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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36#include <linux/kernel.h>
37#include <linux/module.h>
38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
44#include <net/ip.h>
45#include <linux/ip.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <linux/io.h>
57#include <asm/dma.h>
58#include <linux/uaccess.h>
59#include <linux/errno.h>
60#include <linux/netdevice.h>
61#include <linux/inetdevice.h>
62#include <linux/igmp.h>
63#include <linux/etherdevice.h>
64#include <linux/skbuff.h>
65#include <net/sock.h>
66#include <linux/rtnetlink.h>
67#include <linux/smp.h>
68#include <linux/if_ether.h>
69#include <net/arp.h>
70#include <linux/mii.h>
71#include <linux/ethtool.h>
72#include <linux/if_vlan.h>
73#include <linux/if_bonding.h>
74#include <linux/jiffies.h>
75#include <linux/preempt.h>
76#include <net/route.h>
77#include <net/net_namespace.h>
78#include <net/netns/generic.h>
79#include <net/pkt_sched.h>
80#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
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 = BOND_LINK_MON_INTERV;
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 = BOND_LINK_ARP_INTERV;
105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *fail_over_mac;
108static int all_slaves_active = 0;
109static struct bond_params bonding_defaults;
110static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111
112module_param(max_bonds, int, 0);
113MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114module_param(tx_queues, int, 0);
115MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122module_param(miimon, int, 0);
123MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124module_param(updelay, int, 0);
125MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126module_param(downdelay, int, 0);
127MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 "in milliseconds");
129module_param(use_carrier, int, 0);
130MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132module_param(mode, charp, 0);
133MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137module_param(primary, charp, 0);
138MODULE_PARM_DESC(primary, "Primary network device to use");
139module_param(primary_reselect, charp, 0);
140MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "once it comes up; "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
144 "better, "
145 "2 for only on active slave "
146 "failure");
147module_param(lacp_rate, charp, 0);
148MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150module_param(ad_select, charp, 0);
151MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
153 "2 for count");
154module_param(min_links, int, 0);
155MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156
157module_param(xmit_hash_policy, charp, 0);
158MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
160 "2 for layer 2+3");
161module_param(arp_interval, int, 0);
162MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163module_param_array(arp_ip_target, charp, NULL, 0);
164MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165module_param(arp_validate, charp, 0);
166MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169module_param(fail_over_mac, charp, 0);
170MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173module_param(all_slaves_active, int, 0);
174MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177module_param(resend_igmp, int, 0);
178MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179 "link failure");
180
181/*----------------------------- Global variables ----------------------------*/
182
183#ifdef CONFIG_NET_POLL_CONTROLLER
184atomic_t netpoll_block_tx = ATOMIC_INIT(0);
185#endif
186
187int bond_net_id __read_mostly;
188
189static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190static int arp_ip_count;
191static int bond_mode = BOND_MODE_ROUNDROBIN;
192static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193static int lacp_fast;
194
195const struct bond_parm_tbl bond_lacp_tbl[] = {
196{ "slow", AD_LACP_SLOW},
197{ "fast", AD_LACP_FAST},
198{ NULL, -1},
199};
200
201const struct bond_parm_tbl bond_mode_tbl[] = {
202{ "balance-rr", BOND_MODE_ROUNDROBIN},
203{ "active-backup", BOND_MODE_ACTIVEBACKUP},
204{ "balance-xor", BOND_MODE_XOR},
205{ "broadcast", BOND_MODE_BROADCAST},
206{ "802.3ad", BOND_MODE_8023AD},
207{ "balance-tlb", BOND_MODE_TLB},
208{ "balance-alb", BOND_MODE_ALB},
209{ NULL, -1},
210};
211
212const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213{ "layer2", BOND_XMIT_POLICY_LAYER2},
214{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
215{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
216{ NULL, -1},
217};
218
219const struct bond_parm_tbl arp_validate_tbl[] = {
220{ "none", BOND_ARP_VALIDATE_NONE},
221{ "active", BOND_ARP_VALIDATE_ACTIVE},
222{ "backup", BOND_ARP_VALIDATE_BACKUP},
223{ "all", BOND_ARP_VALIDATE_ALL},
224{ NULL, -1},
225};
226
227const struct bond_parm_tbl fail_over_mac_tbl[] = {
228{ "none", BOND_FOM_NONE},
229{ "active", BOND_FOM_ACTIVE},
230{ "follow", BOND_FOM_FOLLOW},
231{ NULL, -1},
232};
233
234const struct bond_parm_tbl pri_reselect_tbl[] = {
235{ "always", BOND_PRI_RESELECT_ALWAYS},
236{ "better", BOND_PRI_RESELECT_BETTER},
237{ "failure", BOND_PRI_RESELECT_FAILURE},
238{ NULL, -1},
239};
240
241struct bond_parm_tbl ad_select_tbl[] = {
242{ "stable", BOND_AD_STABLE},
243{ "bandwidth", BOND_AD_BANDWIDTH},
244{ "count", BOND_AD_COUNT},
245{ NULL, -1},
246};
247
248/*-------------------------- Forward declarations ---------------------------*/
249
250static int bond_init(struct net_device *bond_dev);
251static void bond_uninit(struct net_device *bond_dev);
252
253/*---------------------------- General routines -----------------------------*/
254
255const char *bond_mode_name(int mode)
256{
257 static const char *names[] = {
258 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB] = "transmit load balancing",
264 [BOND_MODE_ALB] = "adaptive load balancing",
265 };
266
267 if (mode < 0 || mode > BOND_MODE_ALB)
268 return "unknown";
269
270 return names[mode];
271}
272
273/*---------------------------------- VLAN -----------------------------------*/
274
275/**
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
279 *
280 * Returns -ENOMEM if allocation failed.
281 */
282static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283{
284 struct vlan_entry *vlan;
285
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond ? bond->dev->name : "None"), vlan_id);
288
289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
290 if (!vlan)
291 return -ENOMEM;
292
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
295
296 write_lock_bh(&bond->lock);
297
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299
300 write_unlock_bh(&bond->lock);
301
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
303
304 return 0;
305}
306
307/**
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
311 *
312 * returns -ENODEV if @vlan_id was not found in @bond.
313 */
314static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315{
316 struct vlan_entry *vlan;
317 int res = -ENODEV;
318
319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
320
321 block_netpoll_tx();
322 write_lock_bh(&bond->lock);
323
324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
327
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vlan_id);
330
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
333
334 kfree(vlan);
335
336 res = 0;
337 goto out;
338 }
339 }
340
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
343
344out:
345 write_unlock_bh(&bond->lock);
346 unblock_netpoll_tx();
347 return res;
348}
349
350/**
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
354 *
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
357 *
358 * Caller must hold bond->lock
359 */
360struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361{
362 struct vlan_entry *next, *last;
363
364 if (list_empty(&bond->vlan_list))
365 return NULL;
366
367 if (!curr) {
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
370 } else {
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
373 if (last == curr) {
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
376 } else {
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
379 }
380 }
381
382 return next;
383}
384
385/**
386 * bond_dev_queue_xmit - Prepare skb for xmit.
387 *
388 * @bond: bond device that got this skb for tx.
389 * @skb: hw accel VLAN tagged skb to transmit
390 * @slave_dev: slave that is supposed to xmit this skbuff
391 */
392int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
393 struct net_device *slave_dev)
394{
395 skb->dev = slave_dev;
396
397 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
398 sizeof(qdisc_skb_cb(skb)->bond_queue_mapping));
399 skb->queue_mapping = qdisc_skb_cb(skb)->bond_queue_mapping;
400
401 if (unlikely(netpoll_tx_running(slave_dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
403 else
404 dev_queue_xmit(skb);
405
406 return 0;
407}
408
409/*
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
416 *
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
424*/
425
426/**
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
430 */
431static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
432{
433 struct bonding *bond = netdev_priv(bond_dev);
434 struct slave *slave, *stop_at;
435 int i, res;
436
437 bond_for_each_slave(bond, slave, i) {
438 res = vlan_vid_add(slave->dev, vid);
439 if (res)
440 goto unwind;
441 }
442
443 res = bond_add_vlan(bond, vid);
444 if (res) {
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev->name, vid);
447 return res;
448 }
449
450 return 0;
451
452unwind:
453 /* unwind from head to the slave that failed */
454 stop_at = slave;
455 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
456 vlan_vid_del(slave->dev, vid);
457
458 return res;
459}
460
461/**
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
465 */
466static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
467{
468 struct bonding *bond = netdev_priv(bond_dev);
469 struct slave *slave;
470 int i, res;
471
472 bond_for_each_slave(bond, slave, i)
473 vlan_vid_del(slave->dev, vid);
474
475 res = bond_del_vlan(bond, vid);
476 if (res) {
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev->name, vid);
479 return res;
480 }
481
482 return 0;
483}
484
485static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
486{
487 struct vlan_entry *vlan;
488 int res;
489
490 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
491 res = vlan_vid_add(slave_dev, vlan->vlan_id);
492 if (res)
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond->dev->name, vlan->vlan_id,
495 slave_dev->name);
496 }
497}
498
499static void bond_del_vlans_from_slave(struct bonding *bond,
500 struct net_device *slave_dev)
501{
502 struct vlan_entry *vlan;
503
504 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
505 if (!vlan->vlan_id)
506 continue;
507 vlan_vid_del(slave_dev, vlan->vlan_id);
508 }
509}
510
511/*------------------------------- Link status -------------------------------*/
512
513/*
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
517 *
518 * Returns zero if carrier state does not change, nonzero if it does.
519 */
520static int bond_set_carrier(struct bonding *bond)
521{
522 struct slave *slave;
523 int i;
524
525 if (bond->slave_cnt == 0)
526 goto down;
527
528 if (bond->params.mode == BOND_MODE_8023AD)
529 return bond_3ad_set_carrier(bond);
530
531 bond_for_each_slave(bond, slave, i) {
532 if (slave->link == BOND_LINK_UP) {
533 if (!netif_carrier_ok(bond->dev)) {
534 netif_carrier_on(bond->dev);
535 return 1;
536 }
537 return 0;
538 }
539 }
540
541down:
542 if (netif_carrier_ok(bond->dev)) {
543 netif_carrier_off(bond->dev);
544 return 1;
545 }
546 return 0;
547}
548
549/*
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
553 * and return.
554 */
555static void bond_update_speed_duplex(struct slave *slave)
556{
557 struct net_device *slave_dev = slave->dev;
558 struct ethtool_cmd ecmd;
559 u32 slave_speed;
560 int res;
561
562 slave->speed = SPEED_UNKNOWN;
563 slave->duplex = DUPLEX_UNKNOWN;
564
565 res = __ethtool_get_settings(slave_dev, &ecmd);
566 if (res < 0)
567 return;
568
569 slave_speed = ethtool_cmd_speed(&ecmd);
570 if (slave_speed == 0 || slave_speed == ((__u32) -1))
571 return;
572
573 switch (ecmd.duplex) {
574 case DUPLEX_FULL:
575 case DUPLEX_HALF:
576 break;
577 default:
578 return;
579 }
580
581 slave->speed = slave_speed;
582 slave->duplex = ecmd.duplex;
583
584 return;
585}
586
587/*
588 * if <dev> supports MII link status reporting, check its link status.
589 *
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
592 *
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
595 * down.
596 *
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
602 */
603static int bond_check_dev_link(struct bonding *bond,
604 struct net_device *slave_dev, int reporting)
605{
606 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
607 int (*ioctl)(struct net_device *, struct ifreq *, int);
608 struct ifreq ifr;
609 struct mii_ioctl_data *mii;
610
611 if (!reporting && !netif_running(slave_dev))
612 return 0;
613
614 if (bond->params.use_carrier)
615 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
616
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev->ethtool_ops) {
619 if (slave_dev->ethtool_ops->get_link) {
620 u32 link;
621
622 link = slave_dev->ethtool_ops->get_link(slave_dev);
623
624 return link ? BMSR_LSTATUS : 0;
625 }
626 }
627
628 /* Ethtool can't be used, fallback to MII ioctls. */
629 ioctl = slave_ops->ndo_do_ioctl;
630 if (ioctl) {
631 /* TODO: set pointer to correct ioctl on a per team member */
632 /* bases to make this more efficient. that is, once */
633 /* we determine the correct ioctl, we will always */
634 /* call it and not the others for that team */
635 /* member. */
636
637 /*
638 * We cannot assume that SIOCGMIIPHY will also read a
639 * register; not all network drivers (e.g., e100)
640 * support that.
641 */
642
643 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
644 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
645 mii = if_mii(&ifr);
646 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
647 mii->reg_num = MII_BMSR;
648 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
649 return mii->val_out & BMSR_LSTATUS;
650 }
651 }
652
653 /*
654 * If reporting, report that either there's no dev->do_ioctl,
655 * or both SIOCGMIIREG and get_link failed (meaning that we
656 * cannot report link status). If not reporting, pretend
657 * we're ok.
658 */
659 return reporting ? -1 : BMSR_LSTATUS;
660}
661
662/*----------------------------- Multicast list ------------------------------*/
663
664/*
665 * Push the promiscuity flag down to appropriate slaves
666 */
667static int bond_set_promiscuity(struct bonding *bond, int inc)
668{
669 int err = 0;
670 if (USES_PRIMARY(bond->params.mode)) {
671 /* write lock already acquired */
672 if (bond->curr_active_slave) {
673 err = dev_set_promiscuity(bond->curr_active_slave->dev,
674 inc);
675 }
676 } else {
677 struct slave *slave;
678 int i;
679 bond_for_each_slave(bond, slave, i) {
680 err = dev_set_promiscuity(slave->dev, inc);
681 if (err)
682 return err;
683 }
684 }
685 return err;
686}
687
688/*
689 * Push the allmulti flag down to all slaves
690 */
691static int bond_set_allmulti(struct bonding *bond, int inc)
692{
693 int err = 0;
694 if (USES_PRIMARY(bond->params.mode)) {
695 /* write lock already acquired */
696 if (bond->curr_active_slave) {
697 err = dev_set_allmulti(bond->curr_active_slave->dev,
698 inc);
699 }
700 } else {
701 struct slave *slave;
702 int i;
703 bond_for_each_slave(bond, slave, i) {
704 err = dev_set_allmulti(slave->dev, inc);
705 if (err)
706 return err;
707 }
708 }
709 return err;
710}
711
712/*
713 * Add a Multicast address to slaves
714 * according to mode
715 */
716static void bond_mc_add(struct bonding *bond, void *addr)
717{
718 if (USES_PRIMARY(bond->params.mode)) {
719 /* write lock already acquired */
720 if (bond->curr_active_slave)
721 dev_mc_add(bond->curr_active_slave->dev, addr);
722 } else {
723 struct slave *slave;
724 int i;
725
726 bond_for_each_slave(bond, slave, i)
727 dev_mc_add(slave->dev, addr);
728 }
729}
730
731/*
732 * Remove a multicast address from slave
733 * according to mode
734 */
735static void bond_mc_del(struct bonding *bond, void *addr)
736{
737 if (USES_PRIMARY(bond->params.mode)) {
738 /* write lock already acquired */
739 if (bond->curr_active_slave)
740 dev_mc_del(bond->curr_active_slave->dev, addr);
741 } else {
742 struct slave *slave;
743 int i;
744 bond_for_each_slave(bond, slave, i) {
745 dev_mc_del(slave->dev, addr);
746 }
747 }
748}
749
750
751static void __bond_resend_igmp_join_requests(struct net_device *dev)
752{
753 struct in_device *in_dev;
754
755 rcu_read_lock();
756 in_dev = __in_dev_get_rcu(dev);
757 if (in_dev)
758 ip_mc_rejoin_groups(in_dev);
759 rcu_read_unlock();
760}
761
762/*
763 * Retrieve the list of registered multicast addresses for the bonding
764 * device and retransmit an IGMP JOIN request to the current active
765 * slave.
766 */
767static void bond_resend_igmp_join_requests(struct bonding *bond)
768{
769 struct net_device *bond_dev, *vlan_dev, *master_dev;
770 struct vlan_entry *vlan;
771
772 read_lock(&bond->lock);
773
774 bond_dev = bond->dev;
775
776 /* rejoin all groups on bond device */
777 __bond_resend_igmp_join_requests(bond_dev);
778
779 /*
780 * if bond is enslaved to a bridge,
781 * then rejoin all groups on its master
782 */
783 master_dev = bond_dev->master;
784 if (master_dev)
785 if ((master_dev->priv_flags & IFF_EBRIDGE)
786 && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
787 __bond_resend_igmp_join_requests(master_dev);
788
789 /* rejoin all groups on vlan devices */
790 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
791 rcu_read_lock();
792 vlan_dev = __vlan_find_dev_deep(bond_dev,
793 vlan->vlan_id);
794 rcu_read_unlock();
795 if (vlan_dev)
796 __bond_resend_igmp_join_requests(vlan_dev);
797 }
798
799 if (--bond->igmp_retrans > 0)
800 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
801
802 read_unlock(&bond->lock);
803}
804
805static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
806{
807 struct bonding *bond = container_of(work, struct bonding,
808 mcast_work.work);
809 bond_resend_igmp_join_requests(bond);
810}
811
812/*
813 * flush all members of flush->mc_list from device dev->mc_list
814 */
815static void bond_mc_list_flush(struct net_device *bond_dev,
816 struct net_device *slave_dev)
817{
818 struct bonding *bond = netdev_priv(bond_dev);
819 struct netdev_hw_addr *ha;
820
821 netdev_for_each_mc_addr(ha, bond_dev)
822 dev_mc_del(slave_dev, ha->addr);
823
824 if (bond->params.mode == BOND_MODE_8023AD) {
825 /* del lacpdu mc addr from mc list */
826 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
827
828 dev_mc_del(slave_dev, lacpdu_multicast);
829 }
830}
831
832/*--------------------------- Active slave change ---------------------------*/
833
834/*
835 * Update the mc list and multicast-related flags for the new and
836 * old active slaves (if any) according to the multicast mode, and
837 * promiscuous flags unconditionally.
838 */
839static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
840 struct slave *old_active)
841{
842 struct netdev_hw_addr *ha;
843
844 if (!USES_PRIMARY(bond->params.mode))
845 /* nothing to do - mc list is already up-to-date on
846 * all slaves
847 */
848 return;
849
850 if (old_active) {
851 if (bond->dev->flags & IFF_PROMISC)
852 dev_set_promiscuity(old_active->dev, -1);
853
854 if (bond->dev->flags & IFF_ALLMULTI)
855 dev_set_allmulti(old_active->dev, -1);
856
857 netdev_for_each_mc_addr(ha, bond->dev)
858 dev_mc_del(old_active->dev, ha->addr);
859 }
860
861 if (new_active) {
862 /* FIXME: Signal errors upstream. */
863 if (bond->dev->flags & IFF_PROMISC)
864 dev_set_promiscuity(new_active->dev, 1);
865
866 if (bond->dev->flags & IFF_ALLMULTI)
867 dev_set_allmulti(new_active->dev, 1);
868
869 netdev_for_each_mc_addr(ha, bond->dev)
870 dev_mc_add(new_active->dev, ha->addr);
871 }
872}
873
874/*
875 * bond_do_fail_over_mac
876 *
877 * Perform special MAC address swapping for fail_over_mac settings
878 *
879 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
880 */
881static void bond_do_fail_over_mac(struct bonding *bond,
882 struct slave *new_active,
883 struct slave *old_active)
884 __releases(&bond->curr_slave_lock)
885 __releases(&bond->lock)
886 __acquires(&bond->lock)
887 __acquires(&bond->curr_slave_lock)
888{
889 u8 tmp_mac[ETH_ALEN];
890 struct sockaddr saddr;
891 int rv;
892
893 switch (bond->params.fail_over_mac) {
894 case BOND_FOM_ACTIVE:
895 if (new_active) {
896 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
897 new_active->dev->addr_len);
898 write_unlock_bh(&bond->curr_slave_lock);
899 read_unlock(&bond->lock);
900 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
901 read_lock(&bond->lock);
902 write_lock_bh(&bond->curr_slave_lock);
903 }
904 break;
905 case BOND_FOM_FOLLOW:
906 /*
907 * if new_active && old_active, swap them
908 * if just old_active, do nothing (going to no active slave)
909 * if just new_active, set new_active to bond's MAC
910 */
911 if (!new_active)
912 return;
913
914 write_unlock_bh(&bond->curr_slave_lock);
915 read_unlock(&bond->lock);
916
917 if (old_active) {
918 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
919 memcpy(saddr.sa_data, old_active->dev->dev_addr,
920 ETH_ALEN);
921 saddr.sa_family = new_active->dev->type;
922 } else {
923 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
924 saddr.sa_family = bond->dev->type;
925 }
926
927 rv = dev_set_mac_address(new_active->dev, &saddr);
928 if (rv) {
929 pr_err("%s: Error %d setting MAC of slave %s\n",
930 bond->dev->name, -rv, new_active->dev->name);
931 goto out;
932 }
933
934 if (!old_active)
935 goto out;
936
937 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
938 saddr.sa_family = old_active->dev->type;
939
940 rv = dev_set_mac_address(old_active->dev, &saddr);
941 if (rv)
942 pr_err("%s: Error %d setting MAC of slave %s\n",
943 bond->dev->name, -rv, new_active->dev->name);
944out:
945 read_lock(&bond->lock);
946 write_lock_bh(&bond->curr_slave_lock);
947 break;
948 default:
949 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
950 bond->dev->name, bond->params.fail_over_mac);
951 break;
952 }
953
954}
955
956static bool bond_should_change_active(struct bonding *bond)
957{
958 struct slave *prim = bond->primary_slave;
959 struct slave *curr = bond->curr_active_slave;
960
961 if (!prim || !curr || curr->link != BOND_LINK_UP)
962 return true;
963 if (bond->force_primary) {
964 bond->force_primary = false;
965 return true;
966 }
967 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
968 (prim->speed < curr->speed ||
969 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
970 return false;
971 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
972 return false;
973 return true;
974}
975
976/**
977 * find_best_interface - select the best available slave to be the active one
978 * @bond: our bonding struct
979 *
980 * Warning: Caller must hold curr_slave_lock for writing.
981 */
982static struct slave *bond_find_best_slave(struct bonding *bond)
983{
984 struct slave *new_active, *old_active;
985 struct slave *bestslave = NULL;
986 int mintime = bond->params.updelay;
987 int i;
988
989 new_active = bond->curr_active_slave;
990
991 if (!new_active) { /* there were no active slaves left */
992 if (bond->slave_cnt > 0) /* found one slave */
993 new_active = bond->first_slave;
994 else
995 return NULL; /* still no slave, return NULL */
996 }
997
998 if ((bond->primary_slave) &&
999 bond->primary_slave->link == BOND_LINK_UP &&
1000 bond_should_change_active(bond)) {
1001 new_active = bond->primary_slave;
1002 }
1003
1004 /* remember where to stop iterating over the slaves */
1005 old_active = new_active;
1006
1007 bond_for_each_slave_from(bond, new_active, i, old_active) {
1008 if (new_active->link == BOND_LINK_UP) {
1009 return new_active;
1010 } else if (new_active->link == BOND_LINK_BACK &&
1011 IS_UP(new_active->dev)) {
1012 /* link up, but waiting for stabilization */
1013 if (new_active->delay < mintime) {
1014 mintime = new_active->delay;
1015 bestslave = new_active;
1016 }
1017 }
1018 }
1019
1020 return bestslave;
1021}
1022
1023static bool bond_should_notify_peers(struct bonding *bond)
1024{
1025 struct slave *slave = bond->curr_active_slave;
1026
1027 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028 bond->dev->name, slave ? slave->dev->name : "NULL");
1029
1030 if (!slave || !bond->send_peer_notif ||
1031 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1032 return false;
1033
1034 bond->send_peer_notif--;
1035 return true;
1036}
1037
1038/**
1039 * change_active_interface - change the active slave into the specified one
1040 * @bond: our bonding struct
1041 * @new: the new slave to make the active one
1042 *
1043 * Set the new slave to the bond's settings and unset them on the old
1044 * curr_active_slave.
1045 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1046 *
1047 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048 * because it is apparently the best available slave we have, even though its
1049 * updelay hasn't timed out yet.
1050 *
1051 * If new_active is not NULL, caller must hold bond->lock for read and
1052 * curr_slave_lock for write_bh.
1053 */
1054void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1055{
1056 struct slave *old_active = bond->curr_active_slave;
1057
1058 if (old_active == new_active)
1059 return;
1060
1061 if (new_active) {
1062 new_active->jiffies = jiffies;
1063
1064 if (new_active->link == BOND_LINK_BACK) {
1065 if (USES_PRIMARY(bond->params.mode)) {
1066 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067 bond->dev->name, new_active->dev->name,
1068 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1069 }
1070
1071 new_active->delay = 0;
1072 new_active->link = BOND_LINK_UP;
1073
1074 if (bond->params.mode == BOND_MODE_8023AD)
1075 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1076
1077 if (bond_is_lb(bond))
1078 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1079 } else {
1080 if (USES_PRIMARY(bond->params.mode)) {
1081 pr_info("%s: making interface %s the new active one.\n",
1082 bond->dev->name, new_active->dev->name);
1083 }
1084 }
1085 }
1086
1087 if (USES_PRIMARY(bond->params.mode))
1088 bond_mc_swap(bond, new_active, old_active);
1089
1090 if (bond_is_lb(bond)) {
1091 bond_alb_handle_active_change(bond, new_active);
1092 if (old_active)
1093 bond_set_slave_inactive_flags(old_active);
1094 if (new_active)
1095 bond_set_slave_active_flags(new_active);
1096 } else {
1097 bond->curr_active_slave = new_active;
1098 }
1099
1100 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1101 if (old_active)
1102 bond_set_slave_inactive_flags(old_active);
1103
1104 if (new_active) {
1105 bool should_notify_peers = false;
1106
1107 bond_set_slave_active_flags(new_active);
1108
1109 if (bond->params.fail_over_mac)
1110 bond_do_fail_over_mac(bond, new_active,
1111 old_active);
1112
1113 if (netif_running(bond->dev)) {
1114 bond->send_peer_notif =
1115 bond->params.num_peer_notif;
1116 should_notify_peers =
1117 bond_should_notify_peers(bond);
1118 }
1119
1120 write_unlock_bh(&bond->curr_slave_lock);
1121 read_unlock(&bond->lock);
1122
1123 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1124 if (should_notify_peers)
1125 netdev_bonding_change(bond->dev,
1126 NETDEV_NOTIFY_PEERS);
1127
1128 read_lock(&bond->lock);
1129 write_lock_bh(&bond->curr_slave_lock);
1130 }
1131 }
1132
1133 /* resend IGMP joins since active slave has changed or
1134 * all were sent on curr_active_slave.
1135 * resend only if bond is brought up with the affected
1136 * bonding modes and the retransmission is enabled */
1137 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1138 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1139 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1140 bond->igmp_retrans = bond->params.resend_igmp;
1141 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1142 }
1143}
1144
1145/**
1146 * bond_select_active_slave - select a new active slave, if needed
1147 * @bond: our bonding struct
1148 *
1149 * This functions should be called when one of the following occurs:
1150 * - The old curr_active_slave has been released or lost its link.
1151 * - The primary_slave has got its link back.
1152 * - A slave has got its link back and there's no old curr_active_slave.
1153 *
1154 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1155 */
1156void bond_select_active_slave(struct bonding *bond)
1157{
1158 struct slave *best_slave;
1159 int rv;
1160
1161 best_slave = bond_find_best_slave(bond);
1162 if (best_slave != bond->curr_active_slave) {
1163 bond_change_active_slave(bond, best_slave);
1164 rv = bond_set_carrier(bond);
1165 if (!rv)
1166 return;
1167
1168 if (netif_carrier_ok(bond->dev)) {
1169 pr_info("%s: first active interface up!\n",
1170 bond->dev->name);
1171 } else {
1172 pr_info("%s: now running without any active interface !\n",
1173 bond->dev->name);
1174 }
1175 }
1176}
1177
1178/*--------------------------- slave list handling ---------------------------*/
1179
1180/*
1181 * This function attaches the slave to the end of list.
1182 *
1183 * bond->lock held for writing by caller.
1184 */
1185static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1186{
1187 if (bond->first_slave == NULL) { /* attaching the first slave */
1188 new_slave->next = new_slave;
1189 new_slave->prev = new_slave;
1190 bond->first_slave = new_slave;
1191 } else {
1192 new_slave->next = bond->first_slave;
1193 new_slave->prev = bond->first_slave->prev;
1194 new_slave->next->prev = new_slave;
1195 new_slave->prev->next = new_slave;
1196 }
1197
1198 bond->slave_cnt++;
1199}
1200
1201/*
1202 * This function detaches the slave from the list.
1203 * WARNING: no check is made to verify if the slave effectively
1204 * belongs to <bond>.
1205 * Nothing is freed on return, structures are just unchained.
1206 * If any slave pointer in bond was pointing to <slave>,
1207 * it should be changed by the calling function.
1208 *
1209 * bond->lock held for writing by caller.
1210 */
1211static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1212{
1213 if (slave->next)
1214 slave->next->prev = slave->prev;
1215
1216 if (slave->prev)
1217 slave->prev->next = slave->next;
1218
1219 if (bond->first_slave == slave) { /* slave is the first slave */
1220 if (bond->slave_cnt > 1) { /* there are more slave */
1221 bond->first_slave = slave->next;
1222 } else {
1223 bond->first_slave = NULL; /* slave was the last one */
1224 }
1225 }
1226
1227 slave->next = NULL;
1228 slave->prev = NULL;
1229 bond->slave_cnt--;
1230}
1231
1232#ifdef CONFIG_NET_POLL_CONTROLLER
1233static inline int slave_enable_netpoll(struct slave *slave)
1234{
1235 struct netpoll *np;
1236 int err = 0;
1237
1238 np = kzalloc(sizeof(*np), GFP_KERNEL);
1239 err = -ENOMEM;
1240 if (!np)
1241 goto out;
1242
1243 np->dev = slave->dev;
1244 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1245 err = __netpoll_setup(np);
1246 if (err) {
1247 kfree(np);
1248 goto out;
1249 }
1250 slave->np = np;
1251out:
1252 return err;
1253}
1254static inline void slave_disable_netpoll(struct slave *slave)
1255{
1256 struct netpoll *np = slave->np;
1257
1258 if (!np)
1259 return;
1260
1261 slave->np = NULL;
1262 synchronize_rcu_bh();
1263 __netpoll_cleanup(np);
1264 kfree(np);
1265}
1266static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1267{
1268 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1269 return false;
1270 if (!slave_dev->netdev_ops->ndo_poll_controller)
1271 return false;
1272 return true;
1273}
1274
1275static void bond_poll_controller(struct net_device *bond_dev)
1276{
1277}
1278
1279static void __bond_netpoll_cleanup(struct bonding *bond)
1280{
1281 struct slave *slave;
1282 int i;
1283
1284 bond_for_each_slave(bond, slave, i)
1285 if (IS_UP(slave->dev))
1286 slave_disable_netpoll(slave);
1287}
1288static void bond_netpoll_cleanup(struct net_device *bond_dev)
1289{
1290 struct bonding *bond = netdev_priv(bond_dev);
1291
1292 read_lock(&bond->lock);
1293 __bond_netpoll_cleanup(bond);
1294 read_unlock(&bond->lock);
1295}
1296
1297static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1298{
1299 struct bonding *bond = netdev_priv(dev);
1300 struct slave *slave;
1301 int i, err = 0;
1302
1303 read_lock(&bond->lock);
1304 bond_for_each_slave(bond, slave, i) {
1305 err = slave_enable_netpoll(slave);
1306 if (err) {
1307 __bond_netpoll_cleanup(bond);
1308 break;
1309 }
1310 }
1311 read_unlock(&bond->lock);
1312 return err;
1313}
1314
1315static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1316{
1317 return bond->dev->npinfo;
1318}
1319
1320#else
1321static inline int slave_enable_netpoll(struct slave *slave)
1322{
1323 return 0;
1324}
1325static inline void slave_disable_netpoll(struct slave *slave)
1326{
1327}
1328static void bond_netpoll_cleanup(struct net_device *bond_dev)
1329{
1330}
1331#endif
1332
1333/*---------------------------------- IOCTL ----------------------------------*/
1334
1335static int bond_sethwaddr(struct net_device *bond_dev,
1336 struct net_device *slave_dev)
1337{
1338 pr_debug("bond_dev=%p\n", bond_dev);
1339 pr_debug("slave_dev=%p\n", slave_dev);
1340 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1341 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1342 return 0;
1343}
1344
1345static netdev_features_t bond_fix_features(struct net_device *dev,
1346 netdev_features_t features)
1347{
1348 struct slave *slave;
1349 struct bonding *bond = netdev_priv(dev);
1350 netdev_features_t mask;
1351 int i;
1352
1353 read_lock(&bond->lock);
1354
1355 if (!bond->first_slave) {
1356 /* Disable adding VLANs to empty bond. But why? --mq */
1357 features |= NETIF_F_VLAN_CHALLENGED;
1358 goto out;
1359 }
1360
1361 mask = features;
1362 features &= ~NETIF_F_ONE_FOR_ALL;
1363 features |= NETIF_F_ALL_FOR_ALL;
1364
1365 bond_for_each_slave(bond, slave, i) {
1366 features = netdev_increment_features(features,
1367 slave->dev->features,
1368 mask);
1369 }
1370
1371out:
1372 read_unlock(&bond->lock);
1373 return features;
1374}
1375
1376#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1377 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1378 NETIF_F_HIGHDMA | NETIF_F_LRO)
1379
1380static void bond_compute_features(struct bonding *bond)
1381{
1382 struct slave *slave;
1383 struct net_device *bond_dev = bond->dev;
1384 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1385 unsigned short max_hard_header_len = ETH_HLEN;
1386 int i;
1387
1388 read_lock(&bond->lock);
1389
1390 if (!bond->first_slave)
1391 goto done;
1392
1393 bond_for_each_slave(bond, slave, i) {
1394 vlan_features = netdev_increment_features(vlan_features,
1395 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1396
1397 if (slave->dev->hard_header_len > max_hard_header_len)
1398 max_hard_header_len = slave->dev->hard_header_len;
1399 }
1400
1401done:
1402 bond_dev->vlan_features = vlan_features;
1403 bond_dev->hard_header_len = max_hard_header_len;
1404
1405 read_unlock(&bond->lock);
1406
1407 netdev_change_features(bond_dev);
1408}
1409
1410static void bond_setup_by_slave(struct net_device *bond_dev,
1411 struct net_device *slave_dev)
1412{
1413 struct bonding *bond = netdev_priv(bond_dev);
1414
1415 bond_dev->header_ops = slave_dev->header_ops;
1416
1417 bond_dev->type = slave_dev->type;
1418 bond_dev->hard_header_len = slave_dev->hard_header_len;
1419 bond_dev->addr_len = slave_dev->addr_len;
1420
1421 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422 slave_dev->addr_len);
1423 bond->setup_by_slave = 1;
1424}
1425
1426/* On bonding slaves other than the currently active slave, suppress
1427 * duplicates except for alb non-mcast/bcast.
1428 */
1429static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1430 struct slave *slave,
1431 struct bonding *bond)
1432{
1433 if (bond_is_slave_inactive(slave)) {
1434 if (bond->params.mode == BOND_MODE_ALB &&
1435 skb->pkt_type != PACKET_BROADCAST &&
1436 skb->pkt_type != PACKET_MULTICAST)
1437 return false;
1438 return true;
1439 }
1440 return false;
1441}
1442
1443static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1444{
1445 struct sk_buff *skb = *pskb;
1446 struct slave *slave;
1447 struct bonding *bond;
1448 int (*recv_probe)(struct sk_buff *, struct bonding *,
1449 struct slave *);
1450 int ret = RX_HANDLER_ANOTHER;
1451
1452 skb = skb_share_check(skb, GFP_ATOMIC);
1453 if (unlikely(!skb))
1454 return RX_HANDLER_CONSUMED;
1455
1456 *pskb = skb;
1457
1458 slave = bond_slave_get_rcu(skb->dev);
1459 bond = slave->bond;
1460
1461 if (bond->params.arp_interval)
1462 slave->dev->last_rx = jiffies;
1463
1464 recv_probe = ACCESS_ONCE(bond->recv_probe);
1465 if (recv_probe) {
1466 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1467
1468 if (likely(nskb)) {
1469 ret = recv_probe(nskb, bond, slave);
1470 dev_kfree_skb(nskb);
1471 if (ret == RX_HANDLER_CONSUMED) {
1472 consume_skb(skb);
1473 return ret;
1474 }
1475 }
1476 }
1477
1478 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1479 return RX_HANDLER_EXACT;
1480 }
1481
1482 skb->dev = bond->dev;
1483
1484 if (bond->params.mode == BOND_MODE_ALB &&
1485 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1486 skb->pkt_type == PACKET_HOST) {
1487
1488 if (unlikely(skb_cow_head(skb,
1489 skb->data - skb_mac_header(skb)))) {
1490 kfree_skb(skb);
1491 return RX_HANDLER_CONSUMED;
1492 }
1493 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494 }
1495
1496 return ret;
1497}
1498
1499/* enslave device <slave> to bond device <master> */
1500int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1501{
1502 struct bonding *bond = netdev_priv(bond_dev);
1503 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1504 struct slave *new_slave = NULL;
1505 struct netdev_hw_addr *ha;
1506 struct sockaddr addr;
1507 int link_reporting;
1508 int res = 0;
1509
1510 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1511 slave_ops->ndo_do_ioctl == NULL) {
1512 pr_warning("%s: Warning: no link monitoring support for %s\n",
1513 bond_dev->name, slave_dev->name);
1514 }
1515
1516 /* already enslaved */
1517 if (slave_dev->flags & IFF_SLAVE) {
1518 pr_debug("Error, Device was already enslaved\n");
1519 return -EBUSY;
1520 }
1521
1522 /* vlan challenged mutual exclusion */
1523 /* no need to lock since we're protected by rtnl_lock */
1524 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1525 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1526 if (bond_vlan_used(bond)) {
1527 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1528 bond_dev->name, slave_dev->name, bond_dev->name);
1529 return -EPERM;
1530 } else {
1531 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1532 bond_dev->name, slave_dev->name,
1533 slave_dev->name, bond_dev->name);
1534 }
1535 } else {
1536 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1537 }
1538
1539 /*
1540 * Old ifenslave binaries are no longer supported. These can
1541 * be identified with moderate accuracy by the state of the slave:
1542 * the current ifenslave will set the interface down prior to
1543 * enslaving it; the old ifenslave will not.
1544 */
1545 if ((slave_dev->flags & IFF_UP)) {
1546 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1547 slave_dev->name);
1548 res = -EPERM;
1549 goto err_undo_flags;
1550 }
1551
1552 /* set bonding device ether type by slave - bonding netdevices are
1553 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1554 * there is a need to override some of the type dependent attribs/funcs.
1555 *
1556 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1557 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1558 */
1559 if (bond->slave_cnt == 0) {
1560 if (bond_dev->type != slave_dev->type) {
1561 pr_debug("%s: change device type from %d to %d\n",
1562 bond_dev->name,
1563 bond_dev->type, slave_dev->type);
1564
1565 res = netdev_bonding_change(bond_dev,
1566 NETDEV_PRE_TYPE_CHANGE);
1567 res = notifier_to_errno(res);
1568 if (res) {
1569 pr_err("%s: refused to change device type\n",
1570 bond_dev->name);
1571 res = -EBUSY;
1572 goto err_undo_flags;
1573 }
1574
1575 /* Flush unicast and multicast addresses */
1576 dev_uc_flush(bond_dev);
1577 dev_mc_flush(bond_dev);
1578
1579 if (slave_dev->type != ARPHRD_ETHER)
1580 bond_setup_by_slave(bond_dev, slave_dev);
1581 else {
1582 ether_setup(bond_dev);
1583 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1584 }
1585
1586 netdev_bonding_change(bond_dev,
1587 NETDEV_POST_TYPE_CHANGE);
1588 }
1589 } else if (bond_dev->type != slave_dev->type) {
1590 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1591 slave_dev->name,
1592 slave_dev->type, bond_dev->type);
1593 res = -EINVAL;
1594 goto err_undo_flags;
1595 }
1596
1597 if (slave_ops->ndo_set_mac_address == NULL) {
1598 if (bond->slave_cnt == 0) {
1599 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1600 bond_dev->name);
1601 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1602 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1603 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1604 bond_dev->name);
1605 res = -EOPNOTSUPP;
1606 goto err_undo_flags;
1607 }
1608 }
1609
1610 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1611
1612 /* If this is the first slave, then we need to set the master's hardware
1613 * address to be the same as the slave's. */
1614 if (is_zero_ether_addr(bond->dev->dev_addr))
1615 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1616 slave_dev->addr_len);
1617
1618
1619 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1620 if (!new_slave) {
1621 res = -ENOMEM;
1622 goto err_undo_flags;
1623 }
1624
1625 /*
1626 * Set the new_slave's queue_id to be zero. Queue ID mapping
1627 * is set via sysfs or module option if desired.
1628 */
1629 new_slave->queue_id = 0;
1630
1631 /* Save slave's original mtu and then set it to match the bond */
1632 new_slave->original_mtu = slave_dev->mtu;
1633 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1634 if (res) {
1635 pr_debug("Error %d calling dev_set_mtu\n", res);
1636 goto err_free;
1637 }
1638
1639 /*
1640 * Save slave's original ("permanent") mac address for modes
1641 * that need it, and for restoring it upon release, and then
1642 * set it to the master's address
1643 */
1644 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1645
1646 if (!bond->params.fail_over_mac) {
1647 /*
1648 * Set slave to master's mac address. The application already
1649 * set the master's mac address to that of the first slave
1650 */
1651 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1652 addr.sa_family = slave_dev->type;
1653 res = dev_set_mac_address(slave_dev, &addr);
1654 if (res) {
1655 pr_debug("Error %d calling set_mac_address\n", res);
1656 goto err_restore_mtu;
1657 }
1658 }
1659
1660 res = netdev_set_bond_master(slave_dev, bond_dev);
1661 if (res) {
1662 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1663 goto err_restore_mac;
1664 }
1665
1666 /* open the slave since the application closed it */
1667 res = dev_open(slave_dev);
1668 if (res) {
1669 pr_debug("Opening slave %s failed\n", slave_dev->name);
1670 goto err_unset_master;
1671 }
1672
1673 new_slave->bond = bond;
1674 new_slave->dev = slave_dev;
1675 slave_dev->priv_flags |= IFF_BONDING;
1676
1677 if (bond_is_lb(bond)) {
1678 /* bond_alb_init_slave() must be called before all other stages since
1679 * it might fail and we do not want to have to undo everything
1680 */
1681 res = bond_alb_init_slave(bond, new_slave);
1682 if (res)
1683 goto err_close;
1684 }
1685
1686 /* If the mode USES_PRIMARY, then the new slave gets the
1687 * master's promisc (and mc) settings only if it becomes the
1688 * curr_active_slave, and that is taken care of later when calling
1689 * bond_change_active()
1690 */
1691 if (!USES_PRIMARY(bond->params.mode)) {
1692 /* set promiscuity level to new slave */
1693 if (bond_dev->flags & IFF_PROMISC) {
1694 res = dev_set_promiscuity(slave_dev, 1);
1695 if (res)
1696 goto err_close;
1697 }
1698
1699 /* set allmulti level to new slave */
1700 if (bond_dev->flags & IFF_ALLMULTI) {
1701 res = dev_set_allmulti(slave_dev, 1);
1702 if (res)
1703 goto err_close;
1704 }
1705
1706 netif_addr_lock_bh(bond_dev);
1707 /* upload master's mc_list to new slave */
1708 netdev_for_each_mc_addr(ha, bond_dev)
1709 dev_mc_add(slave_dev, ha->addr);
1710 netif_addr_unlock_bh(bond_dev);
1711 }
1712
1713 if (bond->params.mode == BOND_MODE_8023AD) {
1714 /* add lacpdu mc addr to mc list */
1715 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1716
1717 dev_mc_add(slave_dev, lacpdu_multicast);
1718 }
1719
1720 bond_add_vlans_on_slave(bond, slave_dev);
1721
1722 write_lock_bh(&bond->lock);
1723
1724 bond_attach_slave(bond, new_slave);
1725
1726 new_slave->delay = 0;
1727 new_slave->link_failure_count = 0;
1728
1729 write_unlock_bh(&bond->lock);
1730
1731 bond_compute_features(bond);
1732
1733 read_lock(&bond->lock);
1734
1735 new_slave->last_arp_rx = jiffies -
1736 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1737
1738 if (bond->params.miimon && !bond->params.use_carrier) {
1739 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1740
1741 if ((link_reporting == -1) && !bond->params.arp_interval) {
1742 /*
1743 * miimon is set but a bonded network driver
1744 * does not support ETHTOOL/MII and
1745 * arp_interval is not set. Note: if
1746 * use_carrier is enabled, we will never go
1747 * here (because netif_carrier is always
1748 * supported); thus, we don't need to change
1749 * the messages for netif_carrier.
1750 */
1751 pr_warning("%s: Warning: 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",
1752 bond_dev->name, slave_dev->name);
1753 } else if (link_reporting == -1) {
1754 /* unable get link status using mii/ethtool */
1755 pr_warning("%s: Warning: 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",
1756 bond_dev->name, slave_dev->name);
1757 }
1758 }
1759
1760 /* check for initial state */
1761 if (bond->params.miimon) {
1762 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1763 if (bond->params.updelay) {
1764 new_slave->link = BOND_LINK_BACK;
1765 new_slave->delay = bond->params.updelay;
1766 } else {
1767 new_slave->link = BOND_LINK_UP;
1768 }
1769 } else {
1770 new_slave->link = BOND_LINK_DOWN;
1771 }
1772 } else if (bond->params.arp_interval) {
1773 new_slave->link = (netif_carrier_ok(slave_dev) ?
1774 BOND_LINK_UP : BOND_LINK_DOWN);
1775 } else {
1776 new_slave->link = BOND_LINK_UP;
1777 }
1778
1779 if (new_slave->link != BOND_LINK_DOWN)
1780 new_slave->jiffies = jiffies;
1781 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1782 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1783 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1784
1785 bond_update_speed_duplex(new_slave);
1786
1787 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1788 /* if there is a primary slave, remember it */
1789 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1790 bond->primary_slave = new_slave;
1791 bond->force_primary = true;
1792 }
1793 }
1794
1795 write_lock_bh(&bond->curr_slave_lock);
1796
1797 switch (bond->params.mode) {
1798 case BOND_MODE_ACTIVEBACKUP:
1799 bond_set_slave_inactive_flags(new_slave);
1800 bond_select_active_slave(bond);
1801 break;
1802 case BOND_MODE_8023AD:
1803 /* in 802.3ad mode, the internal mechanism
1804 * will activate the slaves in the selected
1805 * aggregator
1806 */
1807 bond_set_slave_inactive_flags(new_slave);
1808 /* if this is the first slave */
1809 if (bond->slave_cnt == 1) {
1810 SLAVE_AD_INFO(new_slave).id = 1;
1811 /* Initialize AD with the number of times that the AD timer is called in 1 second
1812 * can be called only after the mac address of the bond is set
1813 */
1814 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1815 } else {
1816 SLAVE_AD_INFO(new_slave).id =
1817 SLAVE_AD_INFO(new_slave->prev).id + 1;
1818 }
1819
1820 bond_3ad_bind_slave(new_slave);
1821 break;
1822 case BOND_MODE_TLB:
1823 case BOND_MODE_ALB:
1824 bond_set_active_slave(new_slave);
1825 bond_set_slave_inactive_flags(new_slave);
1826 bond_select_active_slave(bond);
1827 break;
1828 default:
1829 pr_debug("This slave is always active in trunk mode\n");
1830
1831 /* always active in trunk mode */
1832 bond_set_active_slave(new_slave);
1833
1834 /* In trunking mode there is little meaning to curr_active_slave
1835 * anyway (it holds no special properties of the bond device),
1836 * so we can change it without calling change_active_interface()
1837 */
1838 if (!bond->curr_active_slave)
1839 bond->curr_active_slave = new_slave;
1840
1841 break;
1842 } /* switch(bond_mode) */
1843
1844 write_unlock_bh(&bond->curr_slave_lock);
1845
1846 bond_set_carrier(bond);
1847
1848#ifdef CONFIG_NET_POLL_CONTROLLER
1849 slave_dev->npinfo = bond_netpoll_info(bond);
1850 if (slave_dev->npinfo) {
1851 if (slave_enable_netpoll(new_slave)) {
1852 read_unlock(&bond->lock);
1853 pr_info("Error, %s: master_dev is using netpoll, "
1854 "but new slave device does not support netpoll.\n",
1855 bond_dev->name);
1856 res = -EBUSY;
1857 goto err_detach;
1858 }
1859 }
1860#endif
1861
1862 read_unlock(&bond->lock);
1863
1864 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1865 if (res)
1866 goto err_detach;
1867
1868 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1869 new_slave);
1870 if (res) {
1871 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1872 goto err_dest_symlinks;
1873 }
1874
1875 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1876 bond_dev->name, slave_dev->name,
1877 bond_is_active_slave(new_slave) ? "n active" : " backup",
1878 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1879
1880 /* enslave is successful */
1881 return 0;
1882
1883/* Undo stages on error */
1884err_dest_symlinks:
1885 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1886
1887err_detach:
1888 write_lock_bh(&bond->lock);
1889 bond_detach_slave(bond, new_slave);
1890 write_unlock_bh(&bond->lock);
1891
1892err_close:
1893 dev_close(slave_dev);
1894
1895err_unset_master:
1896 netdev_set_bond_master(slave_dev, NULL);
1897
1898err_restore_mac:
1899 if (!bond->params.fail_over_mac) {
1900 /* XXX TODO - fom follow mode needs to change master's
1901 * MAC if this slave's MAC is in use by the bond, or at
1902 * least print a warning.
1903 */
1904 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1905 addr.sa_family = slave_dev->type;
1906 dev_set_mac_address(slave_dev, &addr);
1907 }
1908
1909err_restore_mtu:
1910 dev_set_mtu(slave_dev, new_slave->original_mtu);
1911
1912err_free:
1913 kfree(new_slave);
1914
1915err_undo_flags:
1916 bond_compute_features(bond);
1917
1918 return res;
1919}
1920
1921/*
1922 * Try to release the slave device <slave> from the bond device <master>
1923 * It is legal to access curr_active_slave without a lock because all the function
1924 * is write-locked.
1925 *
1926 * The rules for slave state should be:
1927 * for Active/Backup:
1928 * Active stays on all backups go down
1929 * for Bonded connections:
1930 * The first up interface should be left on and all others downed.
1931 */
1932int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1933{
1934 struct bonding *bond = netdev_priv(bond_dev);
1935 struct slave *slave, *oldcurrent;
1936 struct sockaddr addr;
1937 netdev_features_t old_features = bond_dev->features;
1938
1939 /* slave is not a slave or master is not master of this slave */
1940 if (!(slave_dev->flags & IFF_SLAVE) ||
1941 (slave_dev->master != bond_dev)) {
1942 pr_err("%s: Error: cannot release %s.\n",
1943 bond_dev->name, slave_dev->name);
1944 return -EINVAL;
1945 }
1946
1947 block_netpoll_tx();
1948 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1949 write_lock_bh(&bond->lock);
1950
1951 slave = bond_get_slave_by_dev(bond, slave_dev);
1952 if (!slave) {
1953 /* not a slave of this bond */
1954 pr_info("%s: %s not enslaved\n",
1955 bond_dev->name, slave_dev->name);
1956 write_unlock_bh(&bond->lock);
1957 unblock_netpoll_tx();
1958 return -EINVAL;
1959 }
1960
1961 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1962 * for this slave anymore.
1963 */
1964 netdev_rx_handler_unregister(slave_dev);
1965 write_unlock_bh(&bond->lock);
1966 synchronize_net();
1967 write_lock_bh(&bond->lock);
1968
1969 if (!bond->params.fail_over_mac) {
1970 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1971 bond->slave_cnt > 1)
1972 pr_warning("%s: Warning: 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",
1973 bond_dev->name, slave_dev->name,
1974 slave->perm_hwaddr,
1975 bond_dev->name, slave_dev->name);
1976 }
1977
1978 /* Inform AD package of unbinding of slave. */
1979 if (bond->params.mode == BOND_MODE_8023AD) {
1980 /* must be called before the slave is
1981 * detached from the list
1982 */
1983 bond_3ad_unbind_slave(slave);
1984 }
1985
1986 pr_info("%s: releasing %s interface %s\n",
1987 bond_dev->name,
1988 bond_is_active_slave(slave) ? "active" : "backup",
1989 slave_dev->name);
1990
1991 oldcurrent = bond->curr_active_slave;
1992
1993 bond->current_arp_slave = NULL;
1994
1995 /* release the slave from its bond */
1996 bond_detach_slave(bond, slave);
1997
1998 if (bond->primary_slave == slave)
1999 bond->primary_slave = NULL;
2000
2001 if (oldcurrent == slave)
2002 bond_change_active_slave(bond, NULL);
2003
2004 if (bond_is_lb(bond)) {
2005 /* Must be called only after the slave has been
2006 * detached from the list and the curr_active_slave
2007 * has been cleared (if our_slave == old_current),
2008 * but before a new active slave is selected.
2009 */
2010 write_unlock_bh(&bond->lock);
2011 bond_alb_deinit_slave(bond, slave);
2012 write_lock_bh(&bond->lock);
2013 }
2014
2015 if (oldcurrent == slave) {
2016 /*
2017 * Note that we hold RTNL over this sequence, so there
2018 * is no concern that another slave add/remove event
2019 * will interfere.
2020 */
2021 write_unlock_bh(&bond->lock);
2022 read_lock(&bond->lock);
2023 write_lock_bh(&bond->curr_slave_lock);
2024
2025 bond_select_active_slave(bond);
2026
2027 write_unlock_bh(&bond->curr_slave_lock);
2028 read_unlock(&bond->lock);
2029 write_lock_bh(&bond->lock);
2030 }
2031
2032 if (bond->slave_cnt == 0) {
2033 bond_set_carrier(bond);
2034
2035 /* if the last slave was removed, zero the mac address
2036 * of the master so it will be set by the application
2037 * to the mac address of the first slave
2038 */
2039 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2040
2041 if (bond_vlan_used(bond)) {
2042 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2043 bond_dev->name, bond_dev->name);
2044 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2045 bond_dev->name);
2046 }
2047 }
2048
2049 write_unlock_bh(&bond->lock);
2050 unblock_netpoll_tx();
2051
2052 if (bond->slave_cnt == 0)
2053 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2054
2055 bond_compute_features(bond);
2056 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2057 (old_features & NETIF_F_VLAN_CHALLENGED))
2058 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2059 bond_dev->name, slave_dev->name, bond_dev->name);
2060
2061 /* must do this from outside any spinlocks */
2062 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2063
2064 bond_del_vlans_from_slave(bond, slave_dev);
2065
2066 /* If the mode USES_PRIMARY, then we should only remove its
2067 * promisc and mc settings if it was the curr_active_slave, but that was
2068 * already taken care of above when we detached the slave
2069 */
2070 if (!USES_PRIMARY(bond->params.mode)) {
2071 /* unset promiscuity level from slave */
2072 if (bond_dev->flags & IFF_PROMISC)
2073 dev_set_promiscuity(slave_dev, -1);
2074
2075 /* unset allmulti level from slave */
2076 if (bond_dev->flags & IFF_ALLMULTI)
2077 dev_set_allmulti(slave_dev, -1);
2078
2079 /* flush master's mc_list from slave */
2080 netif_addr_lock_bh(bond_dev);
2081 bond_mc_list_flush(bond_dev, slave_dev);
2082 netif_addr_unlock_bh(bond_dev);
2083 }
2084
2085 netdev_set_bond_master(slave_dev, NULL);
2086
2087 slave_disable_netpoll(slave);
2088
2089 /* close slave before restoring its mac address */
2090 dev_close(slave_dev);
2091
2092 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2093 /* restore original ("permanent") mac address */
2094 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2095 addr.sa_family = slave_dev->type;
2096 dev_set_mac_address(slave_dev, &addr);
2097 }
2098
2099 dev_set_mtu(slave_dev, slave->original_mtu);
2100
2101 slave_dev->priv_flags &= ~IFF_BONDING;
2102
2103 kfree(slave);
2104
2105 return 0; /* deletion OK */
2106}
2107
2108/*
2109* First release a slave and then destroy the bond if no more slaves are left.
2110* Must be under rtnl_lock when this function is called.
2111*/
2112static int bond_release_and_destroy(struct net_device *bond_dev,
2113 struct net_device *slave_dev)
2114{
2115 struct bonding *bond = netdev_priv(bond_dev);
2116 int ret;
2117
2118 ret = bond_release(bond_dev, slave_dev);
2119 if ((ret == 0) && (bond->slave_cnt == 0)) {
2120 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2121 pr_info("%s: destroying bond %s.\n",
2122 bond_dev->name, bond_dev->name);
2123 unregister_netdevice(bond_dev);
2124 }
2125 return ret;
2126}
2127
2128/*
2129 * This function releases all slaves.
2130 */
2131static int bond_release_all(struct net_device *bond_dev)
2132{
2133 struct bonding *bond = netdev_priv(bond_dev);
2134 struct slave *slave;
2135 struct net_device *slave_dev;
2136 struct sockaddr addr;
2137
2138 write_lock_bh(&bond->lock);
2139
2140 netif_carrier_off(bond_dev);
2141
2142 if (bond->slave_cnt == 0)
2143 goto out;
2144
2145 bond->current_arp_slave = NULL;
2146 bond->primary_slave = NULL;
2147 bond_change_active_slave(bond, NULL);
2148
2149 while ((slave = bond->first_slave) != NULL) {
2150 /* Inform AD package of unbinding of slave
2151 * before slave is detached from the list.
2152 */
2153 if (bond->params.mode == BOND_MODE_8023AD)
2154 bond_3ad_unbind_slave(slave);
2155
2156 slave_dev = slave->dev;
2157 bond_detach_slave(bond, slave);
2158
2159 /* now that the slave is detached, unlock and perform
2160 * all the undo steps that should not be called from
2161 * within a lock.
2162 */
2163 write_unlock_bh(&bond->lock);
2164
2165 /* unregister rx_handler early so bond_handle_frame wouldn't
2166 * be called for this slave anymore.
2167 */
2168 netdev_rx_handler_unregister(slave_dev);
2169 synchronize_net();
2170
2171 if (bond_is_lb(bond)) {
2172 /* must be called only after the slave
2173 * has been detached from the list
2174 */
2175 bond_alb_deinit_slave(bond, slave);
2176 }
2177
2178 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2179 bond_del_vlans_from_slave(bond, slave_dev);
2180
2181 /* If the mode USES_PRIMARY, then we should only remove its
2182 * promisc and mc settings if it was the curr_active_slave, but that was
2183 * already taken care of above when we detached the slave
2184 */
2185 if (!USES_PRIMARY(bond->params.mode)) {
2186 /* unset promiscuity level from slave */
2187 if (bond_dev->flags & IFF_PROMISC)
2188 dev_set_promiscuity(slave_dev, -1);
2189
2190 /* unset allmulti level from slave */
2191 if (bond_dev->flags & IFF_ALLMULTI)
2192 dev_set_allmulti(slave_dev, -1);
2193
2194 /* flush master's mc_list from slave */
2195 netif_addr_lock_bh(bond_dev);
2196 bond_mc_list_flush(bond_dev, slave_dev);
2197 netif_addr_unlock_bh(bond_dev);
2198 }
2199
2200 netdev_set_bond_master(slave_dev, NULL);
2201
2202 slave_disable_netpoll(slave);
2203
2204 /* close slave before restoring its mac address */
2205 dev_close(slave_dev);
2206
2207 if (!bond->params.fail_over_mac) {
2208 /* restore original ("permanent") mac address*/
2209 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2210 addr.sa_family = slave_dev->type;
2211 dev_set_mac_address(slave_dev, &addr);
2212 }
2213
2214 kfree(slave);
2215
2216 /* re-acquire the lock before getting the next slave */
2217 write_lock_bh(&bond->lock);
2218 }
2219
2220 /* zero the mac address of the master so it will be
2221 * set by the application to the mac address of the
2222 * first slave
2223 */
2224 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2225
2226 if (bond_vlan_used(bond)) {
2227 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2228 bond_dev->name, bond_dev->name);
2229 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2230 bond_dev->name);
2231 }
2232
2233 pr_info("%s: released all slaves\n", bond_dev->name);
2234
2235out:
2236 write_unlock_bh(&bond->lock);
2237
2238 bond_compute_features(bond);
2239
2240 return 0;
2241}
2242
2243/*
2244 * This function changes the active slave to slave <slave_dev>.
2245 * It returns -EINVAL in the following cases.
2246 * - <slave_dev> is not found in the list.
2247 * - There is not active slave now.
2248 * - <slave_dev> is already active.
2249 * - The link state of <slave_dev> is not BOND_LINK_UP.
2250 * - <slave_dev> is not running.
2251 * In these cases, this function does nothing.
2252 * In the other cases, current_slave pointer is changed and 0 is returned.
2253 */
2254static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2255{
2256 struct bonding *bond = netdev_priv(bond_dev);
2257 struct slave *old_active = NULL;
2258 struct slave *new_active = NULL;
2259 int res = 0;
2260
2261 if (!USES_PRIMARY(bond->params.mode))
2262 return -EINVAL;
2263
2264 /* Verify that master_dev is indeed the master of slave_dev */
2265 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2266 return -EINVAL;
2267
2268 read_lock(&bond->lock);
2269
2270 read_lock(&bond->curr_slave_lock);
2271 old_active = bond->curr_active_slave;
2272 read_unlock(&bond->curr_slave_lock);
2273
2274 new_active = bond_get_slave_by_dev(bond, slave_dev);
2275
2276 /*
2277 * Changing to the current active: do nothing; return success.
2278 */
2279 if (new_active && (new_active == old_active)) {
2280 read_unlock(&bond->lock);
2281 return 0;
2282 }
2283
2284 if ((new_active) &&
2285 (old_active) &&
2286 (new_active->link == BOND_LINK_UP) &&
2287 IS_UP(new_active->dev)) {
2288 block_netpoll_tx();
2289 write_lock_bh(&bond->curr_slave_lock);
2290 bond_change_active_slave(bond, new_active);
2291 write_unlock_bh(&bond->curr_slave_lock);
2292 unblock_netpoll_tx();
2293 } else
2294 res = -EINVAL;
2295
2296 read_unlock(&bond->lock);
2297
2298 return res;
2299}
2300
2301static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2302{
2303 struct bonding *bond = netdev_priv(bond_dev);
2304
2305 info->bond_mode = bond->params.mode;
2306 info->miimon = bond->params.miimon;
2307
2308 read_lock(&bond->lock);
2309 info->num_slaves = bond->slave_cnt;
2310 read_unlock(&bond->lock);
2311
2312 return 0;
2313}
2314
2315static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2316{
2317 struct bonding *bond = netdev_priv(bond_dev);
2318 struct slave *slave;
2319 int i, res = -ENODEV;
2320
2321 read_lock(&bond->lock);
2322
2323 bond_for_each_slave(bond, slave, i) {
2324 if (i == (int)info->slave_id) {
2325 res = 0;
2326 strcpy(info->slave_name, slave->dev->name);
2327 info->link = slave->link;
2328 info->state = bond_slave_state(slave);
2329 info->link_failure_count = slave->link_failure_count;
2330 break;
2331 }
2332 }
2333
2334 read_unlock(&bond->lock);
2335
2336 return res;
2337}
2338
2339/*-------------------------------- Monitoring -------------------------------*/
2340
2341
2342static int bond_miimon_inspect(struct bonding *bond)
2343{
2344 struct slave *slave;
2345 int i, link_state, commit = 0;
2346 bool ignore_updelay;
2347
2348 ignore_updelay = !bond->curr_active_slave ? true : false;
2349
2350 bond_for_each_slave(bond, slave, i) {
2351 slave->new_link = BOND_LINK_NOCHANGE;
2352
2353 link_state = bond_check_dev_link(bond, slave->dev, 0);
2354
2355 switch (slave->link) {
2356 case BOND_LINK_UP:
2357 if (link_state)
2358 continue;
2359
2360 slave->link = BOND_LINK_FAIL;
2361 slave->delay = bond->params.downdelay;
2362 if (slave->delay) {
2363 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2364 bond->dev->name,
2365 (bond->params.mode ==
2366 BOND_MODE_ACTIVEBACKUP) ?
2367 (bond_is_active_slave(slave) ?
2368 "active " : "backup ") : "",
2369 slave->dev->name,
2370 bond->params.downdelay * bond->params.miimon);
2371 }
2372 /*FALLTHRU*/
2373 case BOND_LINK_FAIL:
2374 if (link_state) {
2375 /*
2376 * recovered before downdelay expired
2377 */
2378 slave->link = BOND_LINK_UP;
2379 slave->jiffies = jiffies;
2380 pr_info("%s: link status up again after %d ms for interface %s.\n",
2381 bond->dev->name,
2382 (bond->params.downdelay - slave->delay) *
2383 bond->params.miimon,
2384 slave->dev->name);
2385 continue;
2386 }
2387
2388 if (slave->delay <= 0) {
2389 slave->new_link = BOND_LINK_DOWN;
2390 commit++;
2391 continue;
2392 }
2393
2394 slave->delay--;
2395 break;
2396
2397 case BOND_LINK_DOWN:
2398 if (!link_state)
2399 continue;
2400
2401 slave->link = BOND_LINK_BACK;
2402 slave->delay = bond->params.updelay;
2403
2404 if (slave->delay) {
2405 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2406 bond->dev->name, slave->dev->name,
2407 ignore_updelay ? 0 :
2408 bond->params.updelay *
2409 bond->params.miimon);
2410 }
2411 /*FALLTHRU*/
2412 case BOND_LINK_BACK:
2413 if (!link_state) {
2414 slave->link = BOND_LINK_DOWN;
2415 pr_info("%s: link status down again after %d ms for interface %s.\n",
2416 bond->dev->name,
2417 (bond->params.updelay - slave->delay) *
2418 bond->params.miimon,
2419 slave->dev->name);
2420
2421 continue;
2422 }
2423
2424 if (ignore_updelay)
2425 slave->delay = 0;
2426
2427 if (slave->delay <= 0) {
2428 slave->new_link = BOND_LINK_UP;
2429 commit++;
2430 ignore_updelay = false;
2431 continue;
2432 }
2433
2434 slave->delay--;
2435 break;
2436 }
2437 }
2438
2439 return commit;
2440}
2441
2442static void bond_miimon_commit(struct bonding *bond)
2443{
2444 struct slave *slave;
2445 int i;
2446
2447 bond_for_each_slave(bond, slave, i) {
2448 switch (slave->new_link) {
2449 case BOND_LINK_NOCHANGE:
2450 continue;
2451
2452 case BOND_LINK_UP:
2453 slave->link = BOND_LINK_UP;
2454 slave->jiffies = jiffies;
2455
2456 if (bond->params.mode == BOND_MODE_8023AD) {
2457 /* prevent it from being the active one */
2458 bond_set_backup_slave(slave);
2459 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2460 /* make it immediately active */
2461 bond_set_active_slave(slave);
2462 } else if (slave != bond->primary_slave) {
2463 /* prevent it from being the active one */
2464 bond_set_backup_slave(slave);
2465 }
2466
2467 bond_update_speed_duplex(slave);
2468
2469 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2470 bond->dev->name, slave->dev->name,
2471 slave->speed, slave->duplex ? "full" : "half");
2472
2473 /* notify ad that the link status has changed */
2474 if (bond->params.mode == BOND_MODE_8023AD)
2475 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2476
2477 if (bond_is_lb(bond))
2478 bond_alb_handle_link_change(bond, slave,
2479 BOND_LINK_UP);
2480
2481 if (!bond->curr_active_slave ||
2482 (slave == bond->primary_slave))
2483 goto do_failover;
2484
2485 continue;
2486
2487 case BOND_LINK_DOWN:
2488 if (slave->link_failure_count < UINT_MAX)
2489 slave->link_failure_count++;
2490
2491 slave->link = BOND_LINK_DOWN;
2492
2493 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2494 bond->params.mode == BOND_MODE_8023AD)
2495 bond_set_slave_inactive_flags(slave);
2496
2497 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2498 bond->dev->name, slave->dev->name);
2499
2500 if (bond->params.mode == BOND_MODE_8023AD)
2501 bond_3ad_handle_link_change(slave,
2502 BOND_LINK_DOWN);
2503
2504 if (bond_is_lb(bond))
2505 bond_alb_handle_link_change(bond, slave,
2506 BOND_LINK_DOWN);
2507
2508 if (slave == bond->curr_active_slave)
2509 goto do_failover;
2510
2511 continue;
2512
2513 default:
2514 pr_err("%s: invalid new link %d on slave %s\n",
2515 bond->dev->name, slave->new_link,
2516 slave->dev->name);
2517 slave->new_link = BOND_LINK_NOCHANGE;
2518
2519 continue;
2520 }
2521
2522do_failover:
2523 ASSERT_RTNL();
2524 block_netpoll_tx();
2525 write_lock_bh(&bond->curr_slave_lock);
2526 bond_select_active_slave(bond);
2527 write_unlock_bh(&bond->curr_slave_lock);
2528 unblock_netpoll_tx();
2529 }
2530
2531 bond_set_carrier(bond);
2532}
2533
2534/*
2535 * bond_mii_monitor
2536 *
2537 * Really a wrapper that splits the mii monitor into two phases: an
2538 * inspection, then (if inspection indicates something needs to be done)
2539 * an acquisition of appropriate locks followed by a commit phase to
2540 * implement whatever link state changes are indicated.
2541 */
2542void bond_mii_monitor(struct work_struct *work)
2543{
2544 struct bonding *bond = container_of(work, struct bonding,
2545 mii_work.work);
2546 bool should_notify_peers = false;
2547 unsigned long delay;
2548
2549 read_lock(&bond->lock);
2550
2551 delay = msecs_to_jiffies(bond->params.miimon);
2552
2553 if (bond->slave_cnt == 0)
2554 goto re_arm;
2555
2556 should_notify_peers = bond_should_notify_peers(bond);
2557
2558 if (bond_miimon_inspect(bond)) {
2559 read_unlock(&bond->lock);
2560
2561 /* Race avoidance with bond_close cancel of workqueue */
2562 if (!rtnl_trylock()) {
2563 read_lock(&bond->lock);
2564 delay = 1;
2565 should_notify_peers = false;
2566 goto re_arm;
2567 }
2568
2569 read_lock(&bond->lock);
2570
2571 bond_miimon_commit(bond);
2572
2573 read_unlock(&bond->lock);
2574 rtnl_unlock(); /* might sleep, hold no other locks */
2575 read_lock(&bond->lock);
2576 }
2577
2578re_arm:
2579 if (bond->params.miimon)
2580 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2581
2582 read_unlock(&bond->lock);
2583
2584 if (should_notify_peers) {
2585 if (!rtnl_trylock()) {
2586 read_lock(&bond->lock);
2587 bond->send_peer_notif++;
2588 read_unlock(&bond->lock);
2589 return;
2590 }
2591 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2592 rtnl_unlock();
2593 }
2594}
2595
2596static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2597{
2598 struct vlan_entry *vlan;
2599 struct net_device *vlan_dev;
2600
2601 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2602 return 1;
2603
2604 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2605 rcu_read_lock();
2606 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2607 rcu_read_unlock();
2608 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2609 return 1;
2610 }
2611
2612 return 0;
2613}
2614
2615/*
2616 * We go to the (large) trouble of VLAN tagging ARP frames because
2617 * switches in VLAN mode (especially if ports are configured as
2618 * "native" to a VLAN) might not pass non-tagged frames.
2619 */
2620static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2621{
2622 struct sk_buff *skb;
2623
2624 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2625 slave_dev->name, dest_ip, src_ip, vlan_id);
2626
2627 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2628 NULL, slave_dev->dev_addr, NULL);
2629
2630 if (!skb) {
2631 pr_err("ARP packet allocation failed\n");
2632 return;
2633 }
2634 if (vlan_id) {
2635 skb = vlan_put_tag(skb, vlan_id);
2636 if (!skb) {
2637 pr_err("failed to insert VLAN tag\n");
2638 return;
2639 }
2640 }
2641 arp_xmit(skb);
2642}
2643
2644
2645static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2646{
2647 int i, vlan_id;
2648 __be32 *targets = bond->params.arp_targets;
2649 struct vlan_entry *vlan;
2650 struct net_device *vlan_dev = NULL;
2651 struct rtable *rt;
2652
2653 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2654 __be32 addr;
2655 if (!targets[i])
2656 break;
2657 pr_debug("basa: target %x\n", targets[i]);
2658 if (!bond_vlan_used(bond)) {
2659 pr_debug("basa: empty vlan: arp_send\n");
2660 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2661 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2662 addr, 0);
2663 continue;
2664 }
2665
2666 /*
2667 * If VLANs are configured, we do a route lookup to
2668 * determine which VLAN interface would be used, so we
2669 * can tag the ARP with the proper VLAN tag.
2670 */
2671 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2672 RTO_ONLINK, 0);
2673 if (IS_ERR(rt)) {
2674 if (net_ratelimit()) {
2675 pr_warning("%s: no route to arp_ip_target %pI4\n",
2676 bond->dev->name, &targets[i]);
2677 }
2678 continue;
2679 }
2680
2681 /*
2682 * This target is not on a VLAN
2683 */
2684 if (rt->dst.dev == bond->dev) {
2685 ip_rt_put(rt);
2686 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2687 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2688 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2689 addr, 0);
2690 continue;
2691 }
2692
2693 vlan_id = 0;
2694 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2695 rcu_read_lock();
2696 vlan_dev = __vlan_find_dev_deep(bond->dev,
2697 vlan->vlan_id);
2698 rcu_read_unlock();
2699 if (vlan_dev == rt->dst.dev) {
2700 vlan_id = vlan->vlan_id;
2701 pr_debug("basa: vlan match on %s %d\n",
2702 vlan_dev->name, vlan_id);
2703 break;
2704 }
2705 }
2706
2707 if (vlan_id && vlan_dev) {
2708 ip_rt_put(rt);
2709 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2710 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2711 addr, vlan_id);
2712 continue;
2713 }
2714
2715 if (net_ratelimit()) {
2716 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2717 bond->dev->name, &targets[i],
2718 rt->dst.dev ? rt->dst.dev->name : "NULL");
2719 }
2720 ip_rt_put(rt);
2721 }
2722}
2723
2724static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2725{
2726 int i;
2727 __be32 *targets = bond->params.arp_targets;
2728
2729 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2730 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2731 &sip, &tip, i, &targets[i],
2732 bond_has_this_ip(bond, tip));
2733 if (sip == targets[i]) {
2734 if (bond_has_this_ip(bond, tip))
2735 slave->last_arp_rx = jiffies;
2736 return;
2737 }
2738 }
2739}
2740
2741static int bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2742 struct slave *slave)
2743{
2744 struct arphdr *arp;
2745 unsigned char *arp_ptr;
2746 __be32 sip, tip;
2747
2748 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2749 return RX_HANDLER_ANOTHER;
2750
2751 read_lock(&bond->lock);
2752
2753 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2754 bond->dev->name, skb->dev->name);
2755
2756 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2757 goto out_unlock;
2758
2759 arp = arp_hdr(skb);
2760 if (arp->ar_hln != bond->dev->addr_len ||
2761 skb->pkt_type == PACKET_OTHERHOST ||
2762 skb->pkt_type == PACKET_LOOPBACK ||
2763 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2764 arp->ar_pro != htons(ETH_P_IP) ||
2765 arp->ar_pln != 4)
2766 goto out_unlock;
2767
2768 arp_ptr = (unsigned char *)(arp + 1);
2769 arp_ptr += bond->dev->addr_len;
2770 memcpy(&sip, arp_ptr, 4);
2771 arp_ptr += 4 + bond->dev->addr_len;
2772 memcpy(&tip, arp_ptr, 4);
2773
2774 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2775 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2776 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2777 &sip, &tip);
2778
2779 /*
2780 * Backup slaves won't see the ARP reply, but do come through
2781 * here for each ARP probe (so we swap the sip/tip to validate
2782 * the probe). In a "redundant switch, common router" type of
2783 * configuration, the ARP probe will (hopefully) travel from
2784 * the active, through one switch, the router, then the other
2785 * switch before reaching the backup.
2786 */
2787 if (bond_is_active_slave(slave))
2788 bond_validate_arp(bond, slave, sip, tip);
2789 else
2790 bond_validate_arp(bond, slave, tip, sip);
2791
2792out_unlock:
2793 read_unlock(&bond->lock);
2794 return RX_HANDLER_ANOTHER;
2795}
2796
2797/*
2798 * this function is called regularly to monitor each slave's link
2799 * ensuring that traffic is being sent and received when arp monitoring
2800 * is used in load-balancing mode. if the adapter has been dormant, then an
2801 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2802 * arp monitoring in active backup mode.
2803 */
2804void bond_loadbalance_arp_mon(struct work_struct *work)
2805{
2806 struct bonding *bond = container_of(work, struct bonding,
2807 arp_work.work);
2808 struct slave *slave, *oldcurrent;
2809 int do_failover = 0;
2810 int delta_in_ticks;
2811 int i;
2812
2813 read_lock(&bond->lock);
2814
2815 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2816
2817 if (bond->slave_cnt == 0)
2818 goto re_arm;
2819
2820 read_lock(&bond->curr_slave_lock);
2821 oldcurrent = bond->curr_active_slave;
2822 read_unlock(&bond->curr_slave_lock);
2823
2824 /* see if any of the previous devices are up now (i.e. they have
2825 * xmt and rcv traffic). the curr_active_slave does not come into
2826 * the picture unless it is null. also, slave->jiffies is not needed
2827 * here because we send an arp on each slave and give a slave as
2828 * long as it needs to get the tx/rx within the delta.
2829 * TODO: what about up/down delay in arp mode? it wasn't here before
2830 * so it can wait
2831 */
2832 bond_for_each_slave(bond, slave, i) {
2833 unsigned long trans_start = dev_trans_start(slave->dev);
2834
2835 if (slave->link != BOND_LINK_UP) {
2836 if (time_in_range(jiffies,
2837 trans_start - delta_in_ticks,
2838 trans_start + delta_in_ticks) &&
2839 time_in_range(jiffies,
2840 slave->dev->last_rx - delta_in_ticks,
2841 slave->dev->last_rx + delta_in_ticks)) {
2842
2843 slave->link = BOND_LINK_UP;
2844 bond_set_active_slave(slave);
2845
2846 /* primary_slave has no meaning in round-robin
2847 * mode. the window of a slave being up and
2848 * curr_active_slave being null after enslaving
2849 * is closed.
2850 */
2851 if (!oldcurrent) {
2852 pr_info("%s: link status definitely up for interface %s, ",
2853 bond->dev->name,
2854 slave->dev->name);
2855 do_failover = 1;
2856 } else {
2857 pr_info("%s: interface %s is now up\n",
2858 bond->dev->name,
2859 slave->dev->name);
2860 }
2861 }
2862 } else {
2863 /* slave->link == BOND_LINK_UP */
2864
2865 /* not all switches will respond to an arp request
2866 * when the source ip is 0, so don't take the link down
2867 * if we don't know our ip yet
2868 */
2869 if (!time_in_range(jiffies,
2870 trans_start - delta_in_ticks,
2871 trans_start + 2 * delta_in_ticks) ||
2872 !time_in_range(jiffies,
2873 slave->dev->last_rx - delta_in_ticks,
2874 slave->dev->last_rx + 2 * delta_in_ticks)) {
2875
2876 slave->link = BOND_LINK_DOWN;
2877 bond_set_backup_slave(slave);
2878
2879 if (slave->link_failure_count < UINT_MAX)
2880 slave->link_failure_count++;
2881
2882 pr_info("%s: interface %s is now down.\n",
2883 bond->dev->name,
2884 slave->dev->name);
2885
2886 if (slave == oldcurrent)
2887 do_failover = 1;
2888 }
2889 }
2890
2891 /* note: if switch is in round-robin mode, all links
2892 * must tx arp to ensure all links rx an arp - otherwise
2893 * links may oscillate or not come up at all; if switch is
2894 * in something like xor mode, there is nothing we can
2895 * do - all replies will be rx'ed on same link causing slaves
2896 * to be unstable during low/no traffic periods
2897 */
2898 if (IS_UP(slave->dev))
2899 bond_arp_send_all(bond, slave);
2900 }
2901
2902 if (do_failover) {
2903 block_netpoll_tx();
2904 write_lock_bh(&bond->curr_slave_lock);
2905
2906 bond_select_active_slave(bond);
2907
2908 write_unlock_bh(&bond->curr_slave_lock);
2909 unblock_netpoll_tx();
2910 }
2911
2912re_arm:
2913 if (bond->params.arp_interval)
2914 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2915
2916 read_unlock(&bond->lock);
2917}
2918
2919/*
2920 * Called to inspect slaves for active-backup mode ARP monitor link state
2921 * changes. Sets new_link in slaves to specify what action should take
2922 * place for the slave. Returns 0 if no changes are found, >0 if changes
2923 * to link states must be committed.
2924 *
2925 * Called with bond->lock held for read.
2926 */
2927static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2928{
2929 struct slave *slave;
2930 int i, commit = 0;
2931 unsigned long trans_start;
2932
2933 bond_for_each_slave(bond, slave, i) {
2934 slave->new_link = BOND_LINK_NOCHANGE;
2935
2936 if (slave->link != BOND_LINK_UP) {
2937 if (time_in_range(jiffies,
2938 slave_last_rx(bond, slave) - delta_in_ticks,
2939 slave_last_rx(bond, slave) + delta_in_ticks)) {
2940
2941 slave->new_link = BOND_LINK_UP;
2942 commit++;
2943 }
2944
2945 continue;
2946 }
2947
2948 /*
2949 * Give slaves 2*delta after being enslaved or made
2950 * active. This avoids bouncing, as the last receive
2951 * times need a full ARP monitor cycle to be updated.
2952 */
2953 if (time_in_range(jiffies,
2954 slave->jiffies - delta_in_ticks,
2955 slave->jiffies + 2 * delta_in_ticks))
2956 continue;
2957
2958 /*
2959 * Backup slave is down if:
2960 * - No current_arp_slave AND
2961 * - more than 3*delta since last receive AND
2962 * - the bond has an IP address
2963 *
2964 * Note: a non-null current_arp_slave indicates
2965 * the curr_active_slave went down and we are
2966 * searching for a new one; under this condition
2967 * we only take the curr_active_slave down - this
2968 * gives each slave a chance to tx/rx traffic
2969 * before being taken out
2970 */
2971 if (!bond_is_active_slave(slave) &&
2972 !bond->current_arp_slave &&
2973 !time_in_range(jiffies,
2974 slave_last_rx(bond, slave) - delta_in_ticks,
2975 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2976
2977 slave->new_link = BOND_LINK_DOWN;
2978 commit++;
2979 }
2980
2981 /*
2982 * Active slave is down if:
2983 * - more than 2*delta since transmitting OR
2984 * - (more than 2*delta since receive AND
2985 * the bond has an IP address)
2986 */
2987 trans_start = dev_trans_start(slave->dev);
2988 if (bond_is_active_slave(slave) &&
2989 (!time_in_range(jiffies,
2990 trans_start - delta_in_ticks,
2991 trans_start + 2 * delta_in_ticks) ||
2992 !time_in_range(jiffies,
2993 slave_last_rx(bond, slave) - delta_in_ticks,
2994 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2995
2996 slave->new_link = BOND_LINK_DOWN;
2997 commit++;
2998 }
2999 }
3000
3001 return commit;
3002}
3003
3004/*
3005 * Called to commit link state changes noted by inspection step of
3006 * active-backup mode ARP monitor.
3007 *
3008 * Called with RTNL and bond->lock for read.
3009 */
3010static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3011{
3012 struct slave *slave;
3013 int i;
3014 unsigned long trans_start;
3015
3016 bond_for_each_slave(bond, slave, i) {
3017 switch (slave->new_link) {
3018 case BOND_LINK_NOCHANGE:
3019 continue;
3020
3021 case BOND_LINK_UP:
3022 trans_start = dev_trans_start(slave->dev);
3023 if ((!bond->curr_active_slave &&
3024 time_in_range(jiffies,
3025 trans_start - delta_in_ticks,
3026 trans_start + delta_in_ticks)) ||
3027 bond->curr_active_slave != slave) {
3028 slave->link = BOND_LINK_UP;
3029 if (bond->current_arp_slave) {
3030 bond_set_slave_inactive_flags(
3031 bond->current_arp_slave);
3032 bond->current_arp_slave = NULL;
3033 }
3034
3035 pr_info("%s: link status definitely up for interface %s.\n",
3036 bond->dev->name, slave->dev->name);
3037
3038 if (!bond->curr_active_slave ||
3039 (slave == bond->primary_slave))
3040 goto do_failover;
3041
3042 }
3043
3044 continue;
3045
3046 case BOND_LINK_DOWN:
3047 if (slave->link_failure_count < UINT_MAX)
3048 slave->link_failure_count++;
3049
3050 slave->link = BOND_LINK_DOWN;
3051 bond_set_slave_inactive_flags(slave);
3052
3053 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3054 bond->dev->name, slave->dev->name);
3055
3056 if (slave == bond->curr_active_slave) {
3057 bond->current_arp_slave = NULL;
3058 goto do_failover;
3059 }
3060
3061 continue;
3062
3063 default:
3064 pr_err("%s: impossible: new_link %d on slave %s\n",
3065 bond->dev->name, slave->new_link,
3066 slave->dev->name);
3067 continue;
3068 }
3069
3070do_failover:
3071 ASSERT_RTNL();
3072 block_netpoll_tx();
3073 write_lock_bh(&bond->curr_slave_lock);
3074 bond_select_active_slave(bond);
3075 write_unlock_bh(&bond->curr_slave_lock);
3076 unblock_netpoll_tx();
3077 }
3078
3079 bond_set_carrier(bond);
3080}
3081
3082/*
3083 * Send ARP probes for active-backup mode ARP monitor.
3084 *
3085 * Called with bond->lock held for read.
3086 */
3087static void bond_ab_arp_probe(struct bonding *bond)
3088{
3089 struct slave *slave;
3090 int i;
3091
3092 read_lock(&bond->curr_slave_lock);
3093
3094 if (bond->current_arp_slave && bond->curr_active_slave)
3095 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3096 bond->current_arp_slave->dev->name,
3097 bond->curr_active_slave->dev->name);
3098
3099 if (bond->curr_active_slave) {
3100 bond_arp_send_all(bond, bond->curr_active_slave);
3101 read_unlock(&bond->curr_slave_lock);
3102 return;
3103 }
3104
3105 read_unlock(&bond->curr_slave_lock);
3106
3107 /* if we don't have a curr_active_slave, search for the next available
3108 * backup slave from the current_arp_slave and make it the candidate
3109 * for becoming the curr_active_slave
3110 */
3111
3112 if (!bond->current_arp_slave) {
3113 bond->current_arp_slave = bond->first_slave;
3114 if (!bond->current_arp_slave)
3115 return;
3116 }
3117
3118 bond_set_slave_inactive_flags(bond->current_arp_slave);
3119
3120 /* search for next candidate */
3121 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3122 if (IS_UP(slave->dev)) {
3123 slave->link = BOND_LINK_BACK;
3124 bond_set_slave_active_flags(slave);
3125 bond_arp_send_all(bond, slave);
3126 slave->jiffies = jiffies;
3127 bond->current_arp_slave = slave;
3128 break;
3129 }
3130
3131 /* if the link state is up at this point, we
3132 * mark it down - this can happen if we have
3133 * simultaneous link failures and
3134 * reselect_active_interface doesn't make this
3135 * one the current slave so it is still marked
3136 * up when it is actually down
3137 */
3138 if (slave->link == BOND_LINK_UP) {
3139 slave->link = BOND_LINK_DOWN;
3140 if (slave->link_failure_count < UINT_MAX)
3141 slave->link_failure_count++;
3142
3143 bond_set_slave_inactive_flags(slave);
3144
3145 pr_info("%s: backup interface %s is now down.\n",
3146 bond->dev->name, slave->dev->name);
3147 }
3148 }
3149}
3150
3151void bond_activebackup_arp_mon(struct work_struct *work)
3152{
3153 struct bonding *bond = container_of(work, struct bonding,
3154 arp_work.work);
3155 bool should_notify_peers = false;
3156 int delta_in_ticks;
3157
3158 read_lock(&bond->lock);
3159
3160 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3161
3162 if (bond->slave_cnt == 0)
3163 goto re_arm;
3164
3165 should_notify_peers = bond_should_notify_peers(bond);
3166
3167 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3168 read_unlock(&bond->lock);
3169
3170 /* Race avoidance with bond_close flush of workqueue */
3171 if (!rtnl_trylock()) {
3172 read_lock(&bond->lock);
3173 delta_in_ticks = 1;
3174 should_notify_peers = false;
3175 goto re_arm;
3176 }
3177
3178 read_lock(&bond->lock);
3179
3180 bond_ab_arp_commit(bond, delta_in_ticks);
3181
3182 read_unlock(&bond->lock);
3183 rtnl_unlock();
3184 read_lock(&bond->lock);
3185 }
3186
3187 bond_ab_arp_probe(bond);
3188
3189re_arm:
3190 if (bond->params.arp_interval)
3191 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3192
3193 read_unlock(&bond->lock);
3194
3195 if (should_notify_peers) {
3196 if (!rtnl_trylock()) {
3197 read_lock(&bond->lock);
3198 bond->send_peer_notif++;
3199 read_unlock(&bond->lock);
3200 return;
3201 }
3202 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3203 rtnl_unlock();
3204 }
3205}
3206
3207/*-------------------------- netdev event handling --------------------------*/
3208
3209/*
3210 * Change device name
3211 */
3212static int bond_event_changename(struct bonding *bond)
3213{
3214 bond_remove_proc_entry(bond);
3215 bond_create_proc_entry(bond);
3216
3217 bond_debug_reregister(bond);
3218
3219 return NOTIFY_DONE;
3220}
3221
3222static int bond_master_netdev_event(unsigned long event,
3223 struct net_device *bond_dev)
3224{
3225 struct bonding *event_bond = netdev_priv(bond_dev);
3226
3227 switch (event) {
3228 case NETDEV_CHANGENAME:
3229 return bond_event_changename(event_bond);
3230 case NETDEV_UNREGISTER:
3231 bond_remove_proc_entry(event_bond);
3232 break;
3233 case NETDEV_REGISTER:
3234 bond_create_proc_entry(event_bond);
3235 break;
3236 default:
3237 break;
3238 }
3239
3240 return NOTIFY_DONE;
3241}
3242
3243static int bond_slave_netdev_event(unsigned long event,
3244 struct net_device *slave_dev)
3245{
3246 struct net_device *bond_dev = slave_dev->master;
3247 struct bonding *bond = netdev_priv(bond_dev);
3248 struct slave *slave = NULL;
3249
3250 switch (event) {
3251 case NETDEV_UNREGISTER:
3252 if (bond_dev) {
3253 if (bond->setup_by_slave)
3254 bond_release_and_destroy(bond_dev, slave_dev);
3255 else
3256 bond_release(bond_dev, slave_dev);
3257 }
3258 break;
3259 case NETDEV_UP:
3260 case NETDEV_CHANGE:
3261 slave = bond_get_slave_by_dev(bond, slave_dev);
3262 if (slave) {
3263 u32 old_speed = slave->speed;
3264 u8 old_duplex = slave->duplex;
3265
3266 bond_update_speed_duplex(slave);
3267
3268 if (bond->params.mode == BOND_MODE_8023AD) {
3269 if (old_speed != slave->speed)
3270 bond_3ad_adapter_speed_changed(slave);
3271 if (old_duplex != slave->duplex)
3272 bond_3ad_adapter_duplex_changed(slave);
3273 }
3274 }
3275
3276 break;
3277 case NETDEV_DOWN:
3278 /*
3279 * ... Or is it this?
3280 */
3281 break;
3282 case NETDEV_CHANGEMTU:
3283 /*
3284 * TODO: Should slaves be allowed to
3285 * independently alter their MTU? For
3286 * an active-backup bond, slaves need
3287 * not be the same type of device, so
3288 * MTUs may vary. For other modes,
3289 * slaves arguably should have the
3290 * same MTUs. To do this, we'd need to
3291 * take over the slave's change_mtu
3292 * function for the duration of their
3293 * servitude.
3294 */
3295 break;
3296 case NETDEV_CHANGENAME:
3297 /*
3298 * TODO: handle changing the primary's name
3299 */
3300 break;
3301 case NETDEV_FEAT_CHANGE:
3302 bond_compute_features(bond);
3303 break;
3304 default:
3305 break;
3306 }
3307
3308 return NOTIFY_DONE;
3309}
3310
3311/*
3312 * bond_netdev_event: handle netdev notifier chain events.
3313 *
3314 * This function receives events for the netdev chain. The caller (an
3315 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3316 * locks for us to safely manipulate the slave devices (RTNL lock,
3317 * dev_probe_lock).
3318 */
3319static int bond_netdev_event(struct notifier_block *this,
3320 unsigned long event, void *ptr)
3321{
3322 struct net_device *event_dev = (struct net_device *)ptr;
3323
3324 pr_debug("event_dev: %s, event: %lx\n",
3325 event_dev ? event_dev->name : "None",
3326 event);
3327
3328 if (!(event_dev->priv_flags & IFF_BONDING))
3329 return NOTIFY_DONE;
3330
3331 if (event_dev->flags & IFF_MASTER) {
3332 pr_debug("IFF_MASTER\n");
3333 return bond_master_netdev_event(event, event_dev);
3334 }
3335
3336 if (event_dev->flags & IFF_SLAVE) {
3337 pr_debug("IFF_SLAVE\n");
3338 return bond_slave_netdev_event(event, event_dev);
3339 }
3340
3341 return NOTIFY_DONE;
3342}
3343
3344static struct notifier_block bond_netdev_notifier = {
3345 .notifier_call = bond_netdev_event,
3346};
3347
3348/*---------------------------- Hashing Policies -----------------------------*/
3349
3350/*
3351 * Hash for the output device based upon layer 2 and layer 3 data. If
3352 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3353 */
3354static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3355{
3356 struct ethhdr *data = (struct ethhdr *)skb->data;
3357 struct iphdr *iph = ip_hdr(skb);
3358
3359 if (skb->protocol == htons(ETH_P_IP)) {
3360 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3361 (data->h_dest[5] ^ data->h_source[5])) % count;
3362 }
3363
3364 return (data->h_dest[5] ^ data->h_source[5]) % count;
3365}
3366
3367/*
3368 * Hash for the output device based upon layer 3 and layer 4 data. If
3369 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3370 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3371 */
3372static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3373{
3374 struct ethhdr *data = (struct ethhdr *)skb->data;
3375 struct iphdr *iph = ip_hdr(skb);
3376 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3377 int layer4_xor = 0;
3378
3379 if (skb->protocol == htons(ETH_P_IP)) {
3380 if (!ip_is_fragment(iph) &&
3381 (iph->protocol == IPPROTO_TCP ||
3382 iph->protocol == IPPROTO_UDP)) {
3383 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3384 }
3385 return (layer4_xor ^
3386 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3387
3388 }
3389
3390 return (data->h_dest[5] ^ data->h_source[5]) % count;
3391}
3392
3393/*
3394 * Hash for the output device based upon layer 2 data
3395 */
3396static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3397{
3398 struct ethhdr *data = (struct ethhdr *)skb->data;
3399
3400 return (data->h_dest[5] ^ data->h_source[5]) % count;
3401}
3402
3403/*-------------------------- Device entry points ----------------------------*/
3404
3405static int bond_open(struct net_device *bond_dev)
3406{
3407 struct bonding *bond = netdev_priv(bond_dev);
3408 struct slave *slave;
3409 int i;
3410
3411 /* reset slave->backup and slave->inactive */
3412 read_lock(&bond->lock);
3413 if (bond->slave_cnt > 0) {
3414 read_lock(&bond->curr_slave_lock);
3415 bond_for_each_slave(bond, slave, i) {
3416 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3417 && (slave != bond->curr_active_slave)) {
3418 bond_set_slave_inactive_flags(slave);
3419 } else {
3420 bond_set_slave_active_flags(slave);
3421 }
3422 }
3423 read_unlock(&bond->curr_slave_lock);
3424 }
3425 read_unlock(&bond->lock);
3426
3427 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3428
3429 if (bond_is_lb(bond)) {
3430 /* bond_alb_initialize must be called before the timer
3431 * is started.
3432 */
3433 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3434 /* something went wrong - fail the open operation */
3435 return -ENOMEM;
3436 }
3437
3438 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3439 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3440 }
3441
3442 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3443 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3444 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3445 }
3446
3447 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3448 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3449 INIT_DELAYED_WORK(&bond->arp_work,
3450 bond_activebackup_arp_mon);
3451 else
3452 INIT_DELAYED_WORK(&bond->arp_work,
3453 bond_loadbalance_arp_mon);
3454
3455 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3456 if (bond->params.arp_validate)
3457 bond->recv_probe = bond_arp_rcv;
3458 }
3459
3460 if (bond->params.mode == BOND_MODE_8023AD) {
3461 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3462 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3463 /* register to receive LACPDUs */
3464 bond->recv_probe = bond_3ad_lacpdu_recv;
3465 bond_3ad_initiate_agg_selection(bond, 1);
3466 }
3467
3468 return 0;
3469}
3470
3471static int bond_close(struct net_device *bond_dev)
3472{
3473 struct bonding *bond = netdev_priv(bond_dev);
3474
3475 write_lock_bh(&bond->lock);
3476
3477 bond->send_peer_notif = 0;
3478
3479 write_unlock_bh(&bond->lock);
3480
3481 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3482 cancel_delayed_work_sync(&bond->mii_work);
3483 }
3484
3485 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3486 cancel_delayed_work_sync(&bond->arp_work);
3487 }
3488
3489 switch (bond->params.mode) {
3490 case BOND_MODE_8023AD:
3491 cancel_delayed_work_sync(&bond->ad_work);
3492 break;
3493 case BOND_MODE_TLB:
3494 case BOND_MODE_ALB:
3495 cancel_delayed_work_sync(&bond->alb_work);
3496 break;
3497 default:
3498 break;
3499 }
3500
3501 if (delayed_work_pending(&bond->mcast_work))
3502 cancel_delayed_work_sync(&bond->mcast_work);
3503
3504 if (bond_is_lb(bond)) {
3505 /* Must be called only after all
3506 * slaves have been released
3507 */
3508 bond_alb_deinitialize(bond);
3509 }
3510 bond->recv_probe = NULL;
3511
3512 return 0;
3513}
3514
3515static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3516 struct rtnl_link_stats64 *stats)
3517{
3518 struct bonding *bond = netdev_priv(bond_dev);
3519 struct rtnl_link_stats64 temp;
3520 struct slave *slave;
3521 int i;
3522
3523 memset(stats, 0, sizeof(*stats));
3524
3525 read_lock_bh(&bond->lock);
3526
3527 bond_for_each_slave(bond, slave, i) {
3528 const struct rtnl_link_stats64 *sstats =
3529 dev_get_stats(slave->dev, &temp);
3530
3531 stats->rx_packets += sstats->rx_packets;
3532 stats->rx_bytes += sstats->rx_bytes;
3533 stats->rx_errors += sstats->rx_errors;
3534 stats->rx_dropped += sstats->rx_dropped;
3535
3536 stats->tx_packets += sstats->tx_packets;
3537 stats->tx_bytes += sstats->tx_bytes;
3538 stats->tx_errors += sstats->tx_errors;
3539 stats->tx_dropped += sstats->tx_dropped;
3540
3541 stats->multicast += sstats->multicast;
3542 stats->collisions += sstats->collisions;
3543
3544 stats->rx_length_errors += sstats->rx_length_errors;
3545 stats->rx_over_errors += sstats->rx_over_errors;
3546 stats->rx_crc_errors += sstats->rx_crc_errors;
3547 stats->rx_frame_errors += sstats->rx_frame_errors;
3548 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3549 stats->rx_missed_errors += sstats->rx_missed_errors;
3550
3551 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3552 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3553 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3554 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3555 stats->tx_window_errors += sstats->tx_window_errors;
3556 }
3557
3558 read_unlock_bh(&bond->lock);
3559
3560 return stats;
3561}
3562
3563static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3564{
3565 struct net_device *slave_dev = NULL;
3566 struct ifbond k_binfo;
3567 struct ifbond __user *u_binfo = NULL;
3568 struct ifslave k_sinfo;
3569 struct ifslave __user *u_sinfo = NULL;
3570 struct mii_ioctl_data *mii = NULL;
3571 int res = 0;
3572
3573 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3574
3575 switch (cmd) {
3576 case SIOCGMIIPHY:
3577 mii = if_mii(ifr);
3578 if (!mii)
3579 return -EINVAL;
3580
3581 mii->phy_id = 0;
3582 /* Fall Through */
3583 case SIOCGMIIREG:
3584 /*
3585 * We do this again just in case we were called by SIOCGMIIREG
3586 * instead of SIOCGMIIPHY.
3587 */
3588 mii = if_mii(ifr);
3589 if (!mii)
3590 return -EINVAL;
3591
3592
3593 if (mii->reg_num == 1) {
3594 struct bonding *bond = netdev_priv(bond_dev);
3595 mii->val_out = 0;
3596 read_lock(&bond->lock);
3597 read_lock(&bond->curr_slave_lock);
3598 if (netif_carrier_ok(bond->dev))
3599 mii->val_out = BMSR_LSTATUS;
3600
3601 read_unlock(&bond->curr_slave_lock);
3602 read_unlock(&bond->lock);
3603 }
3604
3605 return 0;
3606 case BOND_INFO_QUERY_OLD:
3607 case SIOCBONDINFOQUERY:
3608 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3609
3610 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3611 return -EFAULT;
3612
3613 res = bond_info_query(bond_dev, &k_binfo);
3614 if (res == 0 &&
3615 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3616 return -EFAULT;
3617
3618 return res;
3619 case BOND_SLAVE_INFO_QUERY_OLD:
3620 case SIOCBONDSLAVEINFOQUERY:
3621 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3622
3623 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3624 return -EFAULT;
3625
3626 res = bond_slave_info_query(bond_dev, &k_sinfo);
3627 if (res == 0 &&
3628 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3629 return -EFAULT;
3630
3631 return res;
3632 default:
3633 /* Go on */
3634 break;
3635 }
3636
3637 if (!capable(CAP_NET_ADMIN))
3638 return -EPERM;
3639
3640 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3641
3642 pr_debug("slave_dev=%p:\n", slave_dev);
3643
3644 if (!slave_dev)
3645 res = -ENODEV;
3646 else {
3647 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3648 switch (cmd) {
3649 case BOND_ENSLAVE_OLD:
3650 case SIOCBONDENSLAVE:
3651 res = bond_enslave(bond_dev, slave_dev);
3652 break;
3653 case BOND_RELEASE_OLD:
3654 case SIOCBONDRELEASE:
3655 res = bond_release(bond_dev, slave_dev);
3656 break;
3657 case BOND_SETHWADDR_OLD:
3658 case SIOCBONDSETHWADDR:
3659 res = bond_sethwaddr(bond_dev, slave_dev);
3660 break;
3661 case BOND_CHANGE_ACTIVE_OLD:
3662 case SIOCBONDCHANGEACTIVE:
3663 res = bond_ioctl_change_active(bond_dev, slave_dev);
3664 break;
3665 default:
3666 res = -EOPNOTSUPP;
3667 }
3668
3669 dev_put(slave_dev);
3670 }
3671
3672 return res;
3673}
3674
3675static bool bond_addr_in_mc_list(unsigned char *addr,
3676 struct netdev_hw_addr_list *list,
3677 int addrlen)
3678{
3679 struct netdev_hw_addr *ha;
3680
3681 netdev_hw_addr_list_for_each(ha, list)
3682 if (!memcmp(ha->addr, addr, addrlen))
3683 return true;
3684
3685 return false;
3686}
3687
3688static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3689{
3690 struct bonding *bond = netdev_priv(bond_dev);
3691
3692 if (change & IFF_PROMISC)
3693 bond_set_promiscuity(bond,
3694 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3695
3696 if (change & IFF_ALLMULTI)
3697 bond_set_allmulti(bond,
3698 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3699}
3700
3701static void bond_set_multicast_list(struct net_device *bond_dev)
3702{
3703 struct bonding *bond = netdev_priv(bond_dev);
3704 struct netdev_hw_addr *ha;
3705 bool found;
3706
3707 read_lock(&bond->lock);
3708
3709 /* looking for addresses to add to slaves' mc list */
3710 netdev_for_each_mc_addr(ha, bond_dev) {
3711 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3712 bond_dev->addr_len);
3713 if (!found)
3714 bond_mc_add(bond, ha->addr);
3715 }
3716
3717 /* looking for addresses to delete from slaves' list */
3718 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3719 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3720 bond_dev->addr_len);
3721 if (!found)
3722 bond_mc_del(bond, ha->addr);
3723 }
3724
3725 /* save master's multicast list */
3726 __hw_addr_flush(&bond->mc_list);
3727 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3728 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3729
3730 read_unlock(&bond->lock);
3731}
3732
3733static int bond_neigh_init(struct neighbour *n)
3734{
3735 struct bonding *bond = netdev_priv(n->dev);
3736 struct slave *slave = bond->first_slave;
3737 const struct net_device_ops *slave_ops;
3738 struct neigh_parms parms;
3739 int ret;
3740
3741 if (!slave)
3742 return 0;
3743
3744 slave_ops = slave->dev->netdev_ops;
3745
3746 if (!slave_ops->ndo_neigh_setup)
3747 return 0;
3748
3749 parms.neigh_setup = NULL;
3750 parms.neigh_cleanup = NULL;
3751 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3752 if (ret)
3753 return ret;
3754
3755 /*
3756 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3757 * after the last slave has been detached. Assumes that all slaves
3758 * utilize the same neigh_cleanup (true at this writing as only user
3759 * is ipoib).
3760 */
3761 n->parms->neigh_cleanup = parms.neigh_cleanup;
3762
3763 if (!parms.neigh_setup)
3764 return 0;
3765
3766 return parms.neigh_setup(n);
3767}
3768
3769/*
3770 * The bonding ndo_neigh_setup is called at init time beofre any
3771 * slave exists. So we must declare proxy setup function which will
3772 * be used at run time to resolve the actual slave neigh param setup.
3773 */
3774static int bond_neigh_setup(struct net_device *dev,
3775 struct neigh_parms *parms)
3776{
3777 parms->neigh_setup = bond_neigh_init;
3778
3779 return 0;
3780}
3781
3782/*
3783 * Change the MTU of all of a master's slaves to match the master
3784 */
3785static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3786{
3787 struct bonding *bond = netdev_priv(bond_dev);
3788 struct slave *slave, *stop_at;
3789 int res = 0;
3790 int i;
3791
3792 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3793 (bond_dev ? bond_dev->name : "None"), new_mtu);
3794
3795 /* Can't hold bond->lock with bh disabled here since
3796 * some base drivers panic. On the other hand we can't
3797 * hold bond->lock without bh disabled because we'll
3798 * deadlock. The only solution is to rely on the fact
3799 * that we're under rtnl_lock here, and the slaves
3800 * list won't change. This doesn't solve the problem
3801 * of setting the slave's MTU while it is
3802 * transmitting, but the assumption is that the base
3803 * driver can handle that.
3804 *
3805 * TODO: figure out a way to safely iterate the slaves
3806 * list, but without holding a lock around the actual
3807 * call to the base driver.
3808 */
3809
3810 bond_for_each_slave(bond, slave, i) {
3811 pr_debug("s %p s->p %p c_m %p\n",
3812 slave,
3813 slave->prev,
3814 slave->dev->netdev_ops->ndo_change_mtu);
3815
3816 res = dev_set_mtu(slave->dev, new_mtu);
3817
3818 if (res) {
3819 /* If we failed to set the slave's mtu to the new value
3820 * we must abort the operation even in ACTIVE_BACKUP
3821 * mode, because if we allow the backup slaves to have
3822 * different mtu values than the active slave we'll
3823 * need to change their mtu when doing a failover. That
3824 * means changing their mtu from timer context, which
3825 * is probably not a good idea.
3826 */
3827 pr_debug("err %d %s\n", res, slave->dev->name);
3828 goto unwind;
3829 }
3830 }
3831
3832 bond_dev->mtu = new_mtu;
3833
3834 return 0;
3835
3836unwind:
3837 /* unwind from head to the slave that failed */
3838 stop_at = slave;
3839 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3840 int tmp_res;
3841
3842 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3843 if (tmp_res) {
3844 pr_debug("unwind err %d dev %s\n",
3845 tmp_res, slave->dev->name);
3846 }
3847 }
3848
3849 return res;
3850}
3851
3852/*
3853 * Change HW address
3854 *
3855 * Note that many devices must be down to change the HW address, and
3856 * downing the master releases all slaves. We can make bonds full of
3857 * bonding devices to test this, however.
3858 */
3859static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3860{
3861 struct bonding *bond = netdev_priv(bond_dev);
3862 struct sockaddr *sa = addr, tmp_sa;
3863 struct slave *slave, *stop_at;
3864 int res = 0;
3865 int i;
3866
3867 if (bond->params.mode == BOND_MODE_ALB)
3868 return bond_alb_set_mac_address(bond_dev, addr);
3869
3870
3871 pr_debug("bond=%p, name=%s\n",
3872 bond, bond_dev ? bond_dev->name : "None");
3873
3874 /*
3875 * If fail_over_mac is set to active, do nothing and return
3876 * success. Returning an error causes ifenslave to fail.
3877 */
3878 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3879 return 0;
3880
3881 if (!is_valid_ether_addr(sa->sa_data))
3882 return -EADDRNOTAVAIL;
3883
3884 /* Can't hold bond->lock with bh disabled here since
3885 * some base drivers panic. On the other hand we can't
3886 * hold bond->lock without bh disabled because we'll
3887 * deadlock. The only solution is to rely on the fact
3888 * that we're under rtnl_lock here, and the slaves
3889 * list won't change. This doesn't solve the problem
3890 * of setting the slave's hw address while it is
3891 * transmitting, but the assumption is that the base
3892 * driver can handle that.
3893 *
3894 * TODO: figure out a way to safely iterate the slaves
3895 * list, but without holding a lock around the actual
3896 * call to the base driver.
3897 */
3898
3899 bond_for_each_slave(bond, slave, i) {
3900 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3901 pr_debug("slave %p %s\n", slave, slave->dev->name);
3902
3903 if (slave_ops->ndo_set_mac_address == NULL) {
3904 res = -EOPNOTSUPP;
3905 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3906 goto unwind;
3907 }
3908
3909 res = dev_set_mac_address(slave->dev, addr);
3910 if (res) {
3911 /* TODO: consider downing the slave
3912 * and retry ?
3913 * User should expect communications
3914 * breakage anyway until ARP finish
3915 * updating, so...
3916 */
3917 pr_debug("err %d %s\n", res, slave->dev->name);
3918 goto unwind;
3919 }
3920 }
3921
3922 /* success */
3923 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3924 return 0;
3925
3926unwind:
3927 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3928 tmp_sa.sa_family = bond_dev->type;
3929
3930 /* unwind from head to the slave that failed */
3931 stop_at = slave;
3932 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3933 int tmp_res;
3934
3935 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3936 if (tmp_res) {
3937 pr_debug("unwind err %d dev %s\n",
3938 tmp_res, slave->dev->name);
3939 }
3940 }
3941
3942 return res;
3943}
3944
3945static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3946{
3947 struct bonding *bond = netdev_priv(bond_dev);
3948 struct slave *slave, *start_at;
3949 int i, slave_no, res = 1;
3950 struct iphdr *iph = ip_hdr(skb);
3951
3952 /*
3953 * Start with the curr_active_slave that joined the bond as the
3954 * default for sending IGMP traffic. For failover purposes one
3955 * needs to maintain some consistency for the interface that will
3956 * send the join/membership reports. The curr_active_slave found
3957 * will send all of this type of traffic.
3958 */
3959 if ((iph->protocol == IPPROTO_IGMP) &&
3960 (skb->protocol == htons(ETH_P_IP))) {
3961
3962 read_lock(&bond->curr_slave_lock);
3963 slave = bond->curr_active_slave;
3964 read_unlock(&bond->curr_slave_lock);
3965
3966 if (!slave)
3967 goto out;
3968 } else {
3969 /*
3970 * Concurrent TX may collide on rr_tx_counter; we accept
3971 * that as being rare enough not to justify using an
3972 * atomic op here.
3973 */
3974 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3975
3976 bond_for_each_slave(bond, slave, i) {
3977 slave_no--;
3978 if (slave_no < 0)
3979 break;
3980 }
3981 }
3982
3983 start_at = slave;
3984 bond_for_each_slave_from(bond, slave, i, start_at) {
3985 if (IS_UP(slave->dev) &&
3986 (slave->link == BOND_LINK_UP) &&
3987 bond_is_active_slave(slave)) {
3988 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3989 break;
3990 }
3991 }
3992
3993out:
3994 if (res) {
3995 /* no suitable interface, frame not sent */
3996 dev_kfree_skb(skb);
3997 }
3998
3999 return NETDEV_TX_OK;
4000}
4001
4002
4003/*
4004 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4005 * the bond has a usable interface.
4006 */
4007static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4008{
4009 struct bonding *bond = netdev_priv(bond_dev);
4010 int res = 1;
4011
4012 read_lock(&bond->curr_slave_lock);
4013
4014 if (bond->curr_active_slave)
4015 res = bond_dev_queue_xmit(bond, skb,
4016 bond->curr_active_slave->dev);
4017
4018 if (res)
4019 /* no suitable interface, frame not sent */
4020 dev_kfree_skb(skb);
4021
4022 read_unlock(&bond->curr_slave_lock);
4023
4024 return NETDEV_TX_OK;
4025}
4026
4027/*
4028 * In bond_xmit_xor() , we determine the output device by using a pre-
4029 * determined xmit_hash_policy(), If the selected device is not enabled,
4030 * find the next active slave.
4031 */
4032static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4033{
4034 struct bonding *bond = netdev_priv(bond_dev);
4035 struct slave *slave, *start_at;
4036 int slave_no;
4037 int i;
4038 int res = 1;
4039
4040 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4041
4042 bond_for_each_slave(bond, slave, i) {
4043 slave_no--;
4044 if (slave_no < 0)
4045 break;
4046 }
4047
4048 start_at = slave;
4049
4050 bond_for_each_slave_from(bond, slave, i, start_at) {
4051 if (IS_UP(slave->dev) &&
4052 (slave->link == BOND_LINK_UP) &&
4053 bond_is_active_slave(slave)) {
4054 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4055 break;
4056 }
4057 }
4058
4059 if (res) {
4060 /* no suitable interface, frame not sent */
4061 dev_kfree_skb(skb);
4062 }
4063
4064 return NETDEV_TX_OK;
4065}
4066
4067/*
4068 * in broadcast mode, we send everything to all usable interfaces.
4069 */
4070static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4071{
4072 struct bonding *bond = netdev_priv(bond_dev);
4073 struct slave *slave, *start_at;
4074 struct net_device *tx_dev = NULL;
4075 int i;
4076 int res = 1;
4077
4078 read_lock(&bond->curr_slave_lock);
4079 start_at = bond->curr_active_slave;
4080 read_unlock(&bond->curr_slave_lock);
4081
4082 if (!start_at)
4083 goto out;
4084
4085 bond_for_each_slave_from(bond, slave, i, start_at) {
4086 if (IS_UP(slave->dev) &&
4087 (slave->link == BOND_LINK_UP) &&
4088 bond_is_active_slave(slave)) {
4089 if (tx_dev) {
4090 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4091 if (!skb2) {
4092 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4093 bond_dev->name);
4094 continue;
4095 }
4096
4097 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4098 if (res) {
4099 dev_kfree_skb(skb2);
4100 continue;
4101 }
4102 }
4103 tx_dev = slave->dev;
4104 }
4105 }
4106
4107 if (tx_dev)
4108 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4109
4110out:
4111 if (res)
4112 /* no suitable interface, frame not sent */
4113 dev_kfree_skb(skb);
4114
4115 /* frame sent to all suitable interfaces */
4116 return NETDEV_TX_OK;
4117}
4118
4119/*------------------------- Device initialization ---------------------------*/
4120
4121static void bond_set_xmit_hash_policy(struct bonding *bond)
4122{
4123 switch (bond->params.xmit_policy) {
4124 case BOND_XMIT_POLICY_LAYER23:
4125 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4126 break;
4127 case BOND_XMIT_POLICY_LAYER34:
4128 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4129 break;
4130 case BOND_XMIT_POLICY_LAYER2:
4131 default:
4132 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4133 break;
4134 }
4135}
4136
4137/*
4138 * Lookup the slave that corresponds to a qid
4139 */
4140static inline int bond_slave_override(struct bonding *bond,
4141 struct sk_buff *skb)
4142{
4143 int i, res = 1;
4144 struct slave *slave = NULL;
4145 struct slave *check_slave;
4146
4147 if (!skb->queue_mapping)
4148 return 1;
4149
4150 /* Find out if any slaves have the same mapping as this skb. */
4151 bond_for_each_slave(bond, check_slave, i) {
4152 if (check_slave->queue_id == skb->queue_mapping) {
4153 slave = check_slave;
4154 break;
4155 }
4156 }
4157
4158 /* If the slave isn't UP, use default transmit policy. */
4159 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4160 (slave->link == BOND_LINK_UP)) {
4161 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4162 }
4163
4164 return res;
4165}
4166
4167
4168static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4169{
4170 /*
4171 * This helper function exists to help dev_pick_tx get the correct
4172 * destination queue. Using a helper function skips a call to
4173 * skb_tx_hash and will put the skbs in the queue we expect on their
4174 * way down to the bonding driver.
4175 */
4176 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4177
4178 /*
4179 * Save the original txq to restore before passing to the driver
4180 */
4181 qdisc_skb_cb(skb)->bond_queue_mapping = skb->queue_mapping;
4182
4183 if (unlikely(txq >= dev->real_num_tx_queues)) {
4184 do {
4185 txq -= dev->real_num_tx_queues;
4186 } while (txq >= dev->real_num_tx_queues);
4187 }
4188 return txq;
4189}
4190
4191static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4192{
4193 struct bonding *bond = netdev_priv(dev);
4194
4195 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4196 if (!bond_slave_override(bond, skb))
4197 return NETDEV_TX_OK;
4198 }
4199
4200 switch (bond->params.mode) {
4201 case BOND_MODE_ROUNDROBIN:
4202 return bond_xmit_roundrobin(skb, dev);
4203 case BOND_MODE_ACTIVEBACKUP:
4204 return bond_xmit_activebackup(skb, dev);
4205 case BOND_MODE_XOR:
4206 return bond_xmit_xor(skb, dev);
4207 case BOND_MODE_BROADCAST:
4208 return bond_xmit_broadcast(skb, dev);
4209 case BOND_MODE_8023AD:
4210 return bond_3ad_xmit_xor(skb, dev);
4211 case BOND_MODE_ALB:
4212 case BOND_MODE_TLB:
4213 return bond_alb_xmit(skb, dev);
4214 default:
4215 /* Should never happen, mode already checked */
4216 pr_err("%s: Error: Unknown bonding mode %d\n",
4217 dev->name, bond->params.mode);
4218 WARN_ON_ONCE(1);
4219 dev_kfree_skb(skb);
4220 return NETDEV_TX_OK;
4221 }
4222}
4223
4224static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4225{
4226 struct bonding *bond = netdev_priv(dev);
4227 netdev_tx_t ret = NETDEV_TX_OK;
4228
4229 /*
4230 * If we risk deadlock from transmitting this in the
4231 * netpoll path, tell netpoll to queue the frame for later tx
4232 */
4233 if (is_netpoll_tx_blocked(dev))
4234 return NETDEV_TX_BUSY;
4235
4236 read_lock(&bond->lock);
4237
4238 if (bond->slave_cnt)
4239 ret = __bond_start_xmit(skb, dev);
4240 else
4241 dev_kfree_skb(skb);
4242
4243 read_unlock(&bond->lock);
4244
4245 return ret;
4246}
4247
4248/*
4249 * set bond mode specific net device operations
4250 */
4251void bond_set_mode_ops(struct bonding *bond, int mode)
4252{
4253 struct net_device *bond_dev = bond->dev;
4254
4255 switch (mode) {
4256 case BOND_MODE_ROUNDROBIN:
4257 break;
4258 case BOND_MODE_ACTIVEBACKUP:
4259 break;
4260 case BOND_MODE_XOR:
4261 bond_set_xmit_hash_policy(bond);
4262 break;
4263 case BOND_MODE_BROADCAST:
4264 break;
4265 case BOND_MODE_8023AD:
4266 bond_set_xmit_hash_policy(bond);
4267 break;
4268 case BOND_MODE_ALB:
4269 /* FALLTHRU */
4270 case BOND_MODE_TLB:
4271 break;
4272 default:
4273 /* Should never happen, mode already checked */
4274 pr_err("%s: Error: Unknown bonding mode %d\n",
4275 bond_dev->name, mode);
4276 break;
4277 }
4278}
4279
4280static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4281 struct ethtool_drvinfo *drvinfo)
4282{
4283 strncpy(drvinfo->driver, DRV_NAME, 32);
4284 strncpy(drvinfo->version, DRV_VERSION, 32);
4285 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4286}
4287
4288static const struct ethtool_ops bond_ethtool_ops = {
4289 .get_drvinfo = bond_ethtool_get_drvinfo,
4290 .get_link = ethtool_op_get_link,
4291};
4292
4293static const struct net_device_ops bond_netdev_ops = {
4294 .ndo_init = bond_init,
4295 .ndo_uninit = bond_uninit,
4296 .ndo_open = bond_open,
4297 .ndo_stop = bond_close,
4298 .ndo_start_xmit = bond_start_xmit,
4299 .ndo_select_queue = bond_select_queue,
4300 .ndo_get_stats64 = bond_get_stats,
4301 .ndo_do_ioctl = bond_do_ioctl,
4302 .ndo_change_rx_flags = bond_change_rx_flags,
4303 .ndo_set_rx_mode = bond_set_multicast_list,
4304 .ndo_change_mtu = bond_change_mtu,
4305 .ndo_set_mac_address = bond_set_mac_address,
4306 .ndo_neigh_setup = bond_neigh_setup,
4307 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4308 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4309#ifdef CONFIG_NET_POLL_CONTROLLER
4310 .ndo_netpoll_setup = bond_netpoll_setup,
4311 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4312 .ndo_poll_controller = bond_poll_controller,
4313#endif
4314 .ndo_add_slave = bond_enslave,
4315 .ndo_del_slave = bond_release,
4316 .ndo_fix_features = bond_fix_features,
4317};
4318
4319static void bond_destructor(struct net_device *bond_dev)
4320{
4321 struct bonding *bond = netdev_priv(bond_dev);
4322 if (bond->wq)
4323 destroy_workqueue(bond->wq);
4324 free_netdev(bond_dev);
4325}
4326
4327static void bond_setup(struct net_device *bond_dev)
4328{
4329 struct bonding *bond = netdev_priv(bond_dev);
4330
4331 /* initialize rwlocks */
4332 rwlock_init(&bond->lock);
4333 rwlock_init(&bond->curr_slave_lock);
4334
4335 bond->params = bonding_defaults;
4336
4337 /* Initialize pointers */
4338 bond->dev = bond_dev;
4339 INIT_LIST_HEAD(&bond->vlan_list);
4340
4341 /* Initialize the device entry points */
4342 ether_setup(bond_dev);
4343 bond_dev->netdev_ops = &bond_netdev_ops;
4344 bond_dev->ethtool_ops = &bond_ethtool_ops;
4345 bond_set_mode_ops(bond, bond->params.mode);
4346
4347 bond_dev->destructor = bond_destructor;
4348
4349 /* Initialize the device options */
4350 bond_dev->tx_queue_len = 0;
4351 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4352 bond_dev->priv_flags |= IFF_BONDING;
4353 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4354
4355 /* At first, we block adding VLANs. That's the only way to
4356 * prevent problems that occur when adding VLANs over an
4357 * empty bond. The block will be removed once non-challenged
4358 * slaves are enslaved.
4359 */
4360 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4361
4362 /* don't acquire bond device's netif_tx_lock when
4363 * transmitting */
4364 bond_dev->features |= NETIF_F_LLTX;
4365
4366 /* By default, we declare the bond to be fully
4367 * VLAN hardware accelerated capable. Special
4368 * care is taken in the various xmit functions
4369 * when there are slaves that are not hw accel
4370 * capable
4371 */
4372
4373 bond_dev->hw_features = BOND_VLAN_FEATURES |
4374 NETIF_F_HW_VLAN_TX |
4375 NETIF_F_HW_VLAN_RX |
4376 NETIF_F_HW_VLAN_FILTER;
4377
4378 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4379 bond_dev->features |= bond_dev->hw_features;
4380}
4381
4382static void bond_work_cancel_all(struct bonding *bond)
4383{
4384 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4385 cancel_delayed_work_sync(&bond->mii_work);
4386
4387 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4388 cancel_delayed_work_sync(&bond->arp_work);
4389
4390 if (bond->params.mode == BOND_MODE_ALB &&
4391 delayed_work_pending(&bond->alb_work))
4392 cancel_delayed_work_sync(&bond->alb_work);
4393
4394 if (bond->params.mode == BOND_MODE_8023AD &&
4395 delayed_work_pending(&bond->ad_work))
4396 cancel_delayed_work_sync(&bond->ad_work);
4397
4398 if (delayed_work_pending(&bond->mcast_work))
4399 cancel_delayed_work_sync(&bond->mcast_work);
4400}
4401
4402/*
4403* Destroy a bonding device.
4404* Must be under rtnl_lock when this function is called.
4405*/
4406static void bond_uninit(struct net_device *bond_dev)
4407{
4408 struct bonding *bond = netdev_priv(bond_dev);
4409 struct vlan_entry *vlan, *tmp;
4410
4411 bond_netpoll_cleanup(bond_dev);
4412
4413 /* Release the bonded slaves */
4414 bond_release_all(bond_dev);
4415
4416 list_del(&bond->bond_list);
4417
4418 bond_work_cancel_all(bond);
4419
4420 bond_debug_unregister(bond);
4421
4422 __hw_addr_flush(&bond->mc_list);
4423
4424 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4425 list_del(&vlan->vlan_list);
4426 kfree(vlan);
4427 }
4428}
4429
4430/*------------------------- Module initialization ---------------------------*/
4431
4432/*
4433 * Convert string input module parms. Accept either the
4434 * number of the mode or its string name. A bit complicated because
4435 * some mode names are substrings of other names, and calls from sysfs
4436 * may have whitespace in the name (trailing newlines, for example).
4437 */
4438int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4439{
4440 int modeint = -1, i, rv;
4441 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4442
4443 for (p = (char *)buf; *p; p++)
4444 if (!(isdigit(*p) || isspace(*p)))
4445 break;
4446
4447 if (*p)
4448 rv = sscanf(buf, "%20s", modestr);
4449 else
4450 rv = sscanf(buf, "%d", &modeint);
4451
4452 if (!rv)
4453 return -1;
4454
4455 for (i = 0; tbl[i].modename; i++) {
4456 if (modeint == tbl[i].mode)
4457 return tbl[i].mode;
4458 if (strcmp(modestr, tbl[i].modename) == 0)
4459 return tbl[i].mode;
4460 }
4461
4462 return -1;
4463}
4464
4465static int bond_check_params(struct bond_params *params)
4466{
4467 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4468
4469 /*
4470 * Convert string parameters.
4471 */
4472 if (mode) {
4473 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4474 if (bond_mode == -1) {
4475 pr_err("Error: Invalid bonding mode \"%s\"\n",
4476 mode == NULL ? "NULL" : mode);
4477 return -EINVAL;
4478 }
4479 }
4480
4481 if (xmit_hash_policy) {
4482 if ((bond_mode != BOND_MODE_XOR) &&
4483 (bond_mode != BOND_MODE_8023AD)) {
4484 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4485 bond_mode_name(bond_mode));
4486 } else {
4487 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4488 xmit_hashtype_tbl);
4489 if (xmit_hashtype == -1) {
4490 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4491 xmit_hash_policy == NULL ? "NULL" :
4492 xmit_hash_policy);
4493 return -EINVAL;
4494 }
4495 }
4496 }
4497
4498 if (lacp_rate) {
4499 if (bond_mode != BOND_MODE_8023AD) {
4500 pr_info("lacp_rate param is irrelevant in mode %s\n",
4501 bond_mode_name(bond_mode));
4502 } else {
4503 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4504 if (lacp_fast == -1) {
4505 pr_err("Error: Invalid lacp rate \"%s\"\n",
4506 lacp_rate == NULL ? "NULL" : lacp_rate);
4507 return -EINVAL;
4508 }
4509 }
4510 }
4511
4512 if (ad_select) {
4513 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4514 if (params->ad_select == -1) {
4515 pr_err("Error: Invalid ad_select \"%s\"\n",
4516 ad_select == NULL ? "NULL" : ad_select);
4517 return -EINVAL;
4518 }
4519
4520 if (bond_mode != BOND_MODE_8023AD) {
4521 pr_warning("ad_select param only affects 802.3ad mode\n");
4522 }
4523 } else {
4524 params->ad_select = BOND_AD_STABLE;
4525 }
4526
4527 if (max_bonds < 0) {
4528 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4529 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4530 max_bonds = BOND_DEFAULT_MAX_BONDS;
4531 }
4532
4533 if (miimon < 0) {
4534 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4535 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4536 miimon = BOND_LINK_MON_INTERV;
4537 }
4538
4539 if (updelay < 0) {
4540 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4541 updelay, INT_MAX);
4542 updelay = 0;
4543 }
4544
4545 if (downdelay < 0) {
4546 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547 downdelay, INT_MAX);
4548 downdelay = 0;
4549 }
4550
4551 if ((use_carrier != 0) && (use_carrier != 1)) {
4552 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4553 use_carrier);
4554 use_carrier = 1;
4555 }
4556
4557 if (num_peer_notif < 0 || num_peer_notif > 255) {
4558 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4559 num_peer_notif);
4560 num_peer_notif = 1;
4561 }
4562
4563 /* reset values for 802.3ad */
4564 if (bond_mode == BOND_MODE_8023AD) {
4565 if (!miimon) {
4566 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4567 pr_warning("Forcing miimon to 100msec\n");
4568 miimon = 100;
4569 }
4570 }
4571
4572 if (tx_queues < 1 || tx_queues > 255) {
4573 pr_warning("Warning: tx_queues (%d) should be between "
4574 "1 and 255, resetting to %d\n",
4575 tx_queues, BOND_DEFAULT_TX_QUEUES);
4576 tx_queues = BOND_DEFAULT_TX_QUEUES;
4577 }
4578
4579 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4580 pr_warning("Warning: all_slaves_active module parameter (%d), "
4581 "not of valid value (0/1), so it was set to "
4582 "0\n", all_slaves_active);
4583 all_slaves_active = 0;
4584 }
4585
4586 if (resend_igmp < 0 || resend_igmp > 255) {
4587 pr_warning("Warning: resend_igmp (%d) should be between "
4588 "0 and 255, resetting to %d\n",
4589 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4590 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4591 }
4592
4593 /* reset values for TLB/ALB */
4594 if ((bond_mode == BOND_MODE_TLB) ||
4595 (bond_mode == BOND_MODE_ALB)) {
4596 if (!miimon) {
4597 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4598 pr_warning("Forcing miimon to 100msec\n");
4599 miimon = 100;
4600 }
4601 }
4602
4603 if (bond_mode == BOND_MODE_ALB) {
4604 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",
4605 updelay);
4606 }
4607
4608 if (!miimon) {
4609 if (updelay || downdelay) {
4610 /* just warn the user the up/down delay will have
4611 * no effect since miimon is zero...
4612 */
4613 pr_warning("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",
4614 updelay, downdelay);
4615 }
4616 } else {
4617 /* don't allow arp monitoring */
4618 if (arp_interval) {
4619 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4620 miimon, arp_interval);
4621 arp_interval = 0;
4622 }
4623
4624 if ((updelay % miimon) != 0) {
4625 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4626 updelay, miimon,
4627 (updelay / miimon) * miimon);
4628 }
4629
4630 updelay /= miimon;
4631
4632 if ((downdelay % miimon) != 0) {
4633 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4634 downdelay, miimon,
4635 (downdelay / miimon) * miimon);
4636 }
4637
4638 downdelay /= miimon;
4639 }
4640
4641 if (arp_interval < 0) {
4642 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4643 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4644 arp_interval = BOND_LINK_ARP_INTERV;
4645 }
4646
4647 for (arp_ip_count = 0;
4648 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4649 arp_ip_count++) {
4650 /* not complete check, but should be good enough to
4651 catch mistakes */
4652 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4653 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4654 arp_ip_target[arp_ip_count]);
4655 arp_interval = 0;
4656 } else {
4657 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4658 arp_target[arp_ip_count] = ip;
4659 }
4660 }
4661
4662 if (arp_interval && !arp_ip_count) {
4663 /* don't allow arping if no arp_ip_target given... */
4664 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4665 arp_interval);
4666 arp_interval = 0;
4667 }
4668
4669 if (arp_validate) {
4670 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4671 pr_err("arp_validate only supported in active-backup mode\n");
4672 return -EINVAL;
4673 }
4674 if (!arp_interval) {
4675 pr_err("arp_validate requires arp_interval\n");
4676 return -EINVAL;
4677 }
4678
4679 arp_validate_value = bond_parse_parm(arp_validate,
4680 arp_validate_tbl);
4681 if (arp_validate_value == -1) {
4682 pr_err("Error: invalid arp_validate \"%s\"\n",
4683 arp_validate == NULL ? "NULL" : arp_validate);
4684 return -EINVAL;
4685 }
4686 } else
4687 arp_validate_value = 0;
4688
4689 if (miimon) {
4690 pr_info("MII link monitoring set to %d ms\n", miimon);
4691 } else if (arp_interval) {
4692 int i;
4693
4694 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4695 arp_interval,
4696 arp_validate_tbl[arp_validate_value].modename,
4697 arp_ip_count);
4698
4699 for (i = 0; i < arp_ip_count; i++)
4700 pr_info(" %s", arp_ip_target[i]);
4701
4702 pr_info("\n");
4703
4704 } else if (max_bonds) {
4705 /* miimon and arp_interval not set, we need one so things
4706 * work as expected, see bonding.txt for details
4707 */
4708 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");
4709 }
4710
4711 if (primary && !USES_PRIMARY(bond_mode)) {
4712 /* currently, using a primary only makes sense
4713 * in active backup, TLB or ALB modes
4714 */
4715 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4716 primary, bond_mode_name(bond_mode));
4717 primary = NULL;
4718 }
4719
4720 if (primary && primary_reselect) {
4721 primary_reselect_value = bond_parse_parm(primary_reselect,
4722 pri_reselect_tbl);
4723 if (primary_reselect_value == -1) {
4724 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4725 primary_reselect ==
4726 NULL ? "NULL" : primary_reselect);
4727 return -EINVAL;
4728 }
4729 } else {
4730 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4731 }
4732
4733 if (fail_over_mac) {
4734 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4735 fail_over_mac_tbl);
4736 if (fail_over_mac_value == -1) {
4737 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4738 arp_validate == NULL ? "NULL" : arp_validate);
4739 return -EINVAL;
4740 }
4741
4742 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4743 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4744 } else {
4745 fail_over_mac_value = BOND_FOM_NONE;
4746 }
4747
4748 /* fill params struct with the proper values */
4749 params->mode = bond_mode;
4750 params->xmit_policy = xmit_hashtype;
4751 params->miimon = miimon;
4752 params->num_peer_notif = num_peer_notif;
4753 params->arp_interval = arp_interval;
4754 params->arp_validate = arp_validate_value;
4755 params->updelay = updelay;
4756 params->downdelay = downdelay;
4757 params->use_carrier = use_carrier;
4758 params->lacp_fast = lacp_fast;
4759 params->primary[0] = 0;
4760 params->primary_reselect = primary_reselect_value;
4761 params->fail_over_mac = fail_over_mac_value;
4762 params->tx_queues = tx_queues;
4763 params->all_slaves_active = all_slaves_active;
4764 params->resend_igmp = resend_igmp;
4765 params->min_links = min_links;
4766
4767 if (primary) {
4768 strncpy(params->primary, primary, IFNAMSIZ);
4769 params->primary[IFNAMSIZ - 1] = 0;
4770 }
4771
4772 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4773
4774 return 0;
4775}
4776
4777static struct lock_class_key bonding_netdev_xmit_lock_key;
4778static struct lock_class_key bonding_netdev_addr_lock_key;
4779
4780static void bond_set_lockdep_class_one(struct net_device *dev,
4781 struct netdev_queue *txq,
4782 void *_unused)
4783{
4784 lockdep_set_class(&txq->_xmit_lock,
4785 &bonding_netdev_xmit_lock_key);
4786}
4787
4788static void bond_set_lockdep_class(struct net_device *dev)
4789{
4790 lockdep_set_class(&dev->addr_list_lock,
4791 &bonding_netdev_addr_lock_key);
4792 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4793}
4794
4795/*
4796 * Called from registration process
4797 */
4798static int bond_init(struct net_device *bond_dev)
4799{
4800 struct bonding *bond = netdev_priv(bond_dev);
4801 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4802 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4803
4804 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4805
4806 /*
4807 * Initialize locks that may be required during
4808 * en/deslave operations. All of the bond_open work
4809 * (of which this is part) should really be moved to
4810 * a phase prior to dev_open
4811 */
4812 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4813 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4814
4815 bond->wq = create_singlethread_workqueue(bond_dev->name);
4816 if (!bond->wq)
4817 return -ENOMEM;
4818
4819 bond_set_lockdep_class(bond_dev);
4820
4821 list_add_tail(&bond->bond_list, &bn->dev_list);
4822
4823 bond_prepare_sysfs_group(bond);
4824
4825 bond_debug_register(bond);
4826
4827 __hw_addr_init(&bond->mc_list);
4828 return 0;
4829}
4830
4831static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4832{
4833 if (tb[IFLA_ADDRESS]) {
4834 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4835 return -EINVAL;
4836 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4837 return -EADDRNOTAVAIL;
4838 }
4839 return 0;
4840}
4841
4842static int bond_get_tx_queues(struct net *net, struct nlattr *tb[])
4843{
4844 return tx_queues;
4845}
4846
4847static struct rtnl_link_ops bond_link_ops __read_mostly = {
4848 .kind = "bond",
4849 .priv_size = sizeof(struct bonding),
4850 .setup = bond_setup,
4851 .validate = bond_validate,
4852 .get_tx_queues = bond_get_tx_queues,
4853};
4854
4855/* Create a new bond based on the specified name and bonding parameters.
4856 * If name is NULL, obtain a suitable "bond%d" name for us.
4857 * Caller must NOT hold rtnl_lock; we need to release it here before we
4858 * set up our sysfs entries.
4859 */
4860int bond_create(struct net *net, const char *name)
4861{
4862 struct net_device *bond_dev;
4863 int res;
4864
4865 rtnl_lock();
4866
4867 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4868 name ? name : "bond%d",
4869 bond_setup, tx_queues);
4870 if (!bond_dev) {
4871 pr_err("%s: eek! can't alloc netdev!\n", name);
4872 rtnl_unlock();
4873 return -ENOMEM;
4874 }
4875
4876 dev_net_set(bond_dev, net);
4877 bond_dev->rtnl_link_ops = &bond_link_ops;
4878
4879 res = register_netdevice(bond_dev);
4880
4881 netif_carrier_off(bond_dev);
4882
4883 rtnl_unlock();
4884 if (res < 0)
4885 bond_destructor(bond_dev);
4886 return res;
4887}
4888
4889static int __net_init bond_net_init(struct net *net)
4890{
4891 struct bond_net *bn = net_generic(net, bond_net_id);
4892
4893 bn->net = net;
4894 INIT_LIST_HEAD(&bn->dev_list);
4895
4896 bond_create_proc_dir(bn);
4897 bond_create_sysfs(bn);
4898
4899 return 0;
4900}
4901
4902static void __net_exit bond_net_exit(struct net *net)
4903{
4904 struct bond_net *bn = net_generic(net, bond_net_id);
4905
4906 bond_destroy_sysfs(bn);
4907 bond_destroy_proc_dir(bn);
4908}
4909
4910static struct pernet_operations bond_net_ops = {
4911 .init = bond_net_init,
4912 .exit = bond_net_exit,
4913 .id = &bond_net_id,
4914 .size = sizeof(struct bond_net),
4915};
4916
4917static int __init bonding_init(void)
4918{
4919 int i;
4920 int res;
4921
4922 pr_info("%s", bond_version);
4923
4924 res = bond_check_params(&bonding_defaults);
4925 if (res)
4926 goto out;
4927
4928 res = register_pernet_subsys(&bond_net_ops);
4929 if (res)
4930 goto out;
4931
4932 res = rtnl_link_register(&bond_link_ops);
4933 if (res)
4934 goto err_link;
4935
4936 bond_create_debugfs();
4937
4938 for (i = 0; i < max_bonds; i++) {
4939 res = bond_create(&init_net, NULL);
4940 if (res)
4941 goto err;
4942 }
4943
4944 register_netdevice_notifier(&bond_netdev_notifier);
4945out:
4946 return res;
4947err:
4948 rtnl_link_unregister(&bond_link_ops);
4949err_link:
4950 unregister_pernet_subsys(&bond_net_ops);
4951 goto out;
4952
4953}
4954
4955static void __exit bonding_exit(void)
4956{
4957 unregister_netdevice_notifier(&bond_netdev_notifier);
4958
4959 bond_destroy_debugfs();
4960
4961 rtnl_link_unregister(&bond_link_ops);
4962 unregister_pernet_subsys(&bond_net_ops);
4963
4964#ifdef CONFIG_NET_POLL_CONTROLLER
4965 /*
4966 * Make sure we don't have an imbalance on our netpoll blocking
4967 */
4968 WARN_ON(atomic_read(&netpoll_block_tx));
4969#endif
4970}
4971
4972module_init(bonding_init);
4973module_exit(bonding_exit);
4974MODULE_LICENSE("GPL");
4975MODULE_VERSION(DRV_VERSION);
4976MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4977MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4978MODULE_ALIAS_RTNL_LINK("bond");