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