/
570static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
571{
572 struct net_device *bond_dev = xs->xso.dev;
573 struct net_device *real_dev;
574 struct slave *curr_active;
575 struct bonding *bond;
576 int err;
577
578 bond = netdev_priv(bond_dev);
579 rcu_read_lock();
580 curr_active = rcu_dereference(bond->curr_active_slave);
581 real_dev = curr_active->dev;
582
583 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
584 err = false;
585 goto out;
586 }
587
588 if (!xs->xso.real_dev) {
589 err = false;
590 goto out;
591 }
592
593 if (!real_dev->xfrmdev_ops ||
594 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
595 netif_is_bond_master(real_dev)) {
596 err = false;
597 goto out;
598 }
599
600 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
601out:
602 rcu_read_unlock();
603 return err;
604}
605
606static const struct xfrmdev_ops bond_xfrmdev_ops = {
607 .xdo_dev_state_add = bond_ipsec_add_sa,
608 .xdo_dev_state_delete = bond_ipsec_del_sa,
609 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
610};
611#endif /* CONFIG_XFRM_OFFLOAD */
612
613/*------------------------------- Link status -------------------------------*/
614
615/* Set the carrier state for the master according to the state of its
616 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
617 * do special 802.3ad magic.
618 *
619 * Returns zero if carrier state does not change, nonzero if it does.
620 */
621int bond_set_carrier(struct bonding *bond)
622{
623 struct list_head *iter;
624 struct slave *slave;
625
626 if (!bond_has_slaves(bond))
627 goto down;
628
629 if (BOND_MODE(bond) == BOND_MODE_8023AD)
630 return bond_3ad_set_carrier(bond);
631
632 bond_for_each_slave(bond, slave, iter) {
633 if (slave->link == BOND_LINK_UP) {
634 if (!netif_carrier_ok(bond->dev)) {
635 netif_carrier_on(bond->dev);
636 return 1;
637 }
638 return 0;
639 }
640 }
641
642down:
643 if (netif_carrier_ok(bond->dev)) {
644 netif_carrier_off(bond->dev);
645 return 1;
646 }
647 return 0;
648}
649
650/* Get link speed and duplex from the slave's base driver
651 * using ethtool. If for some reason the call fails or the
652 * values are invalid, set speed and duplex to -1,
653 * and return. Return 1 if speed or duplex settings are
654 * UNKNOWN; 0 otherwise.
655 */
656static int bond_update_speed_duplex(struct slave *slave)
657{
658 struct net_device *slave_dev = slave->dev;
659 struct ethtool_link_ksettings ecmd;
660 int res;
661
662 slave->speed = SPEED_UNKNOWN;
663 slave->duplex = DUPLEX_UNKNOWN;
664
665 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
666 if (res < 0)
667 return 1;
668 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
669 return 1;
670 switch (ecmd.base.duplex) {
671 case DUPLEX_FULL:
672 case DUPLEX_HALF:
673 break;
674 default:
675 return 1;
676 }
677
678 slave->speed = ecmd.base.speed;
679 slave->duplex = ecmd.base.duplex;
680
681 return 0;
682}
683
684const char *bond_slave_link_status(s8 link)
685{
686 switch (link) {
687 case BOND_LINK_UP:
688 return "up";
689 case BOND_LINK_FAIL:
690 return "going down";
691 case BOND_LINK_DOWN:
692 return "down";
693 case BOND_LINK_BACK:
694 return "going back";
695 default:
696 return "unknown";
697 }
698}
699
700/* if <dev> supports MII link status reporting, check its link status.
701 *
702 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
703 * depending upon the setting of the use_carrier parameter.
704 *
705 * Return either BMSR_LSTATUS, meaning that the link is up (or we
706 * can't tell and just pretend it is), or 0, meaning that the link is
707 * down.
708 *
709 * If reporting is non-zero, instead of faking link up, return -1 if
710 * both ETHTOOL and MII ioctls fail (meaning the device does not
711 * support them). If use_carrier is set, return whatever it says.
712 * It'd be nice if there was a good way to tell if a driver supports
713 * netif_carrier, but there really isn't.
714 */
715static int bond_check_dev_link(struct bonding *bond,
716 struct net_device *slave_dev, int reporting)
717{
718 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
719 int (*ioctl)(struct net_device *, struct ifreq *, int);
720 struct ifreq ifr;
721 struct mii_ioctl_data *mii;
722
723 if (!reporting && !netif_running(slave_dev))
724 return 0;
725
726 if (bond->params.use_carrier)
727 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
728
729 /* Try to get link status using Ethtool first. */
730 if (slave_dev->ethtool_ops->get_link)
731 return slave_dev->ethtool_ops->get_link(slave_dev) ?
732 BMSR_LSTATUS : 0;
733
734 /* Ethtool can't be used, fallback to MII ioctls. */
735 ioctl = slave_ops->ndo_do_ioctl;
736 if (ioctl) {
737 /* TODO: set pointer to correct ioctl on a per team member
738 * bases to make this more efficient. that is, once
739 * we determine the correct ioctl, we will always
740 * call it and not the others for that team
741 * member.
742 */
743
744 /* We cannot assume that SIOCGMIIPHY will also read a
745 * register; not all network drivers (e.g., e100)
746 * support that.
747 */
748
749 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
750 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
751 mii = if_mii(&ifr);
752 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
753 mii->reg_num = MII_BMSR;
754 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
755 return mii->val_out & BMSR_LSTATUS;
756 }
757 }
758
759 /* If reporting, report that either there's no dev->do_ioctl,
760 * or both SIOCGMIIREG and get_link failed (meaning that we
761 * cannot report link status). If not reporting, pretend
762 * we're ok.
763 */
764 return reporting ? -1 : BMSR_LSTATUS;
765}
766
767/*----------------------------- Multicast list ------------------------------*/
768
769/* Push the promiscuity flag down to appropriate slaves */
770static int bond_set_promiscuity(struct bonding *bond, int inc)
771{
772 struct list_head *iter;
773 int err = 0;
774
775 if (bond_uses_primary(bond)) {
776 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
777
778 if (curr_active)
779 err = dev_set_promiscuity(curr_active->dev, inc);
780 } else {
781 struct slave *slave;
782
783 bond_for_each_slave(bond, slave, iter) {
784 err = dev_set_promiscuity(slave->dev, inc);
785 if (err)
786 return err;
787 }
788 }
789 return err;
790}
791
792/* Push the allmulti flag down to all slaves */
793static int bond_set_allmulti(struct bonding *bond, int inc)
794{
795 struct list_head *iter;
796 int err = 0;
797
798 if (bond_uses_primary(bond)) {
799 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
800
801 if (curr_active)
802 err = dev_set_allmulti(curr_active->dev, inc);
803 } else {
804 struct slave *slave;
805
806 bond_for_each_slave(bond, slave, iter) {
807 err = dev_set_allmulti(slave->dev, inc);
808 if (err)
809 return err;
810 }
811 }
812 return err;
813}
814
815/* Retrieve the list of registered multicast addresses for the bonding
816 * device and retransmit an IGMP JOIN request to the current active
817 * slave.
818 */
819static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
820{
821 struct bonding *bond = container_of(work, struct bonding,
822 mcast_work.work);
823
824 if (!rtnl_trylock()) {
825 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
826 return;
827 }
828 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
829
830 if (bond->igmp_retrans > 1) {
831 bond->igmp_retrans--;
832 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
833 }
834 rtnl_unlock();
835}
836
837/* Flush bond's hardware addresses from slave */
838static void bond_hw_addr_flush(struct net_device *bond_dev,
839 struct net_device *slave_dev)
840{
841 struct bonding *bond = netdev_priv(bond_dev);
842
843 dev_uc_unsync(slave_dev, bond_dev);
844 dev_mc_unsync(slave_dev, bond_dev);
845
846 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
847 /* del lacpdu mc addr from mc list */
848 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
849
850 dev_mc_del(slave_dev, lacpdu_multicast);
851 }
852}
853
854/*--------------------------- Active slave change ---------------------------*/
855
856/* Update the hardware address list and promisc/allmulti for the new and
857 * old active slaves (if any). Modes that are not using primary keep all
858 * slaves up date at all times; only the modes that use primary need to call
859 * this function to swap these settings during a failover.
860 */
861static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
862 struct slave *old_active)
863{
864 if (old_active) {
865 if (bond->dev->flags & IFF_PROMISC)
866 dev_set_promiscuity(old_active->dev, -1);
867
868 if (bond->dev->flags & IFF_ALLMULTI)
869 dev_set_allmulti(old_active->dev, -1);
870
871 bond_hw_addr_flush(bond->dev, old_active->dev);
872 }
873
874 if (new_active) {
875 /* FIXME: Signal errors upstream. */
876 if (bond->dev->flags & IFF_PROMISC)
877 dev_set_promiscuity(new_active->dev, 1);
878
879 if (bond->dev->flags & IFF_ALLMULTI)
880 dev_set_allmulti(new_active->dev, 1);
881
882 netif_addr_lock_bh(bond->dev);
883 dev_uc_sync(new_active->dev, bond->dev);
884 dev_mc_sync(new_active->dev, bond->dev);
885 netif_addr_unlock_bh(bond->dev);
886 }
887}
888
889/**
890 * bond_set_dev_addr - clone slave's address to bond
891 * @bond_dev: bond net device
892 * @slave_dev: slave net device
893 *
894 * Should be called with RTNL held.
895 */
896static int bond_set_dev_addr(struct net_device *bond_dev,
897 struct net_device *slave_dev)
898{
899 int err;
900
901 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
902 bond_dev, slave_dev, slave_dev->addr_len);
903 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
904 if (err)
905 return err;
906
907 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
908 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
909 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
910 return 0;
911}
912
913static struct slave *bond_get_old_active(struct bonding *bond,
914 struct slave *new_active)
915{
916 struct slave *slave;
917 struct list_head *iter;
918
919 bond_for_each_slave(bond, slave, iter) {
920 if (slave == new_active)
921 continue;
922
923 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
924 return slave;
925 }
926
927 return NULL;
928}
929
930/* bond_do_fail_over_mac
931 *
932 * Perform special MAC address swapping for fail_over_mac settings
933 *
934 * Called with RTNL
935 */
936static void bond_do_fail_over_mac(struct bonding *bond,
937 struct slave *new_active,
938 struct slave *old_active)
939{
940 u8 tmp_mac[MAX_ADDR_LEN];
941 struct sockaddr_storage ss;
942 int rv;
943
944 switch (bond->params.fail_over_mac) {
945 case BOND_FOM_ACTIVE:
946 if (new_active) {
947 rv = bond_set_dev_addr(bond->dev, new_active->dev);
948 if (rv)
949 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
950 -rv);
951 }
952 break;
953 case BOND_FOM_FOLLOW:
954 /* if new_active && old_active, swap them
955 * if just old_active, do nothing (going to no active slave)
956 * if just new_active, set new_active to bond's MAC
957 */
958 if (!new_active)
959 return;
960
961 if (!old_active)
962 old_active = bond_get_old_active(bond, new_active);
963
964 if (old_active) {
965 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
966 new_active->dev->addr_len);
967 bond_hw_addr_copy(ss.__data,
968 old_active->dev->dev_addr,
969 old_active->dev->addr_len);
970 ss.ss_family = new_active->dev->type;
971 } else {
972 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
973 bond->dev->addr_len);
974 ss.ss_family = bond->dev->type;
975 }
976
977 rv = dev_set_mac_address(new_active->dev,
978 (struct sockaddr *)&ss, NULL);
979 if (rv) {
980 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
981 -rv);
982 goto out;
983 }
984
985 if (!old_active)
986 goto out;
987
988 bond_hw_addr_copy(ss.__data, tmp_mac,
989 new_active->dev->addr_len);
990 ss.ss_family = old_active->dev->type;
991
992 rv = dev_set_mac_address(old_active->dev,
993 (struct sockaddr *)&ss, NULL);
994 if (rv)
995 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
996 -rv);
997out:
998 break;
999 default:
1000 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1001 bond->params.fail_over_mac);
1002 break;
1003 }
1004
1005}
1006
1007static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1008{
1009 struct slave *prim = rtnl_dereference(bond->primary_slave);
1010 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1011
1012 if (!prim || prim->link != BOND_LINK_UP) {
1013 if (!curr || curr->link != BOND_LINK_UP)
1014 return NULL;
1015 return curr;
1016 }
1017
1018 if (bond->force_primary) {
1019 bond->force_primary = false;
1020 return prim;
1021 }
1022
1023 if (!curr || curr->link != BOND_LINK_UP)
1024 return prim;
1025
1026 /* At this point, prim and curr are both up */
1027 switch (bond->params.primary_reselect) {
1028 case BOND_PRI_RESELECT_ALWAYS:
1029 return prim;
1030 case BOND_PRI_RESELECT_BETTER:
1031 if (prim->speed < curr->speed)
1032 return curr;
1033 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1034 return curr;
1035 return prim;
1036 case BOND_PRI_RESELECT_FAILURE:
1037 return curr;
1038 default:
1039 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1040 bond->params.primary_reselect);
1041 return curr;
1042 }
1043}
1044
1045/**
1046 * bond_find_best_slave - select the best available slave to be the active one
1047 * @bond: our bonding struct
1048 */
1049static struct slave *bond_find_best_slave(struct bonding *bond)
1050{
1051 struct slave *slave, *bestslave = NULL;
1052 struct list_head *iter;
1053 int mintime = bond->params.updelay;
1054
1055 slave = bond_choose_primary_or_current(bond);
1056 if (slave)
1057 return slave;
1058
1059 bond_for_each_slave(bond, slave, iter) {
1060 if (slave->link == BOND_LINK_UP)
1061 return slave;
1062 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1063 slave->delay < mintime) {
1064 mintime = slave->delay;
1065 bestslave = slave;
1066 }
1067 }
1068
1069 return bestslave;
1070}
1071
1072static bool bond_should_notify_peers(struct bonding *bond)
1073{
1074 struct slave *slave;
1075
1076 rcu_read_lock();
1077 slave = rcu_dereference(bond->curr_active_slave);
1078 rcu_read_unlock();
1079
1080 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1081 slave ? slave->dev->name : "NULL");
1082
1083 if (!slave || !bond->send_peer_notif ||
1084 bond->send_peer_notif %
1085 max(1, bond->params.peer_notif_delay) != 0 ||
1086 !netif_carrier_ok(bond->dev) ||
1087 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1088 return false;
1089
1090 return true;
1091}
1092
1093/**
1094 * bond_change_active_slave - change the active slave into the specified one
1095 * @bond: our bonding struct
1096 * @new_active: the new slave to make the active one
1097 *
1098 * Set the new slave to the bond's settings and unset them on the old
1099 * curr_active_slave.
1100 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1101 *
1102 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1103 * because it is apparently the best available slave we have, even though its
1104 * updelay hasn't timed out yet.
1105 *
1106 * Caller must hold RTNL.
1107 */
1108void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1109{
1110 struct slave *old_active;
1111
1112 ASSERT_RTNL();
1113
1114 old_active = rtnl_dereference(bond->curr_active_slave);
1115
1116 if (old_active == new_active)
1117 return;
1118
1119#ifdef CONFIG_XFRM_OFFLOAD
1120 bond_ipsec_del_sa_all(bond);
1121#endif /* CONFIG_XFRM_OFFLOAD */
1122
1123 if (new_active) {
1124 new_active->last_link_up = jiffies;
1125
1126 if (new_active->link == BOND_LINK_BACK) {
1127 if (bond_uses_primary(bond)) {
1128 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1129 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1130 }
1131
1132 new_active->delay = 0;
1133 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1134 BOND_SLAVE_NOTIFY_NOW);
1135
1136 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1137 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1138
1139 if (bond_is_lb(bond))
1140 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1141 } else {
1142 if (bond_uses_primary(bond))
1143 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1144 }
1145 }
1146
1147 if (bond_uses_primary(bond))
1148 bond_hw_addr_swap(bond, new_active, old_active);
1149
1150 if (bond_is_lb(bond)) {
1151 bond_alb_handle_active_change(bond, new_active);
1152 if (old_active)
1153 bond_set_slave_inactive_flags(old_active,
1154 BOND_SLAVE_NOTIFY_NOW);
1155 if (new_active)
1156 bond_set_slave_active_flags(new_active,
1157 BOND_SLAVE_NOTIFY_NOW);
1158 } else {
1159 rcu_assign_pointer(bond->curr_active_slave, new_active);
1160 }
1161
1162 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1163 if (old_active)
1164 bond_set_slave_inactive_flags(old_active,
1165 BOND_SLAVE_NOTIFY_NOW);
1166
1167 if (new_active) {
1168 bool should_notify_peers = false;
1169
1170 bond_set_slave_active_flags(new_active,
1171 BOND_SLAVE_NOTIFY_NOW);
1172
1173 if (bond->params.fail_over_mac)
1174 bond_do_fail_over_mac(bond, new_active,
1175 old_active);
1176
1177 if (netif_running(bond->dev)) {
1178 bond->send_peer_notif =
1179 bond->params.num_peer_notif *
1180 max(1, bond->params.peer_notif_delay);
1181 should_notify_peers =
1182 bond_should_notify_peers(bond);
1183 }
1184
1185 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1186 if (should_notify_peers) {
1187 bond->send_peer_notif--;
1188 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1189 bond->dev);
1190 }
1191 }
1192 }
1193
1194#ifdef CONFIG_XFRM_OFFLOAD
1195 bond_ipsec_add_sa_all(bond);
1196#endif /* CONFIG_XFRM_OFFLOAD */
1197
1198 /* resend IGMP joins since active slave has changed or
1199 * all were sent on curr_active_slave.
1200 * resend only if bond is brought up with the affected
1201 * bonding modes and the retransmission is enabled
1202 */
1203 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1204 ((bond_uses_primary(bond) && new_active) ||
1205 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1206 bond->igmp_retrans = bond->params.resend_igmp;
1207 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1208 }
1209}
1210
1211/**
1212 * bond_select_active_slave - select a new active slave, if needed
1213 * @bond: our bonding struct
1214 *
1215 * This functions should be called when one of the following occurs:
1216 * - The old curr_active_slave has been released or lost its link.
1217 * - The primary_slave has got its link back.
1218 * - A slave has got its link back and there's no old curr_active_slave.
1219 *
1220 * Caller must hold RTNL.
1221 */
1222void bond_select_active_slave(struct bonding *bond)
1223{
1224 struct slave *best_slave;
1225 int rv;
1226
1227 ASSERT_RTNL();
1228
1229 best_slave = bond_find_best_slave(bond);
1230 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1231 bond_change_active_slave(bond, best_slave);
1232 rv = bond_set_carrier(bond);
1233 if (!rv)
1234 return;
1235
1236 if (netif_carrier_ok(bond->dev))
1237 netdev_info(bond->dev, "active interface up!\n");
1238 else
1239 netdev_info(bond->dev, "now running without any active interface!\n");
1240 }
1241}
1242
1243#ifdef CONFIG_NET_POLL_CONTROLLER
1244static inline int slave_enable_netpoll(struct slave *slave)
1245{
1246 struct netpoll *np;
1247 int err = 0;
1248
1249 np = kzalloc(sizeof(*np), GFP_KERNEL);
1250 err = -ENOMEM;
1251 if (!np)
1252 goto out;
1253
1254 err = __netpoll_setup(np, slave->dev);
1255 if (err) {
1256 kfree(np);
1257 goto out;
1258 }
1259 slave->np = np;
1260out:
1261 return err;
1262}
1263static inline void slave_disable_netpoll(struct slave *slave)
1264{
1265 struct netpoll *np = slave->np;
1266
1267 if (!np)
1268 return;
1269
1270 slave->np = NULL;
1271
1272 __netpoll_free(np);
1273}
1274
1275static void bond_poll_controller(struct net_device *bond_dev)
1276{
1277 struct bonding *bond = netdev_priv(bond_dev);
1278 struct slave *slave = NULL;
1279 struct list_head *iter;
1280 struct ad_info ad_info;
1281
1282 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1283 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1284 return;
1285
1286 bond_for_each_slave_rcu(bond, slave, iter) {
1287 if (!bond_slave_is_up(slave))
1288 continue;
1289
1290 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1291 struct aggregator *agg =
1292 SLAVE_AD_INFO(slave)->port.aggregator;
1293
1294 if (agg &&
1295 agg->aggregator_identifier != ad_info.aggregator_id)
1296 continue;
1297 }
1298
1299 netpoll_poll_dev(slave->dev);
1300 }
1301}
1302
1303static void bond_netpoll_cleanup(struct net_device *bond_dev)
1304{
1305 struct bonding *bond = netdev_priv(bond_dev);
1306 struct list_head *iter;
1307 struct slave *slave;
1308
1309 bond_for_each_slave(bond, slave, iter)
1310 if (bond_slave_is_up(slave))
1311 slave_disable_netpoll(slave);
1312}
1313
1314static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1315{
1316 struct bonding *bond = netdev_priv(dev);
1317 struct list_head *iter;
1318 struct slave *slave;
1319 int err = 0;
1320
1321 bond_for_each_slave(bond, slave, iter) {
1322 err = slave_enable_netpoll(slave);
1323 if (err) {
1324 bond_netpoll_cleanup(dev);
1325 break;
1326 }
1327 }
1328 return err;
1329}
1330#else
1331static inline int slave_enable_netpoll(struct slave *slave)
1332{
1333 return 0;
1334}
1335static inline void slave_disable_netpoll(struct slave *slave)
1336{
1337}
1338static void bond_netpoll_cleanup(struct net_device *bond_dev)
1339{
1340}
1341#endif
1342
1343/*---------------------------------- IOCTL ----------------------------------*/
1344
1345static netdev_features_t bond_fix_features(struct net_device *dev,
1346 netdev_features_t features)
1347{
1348 struct bonding *bond = netdev_priv(dev);
1349 struct list_head *iter;
1350 netdev_features_t mask;
1351 struct slave *slave;
1352
1353#if IS_ENABLED(CONFIG_TLS_DEVICE)
1354 if (bond_sk_check(bond))
1355 features |= BOND_TLS_FEATURES;
1356 else
1357 features &= ~BOND_TLS_FEATURES;
1358#endif
1359
1360 mask = features;
1361
1362 features &= ~NETIF_F_ONE_FOR_ALL;
1363 features |= NETIF_F_ALL_FOR_ALL;
1364
1365 bond_for_each_slave(bond, slave, iter) {
1366 features = netdev_increment_features(features,
1367 slave->dev->features,
1368 mask);
1369 }
1370 features = netdev_add_tso_features(features, mask);
1371
1372 return features;
1373}
1374
1375#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1376 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1377 NETIF_F_HIGHDMA | NETIF_F_LRO)
1378
1379#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1380 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1381
1382#define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1383 NETIF_F_GSO_SOFTWARE)
1384
1385
1386static void bond_compute_features(struct bonding *bond)
1387{
1388 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1389 IFF_XMIT_DST_RELEASE_PERM;
1390 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1391 netdev_features_t enc_features = BOND_ENC_FEATURES;
1392#ifdef CONFIG_XFRM_OFFLOAD
1393 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1394#endif /* CONFIG_XFRM_OFFLOAD */
1395 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1396 struct net_device *bond_dev = bond->dev;
1397 struct list_head *iter;
1398 struct slave *slave;
1399 unsigned short max_hard_header_len = ETH_HLEN;
1400 unsigned int gso_max_size = GSO_MAX_SIZE;
1401 u16 gso_max_segs = GSO_MAX_SEGS;
1402
1403 if (!bond_has_slaves(bond))
1404 goto done;
1405 vlan_features &= NETIF_F_ALL_FOR_ALL;
1406 mpls_features &= NETIF_F_ALL_FOR_ALL;
1407
1408 bond_for_each_slave(bond, slave, iter) {
1409 vlan_features = netdev_increment_features(vlan_features,
1410 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1411
1412 enc_features = netdev_increment_features(enc_features,
1413 slave->dev->hw_enc_features,
1414 BOND_ENC_FEATURES);
1415
1416#ifdef CONFIG_XFRM_OFFLOAD
1417 xfrm_features = netdev_increment_features(xfrm_features,
1418 slave->dev->hw_enc_features,
1419 BOND_XFRM_FEATURES);
1420#endif /* CONFIG_XFRM_OFFLOAD */
1421
1422 mpls_features = netdev_increment_features(mpls_features,
1423 slave->dev->mpls_features,
1424 BOND_MPLS_FEATURES);
1425
1426 dst_release_flag &= slave->dev->priv_flags;
1427 if (slave->dev->hard_header_len > max_hard_header_len)
1428 max_hard_header_len = slave->dev->hard_header_len;
1429
1430 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1431 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1432 }
1433 bond_dev->hard_header_len = max_hard_header_len;
1434
1435done:
1436 bond_dev->vlan_features = vlan_features;
1437 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1438 NETIF_F_HW_VLAN_CTAG_TX |
1439 NETIF_F_HW_VLAN_STAG_TX;
1440#ifdef CONFIG_XFRM_OFFLOAD
1441 bond_dev->hw_enc_features |= xfrm_features;
1442#endif /* CONFIG_XFRM_OFFLOAD */
1443 bond_dev->mpls_features = mpls_features;
1444 bond_dev->gso_max_segs = gso_max_segs;
1445 netif_set_gso_max_size(bond_dev, gso_max_size);
1446
1447 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1448 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1449 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1450 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1451
1452 netdev_change_features(bond_dev);
1453}
1454
1455static void bond_setup_by_slave(struct net_device *bond_dev,
1456 struct net_device *slave_dev)
1457{
1458 bond_dev->header_ops = slave_dev->header_ops;
1459
1460 bond_dev->type = slave_dev->type;
1461 bond_dev->hard_header_len = slave_dev->hard_header_len;
1462 bond_dev->needed_headroom = slave_dev->needed_headroom;
1463 bond_dev->addr_len = slave_dev->addr_len;
1464
1465 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1466 slave_dev->addr_len);
1467}
1468
1469/* On bonding slaves other than the currently active slave, suppress
1470 * duplicates except for alb non-mcast/bcast.
1471 */
1472static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1473 struct slave *slave,
1474 struct bonding *bond)
1475{
1476 if (bond_is_slave_inactive(slave)) {
1477 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1478 skb->pkt_type != PACKET_BROADCAST &&
1479 skb->pkt_type != PACKET_MULTICAST)
1480 return false;
1481 return true;
1482 }
1483 return false;
1484}
1485
1486static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1487{
1488 struct sk_buff *skb = *pskb;
1489 struct slave *slave;
1490 struct bonding *bond;
1491 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1492 struct slave *);
1493 int ret = RX_HANDLER_ANOTHER;
1494
1495 skb = skb_share_check(skb, GFP_ATOMIC);
1496 if (unlikely(!skb))
1497 return RX_HANDLER_CONSUMED;
1498
1499 *pskb = skb;
1500
1501 slave = bond_slave_get_rcu(skb->dev);
1502 bond = slave->bond;
1503
1504 recv_probe = READ_ONCE(bond->recv_probe);
1505 if (recv_probe) {
1506 ret = recv_probe(skb, bond, slave);
1507 if (ret == RX_HANDLER_CONSUMED) {
1508 consume_skb(skb);
1509 return ret;
1510 }
1511 }
1512
1513 /*
1514 * For packets determined by bond_should_deliver_exact_match() call to
1515 * be suppressed we want to make an exception for link-local packets.
1516 * This is necessary for e.g. LLDP daemons to be able to monitor
1517 * inactive slave links without being forced to bind to them
1518 * explicitly.
1519 *
1520 * At the same time, packets that are passed to the bonding master
1521 * (including link-local ones) can have their originating interface
1522 * determined via PACKET_ORIGDEV socket option.
1523 */
1524 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1525 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1526 return RX_HANDLER_PASS;
1527 return RX_HANDLER_EXACT;
1528 }
1529
1530 skb->dev = bond->dev;
1531
1532 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1533 netif_is_bridge_port(bond->dev) &&
1534 skb->pkt_type == PACKET_HOST) {
1535
1536 if (unlikely(skb_cow_head(skb,
1537 skb->data - skb_mac_header(skb)))) {
1538 kfree_skb(skb);
1539 return RX_HANDLER_CONSUMED;
1540 }
1541 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1542 bond->dev->addr_len);
1543 }
1544
1545 return ret;
1546}
1547
1548static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1549{
1550 switch (BOND_MODE(bond)) {
1551 case BOND_MODE_ROUNDROBIN:
1552 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1553 case BOND_MODE_ACTIVEBACKUP:
1554 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1555 case BOND_MODE_BROADCAST:
1556 return NETDEV_LAG_TX_TYPE_BROADCAST;
1557 case BOND_MODE_XOR:
1558 case BOND_MODE_8023AD:
1559 return NETDEV_LAG_TX_TYPE_HASH;
1560 default:
1561 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1562 }
1563}
1564
1565static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1566 enum netdev_lag_tx_type type)
1567{
1568 if (type != NETDEV_LAG_TX_TYPE_HASH)
1569 return NETDEV_LAG_HASH_NONE;
1570
1571 switch (bond->params.xmit_policy) {
1572 case BOND_XMIT_POLICY_LAYER2:
1573 return NETDEV_LAG_HASH_L2;
1574 case BOND_XMIT_POLICY_LAYER34:
1575 return NETDEV_LAG_HASH_L34;
1576 case BOND_XMIT_POLICY_LAYER23:
1577 return NETDEV_LAG_HASH_L23;
1578 case BOND_XMIT_POLICY_ENCAP23:
1579 return NETDEV_LAG_HASH_E23;
1580 case BOND_XMIT_POLICY_ENCAP34:
1581 return NETDEV_LAG_HASH_E34;
1582 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1583 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1584 default:
1585 return NETDEV_LAG_HASH_UNKNOWN;
1586 }
1587}
1588
1589static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1590 struct netlink_ext_ack *extack)
1591{
1592 struct netdev_lag_upper_info lag_upper_info;
1593 enum netdev_lag_tx_type type;
1594
1595 type = bond_lag_tx_type(bond);
1596 lag_upper_info.tx_type = type;
1597 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1598
1599 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1600 &lag_upper_info, extack);
1601}
1602
1603static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1604{
1605 netdev_upper_dev_unlink(slave->dev, bond->dev);
1606 slave->dev->flags &= ~IFF_SLAVE;
1607}
1608
1609static void slave_kobj_release(struct kobject *kobj)
1610{
1611 struct slave *slave = to_slave(kobj);
1612 struct bonding *bond = bond_get_bond_by_slave(slave);
1613
1614 cancel_delayed_work_sync(&slave->notify_work);
1615 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1616 kfree(SLAVE_AD_INFO(slave));
1617
1618 kfree(slave);
1619}
1620
1621static struct kobj_type slave_ktype = {
1622 .release = slave_kobj_release,
1623#ifdef CONFIG_SYSFS
1624 .sysfs_ops = &slave_sysfs_ops,
1625#endif
1626};
1627
1628static int bond_kobj_init(struct slave *slave)
1629{
1630 int err;
1631
1632 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1633 &(slave->dev->dev.kobj), "bonding_slave");
1634 if (err)
1635 kobject_put(&slave->kobj);
1636
1637 return err;
1638}
1639
1640static struct slave *bond_alloc_slave(struct bonding *bond,
1641 struct net_device *slave_dev)
1642{
1643 struct slave *slave = NULL;
1644
1645 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1646 if (!slave)
1647 return NULL;
1648
1649 slave->bond = bond;
1650 slave->dev = slave_dev;
1651 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1652
1653 if (bond_kobj_init(slave))
1654 return NULL;
1655
1656 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1657 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1658 GFP_KERNEL);
1659 if (!SLAVE_AD_INFO(slave)) {
1660 kobject_put(&slave->kobj);
1661 return NULL;
1662 }
1663 }
1664
1665 return slave;
1666}
1667
1668static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1669{
1670 info->bond_mode = BOND_MODE(bond);
1671 info->miimon = bond->params.miimon;
1672 info->num_slaves = bond->slave_cnt;
1673}
1674
1675static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1676{
1677 strcpy(info->slave_name, slave->dev->name);
1678 info->link = slave->link;
1679 info->state = bond_slave_state(slave);
1680 info->link_failure_count = slave->link_failure_count;
1681}
1682
1683static void bond_netdev_notify_work(struct work_struct *_work)
1684{
1685 struct slave *slave = container_of(_work, struct slave,
1686 notify_work.work);
1687
1688 if (rtnl_trylock()) {
1689 struct netdev_bonding_info binfo;
1690
1691 bond_fill_ifslave(slave, &binfo.slave);
1692 bond_fill_ifbond(slave->bond, &binfo.master);
1693 netdev_bonding_info_change(slave->dev, &binfo);
1694 rtnl_unlock();
1695 } else {
1696 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1697 }
1698}
1699
1700void bond_queue_slave_event(struct slave *slave)
1701{
1702 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1703}
1704
1705void bond_lower_state_changed(struct slave *slave)
1706{
1707 struct netdev_lag_lower_state_info info;
1708
1709 info.link_up = slave->link == BOND_LINK_UP ||
1710 slave->link == BOND_LINK_FAIL;
1711 info.tx_enabled = bond_is_active_slave(slave);
1712 netdev_lower_state_changed(slave->dev, &info);
1713}
1714
1715/* enslave device <slave> to bond device <master> */
1716int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1717 struct netlink_ext_ack *extack)
1718{
1719 struct bonding *bond = netdev_priv(bond_dev);
1720 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1721 struct slave *new_slave = NULL, *prev_slave;
1722 struct sockaddr_storage ss;
1723 int link_reporting;
1724 int res = 0, i;
1725
1726 if (slave_dev->flags & IFF_MASTER &&
1727 !netif_is_bond_master(slave_dev)) {
1728 NL_SET_ERR_MSG(extack, "Device with IFF_MASTER cannot be enslaved");
1729 netdev_err(bond_dev,
1730 "Error: Device with IFF_MASTER cannot be enslaved\n");
1731 return -EPERM;
1732 }
1733
1734 if (!bond->params.use_carrier &&
1735 slave_dev->ethtool_ops->get_link == NULL &&
1736 slave_ops->ndo_do_ioctl == NULL) {
1737 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1738 }
1739
1740 /* already in-use? */
1741 if (netdev_is_rx_handler_busy(slave_dev)) {
1742 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1743 slave_err(bond_dev, slave_dev,
1744 "Error: Device is in use and cannot be enslaved\n");
1745 return -EBUSY;
1746 }
1747
1748 if (bond_dev == slave_dev) {
1749 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1750 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1751 return -EPERM;
1752 }
1753
1754 /* vlan challenged mutual exclusion */
1755 /* no need to lock since we're protected by rtnl_lock */
1756 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1757 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1758 if (vlan_uses_dev(bond_dev)) {
1759 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1760 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1761 return -EPERM;
1762 } else {
1763 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1764 }
1765 } else {
1766 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1767 }
1768
1769 if (slave_dev->features & NETIF_F_HW_ESP)
1770 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1771
1772 /* Old ifenslave binaries are no longer supported. These can
1773 * be identified with moderate accuracy by the state of the slave:
1774 * the current ifenslave will set the interface down prior to
1775 * enslaving it; the old ifenslave will not.
1776 */
1777 if (slave_dev->flags & IFF_UP) {
1778 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1779 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1780 return -EPERM;
1781 }
1782
1783 /* set bonding device ether type by slave - bonding netdevices are
1784 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1785 * there is a need to override some of the type dependent attribs/funcs.
1786 *
1787 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1788 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1789 */
1790 if (!bond_has_slaves(bond)) {
1791 if (bond_dev->type != slave_dev->type) {
1792 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1793 bond_dev->type, slave_dev->type);
1794
1795 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1796 bond_dev);
1797 res = notifier_to_errno(res);
1798 if (res) {
1799 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1800 return -EBUSY;
1801 }
1802
1803 /* Flush unicast and multicast addresses */
1804 dev_uc_flush(bond_dev);
1805 dev_mc_flush(bond_dev);
1806
1807 if (slave_dev->type != ARPHRD_ETHER)
1808 bond_setup_by_slave(bond_dev, slave_dev);
1809 else {
1810 ether_setup(bond_dev);
1811 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1812 }
1813
1814 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1815 bond_dev);
1816 }
1817 } else if (bond_dev->type != slave_dev->type) {
1818 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1819 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1820 slave_dev->type, bond_dev->type);
1821 return -EINVAL;
1822 }
1823
1824 if (slave_dev->type == ARPHRD_INFINIBAND &&
1825 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1826 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1827 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1828 slave_dev->type);
1829 res = -EOPNOTSUPP;
1830 goto err_undo_flags;
1831 }
1832
1833 if (!slave_ops->ndo_set_mac_address ||
1834 slave_dev->type == ARPHRD_INFINIBAND) {
1835 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1836 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1837 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1838 if (!bond_has_slaves(bond)) {
1839 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1840 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1841 } else {
1842 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1843 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1844 res = -EOPNOTSUPP;
1845 goto err_undo_flags;
1846 }
1847 }
1848 }
1849
1850 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1851
1852 /* If this is the first slave, then we need to set the master's hardware
1853 * address to be the same as the slave's.
1854 */
1855 if (!bond_has_slaves(bond) &&
1856 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1857 res = bond_set_dev_addr(bond->dev, slave_dev);
1858 if (res)
1859 goto err_undo_flags;
1860 }
1861
1862 new_slave = bond_alloc_slave(bond, slave_dev);
1863 if (!new_slave) {
1864 res = -ENOMEM;
1865 goto err_undo_flags;
1866 }
1867
1868 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1869 * is set via sysfs or module option if desired.
1870 */
1871 new_slave->queue_id = 0;
1872
1873 /* Save slave's original mtu and then set it to match the bond */
1874 new_slave->original_mtu = slave_dev->mtu;
1875 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1876 if (res) {
1877 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1878 goto err_free;
1879 }
1880
1881 /* Save slave's original ("permanent") mac address for modes
1882 * that need it, and for restoring it upon release, and then
1883 * set it to the master's address
1884 */
1885 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1886 slave_dev->addr_len);
1887
1888 if (!bond->params.fail_over_mac ||
1889 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1890 /* Set slave to master's mac address. The application already
1891 * set the master's mac address to that of the first slave
1892 */
1893 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1894 ss.ss_family = slave_dev->type;
1895 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1896 extack);
1897 if (res) {
1898 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1899 goto err_restore_mtu;
1900 }
1901 }
1902
1903 /* set slave flag before open to prevent IPv6 addrconf */
1904 slave_dev->flags |= IFF_SLAVE;
1905
1906 /* open the slave since the application closed it */
1907 res = dev_open(slave_dev, extack);
1908 if (res) {
1909 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1910 goto err_restore_mac;
1911 }
1912
1913 slave_dev->priv_flags |= IFF_BONDING;
1914 /* initialize slave stats */
1915 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1916
1917 if (bond_is_lb(bond)) {
1918 /* bond_alb_init_slave() must be called before all other stages since
1919 * it might fail and we do not want to have to undo everything
1920 */
1921 res = bond_alb_init_slave(bond, new_slave);
1922 if (res)
1923 goto err_close;
1924 }
1925
1926 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1927 if (res) {
1928 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1929 goto err_close;
1930 }
1931
1932 prev_slave = bond_last_slave(bond);
1933
1934 new_slave->delay = 0;
1935 new_slave->link_failure_count = 0;
1936
1937 if (bond_update_speed_duplex(new_slave) &&
1938 bond_needs_speed_duplex(bond))
1939 new_slave->link = BOND_LINK_DOWN;
1940
1941 new_slave->last_rx = jiffies -
1942 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1943 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1944 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1945
1946 if (bond->params.miimon && !bond->params.use_carrier) {
1947 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1948
1949 if ((link_reporting == -1) && !bond->params.arp_interval) {
1950 /* miimon is set but a bonded network driver
1951 * does not support ETHTOOL/MII and
1952 * arp_interval is not set. Note: if
1953 * use_carrier is enabled, we will never go
1954 * here (because netif_carrier is always
1955 * supported); thus, we don't need to change
1956 * the messages for netif_carrier.
1957 */
1958 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");
1959 } else if (link_reporting == -1) {
1960 /* unable get link status using mii/ethtool */
1961 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");
1962 }
1963 }
1964
1965 /* check for initial state */
1966 new_slave->link = BOND_LINK_NOCHANGE;
1967 if (bond->params.miimon) {
1968 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1969 if (bond->params.updelay) {
1970 bond_set_slave_link_state(new_slave,
1971 BOND_LINK_BACK,
1972 BOND_SLAVE_NOTIFY_NOW);
1973 new_slave->delay = bond->params.updelay;
1974 } else {
1975 bond_set_slave_link_state(new_slave,
1976 BOND_LINK_UP,
1977 BOND_SLAVE_NOTIFY_NOW);
1978 }
1979 } else {
1980 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1981 BOND_SLAVE_NOTIFY_NOW);
1982 }
1983 } else if (bond->params.arp_interval) {
1984 bond_set_slave_link_state(new_slave,
1985 (netif_carrier_ok(slave_dev) ?
1986 BOND_LINK_UP : BOND_LINK_DOWN),
1987 BOND_SLAVE_NOTIFY_NOW);
1988 } else {
1989 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1990 BOND_SLAVE_NOTIFY_NOW);
1991 }
1992
1993 if (new_slave->link != BOND_LINK_DOWN)
1994 new_slave->last_link_up = jiffies;
1995 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1996 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1997 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1998
1999 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2000 /* if there is a primary slave, remember it */
2001 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2002 rcu_assign_pointer(bond->primary_slave, new_slave);
2003 bond->force_primary = true;
2004 }
2005 }
2006
2007 switch (BOND_MODE(bond)) {
2008 case BOND_MODE_ACTIVEBACKUP:
2009 bond_set_slave_inactive_flags(new_slave,
2010 BOND_SLAVE_NOTIFY_NOW);
2011 break;
2012 case BOND_MODE_8023AD:
2013 /* in 802.3ad mode, the internal mechanism
2014 * will activate the slaves in the selected
2015 * aggregator
2016 */
2017 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2018 /* if this is the first slave */
2019 if (!prev_slave) {
2020 SLAVE_AD_INFO(new_slave)->id = 1;
2021 /* Initialize AD with the number of times that the AD timer is called in 1 second
2022 * can be called only after the mac address of the bond is set
2023 */
2024 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2025 } else {
2026 SLAVE_AD_INFO(new_slave)->id =
2027 SLAVE_AD_INFO(prev_slave)->id + 1;
2028 }
2029
2030 bond_3ad_bind_slave(new_slave);
2031 break;
2032 case BOND_MODE_TLB:
2033 case BOND_MODE_ALB:
2034 bond_set_active_slave(new_slave);
2035 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2036 break;
2037 default:
2038 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2039
2040 /* always active in trunk mode */
2041 bond_set_active_slave(new_slave);
2042
2043 /* In trunking mode there is little meaning to curr_active_slave
2044 * anyway (it holds no special properties of the bond device),
2045 * so we can change it without calling change_active_interface()
2046 */
2047 if (!rcu_access_pointer(bond->curr_active_slave) &&
2048 new_slave->link == BOND_LINK_UP)
2049 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2050
2051 break;
2052 } /* switch(bond_mode) */
2053
2054#ifdef CONFIG_NET_POLL_CONTROLLER
2055 if (bond->dev->npinfo) {
2056 if (slave_enable_netpoll(new_slave)) {
2057 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2058 res = -EBUSY;
2059 goto err_detach;
2060 }
2061 }
2062#endif
2063
2064 if (!(bond_dev->features & NETIF_F_LRO))
2065 dev_disable_lro(slave_dev);
2066
2067 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2068 new_slave);
2069 if (res) {
2070 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2071 goto err_detach;
2072 }
2073
2074 res = bond_master_upper_dev_link(bond, new_slave, extack);
2075 if (res) {
2076 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2077 goto err_unregister;
2078 }
2079
2080 bond_lower_state_changed(new_slave);
2081
2082 res = bond_sysfs_slave_add(new_slave);
2083 if (res) {
2084 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2085 goto err_upper_unlink;
2086 }
2087
2088 /* If the mode uses primary, then the following is handled by
2089 * bond_change_active_slave().
2090 */
2091 if (!bond_uses_primary(bond)) {
2092 /* set promiscuity level to new slave */
2093 if (bond_dev->flags & IFF_PROMISC) {
2094 res = dev_set_promiscuity(slave_dev, 1);
2095 if (res)
2096 goto err_sysfs_del;
2097 }
2098
2099 /* set allmulti level to new slave */
2100 if (bond_dev->flags & IFF_ALLMULTI) {
2101 res = dev_set_allmulti(slave_dev, 1);
2102 if (res) {
2103 if (bond_dev->flags & IFF_PROMISC)
2104 dev_set_promiscuity(slave_dev, -1);
2105 goto err_sysfs_del;
2106 }
2107 }
2108
2109 netif_addr_lock_bh(bond_dev);
2110 dev_mc_sync_multiple(slave_dev, bond_dev);
2111 dev_uc_sync_multiple(slave_dev, bond_dev);
2112 netif_addr_unlock_bh(bond_dev);
2113
2114 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2115 /* add lacpdu mc addr to mc list */
2116 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2117
2118 dev_mc_add(slave_dev, lacpdu_multicast);
2119 }
2120 }
2121
2122 bond->slave_cnt++;
2123 bond_compute_features(bond);
2124 bond_set_carrier(bond);
2125
2126 if (bond_uses_primary(bond)) {
2127 block_netpoll_tx();
2128 bond_select_active_slave(bond);
2129 unblock_netpoll_tx();
2130 }
2131
2132 if (bond_mode_can_use_xmit_hash(bond))
2133 bond_update_slave_arr(bond, NULL);
2134
2135
2136 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2137 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2138 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2139
2140 /* enslave is successful */
2141 bond_queue_slave_event(new_slave);
2142 return 0;
2143
2144/* Undo stages on error */
2145err_sysfs_del:
2146 bond_sysfs_slave_del(new_slave);
2147
2148err_upper_unlink:
2149 bond_upper_dev_unlink(bond, new_slave);
2150
2151err_unregister:
2152 netdev_rx_handler_unregister(slave_dev);
2153
2154err_detach:
2155 vlan_vids_del_by_dev(slave_dev, bond_dev);
2156 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2157 RCU_INIT_POINTER(bond->primary_slave, NULL);
2158 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2159 block_netpoll_tx();
2160 bond_change_active_slave(bond, NULL);
2161 bond_select_active_slave(bond);
2162 unblock_netpoll_tx();
2163 }
2164 /* either primary_slave or curr_active_slave might've changed */
2165 synchronize_rcu();
2166 slave_disable_netpoll(new_slave);
2167
2168err_close:
2169 if (!netif_is_bond_master(slave_dev))
2170 slave_dev->priv_flags &= ~IFF_BONDING;
2171 dev_close(slave_dev);
2172
2173err_restore_mac:
2174 slave_dev->flags &= ~IFF_SLAVE;
2175 if (!bond->params.fail_over_mac ||
2176 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2177 /* XXX TODO - fom follow mode needs to change master's
2178 * MAC if this slave's MAC is in use by the bond, or at
2179 * least print a warning.
2180 */
2181 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2182 new_slave->dev->addr_len);
2183 ss.ss_family = slave_dev->type;
2184 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2185 }
2186
2187err_restore_mtu:
2188 dev_set_mtu(slave_dev, new_slave->original_mtu);
2189
2190err_free:
2191 kobject_put(&new_slave->kobj);
2192
2193err_undo_flags:
2194 /* Enslave of first slave has failed and we need to fix master's mac */
2195 if (!bond_has_slaves(bond)) {
2196 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2197 slave_dev->dev_addr))
2198 eth_hw_addr_random(bond_dev);
2199 if (bond_dev->type != ARPHRD_ETHER) {
2200 dev_close(bond_dev);
2201 ether_setup(bond_dev);
2202 bond_dev->flags |= IFF_MASTER;
2203 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2204 }
2205 }
2206
2207 return res;
2208}
2209
2210/* Try to release the slave device <slave> from the bond device <master>
2211 * It is legal to access curr_active_slave without a lock because all the function
2212 * is RTNL-locked. If "all" is true it means that the function is being called
2213 * while destroying a bond interface and all slaves are being released.
2214 *
2215 * The rules for slave state should be:
2216 * for Active/Backup:
2217 * Active stays on all backups go down
2218 * for Bonded connections:
2219 * The first up interface should be left on and all others downed.
2220 */
2221static int __bond_release_one(struct net_device *bond_dev,
2222 struct net_device *slave_dev,
2223 bool all, bool unregister)
2224{
2225 struct bonding *bond = netdev_priv(bond_dev);
2226 struct slave *slave, *oldcurrent;
2227 struct sockaddr_storage ss;
2228 int old_flags = bond_dev->flags;
2229 netdev_features_t old_features = bond_dev->features;
2230
2231 /* slave is not a slave or master is not master of this slave */
2232 if (!(slave_dev->flags & IFF_SLAVE) ||
2233 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2234 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2235 return -EINVAL;
2236 }
2237
2238 block_netpoll_tx();
2239
2240 slave = bond_get_slave_by_dev(bond, slave_dev);
2241 if (!slave) {
2242 /* not a slave of this bond */
2243 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2244 unblock_netpoll_tx();
2245 return -EINVAL;
2246 }
2247
2248 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2249
2250 bond_sysfs_slave_del(slave);
2251
2252 /* recompute stats just before removing the slave */
2253 bond_get_stats(bond->dev, &bond->bond_stats);
2254
2255 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2256 * for this slave anymore.
2257 */
2258 netdev_rx_handler_unregister(slave_dev);
2259
2260 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2261 bond_3ad_unbind_slave(slave);
2262
2263 bond_upper_dev_unlink(bond, slave);
2264
2265 if (bond_mode_can_use_xmit_hash(bond))
2266 bond_update_slave_arr(bond, slave);
2267
2268 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2269 bond_is_active_slave(slave) ? "active" : "backup");
2270
2271 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2272
2273 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2274
2275 if (!all && (!bond->params.fail_over_mac ||
2276 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2277 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2278 bond_has_slaves(bond))
2279 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",
2280 slave->perm_hwaddr);
2281 }
2282
2283 if (rtnl_dereference(bond->primary_slave) == slave)
2284 RCU_INIT_POINTER(bond->primary_slave, NULL);
2285
2286 if (oldcurrent == slave)
2287 bond_change_active_slave(bond, NULL);
2288
2289 if (bond_is_lb(bond)) {
2290 /* Must be called only after the slave has been
2291 * detached from the list and the curr_active_slave
2292 * has been cleared (if our_slave == old_current),
2293 * but before a new active slave is selected.
2294 */
2295 bond_alb_deinit_slave(bond, slave);
2296 }
2297
2298 if (all) {
2299 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2300 } else if (oldcurrent == slave) {
2301 /* Note that we hold RTNL over this sequence, so there
2302 * is no concern that another slave add/remove event
2303 * will interfere.
2304 */
2305 bond_select_active_slave(bond);
2306 }
2307
2308 if (!bond_has_slaves(bond)) {
2309 bond_set_carrier(bond);
2310 eth_hw_addr_random(bond_dev);
2311 }
2312
2313 unblock_netpoll_tx();
2314 synchronize_rcu();
2315 bond->slave_cnt--;
2316
2317 if (!bond_has_slaves(bond)) {
2318 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2319 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2320 }
2321
2322 bond_compute_features(bond);
2323 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2324 (old_features & NETIF_F_VLAN_CHALLENGED))
2325 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2326
2327 vlan_vids_del_by_dev(slave_dev, bond_dev);
2328
2329 /* If the mode uses primary, then this case was handled above by
2330 * bond_change_active_slave(..., NULL)
2331 */
2332 if (!bond_uses_primary(bond)) {
2333 /* unset promiscuity level from slave
2334 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2335 * of the IFF_PROMISC flag in the bond_dev, but we need the
2336 * value of that flag before that change, as that was the value
2337 * when this slave was attached, so we cache at the start of the
2338 * function and use it here. Same goes for ALLMULTI below
2339 */
2340 if (old_flags & IFF_PROMISC)
2341 dev_set_promiscuity(slave_dev, -1);
2342
2343 /* unset allmulti level from slave */
2344 if (old_flags & IFF_ALLMULTI)
2345 dev_set_allmulti(slave_dev, -1);
2346
2347 bond_hw_addr_flush(bond_dev, slave_dev);
2348 }
2349
2350 slave_disable_netpoll(slave);
2351
2352 /* close slave before restoring its mac address */
2353 dev_close(slave_dev);
2354
2355 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2356 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2357 /* restore original ("permanent") mac address */
2358 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2359 slave->dev->addr_len);
2360 ss.ss_family = slave_dev->type;
2361 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2362 }
2363
2364 if (unregister)
2365 __dev_set_mtu(slave_dev, slave->original_mtu);
2366 else
2367 dev_set_mtu(slave_dev, slave->original_mtu);
2368
2369 if (!netif_is_bond_master(slave_dev))
2370 slave_dev->priv_flags &= ~IFF_BONDING;
2371
2372 kobject_put(&slave->kobj);
2373
2374 return 0;
2375}
2376
2377/* A wrapper used because of ndo_del_link */
2378int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2379{
2380 return __bond_release_one(bond_dev, slave_dev, false, false);
2381}
2382
2383/* First release a slave and then destroy the bond if no more slaves are left.
2384 * Must be under rtnl_lock when this function is called.
2385 */
2386static int bond_release_and_destroy(struct net_device *bond_dev,
2387 struct net_device *slave_dev)
2388{
2389 struct bonding *bond = netdev_priv(bond_dev);
2390 int ret;
2391
2392 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2393 if (ret == 0 && !bond_has_slaves(bond) &&
2394 bond_dev->reg_state != NETREG_UNREGISTERING) {
2395 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2396 netdev_info(bond_dev, "Destroying bond\n");
2397 bond_remove_proc_entry(bond);
2398 unregister_netdevice(bond_dev);
2399 }
2400 return ret;
2401}
2402
2403static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2404{
2405 struct bonding *bond = netdev_priv(bond_dev);
2406
2407 bond_fill_ifbond(bond, info);
2408}
2409
2410static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2411{
2412 struct bonding *bond = netdev_priv(bond_dev);
2413 struct list_head *iter;
2414 int i = 0, res = -ENODEV;
2415 struct slave *slave;
2416
2417 bond_for_each_slave(bond, slave, iter) {
2418 if (i++ == (int)info->slave_id) {
2419 res = 0;
2420 bond_fill_ifslave(slave, info);
2421 break;
2422 }
2423 }
2424
2425 return res;
2426}
2427
2428/*-------------------------------- Monitoring -------------------------------*/
2429
2430/* called with rcu_read_lock() */
2431static int bond_miimon_inspect(struct bonding *bond)
2432{
2433 int link_state, commit = 0;
2434 struct list_head *iter;
2435 struct slave *slave;
2436 bool ignore_updelay;
2437
2438 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2439
2440 bond_for_each_slave_rcu(bond, slave, iter) {
2441 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2442
2443 link_state = bond_check_dev_link(bond, slave->dev, 0);
2444
2445 switch (slave->link) {
2446 case BOND_LINK_UP:
2447 if (link_state)
2448 continue;
2449
2450 bond_propose_link_state(slave, BOND_LINK_FAIL);
2451 commit++;
2452 slave->delay = bond->params.downdelay;
2453 if (slave->delay) {
2454 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2455 (BOND_MODE(bond) ==
2456 BOND_MODE_ACTIVEBACKUP) ?
2457 (bond_is_active_slave(slave) ?
2458 "active " : "backup ") : "",
2459 bond->params.downdelay * bond->params.miimon);
2460 }
2461 fallthrough;
2462 case BOND_LINK_FAIL:
2463 if (link_state) {
2464 /* recovered before downdelay expired */
2465 bond_propose_link_state(slave, BOND_LINK_UP);
2466 slave->last_link_up = jiffies;
2467 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2468 (bond->params.downdelay - slave->delay) *
2469 bond->params.miimon);
2470 commit++;
2471 continue;
2472 }
2473
2474 if (slave->delay <= 0) {
2475 bond_propose_link_state(slave, BOND_LINK_DOWN);
2476 commit++;
2477 continue;
2478 }
2479
2480 slave->delay--;
2481 break;
2482
2483 case BOND_LINK_DOWN:
2484 if (!link_state)
2485 continue;
2486
2487 bond_propose_link_state(slave, BOND_LINK_BACK);
2488 commit++;
2489 slave->delay = bond->params.updelay;
2490
2491 if (slave->delay) {
2492 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2493 ignore_updelay ? 0 :
2494 bond->params.updelay *
2495 bond->params.miimon);
2496 }
2497 fallthrough;
2498 case BOND_LINK_BACK:
2499 if (!link_state) {
2500 bond_propose_link_state(slave, BOND_LINK_DOWN);
2501 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2502 (bond->params.updelay - slave->delay) *
2503 bond->params.miimon);
2504 commit++;
2505 continue;
2506 }
2507
2508 if (ignore_updelay)
2509 slave->delay = 0;
2510
2511 if (slave->delay <= 0) {
2512 bond_propose_link_state(slave, BOND_LINK_UP);
2513 commit++;
2514 ignore_updelay = false;
2515 continue;
2516 }
2517
2518 slave->delay--;
2519 break;
2520 }
2521 }
2522
2523 return commit;
2524}
2525
2526static void bond_miimon_link_change(struct bonding *bond,
2527 struct slave *slave,
2528 char link)
2529{
2530 switch (BOND_MODE(bond)) {
2531 case BOND_MODE_8023AD:
2532 bond_3ad_handle_link_change(slave, link);
2533 break;
2534 case BOND_MODE_TLB:
2535 case BOND_MODE_ALB:
2536 bond_alb_handle_link_change(bond, slave, link);
2537 break;
2538 case BOND_MODE_XOR:
2539 bond_update_slave_arr(bond, NULL);
2540 break;
2541 }
2542}
2543
2544static void bond_miimon_commit(struct bonding *bond)
2545{
2546 struct list_head *iter;
2547 struct slave *slave, *primary;
2548
2549 bond_for_each_slave(bond, slave, iter) {
2550 switch (slave->link_new_state) {
2551 case BOND_LINK_NOCHANGE:
2552 /* For 802.3ad mode, check current slave speed and
2553 * duplex again in case its port was disabled after
2554 * invalid speed/duplex reporting but recovered before
2555 * link monitoring could make a decision on the actual
2556 * link status
2557 */
2558 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2559 slave->link == BOND_LINK_UP)
2560 bond_3ad_adapter_speed_duplex_changed(slave);
2561 continue;
2562
2563 case BOND_LINK_UP:
2564 if (bond_update_speed_duplex(slave) &&
2565 bond_needs_speed_duplex(bond)) {
2566 slave->link = BOND_LINK_DOWN;
2567 if (net_ratelimit())
2568 slave_warn(bond->dev, slave->dev,
2569 "failed to get link speed/duplex\n");
2570 continue;
2571 }
2572 bond_set_slave_link_state(slave, BOND_LINK_UP,
2573 BOND_SLAVE_NOTIFY_NOW);
2574 slave->last_link_up = jiffies;
2575
2576 primary = rtnl_dereference(bond->primary_slave);
2577 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2578 /* prevent it from being the active one */
2579 bond_set_backup_slave(slave);
2580 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2581 /* make it immediately active */
2582 bond_set_active_slave(slave);
2583 }
2584
2585 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2586 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2587 slave->duplex ? "full" : "half");
2588
2589 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2590
2591 if (!bond->curr_active_slave || slave == primary)
2592 goto do_failover;
2593
2594 continue;
2595
2596 case BOND_LINK_DOWN:
2597 if (slave->link_failure_count < UINT_MAX)
2598 slave->link_failure_count++;
2599
2600 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2601 BOND_SLAVE_NOTIFY_NOW);
2602
2603 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2604 BOND_MODE(bond) == BOND_MODE_8023AD)
2605 bond_set_slave_inactive_flags(slave,
2606 BOND_SLAVE_NOTIFY_NOW);
2607
2608 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2609
2610 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2611
2612 if (slave == rcu_access_pointer(bond->curr_active_slave))
2613 goto do_failover;
2614
2615 continue;
2616
2617 default:
2618 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2619 slave->link_new_state);
2620 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2621
2622 continue;
2623 }
2624
2625do_failover:
2626 block_netpoll_tx();
2627 bond_select_active_slave(bond);
2628 unblock_netpoll_tx();
2629 }
2630
2631 bond_set_carrier(bond);
2632}
2633
2634/* bond_mii_monitor
2635 *
2636 * Really a wrapper that splits the mii monitor into two phases: an
2637 * inspection, then (if inspection indicates something needs to be done)
2638 * an acquisition of appropriate locks followed by a commit phase to
2639 * implement whatever link state changes are indicated.
2640 */
2641static void bond_mii_monitor(struct work_struct *work)
2642{
2643 struct bonding *bond = container_of(work, struct bonding,
2644 mii_work.work);
2645 bool should_notify_peers = false;
2646 bool commit;
2647 unsigned long delay;
2648 struct slave *slave;
2649 struct list_head *iter;
2650
2651 delay = msecs_to_jiffies(bond->params.miimon);
2652
2653 if (!bond_has_slaves(bond))
2654 goto re_arm;
2655
2656 rcu_read_lock();
2657 should_notify_peers = bond_should_notify_peers(bond);
2658 commit = !!bond_miimon_inspect(bond);
2659 if (bond->send_peer_notif) {
2660 rcu_read_unlock();
2661 if (rtnl_trylock()) {
2662 bond->send_peer_notif--;
2663 rtnl_unlock();
2664 }
2665 } else {
2666 rcu_read_unlock();
2667 }
2668
2669 if (commit) {
2670 /* Race avoidance with bond_close cancel of workqueue */
2671 if (!rtnl_trylock()) {
2672 delay = 1;
2673 should_notify_peers = false;
2674 goto re_arm;
2675 }
2676
2677 bond_for_each_slave(bond, slave, iter) {
2678 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2679 }
2680 bond_miimon_commit(bond);
2681
2682 rtnl_unlock(); /* might sleep, hold no other locks */
2683 }
2684
2685re_arm:
2686 if (bond->params.miimon)
2687 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2688
2689 if (should_notify_peers) {
2690 if (!rtnl_trylock())
2691 return;
2692 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2693 rtnl_unlock();
2694 }
2695}
2696
2697static int bond_upper_dev_walk(struct net_device *upper,
2698 struct netdev_nested_priv *priv)
2699{
2700 __be32 ip = *(__be32 *)priv->data;
2701
2702 return ip == bond_confirm_addr(upper, 0, ip);
2703}
2704
2705static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2706{
2707 struct netdev_nested_priv priv = {
2708 .data = (void *)&ip,
2709 };
2710 bool ret = false;
2711
2712 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2713 return true;
2714
2715 rcu_read_lock();
2716 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2717 ret = true;
2718 rcu_read_unlock();
2719
2720 return ret;
2721}
2722
2723/* We go to the (large) trouble of VLAN tagging ARP frames because
2724 * switches in VLAN mode (especially if ports are configured as
2725 * "native" to a VLAN) might not pass non-tagged frames.
2726 */
2727static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2728 __be32 src_ip, struct bond_vlan_tag *tags)
2729{
2730 struct sk_buff *skb;
2731 struct bond_vlan_tag *outer_tag = tags;
2732 struct net_device *slave_dev = slave->dev;
2733 struct net_device *bond_dev = slave->bond->dev;
2734
2735 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2736 arp_op, &dest_ip, &src_ip);
2737
2738 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2739 NULL, slave_dev->dev_addr, NULL);
2740
2741 if (!skb) {
2742 net_err_ratelimited("ARP packet allocation failed\n");
2743 return;
2744 }
2745
2746 if (!tags || tags->vlan_proto == VLAN_N_VID)
2747 goto xmit;
2748
2749 tags++;
2750
2751 /* Go through all the tags backwards and add them to the packet */
2752 while (tags->vlan_proto != VLAN_N_VID) {
2753 if (!tags->vlan_id) {
2754 tags++;
2755 continue;
2756 }
2757
2758 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2759 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2760 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2761 tags->vlan_id);
2762 if (!skb) {
2763 net_err_ratelimited("failed to insert inner VLAN tag\n");
2764 return;
2765 }
2766
2767 tags++;
2768 }
2769 /* Set the outer tag */
2770 if (outer_tag->vlan_id) {
2771 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2772 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2773 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2774 outer_tag->vlan_id);
2775 }
2776
2777xmit:
2778 arp_xmit(skb);
2779}
2780
2781/* Validate the device path between the @start_dev and the @end_dev.
2782 * The path is valid if the @end_dev is reachable through device
2783 * stacking.
2784 * When the path is validated, collect any vlan information in the
2785 * path.
2786 */
2787struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2788 struct net_device *end_dev,
2789 int level)
2790{
2791 struct bond_vlan_tag *tags;
2792 struct net_device *upper;
2793 struct list_head *iter;
2794
2795 if (start_dev == end_dev) {
2796 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2797 if (!tags)
2798 return ERR_PTR(-ENOMEM);
2799 tags[level].vlan_proto = VLAN_N_VID;
2800 return tags;
2801 }
2802
2803 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2804 tags = bond_verify_device_path(upper, end_dev, level + 1);
2805 if (IS_ERR_OR_NULL(tags)) {
2806 if (IS_ERR(tags))
2807 return tags;
2808 continue;
2809 }
2810 if (is_vlan_dev(upper)) {
2811 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2812 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2813 }
2814
2815 return tags;
2816 }
2817
2818 return NULL;
2819}
2820
2821static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2822{
2823 struct rtable *rt;
2824 struct bond_vlan_tag *tags;
2825 __be32 *targets = bond->params.arp_targets, addr;
2826 int i;
2827
2828 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2829 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2830 __func__, &targets[i]);
2831 tags = NULL;
2832
2833 /* Find out through which dev should the packet go */
2834 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2835 RTO_ONLINK, 0);
2836 if (IS_ERR(rt)) {
2837 /* there's no route to target - try to send arp
2838 * probe to generate any traffic (arp_validate=0)
2839 */
2840 if (bond->params.arp_validate)
2841 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2842 bond->dev->name,
2843 &targets[i]);
2844 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2845 0, tags);
2846 continue;
2847 }
2848
2849 /* bond device itself */
2850 if (rt->dst.dev == bond->dev)
2851 goto found;
2852
2853 rcu_read_lock();
2854 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2855 rcu_read_unlock();
2856
2857 if (!IS_ERR_OR_NULL(tags))
2858 goto found;
2859
2860 /* Not our device - skip */
2861 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2862 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2863
2864 ip_rt_put(rt);
2865 continue;
2866
2867found:
2868 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2869 ip_rt_put(rt);
2870 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2871 kfree(tags);
2872 }
2873}
2874
2875static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2876{
2877 int i;
2878
2879 if (!sip || !bond_has_this_ip(bond, tip)) {
2880 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2881 __func__, &sip, &tip);
2882 return;
2883 }
2884
2885 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2886 if (i == -1) {
2887 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2888 __func__, &sip);
2889 return;
2890 }
2891 slave->last_rx = jiffies;
2892 slave->target_last_arp_rx[i] = jiffies;
2893}
2894
2895int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2896 struct slave *slave)
2897{
2898 struct arphdr *arp = (struct arphdr *)skb->data;
2899 struct slave *curr_active_slave, *curr_arp_slave;
2900 unsigned char *arp_ptr;
2901 __be32 sip, tip;
2902 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2903 unsigned int alen;
2904
2905 if (!slave_do_arp_validate(bond, slave)) {
2906 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2907 !slave_do_arp_validate_only(bond))
2908 slave->last_rx = jiffies;
2909 return RX_HANDLER_ANOTHER;
2910 } else if (!is_arp) {
2911 return RX_HANDLER_ANOTHER;
2912 }
2913
2914 alen = arp_hdr_len(bond->dev);
2915
2916 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2917 __func__, skb->dev->name);
2918
2919 if (alen > skb_headlen(skb)) {
2920 arp = kmalloc(alen, GFP_ATOMIC);
2921 if (!arp)
2922 goto out_unlock;
2923 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2924 goto out_unlock;
2925 }
2926
2927 if (arp->ar_hln != bond->dev->addr_len ||
2928 skb->pkt_type == PACKET_OTHERHOST ||
2929 skb->pkt_type == PACKET_LOOPBACK ||
2930 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2931 arp->ar_pro != htons(ETH_P_IP) ||
2932 arp->ar_pln != 4)
2933 goto out_unlock;
2934
2935 arp_ptr = (unsigned char *)(arp + 1);
2936 arp_ptr += bond->dev->addr_len;
2937 memcpy(&sip, arp_ptr, 4);
2938 arp_ptr += 4 + bond->dev->addr_len;
2939 memcpy(&tip, arp_ptr, 4);
2940
2941 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2942 __func__, slave->dev->name, bond_slave_state(slave),
2943 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2944 &sip, &tip);
2945
2946 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2947 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2948
2949 /* We 'trust' the received ARP enough to validate it if:
2950 *
2951 * (a) the slave receiving the ARP is active (which includes the
2952 * current ARP slave, if any), or
2953 *
2954 * (b) the receiving slave isn't active, but there is a currently
2955 * active slave and it received valid arp reply(s) after it became
2956 * the currently active slave, or
2957 *
2958 * (c) there is an ARP slave that sent an ARP during the prior ARP
2959 * interval, and we receive an ARP reply on any slave. We accept
2960 * these because switch FDB update delays may deliver the ARP
2961 * reply to a slave other than the sender of the ARP request.
2962 *
2963 * Note: for (b), backup slaves are receiving the broadcast ARP
2964 * request, not a reply. This request passes from the sending
2965 * slave through the L2 switch(es) to the receiving slave. Since
2966 * this is checking the request, sip/tip are swapped for
2967 * validation.
2968 *
2969 * This is done to avoid endless looping when we can't reach the
2970 * arp_ip_target and fool ourselves with our own arp requests.
2971 */
2972 if (bond_is_active_slave(slave))
2973 bond_validate_arp(bond, slave, sip, tip);
2974 else if (curr_active_slave &&
2975 time_after(slave_last_rx(bond, curr_active_slave),
2976 curr_active_slave->last_link_up))
2977 bond_validate_arp(bond, slave, tip, sip);
2978 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2979 bond_time_in_interval(bond,
2980 dev_trans_start(curr_arp_slave->dev), 1))
2981 bond_validate_arp(bond, slave, sip, tip);
2982
2983out_unlock:
2984 if (arp != (struct arphdr *)skb->data)
2985 kfree(arp);
2986 return RX_HANDLER_ANOTHER;
2987}
2988
2989/* function to verify if we're in the arp_interval timeslice, returns true if
2990 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2991 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2992 */
2993static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2994 int mod)
2995{
2996 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2997
2998 return time_in_range(jiffies,
2999 last_act - delta_in_ticks,
3000 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3001}
3002
3003/* This function is called regularly to monitor each slave's link
3004 * ensuring that traffic is being sent and received when arp monitoring
3005 * is used in load-balancing mode. if the adapter has been dormant, then an
3006 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3007 * arp monitoring in active backup mode.
3008 */
3009static void bond_loadbalance_arp_mon(struct bonding *bond)
3010{
3011 struct slave *slave, *oldcurrent;
3012 struct list_head *iter;
3013 int do_failover = 0, slave_state_changed = 0;
3014
3015 if (!bond_has_slaves(bond))
3016 goto re_arm;
3017
3018 rcu_read_lock();
3019
3020 oldcurrent = rcu_dereference(bond->curr_active_slave);
3021 /* see if any of the previous devices are up now (i.e. they have
3022 * xmt and rcv traffic). the curr_active_slave does not come into
3023 * the picture unless it is null. also, slave->last_link_up is not
3024 * needed here because we send an arp on each slave and give a slave
3025 * as long as it needs to get the tx/rx within the delta.
3026 * TODO: what about up/down delay in arp mode? it wasn't here before
3027 * so it can wait
3028 */
3029 bond_for_each_slave_rcu(bond, slave, iter) {
3030 unsigned long trans_start = dev_trans_start(slave->dev);
3031
3032 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3033
3034 if (slave->link != BOND_LINK_UP) {
3035 if (bond_time_in_interval(bond, trans_start, 1) &&
3036 bond_time_in_interval(bond, slave->last_rx, 1)) {
3037
3038 bond_propose_link_state(slave, BOND_LINK_UP);
3039 slave_state_changed = 1;
3040
3041 /* primary_slave has no meaning in round-robin
3042 * mode. the window of a slave being up and
3043 * curr_active_slave being null after enslaving
3044 * is closed.
3045 */
3046 if (!oldcurrent) {
3047 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3048 do_failover = 1;
3049 } else {
3050 slave_info(bond->dev, slave->dev, "interface is now up\n");
3051 }
3052 }
3053 } else {
3054 /* slave->link == BOND_LINK_UP */
3055
3056 /* not all switches will respond to an arp request
3057 * when the source ip is 0, so don't take the link down
3058 * if we don't know our ip yet
3059 */
3060 if (!bond_time_in_interval(bond, trans_start, 2) ||
3061 !bond_time_in_interval(bond, slave->last_rx, 2)) {
3062
3063 bond_propose_link_state(slave, BOND_LINK_DOWN);
3064 slave_state_changed = 1;
3065
3066 if (slave->link_failure_count < UINT_MAX)
3067 slave->link_failure_count++;
3068
3069 slave_info(bond->dev, slave->dev, "interface is now down\n");
3070
3071 if (slave == oldcurrent)
3072 do_failover = 1;
3073 }
3074 }
3075
3076 /* note: if switch is in round-robin mode, all links
3077 * must tx arp to ensure all links rx an arp - otherwise
3078 * links may oscillate or not come up at all; if switch is
3079 * in something like xor mode, there is nothing we can
3080 * do - all replies will be rx'ed on same link causing slaves
3081 * to be unstable during low/no traffic periods
3082 */
3083 if (bond_slave_is_up(slave))
3084 bond_arp_send_all(bond, slave);
3085 }
3086
3087 rcu_read_unlock();
3088
3089 if (do_failover || slave_state_changed) {
3090 if (!rtnl_trylock())
3091 goto re_arm;
3092
3093 bond_for_each_slave(bond, slave, iter) {
3094 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3095 slave->link = slave->link_new_state;
3096 }
3097
3098 if (slave_state_changed) {
3099 bond_slave_state_change(bond);
3100 if (BOND_MODE(bond) == BOND_MODE_XOR)
3101 bond_update_slave_arr(bond, NULL);
3102 }
3103 if (do_failover) {
3104 block_netpoll_tx();
3105 bond_select_active_slave(bond);
3106 unblock_netpoll_tx();
3107 }
3108 rtnl_unlock();
3109 }
3110
3111re_arm:
3112 if (bond->params.arp_interval)
3113 queue_delayed_work(bond->wq, &bond->arp_work,
3114 msecs_to_jiffies(bond->params.arp_interval));
3115}
3116
3117/* Called to inspect slaves for active-backup mode ARP monitor link state
3118 * changes. Sets proposed link state in slaves to specify what action
3119 * should take place for the slave. Returns 0 if no changes are found, >0
3120 * if changes to link states must be committed.
3121 *
3122 * Called with rcu_read_lock held.
3123 */
3124static int bond_ab_arp_inspect(struct bonding *bond)
3125{
3126 unsigned long trans_start, last_rx;
3127 struct list_head *iter;
3128 struct slave *slave;
3129 int commit = 0;
3130
3131 bond_for_each_slave_rcu(bond, slave, iter) {
3132 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3133 last_rx = slave_last_rx(bond, slave);
3134
3135 if (slave->link != BOND_LINK_UP) {
3136 if (bond_time_in_interval(bond, last_rx, 1)) {
3137 bond_propose_link_state(slave, BOND_LINK_UP);
3138 commit++;
3139 } else if (slave->link == BOND_LINK_BACK) {
3140 bond_propose_link_state(slave, BOND_LINK_FAIL);
3141 commit++;
3142 }
3143 continue;
3144 }
3145
3146 /* Give slaves 2*delta after being enslaved or made
3147 * active. This avoids bouncing, as the last receive
3148 * times need a full ARP monitor cycle to be updated.
3149 */
3150 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3151 continue;
3152
3153 /* Backup slave is down if:
3154 * - No current_arp_slave AND
3155 * - more than 3*delta since last receive AND
3156 * - the bond has an IP address
3157 *
3158 * Note: a non-null current_arp_slave indicates
3159 * the curr_active_slave went down and we are
3160 * searching for a new one; under this condition
3161 * we only take the curr_active_slave down - this
3162 * gives each slave a chance to tx/rx traffic
3163 * before being taken out
3164 */
3165 if (!bond_is_active_slave(slave) &&
3166 !rcu_access_pointer(bond->current_arp_slave) &&
3167 !bond_time_in_interval(bond, last_rx, 3)) {
3168 bond_propose_link_state(slave, BOND_LINK_DOWN);
3169 commit++;
3170 }
3171
3172 /* Active slave is down if:
3173 * - more than 2*delta since transmitting OR
3174 * - (more than 2*delta since receive AND
3175 * the bond has an IP address)
3176 */
3177 trans_start = dev_trans_start(slave->dev);
3178 if (bond_is_active_slave(slave) &&
3179 (!bond_time_in_interval(bond, trans_start, 2) ||
3180 !bond_time_in_interval(bond, last_rx, 2))) {
3181 bond_propose_link_state(slave, BOND_LINK_DOWN);
3182 commit++;
3183 }
3184 }
3185
3186 return commit;
3187}
3188
3189/* Called to commit link state changes noted by inspection step of
3190 * active-backup mode ARP monitor.
3191 *
3192 * Called with RTNL hold.
3193 */
3194static void bond_ab_arp_commit(struct bonding *bond)
3195{
3196 unsigned long trans_start;
3197 struct list_head *iter;
3198 struct slave *slave;
3199
3200 bond_for_each_slave(bond, slave, iter) {
3201 switch (slave->link_new_state) {
3202 case BOND_LINK_NOCHANGE:
3203 continue;
3204
3205 case BOND_LINK_UP:
3206 trans_start = dev_trans_start(slave->dev);
3207 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3208 (!rtnl_dereference(bond->curr_active_slave) &&
3209 bond_time_in_interval(bond, trans_start, 1))) {
3210 struct slave *current_arp_slave;
3211
3212 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3213 bond_set_slave_link_state(slave, BOND_LINK_UP,
3214 BOND_SLAVE_NOTIFY_NOW);
3215 if (current_arp_slave) {
3216 bond_set_slave_inactive_flags(
3217 current_arp_slave,
3218 BOND_SLAVE_NOTIFY_NOW);
3219 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3220 }
3221
3222 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3223
3224 if (!rtnl_dereference(bond->curr_active_slave) ||
3225 slave == rtnl_dereference(bond->primary_slave))
3226 goto do_failover;
3227
3228 }
3229
3230 continue;
3231
3232 case BOND_LINK_DOWN:
3233 if (slave->link_failure_count < UINT_MAX)
3234 slave->link_failure_count++;
3235
3236 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3237 BOND_SLAVE_NOTIFY_NOW);
3238 bond_set_slave_inactive_flags(slave,
3239 BOND_SLAVE_NOTIFY_NOW);
3240
3241 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3242
3243 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3244 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3245 goto do_failover;
3246 }
3247
3248 continue;
3249
3250 case BOND_LINK_FAIL:
3251 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3252 BOND_SLAVE_NOTIFY_NOW);
3253 bond_set_slave_inactive_flags(slave,
3254 BOND_SLAVE_NOTIFY_NOW);
3255
3256 /* A slave has just been enslaved and has become
3257 * the current active slave.
3258 */
3259 if (rtnl_dereference(bond->curr_active_slave))
3260 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3261 continue;
3262
3263 default:
3264 slave_err(bond->dev, slave->dev,
3265 "impossible: link_new_state %d on slave\n",
3266 slave->link_new_state);
3267 continue;
3268 }
3269
3270do_failover:
3271 block_netpoll_tx();
3272 bond_select_active_slave(bond);
3273 unblock_netpoll_tx();
3274 }
3275
3276 bond_set_carrier(bond);
3277}
3278
3279/* Send ARP probes for active-backup mode ARP monitor.
3280 *
3281 * Called with rcu_read_lock held.
3282 */
3283static bool bond_ab_arp_probe(struct bonding *bond)
3284{
3285 struct slave *slave, *before = NULL, *new_slave = NULL,
3286 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3287 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3288 struct list_head *iter;
3289 bool found = false;
3290 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3291
3292 if (curr_arp_slave && curr_active_slave)
3293 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3294 curr_arp_slave->dev->name,
3295 curr_active_slave->dev->name);
3296
3297 if (curr_active_slave) {
3298 bond_arp_send_all(bond, curr_active_slave);
3299 return should_notify_rtnl;
3300 }
3301
3302 /* if we don't have a curr_active_slave, search for the next available
3303 * backup slave from the current_arp_slave and make it the candidate
3304 * for becoming the curr_active_slave
3305 */
3306
3307 if (!curr_arp_slave) {
3308 curr_arp_slave = bond_first_slave_rcu(bond);
3309 if (!curr_arp_slave)
3310 return should_notify_rtnl;
3311 }
3312
3313 bond_for_each_slave_rcu(bond, slave, iter) {
3314 if (!found && !before && bond_slave_is_up(slave))
3315 before = slave;
3316
3317 if (found && !new_slave && bond_slave_is_up(slave))
3318 new_slave = slave;
3319 /* if the link state is up at this point, we
3320 * mark it down - this can happen if we have
3321 * simultaneous link failures and
3322 * reselect_active_interface doesn't make this
3323 * one the current slave so it is still marked
3324 * up when it is actually down
3325 */
3326 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3327 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3328 BOND_SLAVE_NOTIFY_LATER);
3329 if (slave->link_failure_count < UINT_MAX)
3330 slave->link_failure_count++;
3331
3332 bond_set_slave_inactive_flags(slave,
3333 BOND_SLAVE_NOTIFY_LATER);
3334
3335 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3336 }
3337 if (slave == curr_arp_slave)
3338 found = true;
3339 }
3340
3341 if (!new_slave && before)
3342 new_slave = before;
3343
3344 if (!new_slave)
3345 goto check_state;
3346
3347 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3348 BOND_SLAVE_NOTIFY_LATER);
3349 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3350 bond_arp_send_all(bond, new_slave);
3351 new_slave->last_link_up = jiffies;
3352 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3353
3354check_state:
3355 bond_for_each_slave_rcu(bond, slave, iter) {
3356 if (slave->should_notify || slave->should_notify_link) {
3357 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3358 break;
3359 }
3360 }
3361 return should_notify_rtnl;
3362}
3363
3364static void bond_activebackup_arp_mon(struct bonding *bond)
3365{
3366 bool should_notify_peers = false;
3367 bool should_notify_rtnl = false;
3368 int delta_in_ticks;
3369
3370 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3371
3372 if (!bond_has_slaves(bond))
3373 goto re_arm;
3374
3375 rcu_read_lock();
3376
3377 should_notify_peers = bond_should_notify_peers(bond);
3378
3379 if (bond_ab_arp_inspect(bond)) {
3380 rcu_read_unlock();
3381
3382 /* Race avoidance with bond_close flush of workqueue */
3383 if (!rtnl_trylock()) {
3384 delta_in_ticks = 1;
3385 should_notify_peers = false;
3386 goto re_arm;
3387 }
3388
3389 bond_ab_arp_commit(bond);
3390
3391 rtnl_unlock();
3392 rcu_read_lock();
3393 }
3394
3395 should_notify_rtnl = bond_ab_arp_probe(bond);
3396 rcu_read_unlock();
3397
3398re_arm:
3399 if (bond->params.arp_interval)
3400 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3401
3402 if (should_notify_peers || should_notify_rtnl) {
3403 if (!rtnl_trylock())
3404 return;
3405
3406 if (should_notify_peers)
3407 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3408 bond->dev);
3409 if (should_notify_rtnl) {
3410 bond_slave_state_notify(bond);
3411 bond_slave_link_notify(bond);
3412 }
3413
3414 rtnl_unlock();
3415 }
3416}
3417
3418static void bond_arp_monitor(struct work_struct *work)
3419{
3420 struct bonding *bond = container_of(work, struct bonding,
3421 arp_work.work);
3422
3423 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3424 bond_activebackup_arp_mon(bond);
3425 else
3426 bond_loadbalance_arp_mon(bond);
3427}
3428
3429/*-------------------------- netdev event handling --------------------------*/
3430
3431/* Change device name */
3432static int bond_event_changename(struct bonding *bond)
3433{
3434 bond_remove_proc_entry(bond);
3435 bond_create_proc_entry(bond);
3436
3437 bond_debug_reregister(bond);
3438
3439 return NOTIFY_DONE;
3440}
3441
3442static int bond_master_netdev_event(unsigned long event,
3443 struct net_device *bond_dev)
3444{
3445 struct bonding *event_bond = netdev_priv(bond_dev);
3446
3447 netdev_dbg(bond_dev, "%s called\n", __func__);
3448
3449 switch (event) {
3450 case NETDEV_CHANGENAME:
3451 return bond_event_changename(event_bond);
3452 case NETDEV_UNREGISTER:
3453 bond_remove_proc_entry(event_bond);
3454#ifdef CONFIG_XFRM_OFFLOAD
3455 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3456#endif /* CONFIG_XFRM_OFFLOAD */
3457 break;
3458 case NETDEV_REGISTER:
3459 bond_create_proc_entry(event_bond);
3460 break;
3461 default:
3462 break;
3463 }
3464
3465 return NOTIFY_DONE;
3466}
3467
3468static int bond_slave_netdev_event(unsigned long event,
3469 struct net_device *slave_dev)
3470{
3471 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3472 struct bonding *bond;
3473 struct net_device *bond_dev;
3474
3475 /* A netdev event can be generated while enslaving a device
3476 * before netdev_rx_handler_register is called in which case
3477 * slave will be NULL
3478 */
3479 if (!slave) {
3480 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3481 return NOTIFY_DONE;
3482 }
3483
3484 bond_dev = slave->bond->dev;
3485 bond = slave->bond;
3486 primary = rtnl_dereference(bond->primary_slave);
3487
3488 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3489
3490 switch (event) {
3491 case NETDEV_UNREGISTER:
3492 if (bond_dev->type != ARPHRD_ETHER)
3493 bond_release_and_destroy(bond_dev, slave_dev);
3494 else
3495 __bond_release_one(bond_dev, slave_dev, false, true);
3496 break;
3497 case NETDEV_UP:
3498 case NETDEV_CHANGE:
3499 /* For 802.3ad mode only:
3500 * Getting invalid Speed/Duplex values here will put slave
3501 * in weird state. Mark it as link-fail if the link was
3502 * previously up or link-down if it hasn't yet come up, and
3503 * let link-monitoring (miimon) set it right when correct
3504 * speeds/duplex are available.
3505 */
3506 if (bond_update_speed_duplex(slave) &&
3507 BOND_MODE(bond) == BOND_MODE_8023AD) {
3508 if (slave->last_link_up)
3509 slave->link = BOND_LINK_FAIL;
3510 else
3511 slave->link = BOND_LINK_DOWN;
3512 }
3513
3514 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3515 bond_3ad_adapter_speed_duplex_changed(slave);
3516 fallthrough;
3517 case NETDEV_DOWN:
3518 /* Refresh slave-array if applicable!
3519 * If the setup does not use miimon or arpmon (mode-specific!),
3520 * then these events will not cause the slave-array to be
3521 * refreshed. This will cause xmit to use a slave that is not
3522 * usable. Avoid such situation by refeshing the array at these
3523 * events. If these (miimon/arpmon) parameters are configured
3524 * then array gets refreshed twice and that should be fine!
3525 */
3526 if (bond_mode_can_use_xmit_hash(bond))
3527 bond_update_slave_arr(bond, NULL);
3528 break;
3529 case NETDEV_CHANGEMTU:
3530 /* TODO: Should slaves be allowed to
3531 * independently alter their MTU? For
3532 * an active-backup bond, slaves need
3533 * not be the same type of device, so
3534 * MTUs may vary. For other modes,
3535 * slaves arguably should have the
3536 * same MTUs. To do this, we'd need to
3537 * take over the slave's change_mtu
3538 * function for the duration of their
3539 * servitude.
3540 */
3541 break;
3542 case NETDEV_CHANGENAME:
3543 /* we don't care if we don't have primary set */
3544 if (!bond_uses_primary(bond) ||
3545 !bond->params.primary[0])
3546 break;
3547
3548 if (slave == primary) {
3549 /* slave's name changed - he's no longer primary */
3550 RCU_INIT_POINTER(bond->primary_slave, NULL);
3551 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3552 /* we have a new primary slave */
3553 rcu_assign_pointer(bond->primary_slave, slave);
3554 } else { /* we didn't change primary - exit */
3555 break;
3556 }
3557
3558 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3559 primary ? slave_dev->name : "none");
3560
3561 block_netpoll_tx();
3562 bond_select_active_slave(bond);
3563 unblock_netpoll_tx();
3564 break;
3565 case NETDEV_FEAT_CHANGE:
3566 bond_compute_features(bond);
3567 break;
3568 case NETDEV_RESEND_IGMP:
3569 /* Propagate to master device */
3570 call_netdevice_notifiers(event, slave->bond->dev);
3571 break;
3572 default:
3573 break;
3574 }
3575
3576 return NOTIFY_DONE;
3577}
3578
3579/* bond_netdev_event: handle netdev notifier chain events.
3580 *
3581 * This function receives events for the netdev chain. The caller (an
3582 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3583 * locks for us to safely manipulate the slave devices (RTNL lock,
3584 * dev_probe_lock).
3585 */
3586static int bond_netdev_event(struct notifier_block *this,
3587 unsigned long event, void *ptr)
3588{
3589 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3590
3591 netdev_dbg(event_dev, "%s received %s\n",
3592 __func__, netdev_cmd_to_name(event));
3593
3594 if (!(event_dev->priv_flags & IFF_BONDING))
3595 return NOTIFY_DONE;
3596
3597 if (event_dev->flags & IFF_MASTER) {
3598 int ret;
3599
3600 ret = bond_master_netdev_event(event, event_dev);
3601 if (ret != NOTIFY_DONE)
3602 return ret;
3603 }
3604
3605 if (event_dev->flags & IFF_SLAVE)
3606 return bond_slave_netdev_event(event, event_dev);
3607
3608 return NOTIFY_DONE;
3609}
3610
3611static struct notifier_block bond_netdev_notifier = {
3612 .notifier_call = bond_netdev_event,
3613};
3614
3615/*---------------------------- Hashing Policies -----------------------------*/
3616
3617/* L2 hash helper */
3618static inline u32 bond_eth_hash(struct sk_buff *skb)
3619{
3620 struct ethhdr *ep, hdr_tmp;
3621
3622 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3623 if (ep)
3624 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3625 return 0;
3626}
3627
3628static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3629 int *noff, int *proto, bool l34)
3630{
3631 const struct ipv6hdr *iph6;
3632 const struct iphdr *iph;
3633
3634 if (skb->protocol == htons(ETH_P_IP)) {
3635 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3636 return false;
3637 iph = (const struct iphdr *)(skb->data + *noff);
3638 iph_to_flow_copy_v4addrs(fk, iph);
3639 *noff += iph->ihl << 2;
3640 if (!ip_is_fragment(iph))
3641 *proto = iph->protocol;
3642 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3643 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3644 return false;
3645 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3646 iph_to_flow_copy_v6addrs(fk, iph6);
3647 *noff += sizeof(*iph6);
3648 *proto = iph6->nexthdr;
3649 } else {
3650 return false;
3651 }
3652
3653 if (l34 && *proto >= 0)
3654 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3655
3656 return true;
3657}
3658
3659static u32 bond_vlan_srcmac_hash(struct sk_buff *skb)
3660{
3661 struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
3662 u32 srcmac_vendor = 0, srcmac_dev = 0;
3663 u16 vlan;
3664 int i;
3665
3666 for (i = 0; i < 3; i++)
3667 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3668
3669 for (i = 3; i < ETH_ALEN; i++)
3670 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3671
3672 if (!skb_vlan_tag_present(skb))
3673 return srcmac_vendor ^ srcmac_dev;
3674
3675 vlan = skb_vlan_tag_get(skb);
3676
3677 return vlan ^ srcmac_vendor ^ srcmac_dev;
3678}
3679
3680/* Extract the appropriate headers based on bond's xmit policy */
3681static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3682 struct flow_keys *fk)
3683{
3684 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3685 int noff, proto = -1;
3686
3687 switch (bond->params.xmit_policy) {
3688 case BOND_XMIT_POLICY_ENCAP23:
3689 case BOND_XMIT_POLICY_ENCAP34:
3690 memset(fk, 0, sizeof(*fk));
3691 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3692 fk, NULL, 0, 0, 0, 0);
3693 default:
3694 break;
3695 }
3696
3697 fk->ports.ports = 0;
3698 memset(&fk->icmp, 0, sizeof(fk->icmp));
3699 noff = skb_network_offset(skb);
3700 if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3701 return false;
3702
3703 /* ICMP error packets contains at least 8 bytes of the header
3704 * of the packet which generated the error. Use this information
3705 * to correlate ICMP error packets within the same flow which
3706 * generated the error.
3707 */
3708 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3709 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3710 skb_transport_offset(skb),
3711 skb_headlen(skb));
3712 if (proto == IPPROTO_ICMP) {
3713 if (!icmp_is_err(fk->icmp.type))
3714 return true;
3715
3716 noff += sizeof(struct icmphdr);
3717 } else if (proto == IPPROTO_ICMPV6) {
3718 if (!icmpv6_is_err(fk->icmp.type))
3719 return true;
3720
3721 noff += sizeof(struct icmp6hdr);
3722 }
3723 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3724 }
3725
3726 return true;
3727}
3728
3729static u32 bond_ip_hash(u32 hash, struct flow_keys *flow)
3730{
3731 hash ^= (__force u32)flow_get_u32_dst(flow) ^
3732 (__force u32)flow_get_u32_src(flow);
3733 hash ^= (hash >> 16);
3734 hash ^= (hash >> 8);
3735 /* discard lowest hash bit to deal with the common even ports pattern */
3736 return hash >> 1;
3737}
3738
3739/**
3740 * bond_xmit_hash - generate a hash value based on the xmit policy
3741 * @bond: bonding device
3742 * @skb: buffer to use for headers
3743 *
3744 * This function will extract the necessary headers from the skb buffer and use
3745 * them to generate a hash based on the xmit_policy set in the bonding device
3746 */
3747u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3748{
3749 struct flow_keys flow;
3750 u32 hash;
3751
3752 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3753 skb->l4_hash)
3754 return skb->hash;
3755
3756 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
3757 return bond_vlan_srcmac_hash(skb);
3758
3759 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3760 !bond_flow_dissect(bond, skb, &flow))
3761 return bond_eth_hash(skb);
3762
3763 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3764 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3765 hash = bond_eth_hash(skb);
3766 } else {
3767 if (flow.icmp.id)
3768 memcpy(&hash, &flow.icmp, sizeof(hash));
3769 else
3770 memcpy(&hash, &flow.ports.ports, sizeof(hash));
3771 }
3772
3773 return bond_ip_hash(hash, &flow);
3774}
3775
3776/*-------------------------- Device entry points ----------------------------*/
3777
3778void bond_work_init_all(struct bonding *bond)
3779{
3780 INIT_DELAYED_WORK(&bond->mcast_work,
3781 bond_resend_igmp_join_requests_delayed);
3782 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3783 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3784 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3785 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3786 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3787}
3788
3789static void bond_work_cancel_all(struct bonding *bond)
3790{
3791 cancel_delayed_work_sync(&bond->mii_work);
3792 cancel_delayed_work_sync(&bond->arp_work);
3793 cancel_delayed_work_sync(&bond->alb_work);
3794 cancel_delayed_work_sync(&bond->ad_work);
3795 cancel_delayed_work_sync(&bond->mcast_work);
3796 cancel_delayed_work_sync(&bond->slave_arr_work);
3797}
3798
3799static int bond_open(struct net_device *bond_dev)
3800{
3801 struct bonding *bond = netdev_priv(bond_dev);
3802 struct list_head *iter;
3803 struct slave *slave;
3804
3805 /* reset slave->backup and slave->inactive */
3806 if (bond_has_slaves(bond)) {
3807 bond_for_each_slave(bond, slave, iter) {
3808 if (bond_uses_primary(bond) &&
3809 slave != rcu_access_pointer(bond->curr_active_slave)) {
3810 bond_set_slave_inactive_flags(slave,
3811 BOND_SLAVE_NOTIFY_NOW);
3812 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3813 bond_set_slave_active_flags(slave,
3814 BOND_SLAVE_NOTIFY_NOW);
3815 }
3816 }
3817 }
3818
3819 if (bond_is_lb(bond)) {
3820 /* bond_alb_initialize must be called before the timer
3821 * is started.
3822 */
3823 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3824 return -ENOMEM;
3825 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3826 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3827 }
3828
3829 if (bond->params.miimon) /* link check interval, in milliseconds. */
3830 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3831
3832 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3833 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3834 bond->recv_probe = bond_arp_rcv;
3835 }
3836
3837 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3838 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3839 /* register to receive LACPDUs */
3840 bond->recv_probe = bond_3ad_lacpdu_recv;
3841 bond_3ad_initiate_agg_selection(bond, 1);
3842 }
3843
3844 if (bond_mode_can_use_xmit_hash(bond))
3845 bond_update_slave_arr(bond, NULL);
3846
3847 return 0;
3848}
3849
3850static int bond_close(struct net_device *bond_dev)
3851{
3852 struct bonding *bond = netdev_priv(bond_dev);
3853
3854 bond_work_cancel_all(bond);
3855 bond->send_peer_notif = 0;
3856 if (bond_is_lb(bond))
3857 bond_alb_deinitialize(bond);
3858 bond->recv_probe = NULL;
3859
3860 return 0;
3861}
3862
3863/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3864 * that some drivers can provide 32bit values only.
3865 */
3866static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3867 const struct rtnl_link_stats64 *_new,
3868 const struct rtnl_link_stats64 *_old)
3869{
3870 const u64 *new = (const u64 *)_new;
3871 const u64 *old = (const u64 *)_old;
3872 u64 *res = (u64 *)_res;
3873 int i;
3874
3875 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3876 u64 nv = new[i];
3877 u64 ov = old[i];
3878 s64 delta = nv - ov;
3879
3880 /* detects if this particular field is 32bit only */
3881 if (((nv | ov) >> 32) == 0)
3882 delta = (s64)(s32)((u32)nv - (u32)ov);
3883
3884 /* filter anomalies, some drivers reset their stats
3885 * at down/up events.
3886 */
3887 if (delta > 0)
3888 res[i] += delta;
3889 }
3890}
3891
3892#ifdef CONFIG_LOCKDEP
3893static int bond_get_lowest_level_rcu(struct net_device *dev)
3894{
3895 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3896 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3897 int cur = 0, max = 0;
3898
3899 now = dev;
3900 iter = &dev->adj_list.lower;
3901
3902 while (1) {
3903 next = NULL;
3904 while (1) {
3905 ldev = netdev_next_lower_dev_rcu(now, &iter);
3906 if (!ldev)
3907 break;
3908
3909 next = ldev;
3910 niter = &ldev->adj_list.lower;
3911 dev_stack[cur] = now;
3912 iter_stack[cur++] = iter;
3913 if (max <= cur)
3914 max = cur;
3915 break;
3916 }
3917
3918 if (!next) {
3919 if (!cur)
3920 return max;
3921 next = dev_stack[--cur];
3922 niter = iter_stack[cur];
3923 }
3924
3925 now = next;
3926 iter = niter;
3927 }
3928
3929 return max;
3930}
3931#endif
3932
3933static void bond_get_stats(struct net_device *bond_dev,
3934 struct rtnl_link_stats64 *stats)
3935{
3936 struct bonding *bond = netdev_priv(bond_dev);
3937 struct rtnl_link_stats64 temp;
3938 struct list_head *iter;
3939 struct slave *slave;
3940 int nest_level = 0;
3941
3942
3943 rcu_read_lock();
3944#ifdef CONFIG_LOCKDEP
3945 nest_level = bond_get_lowest_level_rcu(bond_dev);
3946#endif
3947
3948 spin_lock_nested(&bond->stats_lock, nest_level);
3949 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3950
3951 bond_for_each_slave_rcu(bond, slave, iter) {
3952 const struct rtnl_link_stats64 *new =
3953 dev_get_stats(slave->dev, &temp);
3954
3955 bond_fold_stats(stats, new, &slave->slave_stats);
3956
3957 /* save off the slave stats for the next run */
3958 memcpy(&slave->slave_stats, new, sizeof(*new));
3959 }
3960
3961 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3962 spin_unlock(&bond->stats_lock);
3963 rcu_read_unlock();
3964}
3965
3966static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3967{
3968 struct bonding *bond = netdev_priv(bond_dev);
3969 struct net_device *slave_dev = NULL;
3970 struct ifbond k_binfo;
3971 struct ifbond __user *u_binfo = NULL;
3972 struct ifslave k_sinfo;
3973 struct ifslave __user *u_sinfo = NULL;
3974 struct mii_ioctl_data *mii = NULL;
3975 struct bond_opt_value newval;
3976 struct net *net;
3977 int res = 0;
3978
3979 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3980
3981 switch (cmd) {
3982 case SIOCGMIIPHY:
3983 mii = if_mii(ifr);
3984 if (!mii)
3985 return -EINVAL;
3986
3987 mii->phy_id = 0;
3988 fallthrough;
3989 case SIOCGMIIREG:
3990 /* We do this again just in case we were called by SIOCGMIIREG
3991 * instead of SIOCGMIIPHY.
3992 */
3993 mii = if_mii(ifr);
3994 if (!mii)
3995 return -EINVAL;
3996
3997 if (mii->reg_num == 1) {
3998 mii->val_out = 0;
3999 if (netif_carrier_ok(bond->dev))
4000 mii->val_out = BMSR_LSTATUS;
4001 }
4002
4003 return 0;
4004 case BOND_INFO_QUERY_OLD:
4005 case SIOCBONDINFOQUERY:
4006 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4007
4008 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4009 return -EFAULT;
4010
4011 bond_info_query(bond_dev, &k_binfo);
4012 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4013 return -EFAULT;
4014
4015 return 0;
4016 case BOND_SLAVE_INFO_QUERY_OLD:
4017 case SIOCBONDSLAVEINFOQUERY:
4018 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4019
4020 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4021 return -EFAULT;
4022
4023 res = bond_slave_info_query(bond_dev, &k_sinfo);
4024 if (res == 0 &&
4025 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4026 return -EFAULT;
4027
4028 return res;
4029 default:
4030 break;
4031 }
4032
4033 net = dev_net(bond_dev);
4034
4035 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4036 return -EPERM;
4037
4038 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4039
4040 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4041
4042 if (!slave_dev)
4043 return -ENODEV;
4044
4045 switch (cmd) {
4046 case BOND_ENSLAVE_OLD:
4047 case SIOCBONDENSLAVE:
4048 res = bond_enslave(bond_dev, slave_dev, NULL);
4049 break;
4050 case BOND_RELEASE_OLD:
4051 case SIOCBONDRELEASE:
4052 res = bond_release(bond_dev, slave_dev);
4053 break;
4054 case BOND_SETHWADDR_OLD:
4055 case SIOCBONDSETHWADDR:
4056 res = bond_set_dev_addr(bond_dev, slave_dev);
4057 break;
4058 case BOND_CHANGE_ACTIVE_OLD:
4059 case SIOCBONDCHANGEACTIVE:
4060 bond_opt_initstr(&newval, slave_dev->name);
4061 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4062 &newval);
4063 break;
4064 default:
4065 res = -EOPNOTSUPP;
4066 }
4067
4068 return res;
4069}
4070
4071static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4072{
4073 struct bonding *bond = netdev_priv(bond_dev);
4074
4075 if (change & IFF_PROMISC)
4076 bond_set_promiscuity(bond,
4077 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4078
4079 if (change & IFF_ALLMULTI)
4080 bond_set_allmulti(bond,
4081 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4082}
4083
4084static void bond_set_rx_mode(struct net_device *bond_dev)
4085{
4086 struct bonding *bond = netdev_priv(bond_dev);
4087 struct list_head *iter;
4088 struct slave *slave;
4089
4090 rcu_read_lock();
4091 if (bond_uses_primary(bond)) {
4092 slave = rcu_dereference(bond->curr_active_slave);
4093 if (slave) {
4094 dev_uc_sync(slave->dev, bond_dev);
4095 dev_mc_sync(slave->dev, bond_dev);
4096 }
4097 } else {
4098 bond_for_each_slave_rcu(bond, slave, iter) {
4099 dev_uc_sync_multiple(slave->dev, bond_dev);
4100 dev_mc_sync_multiple(slave->dev, bond_dev);
4101 }
4102 }
4103 rcu_read_unlock();
4104}
4105
4106static int bond_neigh_init(struct neighbour *n)
4107{
4108 struct bonding *bond = netdev_priv(n->dev);
4109 const struct net_device_ops *slave_ops;
4110 struct neigh_parms parms;
4111 struct slave *slave;
4112 int ret = 0;
4113
4114 rcu_read_lock();
4115 slave = bond_first_slave_rcu(bond);
4116 if (!slave)
4117 goto out;
4118 slave_ops = slave->dev->netdev_ops;
4119 if (!slave_ops->ndo_neigh_setup)
4120 goto out;
4121
4122 /* TODO: find another way [1] to implement this.
4123 * Passing a zeroed structure is fragile,
4124 * but at least we do not pass garbage.
4125 *
4126 * [1] One way would be that ndo_neigh_setup() never touch
4127 * struct neigh_parms, but propagate the new neigh_setup()
4128 * back to ___neigh_create() / neigh_parms_alloc()
4129 */
4130 memset(&parms, 0, sizeof(parms));
4131 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4132
4133 if (ret)
4134 goto out;
4135
4136 if (parms.neigh_setup)
4137 ret = parms.neigh_setup(n);
4138out:
4139 rcu_read_unlock();
4140 return ret;
4141}
4142
4143/* The bonding ndo_neigh_setup is called at init time beofre any
4144 * slave exists. So we must declare proxy setup function which will
4145 * be used at run time to resolve the actual slave neigh param setup.
4146 *
4147 * It's also called by master devices (such as vlans) to setup their
4148 * underlying devices. In that case - do nothing, we're already set up from
4149 * our init.
4150 */
4151static int bond_neigh_setup(struct net_device *dev,
4152 struct neigh_parms *parms)
4153{
4154 /* modify only our neigh_parms */
4155 if (parms->dev == dev)
4156 parms->neigh_setup = bond_neigh_init;
4157
4158 return 0;
4159}
4160
4161/* Change the MTU of all of a master's slaves to match the master */
4162static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4163{
4164 struct bonding *bond = netdev_priv(bond_dev);
4165 struct slave *slave, *rollback_slave;
4166 struct list_head *iter;
4167 int res = 0;
4168
4169 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4170
4171 bond_for_each_slave(bond, slave, iter) {
4172 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4173 slave, slave->dev->netdev_ops->ndo_change_mtu);
4174
4175 res = dev_set_mtu(slave->dev, new_mtu);
4176
4177 if (res) {
4178 /* If we failed to set the slave's mtu to the new value
4179 * we must abort the operation even in ACTIVE_BACKUP
4180 * mode, because if we allow the backup slaves to have
4181 * different mtu values than the active slave we'll
4182 * need to change their mtu when doing a failover. That
4183 * means changing their mtu from timer context, which
4184 * is probably not a good idea.
4185 */
4186 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4187 res, new_mtu);
4188 goto unwind;
4189 }
4190 }
4191
4192 bond_dev->mtu = new_mtu;
4193
4194 return 0;
4195
4196unwind:
4197 /* unwind from head to the slave that failed */
4198 bond_for_each_slave(bond, rollback_slave, iter) {
4199 int tmp_res;
4200
4201 if (rollback_slave == slave)
4202 break;
4203
4204 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4205 if (tmp_res)
4206 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4207 tmp_res);
4208 }
4209
4210 return res;
4211}
4212
4213/* Change HW address
4214 *
4215 * Note that many devices must be down to change the HW address, and
4216 * downing the master releases all slaves. We can make bonds full of
4217 * bonding devices to test this, however.
4218 */
4219static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4220{
4221 struct bonding *bond = netdev_priv(bond_dev);
4222 struct slave *slave, *rollback_slave;
4223 struct sockaddr_storage *ss = addr, tmp_ss;
4224 struct list_head *iter;
4225 int res = 0;
4226
4227 if (BOND_MODE(bond) == BOND_MODE_ALB)
4228 return bond_alb_set_mac_address(bond_dev, addr);
4229
4230
4231 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4232
4233 /* If fail_over_mac is enabled, do nothing and return success.
4234 * Returning an error causes ifenslave to fail.
4235 */
4236 if (bond->params.fail_over_mac &&
4237 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4238 return 0;
4239
4240 if (!is_valid_ether_addr(ss->__data))
4241 return -EADDRNOTAVAIL;
4242
4243 bond_for_each_slave(bond, slave, iter) {
4244 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4245 __func__, slave);
4246 res = dev_set_mac_address(slave->dev, addr, NULL);
4247 if (res) {
4248 /* TODO: consider downing the slave
4249 * and retry ?
4250 * User should expect communications
4251 * breakage anyway until ARP finish
4252 * updating, so...
4253 */
4254 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4255 __func__, res);
4256 goto unwind;
4257 }
4258 }
4259
4260 /* success */
4261 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4262 return 0;
4263
4264unwind:
4265 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4266 tmp_ss.ss_family = bond_dev->type;
4267
4268 /* unwind from head to the slave that failed */
4269 bond_for_each_slave(bond, rollback_slave, iter) {
4270 int tmp_res;
4271
4272 if (rollback_slave == slave)
4273 break;
4274
4275 tmp_res = dev_set_mac_address(rollback_slave->dev,
4276 (struct sockaddr *)&tmp_ss, NULL);
4277 if (tmp_res) {
4278 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4279 __func__, tmp_res);
4280 }
4281 }
4282
4283 return res;
4284}
4285
4286/**
4287 * bond_get_slave_by_id - get xmit slave with slave_id
4288 * @bond: bonding device that is transmitting
4289 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4290 *
4291 * This function tries to get slave with slave_id but in case
4292 * it fails, it tries to find the first available slave for transmission.
4293 */
4294static struct slave *bond_get_slave_by_id(struct bonding *bond,
4295 int slave_id)
4296{
4297 struct list_head *iter;
4298 struct slave *slave;
4299 int i = slave_id;
4300
4301 /* Here we start from the slave with slave_id */
4302 bond_for_each_slave_rcu(bond, slave, iter) {
4303 if (--i < 0) {
4304 if (bond_slave_can_tx(slave))
4305 return slave;
4306 }
4307 }
4308
4309 /* Here we start from the first slave up to slave_id */
4310 i = slave_id;
4311 bond_for_each_slave_rcu(bond, slave, iter) {
4312 if (--i < 0)
4313 break;
4314 if (bond_slave_can_tx(slave))
4315 return slave;
4316 }
4317 /* no slave that can tx has been found */
4318 return NULL;
4319}
4320
4321/**
4322 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4323 * @bond: bonding device to use
4324 *
4325 * Based on the value of the bonding device's packets_per_slave parameter
4326 * this function generates a slave id, which is usually used as the next
4327 * slave to transmit through.
4328 */
4329static u32 bond_rr_gen_slave_id(struct bonding *bond)
4330{
4331 u32 slave_id;
4332 struct reciprocal_value reciprocal_packets_per_slave;
4333 int packets_per_slave = bond->params.packets_per_slave;
4334
4335 switch (packets_per_slave) {
4336 case 0:
4337 slave_id = prandom_u32();
4338 break;
4339 case 1:
4340 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4341 break;
4342 default:
4343 reciprocal_packets_per_slave =
4344 bond->params.reciprocal_packets_per_slave;
4345 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4346 slave_id = reciprocal_divide(slave_id,
4347 reciprocal_packets_per_slave);
4348 break;
4349 }
4350
4351 return slave_id;
4352}
4353
4354static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4355 struct sk_buff *skb)
4356{
4357 struct slave *slave;
4358 int slave_cnt;
4359 u32 slave_id;
4360
4361 /* Start with the curr_active_slave that joined the bond as the
4362 * default for sending IGMP traffic. For failover purposes one
4363 * needs to maintain some consistency for the interface that will
4364 * send the join/membership reports. The curr_active_slave found
4365 * will send all of this type of traffic.
4366 */
4367 if (skb->protocol == htons(ETH_P_IP)) {
4368 int noff = skb_network_offset(skb);
4369 struct iphdr *iph;
4370
4371 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4372 goto non_igmp;
4373
4374 iph = ip_hdr(skb);
4375 if (iph->protocol == IPPROTO_IGMP) {
4376 slave = rcu_dereference(bond->curr_active_slave);
4377 if (slave)
4378 return slave;
4379 return bond_get_slave_by_id(bond, 0);
4380 }
4381 }
4382
4383non_igmp:
4384 slave_cnt = READ_ONCE(bond->slave_cnt);
4385 if (likely(slave_cnt)) {
4386 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4387 return bond_get_slave_by_id(bond, slave_id);
4388 }
4389 return NULL;
4390}
4391
4392static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4393 struct net_device *bond_dev)
4394{
4395 struct bonding *bond = netdev_priv(bond_dev);
4396 struct slave *slave;
4397
4398 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4399 if (likely(slave))
4400 return bond_dev_queue_xmit(bond, skb, slave->dev);
4401
4402 return bond_tx_drop(bond_dev, skb);
4403}
4404
4405static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4406 struct sk_buff *skb)
4407{
4408 return rcu_dereference(bond->curr_active_slave);
4409}
4410
4411/* In active-backup mode, we know that bond->curr_active_slave is always valid if
4412 * the bond has a usable interface.
4413 */
4414static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4415 struct net_device *bond_dev)
4416{
4417 struct bonding *bond = netdev_priv(bond_dev);
4418 struct slave *slave;
4419
4420 slave = bond_xmit_activebackup_slave_get(bond, skb);
4421 if (slave)
4422 return bond_dev_queue_xmit(bond, skb, slave->dev);
4423
4424 return bond_tx_drop(bond_dev, skb);
4425}
4426
4427/* Use this to update slave_array when (a) it's not appropriate to update
4428 * slave_array right away (note that update_slave_array() may sleep)
4429 * and / or (b) RTNL is not held.
4430 */
4431void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4432{
4433 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4434}
4435
4436/* Slave array work handler. Holds only RTNL */
4437static void bond_slave_arr_handler(struct work_struct *work)
4438{
4439 struct bonding *bond = container_of(work, struct bonding,
4440 slave_arr_work.work);
4441 int ret;
4442
4443 if (!rtnl_trylock())
4444 goto err;
4445
4446 ret = bond_update_slave_arr(bond, NULL);
4447 rtnl_unlock();
4448 if (ret) {
4449 pr_warn_ratelimited("Failed to update slave array from WT\n");
4450 goto err;
4451 }
4452 return;
4453
4454err:
4455 bond_slave_arr_work_rearm(bond, 1);
4456}
4457
4458static void bond_skip_slave(struct bond_up_slave *slaves,
4459 struct slave *skipslave)
4460{
4461 int idx;
4462
4463 /* Rare situation where caller has asked to skip a specific
4464 * slave but allocation failed (most likely!). BTW this is
4465 * only possible when the call is initiated from
4466 * __bond_release_one(). In this situation; overwrite the
4467 * skipslave entry in the array with the last entry from the
4468 * array to avoid a situation where the xmit path may choose
4469 * this to-be-skipped slave to send a packet out.
4470 */
4471 for (idx = 0; slaves && idx < slaves->count; idx++) {
4472 if (skipslave == slaves->arr[idx]) {
4473 slaves->arr[idx] =
4474 slaves->arr[slaves->count - 1];
4475 slaves->count--;
4476 break;
4477 }
4478 }
4479}
4480
4481static void bond_set_slave_arr(struct bonding *bond,
4482 struct bond_up_slave *usable_slaves,
4483 struct bond_up_slave *all_slaves)
4484{
4485 struct bond_up_slave *usable, *all;
4486
4487 usable = rtnl_dereference(bond->usable_slaves);
4488 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4489 kfree_rcu(usable, rcu);
4490
4491 all = rtnl_dereference(bond->all_slaves);
4492 rcu_assign_pointer(bond->all_slaves, all_slaves);
4493 kfree_rcu(all, rcu);
4494}
4495
4496static void bond_reset_slave_arr(struct bonding *bond)
4497{
4498 struct bond_up_slave *usable, *all;
4499
4500 usable = rtnl_dereference(bond->usable_slaves);
4501 if (usable) {
4502 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4503 kfree_rcu(usable, rcu);
4504 }
4505
4506 all = rtnl_dereference(bond->all_slaves);
4507 if (all) {
4508 RCU_INIT_POINTER(bond->all_slaves, NULL);
4509 kfree_rcu(all, rcu);
4510 }
4511}
4512
4513/* Build the usable slaves array in control path for modes that use xmit-hash
4514 * to determine the slave interface -
4515 * (a) BOND_MODE_8023AD
4516 * (b) BOND_MODE_XOR
4517 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4518 *
4519 * The caller is expected to hold RTNL only and NO other lock!
4520 */
4521int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4522{
4523 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4524 struct slave *slave;
4525 struct list_head *iter;
4526 int agg_id = 0;
4527 int ret = 0;
4528
4529 might_sleep();
4530
4531 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4532 bond->slave_cnt), GFP_KERNEL);
4533 all_slaves = kzalloc(struct_size(all_slaves, arr,
4534 bond->slave_cnt), GFP_KERNEL);
4535 if (!usable_slaves || !all_slaves) {
4536 ret = -ENOMEM;
4537 goto out;
4538 }
4539 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4540 struct ad_info ad_info;
4541
4542 spin_lock_bh(&bond->mode_lock);
4543 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4544 spin_unlock_bh(&bond->mode_lock);
4545 pr_debug("bond_3ad_get_active_agg_info failed\n");
4546 /* No active aggragator means it's not safe to use
4547 * the previous array.
4548 */
4549 bond_reset_slave_arr(bond);
4550 goto out;
4551 }
4552 spin_unlock_bh(&bond->mode_lock);
4553 agg_id = ad_info.aggregator_id;
4554 }
4555 bond_for_each_slave(bond, slave, iter) {
4556 if (skipslave == slave)
4557 continue;
4558
4559 all_slaves->arr[all_slaves->count++] = slave;
4560 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4561 struct aggregator *agg;
4562
4563 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4564 if (!agg || agg->aggregator_identifier != agg_id)
4565 continue;
4566 }
4567 if (!bond_slave_can_tx(slave))
4568 continue;
4569
4570 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4571 usable_slaves->count);
4572
4573 usable_slaves->arr[usable_slaves->count++] = slave;
4574 }
4575
4576 bond_set_slave_arr(bond, usable_slaves, all_slaves);
4577 return ret;
4578out:
4579 if (ret != 0 && skipslave) {
4580 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4581 skipslave);
4582 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4583 skipslave);
4584 }
4585 kfree_rcu(all_slaves, rcu);
4586 kfree_rcu(usable_slaves, rcu);
4587
4588 return ret;
4589}
4590
4591static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4592 struct sk_buff *skb,
4593 struct bond_up_slave *slaves)
4594{
4595 struct slave *slave;
4596 unsigned int count;
4597 u32 hash;
4598
4599 hash = bond_xmit_hash(bond, skb);
4600 count = slaves ? READ_ONCE(slaves->count) : 0;
4601 if (unlikely(!count))
4602 return NULL;
4603
4604 slave = slaves->arr[hash % count];
4605 return slave;
4606}
4607
4608/* Use this Xmit function for 3AD as well as XOR modes. The current
4609 * usable slave array is formed in the control path. The xmit function
4610 * just calculates hash and sends the packet out.
4611 */
4612static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4613 struct net_device *dev)
4614{
4615 struct bonding *bond = netdev_priv(dev);
4616 struct bond_up_slave *slaves;
4617 struct slave *slave;
4618
4619 slaves = rcu_dereference(bond->usable_slaves);
4620 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4621 if (likely(slave))
4622 return bond_dev_queue_xmit(bond, skb, slave->dev);
4623
4624 return bond_tx_drop(dev, skb);
4625}
4626
4627/* in broadcast mode, we send everything to all usable interfaces. */
4628static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4629 struct net_device *bond_dev)
4630{
4631 struct bonding *bond = netdev_priv(bond_dev);
4632 struct slave *slave = NULL;
4633 struct list_head *iter;
4634
4635 bond_for_each_slave_rcu(bond, slave, iter) {
4636 if (bond_is_last_slave(bond, slave))
4637 break;
4638 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4639 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4640
4641 if (!skb2) {
4642 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4643 bond_dev->name, __func__);
4644 continue;
4645 }
4646 bond_dev_queue_xmit(bond, skb2, slave->dev);
4647 }
4648 }
4649 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4650 return bond_dev_queue_xmit(bond, skb, slave->dev);
4651
4652 return bond_tx_drop(bond_dev, skb);
4653}
4654
4655/*------------------------- Device initialization ---------------------------*/
4656
4657/* Lookup the slave that corresponds to a qid */
4658static inline int bond_slave_override(struct bonding *bond,
4659 struct sk_buff *skb)
4660{
4661 struct slave *slave = NULL;
4662 struct list_head *iter;
4663
4664 if (!skb_rx_queue_recorded(skb))
4665 return 1;
4666
4667 /* Find out if any slaves have the same mapping as this skb. */
4668 bond_for_each_slave_rcu(bond, slave, iter) {
4669 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4670 if (bond_slave_is_up(slave) &&
4671 slave->link == BOND_LINK_UP) {
4672 bond_dev_queue_xmit(bond, skb, slave->dev);
4673 return 0;
4674 }
4675 /* If the slave isn't UP, use default transmit policy. */
4676 break;
4677 }
4678 }
4679
4680 return 1;
4681}
4682
4683
4684static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4685 struct net_device *sb_dev)
4686{
4687 /* This helper function exists to help dev_pick_tx get the correct
4688 * destination queue. Using a helper function skips a call to
4689 * skb_tx_hash and will put the skbs in the queue we expect on their
4690 * way down to the bonding driver.
4691 */
4692 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4693
4694 /* Save the original txq to restore before passing to the driver */
4695 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4696
4697 if (unlikely(txq >= dev->real_num_tx_queues)) {
4698 do {
4699 txq -= dev->real_num_tx_queues;
4700 } while (txq >= dev->real_num_tx_queues);
4701 }
4702 return txq;
4703}
4704
4705static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4706 struct sk_buff *skb,
4707 bool all_slaves)
4708{
4709 struct bonding *bond = netdev_priv(master_dev);
4710 struct bond_up_slave *slaves;
4711 struct slave *slave = NULL;
4712
4713 switch (BOND_MODE(bond)) {
4714 case BOND_MODE_ROUNDROBIN:
4715 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4716 break;
4717 case BOND_MODE_ACTIVEBACKUP:
4718 slave = bond_xmit_activebackup_slave_get(bond, skb);
4719 break;
4720 case BOND_MODE_8023AD:
4721 case BOND_MODE_XOR:
4722 if (all_slaves)
4723 slaves = rcu_dereference(bond->all_slaves);
4724 else
4725 slaves = rcu_dereference(bond->usable_slaves);
4726 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4727 break;
4728 case BOND_MODE_BROADCAST:
4729 break;
4730 case BOND_MODE_ALB:
4731 slave = bond_xmit_alb_slave_get(bond, skb);
4732 break;
4733 case BOND_MODE_TLB:
4734 slave = bond_xmit_tlb_slave_get(bond, skb);
4735 break;
4736 default:
4737 /* Should never happen, mode already checked */
4738 WARN_ONCE(true, "Unknown bonding mode");
4739 break;
4740 }
4741
4742 if (slave)
4743 return slave->dev;
4744 return NULL;
4745}
4746
4747static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
4748{
4749 switch (sk->sk_family) {
4750#if IS_ENABLED(CONFIG_IPV6)
4751 case AF_INET6:
4752 if (sk->sk_ipv6only ||
4753 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
4754 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
4755 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
4756 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
4757 break;
4758 }
4759 fallthrough;
4760#endif
4761 default: /* AF_INET */
4762 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
4763 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
4764 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
4765 break;
4766 }
4767
4768 flow->ports.src = inet_sk(sk)->inet_sport;
4769 flow->ports.dst = inet_sk(sk)->inet_dport;
4770}
4771
4772/**
4773 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
4774 * @sk: socket to use for headers
4775 *
4776 * This function will extract the necessary field from the socket and use
4777 * them to generate a hash based on the LAYER34 xmit_policy.
4778 * Assumes that sk is a TCP or UDP socket.
4779 */
4780static u32 bond_sk_hash_l34(struct sock *sk)
4781{
4782 struct flow_keys flow;
4783 u32 hash;
4784
4785 bond_sk_to_flow(sk, &flow);
4786
4787 /* L4 */
4788 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4789 /* L3 */
4790 return bond_ip_hash(hash, &flow);
4791}
4792
4793static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
4794 struct sock *sk)
4795{
4796 struct bond_up_slave *slaves;
4797 struct slave *slave;
4798 unsigned int count;
4799 u32 hash;
4800
4801 slaves = rcu_dereference(bond->usable_slaves);
4802 count = slaves ? READ_ONCE(slaves->count) : 0;
4803 if (unlikely(!count))
4804 return NULL;
4805
4806 hash = bond_sk_hash_l34(sk);
4807 slave = slaves->arr[hash % count];
4808
4809 return slave->dev;
4810}
4811
4812static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
4813 struct sock *sk)
4814{
4815 struct bonding *bond = netdev_priv(dev);
4816 struct net_device *lower = NULL;
4817
4818 rcu_read_lock();
4819 if (bond_sk_check(bond))
4820 lower = __bond_sk_get_lower_dev(bond, sk);
4821 rcu_read_unlock();
4822
4823 return lower;
4824}
4825
4826#if IS_ENABLED(CONFIG_TLS_DEVICE)
4827static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
4828 struct net_device *dev)
4829{
4830 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
4831 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
4832 return bond_tx_drop(dev, skb);
4833}
4834#endif
4835
4836static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4837{
4838 struct bonding *bond = netdev_priv(dev);
4839
4840 if (bond_should_override_tx_queue(bond) &&
4841 !bond_slave_override(bond, skb))
4842 return NETDEV_TX_OK;
4843
4844#if IS_ENABLED(CONFIG_TLS_DEVICE)
4845 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
4846 return bond_tls_device_xmit(bond, skb, dev);
4847#endif
4848
4849 switch (BOND_MODE(bond)) {
4850 case BOND_MODE_ROUNDROBIN:
4851 return bond_xmit_roundrobin(skb, dev);
4852 case BOND_MODE_ACTIVEBACKUP:
4853 return bond_xmit_activebackup(skb, dev);
4854 case BOND_MODE_8023AD:
4855 case BOND_MODE_XOR:
4856 return bond_3ad_xor_xmit(skb, dev);
4857 case BOND_MODE_BROADCAST:
4858 return bond_xmit_broadcast(skb, dev);
4859 case BOND_MODE_ALB:
4860 return bond_alb_xmit(skb, dev);
4861 case BOND_MODE_TLB:
4862 return bond_tlb_xmit(skb, dev);
4863 default:
4864 /* Should never happen, mode already checked */
4865 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4866 WARN_ON_ONCE(1);
4867 return bond_tx_drop(dev, skb);
4868 }
4869}
4870
4871static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4872{
4873 struct bonding *bond = netdev_priv(dev);
4874 netdev_tx_t ret = NETDEV_TX_OK;
4875
4876 /* If we risk deadlock from transmitting this in the
4877 * netpoll path, tell netpoll to queue the frame for later tx
4878 */
4879 if (unlikely(is_netpoll_tx_blocked(dev)))
4880 return NETDEV_TX_BUSY;
4881
4882 rcu_read_lock();
4883 if (bond_has_slaves(bond))
4884 ret = __bond_start_xmit(skb, dev);
4885 else
4886 ret = bond_tx_drop(dev, skb);
4887 rcu_read_unlock();
4888
4889 return ret;
4890}
4891
4892static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4893{
4894 if (speed == 0 || speed == SPEED_UNKNOWN)
4895 speed = slave->speed;
4896 else
4897 speed = min(speed, slave->speed);
4898
4899 return speed;
4900}
4901
4902static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4903 struct ethtool_link_ksettings *cmd)
4904{
4905 struct bonding *bond = netdev_priv(bond_dev);
4906 struct list_head *iter;
4907 struct slave *slave;
4908 u32 speed = 0;
4909
4910 cmd->base.duplex = DUPLEX_UNKNOWN;
4911 cmd->base.port = PORT_OTHER;
4912
4913 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4914 * do not need to check mode. Though link speed might not represent
4915 * the true receive or transmit bandwidth (not all modes are symmetric)
4916 * this is an accurate maximum.
4917 */
4918 bond_for_each_slave(bond, slave, iter) {
4919 if (bond_slave_can_tx(slave)) {
4920 if (slave->speed != SPEED_UNKNOWN) {
4921 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4922 speed = bond_mode_bcast_speed(slave,
4923 speed);
4924 else
4925 speed += slave->speed;
4926 }
4927 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4928 slave->duplex != DUPLEX_UNKNOWN)
4929 cmd->base.duplex = slave->duplex;
4930 }
4931 }
4932 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4933
4934 return 0;
4935}
4936
4937static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4938 struct ethtool_drvinfo *drvinfo)
4939{
4940 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4941 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4942 BOND_ABI_VERSION);
4943}
4944
4945static const struct ethtool_ops bond_ethtool_ops = {
4946 .get_drvinfo = bond_ethtool_get_drvinfo,
4947 .get_link = ethtool_op_get_link,
4948 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4949};
4950
4951static const struct net_device_ops bond_netdev_ops = {
4952 .ndo_init = bond_init,
4953 .ndo_uninit = bond_uninit,
4954 .ndo_open = bond_open,
4955 .ndo_stop = bond_close,
4956 .ndo_start_xmit = bond_start_xmit,
4957 .ndo_select_queue = bond_select_queue,
4958 .ndo_get_stats64 = bond_get_stats,
4959 .ndo_do_ioctl = bond_do_ioctl,
4960 .ndo_change_rx_flags = bond_change_rx_flags,
4961 .ndo_set_rx_mode = bond_set_rx_mode,
4962 .ndo_change_mtu = bond_change_mtu,
4963 .ndo_set_mac_address = bond_set_mac_address,
4964 .ndo_neigh_setup = bond_neigh_setup,
4965 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4966 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4967#ifdef CONFIG_NET_POLL_CONTROLLER
4968 .ndo_netpoll_setup = bond_netpoll_setup,
4969 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4970 .ndo_poll_controller = bond_poll_controller,
4971#endif
4972 .ndo_add_slave = bond_enslave,
4973 .ndo_del_slave = bond_release,
4974 .ndo_fix_features = bond_fix_features,
4975 .ndo_features_check = passthru_features_check,
4976 .ndo_get_xmit_slave = bond_xmit_get_slave,
4977 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
4978};
4979
4980static const struct device_type bond_type = {
4981 .name = "bond",
4982};
4983
4984static void bond_destructor(struct net_device *bond_dev)
4985{
4986 struct bonding *bond = netdev_priv(bond_dev);
4987
4988 if (bond->wq)
4989 destroy_workqueue(bond->wq);
4990
4991 if (bond->rr_tx_counter)
4992 free_percpu(bond->rr_tx_counter);
4993}
4994
4995void bond_setup(struct net_device *bond_dev)
4996{
4997 struct bonding *bond = netdev_priv(bond_dev);
4998
4999 spin_lock_init(&bond->mode_lock);
5000 bond->params = bonding_defaults;
5001
5002 /* Initialize pointers */
5003 bond->dev = bond_dev;
5004
5005 /* Initialize the device entry points */
5006 ether_setup(bond_dev);
5007 bond_dev->max_mtu = ETH_MAX_MTU;
5008 bond_dev->netdev_ops = &bond_netdev_ops;
5009 bond_dev->ethtool_ops = &bond_ethtool_ops;
5010
5011 bond_dev->needs_free_netdev = true;
5012 bond_dev->priv_destructor = bond_destructor;
5013
5014 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5015
5016 /* Initialize the device options */
5017 bond_dev->flags |= IFF_MASTER;
5018 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5019 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5020
5021#ifdef CONFIG_XFRM_OFFLOAD
5022 /* set up xfrm device ops (only supported in active-backup right now) */
5023 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5024 INIT_LIST_HEAD(&bond->ipsec_list);
5025 spin_lock_init(&bond->ipsec_lock);
5026#endif /* CONFIG_XFRM_OFFLOAD */
5027
5028 /* don't acquire bond device's netif_tx_lock when transmitting */
5029 bond_dev->features |= NETIF_F_LLTX;
5030
5031 /* By default, we declare the bond to be fully
5032 * VLAN hardware accelerated capable. Special
5033 * care is taken in the various xmit functions
5034 * when there are slaves that are not hw accel
5035 * capable
5036 */
5037
5038 /* Don't allow bond devices to change network namespaces. */
5039 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5040
5041 bond_dev->hw_features = BOND_VLAN_FEATURES |
5042 NETIF_F_HW_VLAN_CTAG_RX |
5043 NETIF_F_HW_VLAN_CTAG_FILTER;
5044
5045 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5046 bond_dev->features |= bond_dev->hw_features;
5047 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5048#ifdef CONFIG_XFRM_OFFLOAD
5049 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5050 /* Only enable XFRM features if this is an active-backup config */
5051 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5052 bond_dev->features |= BOND_XFRM_FEATURES;
5053#endif /* CONFIG_XFRM_OFFLOAD */
5054#if IS_ENABLED(CONFIG_TLS_DEVICE)
5055 if (bond_sk_check(bond))
5056 bond_dev->features |= BOND_TLS_FEATURES;
5057#endif
5058}
5059
5060/* Destroy a bonding device.
5061 * Must be under rtnl_lock when this function is called.
5062 */
5063static void bond_uninit(struct net_device *bond_dev)
5064{
5065 struct bonding *bond = netdev_priv(bond_dev);
5066 struct bond_up_slave *usable, *all;
5067 struct list_head *iter;
5068 struct slave *slave;
5069
5070 bond_netpoll_cleanup(bond_dev);
5071
5072 /* Release the bonded slaves */
5073 bond_for_each_slave(bond, slave, iter)
5074 __bond_release_one(bond_dev, slave->dev, true, true);
5075 netdev_info(bond_dev, "Released all slaves\n");
5076
5077 usable = rtnl_dereference(bond->usable_slaves);
5078 if (usable) {
5079 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5080 kfree_rcu(usable, rcu);
5081 }
5082
5083 all = rtnl_dereference(bond->all_slaves);
5084 if (all) {
5085 RCU_INIT_POINTER(bond->all_slaves, NULL);
5086 kfree_rcu(all, rcu);
5087 }
5088
5089 list_del(&bond->bond_list);
5090
5091 bond_debug_unregister(bond);
5092}
5093
5094/*------------------------- Module initialization ---------------------------*/
5095
5096static int bond_check_params(struct bond_params *params)
5097{
5098 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5099 struct bond_opt_value newval;
5100 const struct bond_opt_value *valptr;
5101 int arp_all_targets_value = 0;
5102 u16 ad_actor_sys_prio = 0;
5103 u16 ad_user_port_key = 0;
5104 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5105 int arp_ip_count;
5106 int bond_mode = BOND_MODE_ROUNDROBIN;
5107 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5108 int lacp_fast = 0;
5109 int tlb_dynamic_lb;
5110
5111 /* Convert string parameters. */
5112 if (mode) {
5113 bond_opt_initstr(&newval, mode);
5114 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5115 if (!valptr) {
5116 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5117 return -EINVAL;
5118 }
5119 bond_mode = valptr->value;
5120 }
5121
5122 if (xmit_hash_policy) {
5123 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5124 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5125 bond_mode == BOND_MODE_BROADCAST) {
5126 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5127 bond_mode_name(bond_mode));
5128 } else {
5129 bond_opt_initstr(&newval, xmit_hash_policy);
5130 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5131 &newval);
5132 if (!valptr) {
5133 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5134 xmit_hash_policy);
5135 return -EINVAL;
5136 }
5137 xmit_hashtype = valptr->value;
5138 }
5139 }
5140
5141 if (lacp_rate) {
5142 if (bond_mode != BOND_MODE_8023AD) {
5143 pr_info("lacp_rate param is irrelevant in mode %s\n",
5144 bond_mode_name(bond_mode));
5145 } else {
5146 bond_opt_initstr(&newval, lacp_rate);
5147 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5148 &newval);
5149 if (!valptr) {
5150 pr_err("Error: Invalid lacp rate \"%s\"\n",
5151 lacp_rate);
5152 return -EINVAL;
5153 }
5154 lacp_fast = valptr->value;
5155 }
5156 }
5157
5158 if (ad_select) {
5159 bond_opt_initstr(&newval, ad_select);
5160 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5161 &newval);
5162 if (!valptr) {
5163 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5164 return -EINVAL;
5165 }
5166 params->ad_select = valptr->value;
5167 if (bond_mode != BOND_MODE_8023AD)
5168 pr_warn("ad_select param only affects 802.3ad mode\n");
5169 } else {
5170 params->ad_select = BOND_AD_STABLE;
5171 }
5172
5173 if (max_bonds < 0) {
5174 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5175 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5176 max_bonds = BOND_DEFAULT_MAX_BONDS;
5177 }
5178
5179 if (miimon < 0) {
5180 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5181 miimon, INT_MAX);
5182 miimon = 0;
5183 }
5184
5185 if (updelay < 0) {
5186 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5187 updelay, INT_MAX);
5188 updelay = 0;
5189 }
5190
5191 if (downdelay < 0) {
5192 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5193 downdelay, INT_MAX);
5194 downdelay = 0;
5195 }
5196
5197 if ((use_carrier != 0) && (use_carrier != 1)) {
5198 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5199 use_carrier);
5200 use_carrier = 1;
5201 }
5202
5203 if (num_peer_notif < 0 || num_peer_notif > 255) {
5204 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5205 num_peer_notif);
5206 num_peer_notif = 1;
5207 }
5208
5209 /* reset values for 802.3ad/TLB/ALB */
5210 if (!bond_mode_uses_arp(bond_mode)) {
5211 if (!miimon) {
5212 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");
5213 pr_warn("Forcing miimon to 100msec\n");
5214 miimon = BOND_DEFAULT_MIIMON;
5215 }
5216 }
5217
5218 if (tx_queues < 1 || tx_queues > 255) {
5219 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5220 tx_queues, BOND_DEFAULT_TX_QUEUES);
5221 tx_queues = BOND_DEFAULT_TX_QUEUES;
5222 }
5223
5224 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5225 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5226 all_slaves_active);
5227 all_slaves_active = 0;
5228 }
5229
5230 if (resend_igmp < 0 || resend_igmp > 255) {
5231 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5232 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5233 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5234 }
5235
5236 bond_opt_initval(&newval, packets_per_slave);
5237 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5238 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5239 packets_per_slave, USHRT_MAX);
5240 packets_per_slave = 1;
5241 }
5242
5243 if (bond_mode == BOND_MODE_ALB) {
5244 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",
5245 updelay);
5246 }
5247
5248 if (!miimon) {
5249 if (updelay || downdelay) {
5250 /* just warn the user the up/down delay will have
5251 * no effect since miimon is zero...
5252 */
5253 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",
5254 updelay, downdelay);
5255 }
5256 } else {
5257 /* don't allow arp monitoring */
5258 if (arp_interval) {
5259 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5260 miimon, arp_interval);
5261 arp_interval = 0;
5262 }
5263
5264 if ((updelay % miimon) != 0) {
5265 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5266 updelay, miimon, (updelay / miimon) * miimon);
5267 }
5268
5269 updelay /= miimon;
5270
5271 if ((downdelay % miimon) != 0) {
5272 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5273 downdelay, miimon,
5274 (downdelay / miimon) * miimon);
5275 }
5276
5277 downdelay /= miimon;
5278 }
5279
5280 if (arp_interval < 0) {
5281 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5282 arp_interval, INT_MAX);
5283 arp_interval = 0;
5284 }
5285
5286 for (arp_ip_count = 0, i = 0;
5287 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5288 __be32 ip;
5289
5290 /* not a complete check, but good enough to catch mistakes */
5291 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5292 !bond_is_ip_target_ok(ip)) {
5293 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5294 arp_ip_target[i]);
5295 arp_interval = 0;
5296 } else {
5297 if (bond_get_targets_ip(arp_target, ip) == -1)
5298 arp_target[arp_ip_count++] = ip;
5299 else
5300 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5301 &ip);
5302 }
5303 }
5304
5305 if (arp_interval && !arp_ip_count) {
5306 /* don't allow arping if no arp_ip_target given... */
5307 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5308 arp_interval);
5309 arp_interval = 0;
5310 }
5311
5312 if (arp_validate) {
5313 if (!arp_interval) {
5314 pr_err("arp_validate requires arp_interval\n");
5315 return -EINVAL;
5316 }
5317
5318 bond_opt_initstr(&newval, arp_validate);
5319 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5320 &newval);
5321 if (!valptr) {
5322 pr_err("Error: invalid arp_validate \"%s\"\n",
5323 arp_validate);
5324 return -EINVAL;
5325 }
5326 arp_validate_value = valptr->value;
5327 } else {
5328 arp_validate_value = 0;
5329 }
5330
5331 if (arp_all_targets) {
5332 bond_opt_initstr(&newval, arp_all_targets);
5333 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5334 &newval);
5335 if (!valptr) {
5336 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5337 arp_all_targets);
5338 arp_all_targets_value = 0;
5339 } else {
5340 arp_all_targets_value = valptr->value;
5341 }
5342 }
5343
5344 if (miimon) {
5345 pr_info("MII link monitoring set to %d ms\n", miimon);
5346 } else if (arp_interval) {
5347 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5348 arp_validate_value);
5349 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5350 arp_interval, valptr->string, arp_ip_count);
5351
5352 for (i = 0; i < arp_ip_count; i++)
5353 pr_cont(" %s", arp_ip_target[i]);
5354
5355 pr_cont("\n");
5356
5357 } else if (max_bonds) {
5358 /* miimon and arp_interval not set, we need one so things
5359 * work as expected, see bonding.txt for details
5360 */
5361 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");
5362 }
5363
5364 if (primary && !bond_mode_uses_primary(bond_mode)) {
5365 /* currently, using a primary only makes sense
5366 * in active backup, TLB or ALB modes
5367 */
5368 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5369 primary, bond_mode_name(bond_mode));
5370 primary = NULL;
5371 }
5372
5373 if (primary && primary_reselect) {
5374 bond_opt_initstr(&newval, primary_reselect);
5375 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5376 &newval);
5377 if (!valptr) {
5378 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5379 primary_reselect);
5380 return -EINVAL;
5381 }
5382 primary_reselect_value = valptr->value;
5383 } else {
5384 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5385 }
5386
5387 if (fail_over_mac) {
5388 bond_opt_initstr(&newval, fail_over_mac);
5389 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5390 &newval);
5391 if (!valptr) {
5392 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5393 fail_over_mac);
5394 return -EINVAL;
5395 }
5396 fail_over_mac_value = valptr->value;
5397 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5398 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5399 } else {
5400 fail_over_mac_value = BOND_FOM_NONE;
5401 }
5402
5403 bond_opt_initstr(&newval, "default");
5404 valptr = bond_opt_parse(
5405 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5406 &newval);
5407 if (!valptr) {
5408 pr_err("Error: No ad_actor_sys_prio default value");
5409 return -EINVAL;
5410 }
5411 ad_actor_sys_prio = valptr->value;
5412
5413 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5414 &newval);
5415 if (!valptr) {
5416 pr_err("Error: No ad_user_port_key default value");
5417 return -EINVAL;
5418 }
5419 ad_user_port_key = valptr->value;
5420
5421 bond_opt_initstr(&newval, "default");
5422 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5423 if (!valptr) {
5424 pr_err("Error: No tlb_dynamic_lb default value");
5425 return -EINVAL;
5426 }
5427 tlb_dynamic_lb = valptr->value;
5428
5429 if (lp_interval == 0) {
5430 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5431 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5432 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5433 }
5434
5435 /* fill params struct with the proper values */
5436 params->mode = bond_mode;
5437 params->xmit_policy = xmit_hashtype;
5438 params->miimon = miimon;
5439 params->num_peer_notif = num_peer_notif;
5440 params->arp_interval = arp_interval;
5441 params->arp_validate = arp_validate_value;
5442 params->arp_all_targets = arp_all_targets_value;
5443 params->updelay = updelay;
5444 params->downdelay = downdelay;
5445 params->peer_notif_delay = 0;
5446 params->use_carrier = use_carrier;
5447 params->lacp_fast = lacp_fast;
5448 params->primary[0] = 0;
5449 params->primary_reselect = primary_reselect_value;
5450 params->fail_over_mac = fail_over_mac_value;
5451 params->tx_queues = tx_queues;
5452 params->all_slaves_active = all_slaves_active;
5453 params->resend_igmp = resend_igmp;
5454 params->min_links = min_links;
5455 params->lp_interval = lp_interval;
5456 params->packets_per_slave = packets_per_slave;
5457 params->tlb_dynamic_lb = tlb_dynamic_lb;
5458 params->ad_actor_sys_prio = ad_actor_sys_prio;
5459 eth_zero_addr(params->ad_actor_system);
5460 params->ad_user_port_key = ad_user_port_key;
5461 if (packets_per_slave > 0) {
5462 params->reciprocal_packets_per_slave =
5463 reciprocal_value(packets_per_slave);
5464 } else {
5465 /* reciprocal_packets_per_slave is unused if
5466 * packets_per_slave is 0 or 1, just initialize it
5467 */
5468 params->reciprocal_packets_per_slave =
5469 (struct reciprocal_value) { 0 };
5470 }
5471
5472 if (primary)
5473 strscpy_pad(params->primary, primary, sizeof(params->primary));
5474
5475 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5476
5477 return 0;
5478}
5479
5480/* Called from registration process */
5481static int bond_init(struct net_device *bond_dev)
5482{
5483 struct bonding *bond = netdev_priv(bond_dev);
5484 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5485
5486 netdev_dbg(bond_dev, "Begin bond_init\n");
5487
5488 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5489 if (!bond->wq)
5490 return -ENOMEM;
5491
5492 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
5493 bond->rr_tx_counter = alloc_percpu(u32);
5494 if (!bond->rr_tx_counter) {
5495 destroy_workqueue(bond->wq);
5496 bond->wq = NULL;
5497 return -ENOMEM;
5498 }
5499 }
5500
5501 spin_lock_init(&bond->stats_lock);
5502 netdev_lockdep_set_classes(bond_dev);
5503
5504 list_add_tail(&bond->bond_list, &bn->dev_list);
5505
5506 bond_prepare_sysfs_group(bond);
5507
5508 bond_debug_register(bond);
5509
5510 /* Ensure valid dev_addr */
5511 if (is_zero_ether_addr(bond_dev->dev_addr) &&
5512 bond_dev->addr_assign_type == NET_ADDR_PERM)
5513 eth_hw_addr_random(bond_dev);
5514
5515 return 0;
5516}
5517
5518unsigned int bond_get_num_tx_queues(void)
5519{
5520 return tx_queues;
5521}
5522
5523/* Create a new bond based on the specified name and bonding parameters.
5524 * If name is NULL, obtain a suitable "bond%d" name for us.
5525 * Caller must NOT hold rtnl_lock; we need to release it here before we
5526 * set up our sysfs entries.
5527 */
5528int bond_create(struct net *net, const char *name)
5529{
5530 struct net_device *bond_dev;
5531 struct bonding *bond;
5532 struct alb_bond_info *bond_info;
5533 int res;
5534
5535 rtnl_lock();
5536
5537 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5538 name ? name : "bond%d", NET_NAME_UNKNOWN,
5539 bond_setup, tx_queues);
5540 if (!bond_dev) {
5541 pr_err("%s: eek! can't alloc netdev!\n", name);
5542 rtnl_unlock();
5543 return -ENOMEM;
5544 }
5545
5546 /*
5547 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5548 * It is set to 0 by default which is wrong.
5549 */
5550 bond = netdev_priv(bond_dev);
5551 bond_info = &(BOND_ALB_INFO(bond));
5552 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5553
5554 dev_net_set(bond_dev, net);
5555 bond_dev->rtnl_link_ops = &bond_link_ops;
5556
5557 res = register_netdevice(bond_dev);
5558 if (res < 0) {
5559 free_netdev(bond_dev);
5560 rtnl_unlock();
5561
5562 return res;
5563 }
5564
5565 netif_carrier_off(bond_dev);
5566
5567 bond_work_init_all(bond);
5568
5569 rtnl_unlock();
5570 return 0;
5571}
5572
5573static int __net_init bond_net_init(struct net *net)
5574{
5575 struct bond_net *bn = net_generic(net, bond_net_id);
5576
5577 bn->net = net;
5578 INIT_LIST_HEAD(&bn->dev_list);
5579
5580 bond_create_proc_dir(bn);
5581 bond_create_sysfs(bn);
5582
5583 return 0;
5584}
5585
5586static void __net_exit bond_net_exit(struct net *net)
5587{
5588 struct bond_net *bn = net_generic(net, bond_net_id);
5589 struct bonding *bond, *tmp_bond;
5590 LIST_HEAD(list);
5591
5592 bond_destroy_sysfs(bn);
5593
5594 /* Kill off any bonds created after unregistering bond rtnl ops */
5595 rtnl_lock();
5596 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5597 unregister_netdevice_queue(bond->dev, &list);
5598 unregister_netdevice_many(&list);
5599 rtnl_unlock();
5600
5601 bond_destroy_proc_dir(bn);
5602}
5603
5604static struct pernet_operations bond_net_ops = {
5605 .init = bond_net_init,
5606 .exit = bond_net_exit,
5607 .id = &bond_net_id,
5608 .size = sizeof(struct bond_net),
5609};
5610
5611static int __init bonding_init(void)
5612{
5613 int i;
5614 int res;
5615
5616 res = bond_check_params(&bonding_defaults);
5617 if (res)
5618 goto out;
5619
5620 res = register_pernet_subsys(&bond_net_ops);
5621 if (res)
5622 goto out;
5623
5624 res = bond_netlink_init();
5625 if (res)
5626 goto err_link;
5627
5628 bond_create_debugfs();
5629
5630 for (i = 0; i < max_bonds; i++) {
5631 res = bond_create(&init_net, NULL);
5632 if (res)
5633 goto err;
5634 }
5635
5636 skb_flow_dissector_init(&flow_keys_bonding,
5637 flow_keys_bonding_keys,
5638 ARRAY_SIZE(flow_keys_bonding_keys));
5639
5640 register_netdevice_notifier(&bond_netdev_notifier);
5641out:
5642 return res;
5643err:
5644 bond_destroy_debugfs();
5645 bond_netlink_fini();
5646err_link:
5647 unregister_pernet_subsys(&bond_net_ops);
5648 goto out;
5649
5650}
5651
5652static void __exit bonding_exit(void)
5653{
5654 unregister_netdevice_notifier(&bond_netdev_notifier);
5655
5656 bond_destroy_debugfs();
5657
5658 bond_netlink_fini();
5659 unregister_pernet_subsys(&bond_net_ops);
5660
5661#ifdef CONFIG_NET_POLL_CONTROLLER
5662 /* Make sure we don't have an imbalance on our netpoll blocking */
5663 WARN_ON(atomic_read(&netpoll_block_tx));
5664#endif
5665}
5666
5667module_init(bonding_init);
5668module_exit(bonding_exit);
5669MODULE_LICENSE("GPL");
5670MODULE_DESCRIPTION(DRV_DESCRIPTION);
5671MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");