Loading...
1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
32 */
33
34#include <linux/kernel.h>
35#include <linux/module.h>
36#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
42#include <net/ip.h>
43#include <linux/ip.h>
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
54#include <linux/io.h>
55#include <asm/dma.h>
56#include <linux/uaccess.h>
57#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
60#include <linux/igmp.h>
61#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
65#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
72#include <linux/jiffies.h>
73#include <linux/preempt.h>
74#include <net/route.h>
75#include <net/net_namespace.h>
76#include <net/netns/generic.h>
77#include <net/pkt_sched.h>
78#include <linux/rculist.h>
79#include <net/flow_dissector.h>
80#include <net/switchdev.h>
81#include <net/bonding.h>
82#include <net/bond_3ad.h>
83#include <net/bond_alb.h>
84
85#include "bonding_priv.h"
86
87/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
90
91static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93static int num_peer_notif = 1;
94static int miimon;
95static int updelay;
96static int downdelay;
97static int use_carrier = 1;
98static char *mode;
99static char *primary;
100static char *primary_reselect;
101static char *lacp_rate;
102static int min_links;
103static char *ad_select;
104static char *xmit_hash_policy;
105static int arp_interval;
106static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107static char *arp_validate;
108static char *arp_all_targets;
109static char *fail_over_mac;
110static int all_slaves_active;
111static struct bond_params bonding_defaults;
112static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113static int packets_per_slave = 1;
114static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115
116module_param(max_bonds, int, 0);
117MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118module_param(tx_queues, int, 0);
119MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126module_param(miimon, int, 0);
127MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128module_param(updelay, int, 0);
129MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130module_param(downdelay, int, 0);
131MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
133module_param(use_carrier, int, 0);
134MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136module_param(mode, charp, 0);
137MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141module_param(primary, charp, 0);
142MODULE_PARM_DESC(primary, "Primary network device to use");
143module_param(primary_reselect, charp, 0);
144MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "once it comes up; "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
148 "better, "
149 "2 for only on active slave "
150 "failure");
151module_param(lacp_rate, charp, 0);
152MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154module_param(ad_select, charp, 0);
155MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
157 "2 for count");
158module_param(min_links, int, 0);
159MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160
161module_param(xmit_hash_policy, charp, 0);
162MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166module_param(arp_interval, int, 0);
167MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168module_param_array(arp_ip_target, charp, NULL, 0);
169MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170module_param(arp_validate, charp, 0);
171MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174module_param(arp_all_targets, charp, 0);
175MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176module_param(fail_over_mac, charp, 0);
177MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180module_param(all_slaves_active, int, 0);
181MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184module_param(resend_igmp, int, 0);
185MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
187module_param(packets_per_slave, int, 0);
188MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191module_param(lp_interval, uint, 0);
192MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
195
196/*----------------------------- Global variables ----------------------------*/
197
198#ifdef CONFIG_NET_POLL_CONTROLLER
199atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200#endif
201
202unsigned int bond_net_id __read_mostly;
203
204/*-------------------------- Forward declarations ---------------------------*/
205
206static int bond_init(struct net_device *bond_dev);
207static void bond_uninit(struct net_device *bond_dev);
208static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210static void bond_slave_arr_handler(struct work_struct *work);
211static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 int mod);
213
214/*---------------------------- General routines -----------------------------*/
215
216const char *bond_mode_name(int mode)
217{
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
226 };
227
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
229 return "unknown";
230
231 return names[mode];
232}
233
234/*---------------------------------- VLAN -----------------------------------*/
235
236/**
237 * bond_dev_queue_xmit - Prepare skb for xmit.
238 *
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
242 */
243void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
245{
246 skb->dev = slave_dev;
247
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
251
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
254 else
255 dev_queue_xmit(skb);
256}
257
258/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259 * We don't protect the slave list iteration with a lock because:
260 * a. This operation is performed in IOCTL context,
261 * b. The operation is protected by the RTNL semaphore in the 8021q code,
262 * c. Holding a lock with BH disabled while directly calling a base driver
263 * entry point is generally a BAD idea.
264 *
265 * The design of synchronization/protection for this operation in the 8021q
266 * module is good for one or more VLAN devices over a single physical device
267 * and cannot be extended for a teaming solution like bonding, so there is a
268 * potential race condition here where a net device from the vlan group might
269 * be referenced (either by a base driver or the 8021q code) while it is being
270 * removed from the system. However, it turns out we're not making matters
271 * worse, and if it works for regular VLAN usage it will work here too.
272*/
273
274/**
275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276 * @bond_dev: bonding net device that got called
277 * @vid: vlan id being added
278 */
279static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 __be16 proto, u16 vid)
281{
282 struct bonding *bond = netdev_priv(bond_dev);
283 struct slave *slave, *rollback_slave;
284 struct list_head *iter;
285 int res;
286
287 bond_for_each_slave(bond, slave, iter) {
288 res = vlan_vid_add(slave->dev, proto, vid);
289 if (res)
290 goto unwind;
291 }
292
293 return 0;
294
295unwind:
296 /* unwind to the slave that failed */
297 bond_for_each_slave(bond, rollback_slave, iter) {
298 if (rollback_slave == slave)
299 break;
300
301 vlan_vid_del(rollback_slave->dev, proto, vid);
302 }
303
304 return res;
305}
306
307/**
308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309 * @bond_dev: bonding net device that got called
310 * @vid: vlan id being removed
311 */
312static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 __be16 proto, u16 vid)
314{
315 struct bonding *bond = netdev_priv(bond_dev);
316 struct list_head *iter;
317 struct slave *slave;
318
319 bond_for_each_slave(bond, slave, iter)
320 vlan_vid_del(slave->dev, proto, vid);
321
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vid);
324
325 return 0;
326}
327
328/*------------------------------- Link status -------------------------------*/
329
330/* Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
333 *
334 * Returns zero if carrier state does not change, nonzero if it does.
335 */
336int bond_set_carrier(struct bonding *bond)
337{
338 struct list_head *iter;
339 struct slave *slave;
340
341 if (!bond_has_slaves(bond))
342 goto down;
343
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
346
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
351 return 1;
352 }
353 return 0;
354 }
355 }
356
357down:
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
360 return 1;
361 }
362 return 0;
363}
364
365/* Get link speed and duplex from the slave's base driver
366 * using ethtool. If for some reason the call fails or the
367 * values are invalid, set speed and duplex to -1,
368 * and return. Return 1 if speed or duplex settings are
369 * UNKNOWN; 0 otherwise.
370 */
371static int bond_update_speed_duplex(struct slave *slave)
372{
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_link_ksettings ecmd;
375 int res;
376
377 slave->speed = SPEED_UNKNOWN;
378 slave->duplex = DUPLEX_UNKNOWN;
379
380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
381 if (res < 0)
382 return 1;
383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
384 return 1;
385 switch (ecmd.base.duplex) {
386 case DUPLEX_FULL:
387 case DUPLEX_HALF:
388 break;
389 default:
390 return 1;
391 }
392
393 slave->speed = ecmd.base.speed;
394 slave->duplex = ecmd.base.duplex;
395
396 return 0;
397}
398
399const char *bond_slave_link_status(s8 link)
400{
401 switch (link) {
402 case BOND_LINK_UP:
403 return "up";
404 case BOND_LINK_FAIL:
405 return "going down";
406 case BOND_LINK_DOWN:
407 return "down";
408 case BOND_LINK_BACK:
409 return "going back";
410 default:
411 return "unknown";
412 }
413}
414
415/* if <dev> supports MII link status reporting, check its link status.
416 *
417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418 * depending upon the setting of the use_carrier parameter.
419 *
420 * Return either BMSR_LSTATUS, meaning that the link is up (or we
421 * can't tell and just pretend it is), or 0, meaning that the link is
422 * down.
423 *
424 * If reporting is non-zero, instead of faking link up, return -1 if
425 * both ETHTOOL and MII ioctls fail (meaning the device does not
426 * support them). If use_carrier is set, return whatever it says.
427 * It'd be nice if there was a good way to tell if a driver supports
428 * netif_carrier, but there really isn't.
429 */
430static int bond_check_dev_link(struct bonding *bond,
431 struct net_device *slave_dev, int reporting)
432{
433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 int (*ioctl)(struct net_device *, struct ifreq *, int);
435 struct ifreq ifr;
436 struct mii_ioctl_data *mii;
437
438 if (!reporting && !netif_running(slave_dev))
439 return 0;
440
441 if (bond->params.use_carrier)
442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
443
444 /* Try to get link status using Ethtool first. */
445 if (slave_dev->ethtool_ops->get_link)
446 return slave_dev->ethtool_ops->get_link(slave_dev) ?
447 BMSR_LSTATUS : 0;
448
449 /* Ethtool can't be used, fallback to MII ioctls. */
450 ioctl = slave_ops->ndo_do_ioctl;
451 if (ioctl) {
452 /* TODO: set pointer to correct ioctl on a per team member
453 * bases to make this more efficient. that is, once
454 * we determine the correct ioctl, we will always
455 * call it and not the others for that team
456 * member.
457 */
458
459 /* We cannot assume that SIOCGMIIPHY will also read a
460 * register; not all network drivers (e.g., e100)
461 * support that.
462 */
463
464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
466 mii = if_mii(&ifr);
467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 mii->reg_num = MII_BMSR;
469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 return mii->val_out & BMSR_LSTATUS;
471 }
472 }
473
474 /* If reporting, report that either there's no dev->do_ioctl,
475 * or both SIOCGMIIREG and get_link failed (meaning that we
476 * cannot report link status). If not reporting, pretend
477 * we're ok.
478 */
479 return reporting ? -1 : BMSR_LSTATUS;
480}
481
482/*----------------------------- Multicast list ------------------------------*/
483
484/* Push the promiscuity flag down to appropriate slaves */
485static int bond_set_promiscuity(struct bonding *bond, int inc)
486{
487 struct list_head *iter;
488 int err = 0;
489
490 if (bond_uses_primary(bond)) {
491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
492
493 if (curr_active)
494 err = dev_set_promiscuity(curr_active->dev, inc);
495 } else {
496 struct slave *slave;
497
498 bond_for_each_slave(bond, slave, iter) {
499 err = dev_set_promiscuity(slave->dev, inc);
500 if (err)
501 return err;
502 }
503 }
504 return err;
505}
506
507/* Push the allmulti flag down to all slaves */
508static int bond_set_allmulti(struct bonding *bond, int inc)
509{
510 struct list_head *iter;
511 int err = 0;
512
513 if (bond_uses_primary(bond)) {
514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
515
516 if (curr_active)
517 err = dev_set_allmulti(curr_active->dev, inc);
518 } else {
519 struct slave *slave;
520
521 bond_for_each_slave(bond, slave, iter) {
522 err = dev_set_allmulti(slave->dev, inc);
523 if (err)
524 return err;
525 }
526 }
527 return err;
528}
529
530/* Retrieve the list of registered multicast addresses for the bonding
531 * device and retransmit an IGMP JOIN request to the current active
532 * slave.
533 */
534static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
535{
536 struct bonding *bond = container_of(work, struct bonding,
537 mcast_work.work);
538
539 if (!rtnl_trylock()) {
540 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
541 return;
542 }
543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
544
545 if (bond->igmp_retrans > 1) {
546 bond->igmp_retrans--;
547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
548 }
549 rtnl_unlock();
550}
551
552/* Flush bond's hardware addresses from slave */
553static void bond_hw_addr_flush(struct net_device *bond_dev,
554 struct net_device *slave_dev)
555{
556 struct bonding *bond = netdev_priv(bond_dev);
557
558 dev_uc_unsync(slave_dev, bond_dev);
559 dev_mc_unsync(slave_dev, bond_dev);
560
561 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 /* del lacpdu mc addr from mc list */
563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
564
565 dev_mc_del(slave_dev, lacpdu_multicast);
566 }
567}
568
569/*--------------------------- Active slave change ---------------------------*/
570
571/* Update the hardware address list and promisc/allmulti for the new and
572 * old active slaves (if any). Modes that are not using primary keep all
573 * slaves up date at all times; only the modes that use primary need to call
574 * this function to swap these settings during a failover.
575 */
576static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 struct slave *old_active)
578{
579 if (old_active) {
580 if (bond->dev->flags & IFF_PROMISC)
581 dev_set_promiscuity(old_active->dev, -1);
582
583 if (bond->dev->flags & IFF_ALLMULTI)
584 dev_set_allmulti(old_active->dev, -1);
585
586 bond_hw_addr_flush(bond->dev, old_active->dev);
587 }
588
589 if (new_active) {
590 /* FIXME: Signal errors upstream. */
591 if (bond->dev->flags & IFF_PROMISC)
592 dev_set_promiscuity(new_active->dev, 1);
593
594 if (bond->dev->flags & IFF_ALLMULTI)
595 dev_set_allmulti(new_active->dev, 1);
596
597 netif_addr_lock_bh(bond->dev);
598 dev_uc_sync(new_active->dev, bond->dev);
599 dev_mc_sync(new_active->dev, bond->dev);
600 netif_addr_unlock_bh(bond->dev);
601 }
602}
603
604/**
605 * bond_set_dev_addr - clone slave's address to bond
606 * @bond_dev: bond net device
607 * @slave_dev: slave net device
608 *
609 * Should be called with RTNL held.
610 */
611static void bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
613{
614 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
615 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
616 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
617 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
618 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
619}
620
621static struct slave *bond_get_old_active(struct bonding *bond,
622 struct slave *new_active)
623{
624 struct slave *slave;
625 struct list_head *iter;
626
627 bond_for_each_slave(bond, slave, iter) {
628 if (slave == new_active)
629 continue;
630
631 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
632 return slave;
633 }
634
635 return NULL;
636}
637
638/* bond_do_fail_over_mac
639 *
640 * Perform special MAC address swapping for fail_over_mac settings
641 *
642 * Called with RTNL
643 */
644static void bond_do_fail_over_mac(struct bonding *bond,
645 struct slave *new_active,
646 struct slave *old_active)
647{
648 u8 tmp_mac[MAX_ADDR_LEN];
649 struct sockaddr_storage ss;
650 int rv;
651
652 switch (bond->params.fail_over_mac) {
653 case BOND_FOM_ACTIVE:
654 if (new_active)
655 bond_set_dev_addr(bond->dev, new_active->dev);
656 break;
657 case BOND_FOM_FOLLOW:
658 /* if new_active && old_active, swap them
659 * if just old_active, do nothing (going to no active slave)
660 * if just new_active, set new_active to bond's MAC
661 */
662 if (!new_active)
663 return;
664
665 if (!old_active)
666 old_active = bond_get_old_active(bond, new_active);
667
668 if (old_active) {
669 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
670 new_active->dev->addr_len);
671 bond_hw_addr_copy(ss.__data,
672 old_active->dev->dev_addr,
673 old_active->dev->addr_len);
674 ss.ss_family = new_active->dev->type;
675 } else {
676 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
677 bond->dev->addr_len);
678 ss.ss_family = bond->dev->type;
679 }
680
681 rv = dev_set_mac_address(new_active->dev,
682 (struct sockaddr *)&ss);
683 if (rv) {
684 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
685 -rv, new_active->dev->name);
686 goto out;
687 }
688
689 if (!old_active)
690 goto out;
691
692 bond_hw_addr_copy(ss.__data, tmp_mac,
693 new_active->dev->addr_len);
694 ss.ss_family = old_active->dev->type;
695
696 rv = dev_set_mac_address(old_active->dev,
697 (struct sockaddr *)&ss);
698 if (rv)
699 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
700 -rv, new_active->dev->name);
701out:
702 break;
703 default:
704 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
705 bond->params.fail_over_mac);
706 break;
707 }
708
709}
710
711static struct slave *bond_choose_primary_or_current(struct bonding *bond)
712{
713 struct slave *prim = rtnl_dereference(bond->primary_slave);
714 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
715
716 if (!prim || prim->link != BOND_LINK_UP) {
717 if (!curr || curr->link != BOND_LINK_UP)
718 return NULL;
719 return curr;
720 }
721
722 if (bond->force_primary) {
723 bond->force_primary = false;
724 return prim;
725 }
726
727 if (!curr || curr->link != BOND_LINK_UP)
728 return prim;
729
730 /* At this point, prim and curr are both up */
731 switch (bond->params.primary_reselect) {
732 case BOND_PRI_RESELECT_ALWAYS:
733 return prim;
734 case BOND_PRI_RESELECT_BETTER:
735 if (prim->speed < curr->speed)
736 return curr;
737 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
738 return curr;
739 return prim;
740 case BOND_PRI_RESELECT_FAILURE:
741 return curr;
742 default:
743 netdev_err(bond->dev, "impossible primary_reselect %d\n",
744 bond->params.primary_reselect);
745 return curr;
746 }
747}
748
749/**
750 * bond_find_best_slave - select the best available slave to be the active one
751 * @bond: our bonding struct
752 */
753static struct slave *bond_find_best_slave(struct bonding *bond)
754{
755 struct slave *slave, *bestslave = NULL;
756 struct list_head *iter;
757 int mintime = bond->params.updelay;
758
759 slave = bond_choose_primary_or_current(bond);
760 if (slave)
761 return slave;
762
763 bond_for_each_slave(bond, slave, iter) {
764 if (slave->link == BOND_LINK_UP)
765 return slave;
766 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
767 slave->delay < mintime) {
768 mintime = slave->delay;
769 bestslave = slave;
770 }
771 }
772
773 return bestslave;
774}
775
776static bool bond_should_notify_peers(struct bonding *bond)
777{
778 struct slave *slave;
779
780 rcu_read_lock();
781 slave = rcu_dereference(bond->curr_active_slave);
782 rcu_read_unlock();
783
784 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
785 slave ? slave->dev->name : "NULL");
786
787 if (!slave || !bond->send_peer_notif ||
788 !netif_carrier_ok(bond->dev) ||
789 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
790 return false;
791
792 return true;
793}
794
795/**
796 * change_active_interface - change the active slave into the specified one
797 * @bond: our bonding struct
798 * @new: the new slave to make the active one
799 *
800 * Set the new slave to the bond's settings and unset them on the old
801 * curr_active_slave.
802 * Setting include flags, mc-list, promiscuity, allmulti, etc.
803 *
804 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
805 * because it is apparently the best available slave we have, even though its
806 * updelay hasn't timed out yet.
807 *
808 * Caller must hold RTNL.
809 */
810void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
811{
812 struct slave *old_active;
813
814 ASSERT_RTNL();
815
816 old_active = rtnl_dereference(bond->curr_active_slave);
817
818 if (old_active == new_active)
819 return;
820
821 if (new_active) {
822 new_active->last_link_up = jiffies;
823
824 if (new_active->link == BOND_LINK_BACK) {
825 if (bond_uses_primary(bond)) {
826 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
827 new_active->dev->name,
828 (bond->params.updelay - new_active->delay) * bond->params.miimon);
829 }
830
831 new_active->delay = 0;
832 bond_set_slave_link_state(new_active, BOND_LINK_UP,
833 BOND_SLAVE_NOTIFY_NOW);
834
835 if (BOND_MODE(bond) == BOND_MODE_8023AD)
836 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
837
838 if (bond_is_lb(bond))
839 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
840 } else {
841 if (bond_uses_primary(bond)) {
842 netdev_info(bond->dev, "making interface %s the new active one\n",
843 new_active->dev->name);
844 }
845 }
846 }
847
848 if (bond_uses_primary(bond))
849 bond_hw_addr_swap(bond, new_active, old_active);
850
851 if (bond_is_lb(bond)) {
852 bond_alb_handle_active_change(bond, new_active);
853 if (old_active)
854 bond_set_slave_inactive_flags(old_active,
855 BOND_SLAVE_NOTIFY_NOW);
856 if (new_active)
857 bond_set_slave_active_flags(new_active,
858 BOND_SLAVE_NOTIFY_NOW);
859 } else {
860 rcu_assign_pointer(bond->curr_active_slave, new_active);
861 }
862
863 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
864 if (old_active)
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
867
868 if (new_active) {
869 bool should_notify_peers = false;
870
871 bond_set_slave_active_flags(new_active,
872 BOND_SLAVE_NOTIFY_NOW);
873
874 if (bond->params.fail_over_mac)
875 bond_do_fail_over_mac(bond, new_active,
876 old_active);
877
878 if (netif_running(bond->dev)) {
879 bond->send_peer_notif =
880 bond->params.num_peer_notif;
881 should_notify_peers =
882 bond_should_notify_peers(bond);
883 }
884
885 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
886 if (should_notify_peers)
887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
888 bond->dev);
889 }
890 }
891
892 /* resend IGMP joins since active slave has changed or
893 * all were sent on curr_active_slave.
894 * resend only if bond is brought up with the affected
895 * bonding modes and the retransmission is enabled
896 */
897 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
898 ((bond_uses_primary(bond) && new_active) ||
899 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
900 bond->igmp_retrans = bond->params.resend_igmp;
901 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
902 }
903}
904
905/**
906 * bond_select_active_slave - select a new active slave, if needed
907 * @bond: our bonding struct
908 *
909 * This functions should be called when one of the following occurs:
910 * - The old curr_active_slave has been released or lost its link.
911 * - The primary_slave has got its link back.
912 * - A slave has got its link back and there's no old curr_active_slave.
913 *
914 * Caller must hold RTNL.
915 */
916void bond_select_active_slave(struct bonding *bond)
917{
918 struct slave *best_slave;
919 int rv;
920
921 ASSERT_RTNL();
922
923 best_slave = bond_find_best_slave(bond);
924 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
925 bond_change_active_slave(bond, best_slave);
926 rv = bond_set_carrier(bond);
927 if (!rv)
928 return;
929
930 if (netif_carrier_ok(bond->dev))
931 netdev_info(bond->dev, "first active interface up!\n");
932 else
933 netdev_info(bond->dev, "now running without any active interface!\n");
934 }
935}
936
937#ifdef CONFIG_NET_POLL_CONTROLLER
938static inline int slave_enable_netpoll(struct slave *slave)
939{
940 struct netpoll *np;
941 int err = 0;
942
943 np = kzalloc(sizeof(*np), GFP_KERNEL);
944 err = -ENOMEM;
945 if (!np)
946 goto out;
947
948 err = __netpoll_setup(np, slave->dev);
949 if (err) {
950 kfree(np);
951 goto out;
952 }
953 slave->np = np;
954out:
955 return err;
956}
957static inline void slave_disable_netpoll(struct slave *slave)
958{
959 struct netpoll *np = slave->np;
960
961 if (!np)
962 return;
963
964 slave->np = NULL;
965 __netpoll_free_async(np);
966}
967
968static void bond_poll_controller(struct net_device *bond_dev)
969{
970 struct bonding *bond = netdev_priv(bond_dev);
971 struct slave *slave = NULL;
972 struct list_head *iter;
973 struct ad_info ad_info;
974 struct netpoll_info *ni;
975 const struct net_device_ops *ops;
976
977 if (BOND_MODE(bond) == BOND_MODE_8023AD)
978 if (bond_3ad_get_active_agg_info(bond, &ad_info))
979 return;
980
981 bond_for_each_slave_rcu(bond, slave, iter) {
982 ops = slave->dev->netdev_ops;
983 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
984 continue;
985
986 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
987 struct aggregator *agg =
988 SLAVE_AD_INFO(slave)->port.aggregator;
989
990 if (agg &&
991 agg->aggregator_identifier != ad_info.aggregator_id)
992 continue;
993 }
994
995 ni = rcu_dereference_bh(slave->dev->npinfo);
996 if (down_trylock(&ni->dev_lock))
997 continue;
998 ops->ndo_poll_controller(slave->dev);
999 up(&ni->dev_lock);
1000 }
1001}
1002
1003static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004{
1005 struct bonding *bond = netdev_priv(bond_dev);
1006 struct list_head *iter;
1007 struct slave *slave;
1008
1009 bond_for_each_slave(bond, slave, iter)
1010 if (bond_slave_is_up(slave))
1011 slave_disable_netpoll(slave);
1012}
1013
1014static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1015{
1016 struct bonding *bond = netdev_priv(dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1019 int err = 0;
1020
1021 bond_for_each_slave(bond, slave, iter) {
1022 err = slave_enable_netpoll(slave);
1023 if (err) {
1024 bond_netpoll_cleanup(dev);
1025 break;
1026 }
1027 }
1028 return err;
1029}
1030#else
1031static inline int slave_enable_netpoll(struct slave *slave)
1032{
1033 return 0;
1034}
1035static inline void slave_disable_netpoll(struct slave *slave)
1036{
1037}
1038static void bond_netpoll_cleanup(struct net_device *bond_dev)
1039{
1040}
1041#endif
1042
1043/*---------------------------------- IOCTL ----------------------------------*/
1044
1045static netdev_features_t bond_fix_features(struct net_device *dev,
1046 netdev_features_t features)
1047{
1048 struct bonding *bond = netdev_priv(dev);
1049 struct list_head *iter;
1050 netdev_features_t mask;
1051 struct slave *slave;
1052
1053 mask = features;
1054
1055 features &= ~NETIF_F_ONE_FOR_ALL;
1056 features |= NETIF_F_ALL_FOR_ALL;
1057
1058 bond_for_each_slave(bond, slave, iter) {
1059 features = netdev_increment_features(features,
1060 slave->dev->features,
1061 mask);
1062 }
1063 features = netdev_add_tso_features(features, mask);
1064
1065 return features;
1066}
1067
1068#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1069 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1070 NETIF_F_HIGHDMA | NETIF_F_LRO)
1071
1072#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1074
1075static void bond_compute_features(struct bonding *bond)
1076{
1077 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1078 IFF_XMIT_DST_RELEASE_PERM;
1079 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1080 netdev_features_t enc_features = BOND_ENC_FEATURES;
1081 struct net_device *bond_dev = bond->dev;
1082 struct list_head *iter;
1083 struct slave *slave;
1084 unsigned short max_hard_header_len = ETH_HLEN;
1085 unsigned int gso_max_size = GSO_MAX_SIZE;
1086 u16 gso_max_segs = GSO_MAX_SEGS;
1087
1088 if (!bond_has_slaves(bond))
1089 goto done;
1090 vlan_features &= NETIF_F_ALL_FOR_ALL;
1091
1092 bond_for_each_slave(bond, slave, iter) {
1093 vlan_features = netdev_increment_features(vlan_features,
1094 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1095
1096 enc_features = netdev_increment_features(enc_features,
1097 slave->dev->hw_enc_features,
1098 BOND_ENC_FEATURES);
1099 dst_release_flag &= slave->dev->priv_flags;
1100 if (slave->dev->hard_header_len > max_hard_header_len)
1101 max_hard_header_len = slave->dev->hard_header_len;
1102
1103 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1104 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1105 }
1106 bond_dev->hard_header_len = max_hard_header_len;
1107
1108done:
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 bond_dev->gso_max_segs = gso_max_segs;
1112 netif_set_gso_max_size(bond_dev, gso_max_size);
1113
1114 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1115 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1116 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1117 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1118
1119 netdev_change_features(bond_dev);
1120}
1121
1122static void bond_setup_by_slave(struct net_device *bond_dev,
1123 struct net_device *slave_dev)
1124{
1125 bond_dev->header_ops = slave_dev->header_ops;
1126
1127 bond_dev->type = slave_dev->type;
1128 bond_dev->hard_header_len = slave_dev->hard_header_len;
1129 bond_dev->addr_len = slave_dev->addr_len;
1130
1131 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1132 slave_dev->addr_len);
1133}
1134
1135/* On bonding slaves other than the currently active slave, suppress
1136 * duplicates except for alb non-mcast/bcast.
1137 */
1138static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1139 struct slave *slave,
1140 struct bonding *bond)
1141{
1142 if (bond_is_slave_inactive(slave)) {
1143 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1144 skb->pkt_type != PACKET_BROADCAST &&
1145 skb->pkt_type != PACKET_MULTICAST)
1146 return false;
1147 return true;
1148 }
1149 return false;
1150}
1151
1152static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1153{
1154 struct sk_buff *skb = *pskb;
1155 struct slave *slave;
1156 struct bonding *bond;
1157 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1158 struct slave *);
1159 int ret = RX_HANDLER_ANOTHER;
1160
1161 skb = skb_share_check(skb, GFP_ATOMIC);
1162 if (unlikely(!skb))
1163 return RX_HANDLER_CONSUMED;
1164
1165 *pskb = skb;
1166
1167 slave = bond_slave_get_rcu(skb->dev);
1168 bond = slave->bond;
1169
1170 recv_probe = READ_ONCE(bond->recv_probe);
1171 if (recv_probe) {
1172 ret = recv_probe(skb, bond, slave);
1173 if (ret == RX_HANDLER_CONSUMED) {
1174 consume_skb(skb);
1175 return ret;
1176 }
1177 }
1178
1179 /* don't change skb->dev for link-local packets */
1180 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1181 return RX_HANDLER_PASS;
1182 if (bond_should_deliver_exact_match(skb, slave, bond))
1183 return RX_HANDLER_EXACT;
1184
1185 skb->dev = bond->dev;
1186
1187 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189 skb->pkt_type == PACKET_HOST) {
1190
1191 if (unlikely(skb_cow_head(skb,
1192 skb->data - skb_mac_header(skb)))) {
1193 kfree_skb(skb);
1194 return RX_HANDLER_CONSUMED;
1195 }
1196 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1197 bond->dev->addr_len);
1198 }
1199
1200 return ret;
1201}
1202
1203static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1204{
1205 switch (BOND_MODE(bond)) {
1206 case BOND_MODE_ROUNDROBIN:
1207 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1208 case BOND_MODE_ACTIVEBACKUP:
1209 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1210 case BOND_MODE_BROADCAST:
1211 return NETDEV_LAG_TX_TYPE_BROADCAST;
1212 case BOND_MODE_XOR:
1213 case BOND_MODE_8023AD:
1214 return NETDEV_LAG_TX_TYPE_HASH;
1215 default:
1216 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1217 }
1218}
1219
1220static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1221 struct netlink_ext_ack *extack)
1222{
1223 struct netdev_lag_upper_info lag_upper_info;
1224
1225 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1226
1227 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1228 &lag_upper_info, extack);
1229}
1230
1231static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232{
1233 netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 slave->dev->flags &= ~IFF_SLAVE;
1235}
1236
1237static struct slave *bond_alloc_slave(struct bonding *bond)
1238{
1239 struct slave *slave = NULL;
1240
1241 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1242 if (!slave)
1243 return NULL;
1244
1245 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1246 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1247 GFP_KERNEL);
1248 if (!SLAVE_AD_INFO(slave)) {
1249 kfree(slave);
1250 return NULL;
1251 }
1252 }
1253 return slave;
1254}
1255
1256static void bond_free_slave(struct slave *slave)
1257{
1258 struct bonding *bond = bond_get_bond_by_slave(slave);
1259
1260 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1261 kfree(SLAVE_AD_INFO(slave));
1262
1263 kfree(slave);
1264}
1265
1266static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1267{
1268 info->bond_mode = BOND_MODE(bond);
1269 info->miimon = bond->params.miimon;
1270 info->num_slaves = bond->slave_cnt;
1271}
1272
1273static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1274{
1275 strcpy(info->slave_name, slave->dev->name);
1276 info->link = slave->link;
1277 info->state = bond_slave_state(slave);
1278 info->link_failure_count = slave->link_failure_count;
1279}
1280
1281static void bond_netdev_notify(struct net_device *dev,
1282 struct netdev_bonding_info *info)
1283{
1284 rtnl_lock();
1285 netdev_bonding_info_change(dev, info);
1286 rtnl_unlock();
1287}
1288
1289static void bond_netdev_notify_work(struct work_struct *_work)
1290{
1291 struct netdev_notify_work *w =
1292 container_of(_work, struct netdev_notify_work, work.work);
1293
1294 bond_netdev_notify(w->dev, &w->bonding_info);
1295 dev_put(w->dev);
1296 kfree(w);
1297}
1298
1299void bond_queue_slave_event(struct slave *slave)
1300{
1301 struct bonding *bond = slave->bond;
1302 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1303
1304 if (!nnw)
1305 return;
1306
1307 dev_hold(slave->dev);
1308 nnw->dev = slave->dev;
1309 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1310 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1311 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1312
1313 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1314}
1315
1316void bond_lower_state_changed(struct slave *slave)
1317{
1318 struct netdev_lag_lower_state_info info;
1319
1320 info.link_up = slave->link == BOND_LINK_UP ||
1321 slave->link == BOND_LINK_FAIL;
1322 info.tx_enabled = bond_is_active_slave(slave);
1323 netdev_lower_state_changed(slave->dev, &info);
1324}
1325
1326/* enslave device <slave> to bond device <master> */
1327int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1328 struct netlink_ext_ack *extack)
1329{
1330 struct bonding *bond = netdev_priv(bond_dev);
1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 struct slave *new_slave = NULL, *prev_slave;
1333 struct sockaddr_storage ss;
1334 int link_reporting;
1335 int res = 0, i;
1336
1337 if (!bond->params.use_carrier &&
1338 slave_dev->ethtool_ops->get_link == NULL &&
1339 slave_ops->ndo_do_ioctl == NULL) {
1340 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341 slave_dev->name);
1342 }
1343
1344 /* already in-use? */
1345 if (netdev_is_rx_handler_busy(slave_dev)) {
1346 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1347 netdev_err(bond_dev,
1348 "Error: Device is in use and cannot be enslaved\n");
1349 return -EBUSY;
1350 }
1351
1352 if (bond_dev == slave_dev) {
1353 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1354 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1355 return -EPERM;
1356 }
1357
1358 /* vlan challenged mutual exclusion */
1359 /* no need to lock since we're protected by rtnl_lock */
1360 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1361 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1362 slave_dev->name);
1363 if (vlan_uses_dev(bond_dev)) {
1364 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1365 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366 slave_dev->name, bond_dev->name);
1367 return -EPERM;
1368 } else {
1369 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370 slave_dev->name, slave_dev->name,
1371 bond_dev->name);
1372 }
1373 } else {
1374 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1375 slave_dev->name);
1376 }
1377
1378 /* Old ifenslave binaries are no longer supported. These can
1379 * be identified with moderate accuracy by the state of the slave:
1380 * the current ifenslave will set the interface down prior to
1381 * enslaving it; the old ifenslave will not.
1382 */
1383 if (slave_dev->flags & IFF_UP) {
1384 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1385 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1386 slave_dev->name);
1387 return -EPERM;
1388 }
1389
1390 /* set bonding device ether type by slave - bonding netdevices are
1391 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1392 * there is a need to override some of the type dependent attribs/funcs.
1393 *
1394 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1395 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396 */
1397 if (!bond_has_slaves(bond)) {
1398 if (bond_dev->type != slave_dev->type) {
1399 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1400 bond_dev->type, slave_dev->type);
1401
1402 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403 bond_dev);
1404 res = notifier_to_errno(res);
1405 if (res) {
1406 netdev_err(bond_dev, "refused to change device type\n");
1407 return -EBUSY;
1408 }
1409
1410 /* Flush unicast and multicast addresses */
1411 dev_uc_flush(bond_dev);
1412 dev_mc_flush(bond_dev);
1413
1414 if (slave_dev->type != ARPHRD_ETHER)
1415 bond_setup_by_slave(bond_dev, slave_dev);
1416 else {
1417 ether_setup(bond_dev);
1418 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 }
1420
1421 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1422 bond_dev);
1423 }
1424 } else if (bond_dev->type != slave_dev->type) {
1425 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1426 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1427 slave_dev->name, slave_dev->type, bond_dev->type);
1428 return -EINVAL;
1429 }
1430
1431 if (slave_dev->type == ARPHRD_INFINIBAND &&
1432 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1433 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1434 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1435 slave_dev->type);
1436 res = -EOPNOTSUPP;
1437 goto err_undo_flags;
1438 }
1439
1440 if (!slave_ops->ndo_set_mac_address ||
1441 slave_dev->type == ARPHRD_INFINIBAND) {
1442 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1443 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1444 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1445 if (!bond_has_slaves(bond)) {
1446 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1447 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1448 } else {
1449 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1450 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1451 res = -EOPNOTSUPP;
1452 goto err_undo_flags;
1453 }
1454 }
1455 }
1456
1457 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1458
1459 /* If this is the first slave, then we need to set the master's hardware
1460 * address to be the same as the slave's.
1461 */
1462 if (!bond_has_slaves(bond) &&
1463 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1464 bond_set_dev_addr(bond->dev, slave_dev);
1465
1466 new_slave = bond_alloc_slave(bond);
1467 if (!new_slave) {
1468 res = -ENOMEM;
1469 goto err_undo_flags;
1470 }
1471
1472 new_slave->bond = bond;
1473 new_slave->dev = slave_dev;
1474 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1475 * is set via sysfs or module option if desired.
1476 */
1477 new_slave->queue_id = 0;
1478
1479 /* Save slave's original mtu and then set it to match the bond */
1480 new_slave->original_mtu = slave_dev->mtu;
1481 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1482 if (res) {
1483 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1484 goto err_free;
1485 }
1486
1487 /* Save slave's original ("permanent") mac address for modes
1488 * that need it, and for restoring it upon release, and then
1489 * set it to the master's address
1490 */
1491 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1492 slave_dev->addr_len);
1493
1494 if (!bond->params.fail_over_mac ||
1495 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1496 /* Set slave to master's mac address. The application already
1497 * set the master's mac address to that of the first slave
1498 */
1499 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1500 ss.ss_family = slave_dev->type;
1501 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1502 if (res) {
1503 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1504 goto err_restore_mtu;
1505 }
1506 }
1507
1508 /* set slave flag before open to prevent IPv6 addrconf */
1509 slave_dev->flags |= IFF_SLAVE;
1510
1511 /* open the slave since the application closed it */
1512 res = dev_open(slave_dev);
1513 if (res) {
1514 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1515 goto err_restore_mac;
1516 }
1517
1518 slave_dev->priv_flags |= IFF_BONDING;
1519 /* initialize slave stats */
1520 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1521
1522 if (bond_is_lb(bond)) {
1523 /* bond_alb_init_slave() must be called before all other stages since
1524 * it might fail and we do not want to have to undo everything
1525 */
1526 res = bond_alb_init_slave(bond, new_slave);
1527 if (res)
1528 goto err_close;
1529 }
1530
1531 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1532 if (res) {
1533 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1534 slave_dev->name);
1535 goto err_close;
1536 }
1537
1538 prev_slave = bond_last_slave(bond);
1539
1540 new_slave->delay = 0;
1541 new_slave->link_failure_count = 0;
1542
1543 if (bond_update_speed_duplex(new_slave) &&
1544 bond_needs_speed_duplex(bond))
1545 new_slave->link = BOND_LINK_DOWN;
1546
1547 new_slave->last_rx = jiffies -
1548 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1549 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1550 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1551
1552 if (bond->params.miimon && !bond->params.use_carrier) {
1553 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1554
1555 if ((link_reporting == -1) && !bond->params.arp_interval) {
1556 /* miimon is set but a bonded network driver
1557 * does not support ETHTOOL/MII and
1558 * arp_interval is not set. Note: if
1559 * use_carrier is enabled, we will never go
1560 * here (because netif_carrier is always
1561 * supported); thus, we don't need to change
1562 * the messages for netif_carrier.
1563 */
1564 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1565 slave_dev->name);
1566 } else if (link_reporting == -1) {
1567 /* unable get link status using mii/ethtool */
1568 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1569 slave_dev->name);
1570 }
1571 }
1572
1573 /* check for initial state */
1574 new_slave->link = BOND_LINK_NOCHANGE;
1575 if (bond->params.miimon) {
1576 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1577 if (bond->params.updelay) {
1578 bond_set_slave_link_state(new_slave,
1579 BOND_LINK_BACK,
1580 BOND_SLAVE_NOTIFY_NOW);
1581 new_slave->delay = bond->params.updelay;
1582 } else {
1583 bond_set_slave_link_state(new_slave,
1584 BOND_LINK_UP,
1585 BOND_SLAVE_NOTIFY_NOW);
1586 }
1587 } else {
1588 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1589 BOND_SLAVE_NOTIFY_NOW);
1590 }
1591 } else if (bond->params.arp_interval) {
1592 bond_set_slave_link_state(new_slave,
1593 (netif_carrier_ok(slave_dev) ?
1594 BOND_LINK_UP : BOND_LINK_DOWN),
1595 BOND_SLAVE_NOTIFY_NOW);
1596 } else {
1597 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1598 BOND_SLAVE_NOTIFY_NOW);
1599 }
1600
1601 if (new_slave->link != BOND_LINK_DOWN)
1602 new_slave->last_link_up = jiffies;
1603 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1604 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1605 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1606
1607 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1608 /* if there is a primary slave, remember it */
1609 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1610 rcu_assign_pointer(bond->primary_slave, new_slave);
1611 bond->force_primary = true;
1612 }
1613 }
1614
1615 switch (BOND_MODE(bond)) {
1616 case BOND_MODE_ACTIVEBACKUP:
1617 bond_set_slave_inactive_flags(new_slave,
1618 BOND_SLAVE_NOTIFY_NOW);
1619 break;
1620 case BOND_MODE_8023AD:
1621 /* in 802.3ad mode, the internal mechanism
1622 * will activate the slaves in the selected
1623 * aggregator
1624 */
1625 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1626 /* if this is the first slave */
1627 if (!prev_slave) {
1628 SLAVE_AD_INFO(new_slave)->id = 1;
1629 /* Initialize AD with the number of times that the AD timer is called in 1 second
1630 * can be called only after the mac address of the bond is set
1631 */
1632 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1633 } else {
1634 SLAVE_AD_INFO(new_slave)->id =
1635 SLAVE_AD_INFO(prev_slave)->id + 1;
1636 }
1637
1638 bond_3ad_bind_slave(new_slave);
1639 break;
1640 case BOND_MODE_TLB:
1641 case BOND_MODE_ALB:
1642 bond_set_active_slave(new_slave);
1643 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1644 break;
1645 default:
1646 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1647
1648 /* always active in trunk mode */
1649 bond_set_active_slave(new_slave);
1650
1651 /* In trunking mode there is little meaning to curr_active_slave
1652 * anyway (it holds no special properties of the bond device),
1653 * so we can change it without calling change_active_interface()
1654 */
1655 if (!rcu_access_pointer(bond->curr_active_slave) &&
1656 new_slave->link == BOND_LINK_UP)
1657 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1658
1659 break;
1660 } /* switch(bond_mode) */
1661
1662#ifdef CONFIG_NET_POLL_CONTROLLER
1663 if (bond->dev->npinfo) {
1664 if (slave_enable_netpoll(new_slave)) {
1665 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1666 res = -EBUSY;
1667 goto err_detach;
1668 }
1669 }
1670#endif
1671
1672 if (!(bond_dev->features & NETIF_F_LRO))
1673 dev_disable_lro(slave_dev);
1674
1675 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1676 new_slave);
1677 if (res) {
1678 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1679 goto err_detach;
1680 }
1681
1682 res = bond_master_upper_dev_link(bond, new_slave, extack);
1683 if (res) {
1684 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1685 goto err_unregister;
1686 }
1687
1688 res = bond_sysfs_slave_add(new_slave);
1689 if (res) {
1690 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1691 goto err_upper_unlink;
1692 }
1693
1694 /* If the mode uses primary, then the following is handled by
1695 * bond_change_active_slave().
1696 */
1697 if (!bond_uses_primary(bond)) {
1698 /* set promiscuity level to new slave */
1699 if (bond_dev->flags & IFF_PROMISC) {
1700 res = dev_set_promiscuity(slave_dev, 1);
1701 if (res)
1702 goto err_sysfs_del;
1703 }
1704
1705 /* set allmulti level to new slave */
1706 if (bond_dev->flags & IFF_ALLMULTI) {
1707 res = dev_set_allmulti(slave_dev, 1);
1708 if (res) {
1709 if (bond_dev->flags & IFF_PROMISC)
1710 dev_set_promiscuity(slave_dev, -1);
1711 goto err_sysfs_del;
1712 }
1713 }
1714
1715 netif_addr_lock_bh(bond_dev);
1716 dev_mc_sync_multiple(slave_dev, bond_dev);
1717 dev_uc_sync_multiple(slave_dev, bond_dev);
1718 netif_addr_unlock_bh(bond_dev);
1719
1720 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1721 /* add lacpdu mc addr to mc list */
1722 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1723
1724 dev_mc_add(slave_dev, lacpdu_multicast);
1725 }
1726 }
1727
1728 bond->slave_cnt++;
1729 bond_compute_features(bond);
1730 bond_set_carrier(bond);
1731
1732 if (bond_uses_primary(bond)) {
1733 block_netpoll_tx();
1734 bond_select_active_slave(bond);
1735 unblock_netpoll_tx();
1736 }
1737
1738 if (bond_mode_uses_xmit_hash(bond))
1739 bond_update_slave_arr(bond, NULL);
1740
1741 bond->nest_level = dev_get_nest_level(bond_dev);
1742
1743 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1744 slave_dev->name,
1745 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1746 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1747
1748 /* enslave is successful */
1749 bond_queue_slave_event(new_slave);
1750 return 0;
1751
1752/* Undo stages on error */
1753err_sysfs_del:
1754 bond_sysfs_slave_del(new_slave);
1755
1756err_upper_unlink:
1757 bond_upper_dev_unlink(bond, new_slave);
1758
1759err_unregister:
1760 netdev_rx_handler_unregister(slave_dev);
1761
1762err_detach:
1763 vlan_vids_del_by_dev(slave_dev, bond_dev);
1764 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1765 RCU_INIT_POINTER(bond->primary_slave, NULL);
1766 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1767 block_netpoll_tx();
1768 bond_change_active_slave(bond, NULL);
1769 bond_select_active_slave(bond);
1770 unblock_netpoll_tx();
1771 }
1772 /* either primary_slave or curr_active_slave might've changed */
1773 synchronize_rcu();
1774 slave_disable_netpoll(new_slave);
1775
1776err_close:
1777 slave_dev->priv_flags &= ~IFF_BONDING;
1778 dev_close(slave_dev);
1779
1780err_restore_mac:
1781 slave_dev->flags &= ~IFF_SLAVE;
1782 if (!bond->params.fail_over_mac ||
1783 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1784 /* XXX TODO - fom follow mode needs to change master's
1785 * MAC if this slave's MAC is in use by the bond, or at
1786 * least print a warning.
1787 */
1788 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1789 new_slave->dev->addr_len);
1790 ss.ss_family = slave_dev->type;
1791 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1792 }
1793
1794err_restore_mtu:
1795 dev_set_mtu(slave_dev, new_slave->original_mtu);
1796
1797err_free:
1798 bond_free_slave(new_slave);
1799
1800err_undo_flags:
1801 /* Enslave of first slave has failed and we need to fix master's mac */
1802 if (!bond_has_slaves(bond)) {
1803 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1804 slave_dev->dev_addr))
1805 eth_hw_addr_random(bond_dev);
1806 if (bond_dev->type != ARPHRD_ETHER) {
1807 dev_close(bond_dev);
1808 ether_setup(bond_dev);
1809 bond_dev->flags |= IFF_MASTER;
1810 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1811 }
1812 }
1813
1814 return res;
1815}
1816
1817/* Try to release the slave device <slave> from the bond device <master>
1818 * It is legal to access curr_active_slave without a lock because all the function
1819 * is RTNL-locked. If "all" is true it means that the function is being called
1820 * while destroying a bond interface and all slaves are being released.
1821 *
1822 * The rules for slave state should be:
1823 * for Active/Backup:
1824 * Active stays on all backups go down
1825 * for Bonded connections:
1826 * The first up interface should be left on and all others downed.
1827 */
1828static int __bond_release_one(struct net_device *bond_dev,
1829 struct net_device *slave_dev,
1830 bool all, bool unregister)
1831{
1832 struct bonding *bond = netdev_priv(bond_dev);
1833 struct slave *slave, *oldcurrent;
1834 struct sockaddr_storage ss;
1835 int old_flags = bond_dev->flags;
1836 netdev_features_t old_features = bond_dev->features;
1837
1838 /* slave is not a slave or master is not master of this slave */
1839 if (!(slave_dev->flags & IFF_SLAVE) ||
1840 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1841 netdev_dbg(bond_dev, "cannot release %s\n",
1842 slave_dev->name);
1843 return -EINVAL;
1844 }
1845
1846 block_netpoll_tx();
1847
1848 slave = bond_get_slave_by_dev(bond, slave_dev);
1849 if (!slave) {
1850 /* not a slave of this bond */
1851 netdev_info(bond_dev, "%s not enslaved\n",
1852 slave_dev->name);
1853 unblock_netpoll_tx();
1854 return -EINVAL;
1855 }
1856
1857 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1858
1859 bond_sysfs_slave_del(slave);
1860
1861 /* recompute stats just before removing the slave */
1862 bond_get_stats(bond->dev, &bond->bond_stats);
1863
1864 bond_upper_dev_unlink(bond, slave);
1865 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1866 * for this slave anymore.
1867 */
1868 netdev_rx_handler_unregister(slave_dev);
1869
1870 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1871 bond_3ad_unbind_slave(slave);
1872
1873 if (bond_mode_uses_xmit_hash(bond))
1874 bond_update_slave_arr(bond, slave);
1875
1876 netdev_info(bond_dev, "Releasing %s interface %s\n",
1877 bond_is_active_slave(slave) ? "active" : "backup",
1878 slave_dev->name);
1879
1880 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1881
1882 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1883
1884 if (!all && (!bond->params.fail_over_mac ||
1885 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1886 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1887 bond_has_slaves(bond))
1888 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1889 slave_dev->name, slave->perm_hwaddr,
1890 bond_dev->name, slave_dev->name);
1891 }
1892
1893 if (rtnl_dereference(bond->primary_slave) == slave)
1894 RCU_INIT_POINTER(bond->primary_slave, NULL);
1895
1896 if (oldcurrent == slave)
1897 bond_change_active_slave(bond, NULL);
1898
1899 if (bond_is_lb(bond)) {
1900 /* Must be called only after the slave has been
1901 * detached from the list and the curr_active_slave
1902 * has been cleared (if our_slave == old_current),
1903 * but before a new active slave is selected.
1904 */
1905 bond_alb_deinit_slave(bond, slave);
1906 }
1907
1908 if (all) {
1909 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1910 } else if (oldcurrent == slave) {
1911 /* Note that we hold RTNL over this sequence, so there
1912 * is no concern that another slave add/remove event
1913 * will interfere.
1914 */
1915 bond_select_active_slave(bond);
1916 }
1917
1918 if (!bond_has_slaves(bond)) {
1919 bond_set_carrier(bond);
1920 eth_hw_addr_random(bond_dev);
1921 }
1922
1923 unblock_netpoll_tx();
1924 synchronize_rcu();
1925 bond->slave_cnt--;
1926
1927 if (!bond_has_slaves(bond)) {
1928 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1929 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1930 }
1931
1932 bond_compute_features(bond);
1933 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1934 (old_features & NETIF_F_VLAN_CHALLENGED))
1935 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1936 slave_dev->name, bond_dev->name);
1937
1938 vlan_vids_del_by_dev(slave_dev, bond_dev);
1939
1940 /* If the mode uses primary, then this case was handled above by
1941 * bond_change_active_slave(..., NULL)
1942 */
1943 if (!bond_uses_primary(bond)) {
1944 /* unset promiscuity level from slave
1945 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1946 * of the IFF_PROMISC flag in the bond_dev, but we need the
1947 * value of that flag before that change, as that was the value
1948 * when this slave was attached, so we cache at the start of the
1949 * function and use it here. Same goes for ALLMULTI below
1950 */
1951 if (old_flags & IFF_PROMISC)
1952 dev_set_promiscuity(slave_dev, -1);
1953
1954 /* unset allmulti level from slave */
1955 if (old_flags & IFF_ALLMULTI)
1956 dev_set_allmulti(slave_dev, -1);
1957
1958 bond_hw_addr_flush(bond_dev, slave_dev);
1959 }
1960
1961 slave_disable_netpoll(slave);
1962
1963 /* close slave before restoring its mac address */
1964 dev_close(slave_dev);
1965
1966 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1967 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1968 /* restore original ("permanent") mac address */
1969 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1970 slave->dev->addr_len);
1971 ss.ss_family = slave_dev->type;
1972 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1973 }
1974
1975 if (unregister)
1976 __dev_set_mtu(slave_dev, slave->original_mtu);
1977 else
1978 dev_set_mtu(slave_dev, slave->original_mtu);
1979
1980 slave_dev->priv_flags &= ~IFF_BONDING;
1981
1982 bond_free_slave(slave);
1983
1984 return 0;
1985}
1986
1987/* A wrapper used because of ndo_del_link */
1988int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1989{
1990 return __bond_release_one(bond_dev, slave_dev, false, false);
1991}
1992
1993/* First release a slave and then destroy the bond if no more slaves are left.
1994 * Must be under rtnl_lock when this function is called.
1995 */
1996static int bond_release_and_destroy(struct net_device *bond_dev,
1997 struct net_device *slave_dev)
1998{
1999 struct bonding *bond = netdev_priv(bond_dev);
2000 int ret;
2001
2002 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2003 if (ret == 0 && !bond_has_slaves(bond)) {
2004 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2005 netdev_info(bond_dev, "Destroying bond %s\n",
2006 bond_dev->name);
2007 bond_remove_proc_entry(bond);
2008 unregister_netdevice(bond_dev);
2009 }
2010 return ret;
2011}
2012
2013static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2014{
2015 struct bonding *bond = netdev_priv(bond_dev);
2016 bond_fill_ifbond(bond, info);
2017}
2018
2019static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2020{
2021 struct bonding *bond = netdev_priv(bond_dev);
2022 struct list_head *iter;
2023 int i = 0, res = -ENODEV;
2024 struct slave *slave;
2025
2026 bond_for_each_slave(bond, slave, iter) {
2027 if (i++ == (int)info->slave_id) {
2028 res = 0;
2029 bond_fill_ifslave(slave, info);
2030 break;
2031 }
2032 }
2033
2034 return res;
2035}
2036
2037/*-------------------------------- Monitoring -------------------------------*/
2038
2039/* called with rcu_read_lock() */
2040static int bond_miimon_inspect(struct bonding *bond)
2041{
2042 int link_state, commit = 0;
2043 struct list_head *iter;
2044 struct slave *slave;
2045 bool ignore_updelay;
2046
2047 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2048
2049 bond_for_each_slave_rcu(bond, slave, iter) {
2050 slave->new_link = BOND_LINK_NOCHANGE;
2051 slave->link_new_state = slave->link;
2052
2053 link_state = bond_check_dev_link(bond, slave->dev, 0);
2054
2055 switch (slave->link) {
2056 case BOND_LINK_UP:
2057 if (link_state)
2058 continue;
2059
2060 bond_propose_link_state(slave, BOND_LINK_FAIL);
2061 commit++;
2062 slave->delay = bond->params.downdelay;
2063 if (slave->delay) {
2064 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2065 (BOND_MODE(bond) ==
2066 BOND_MODE_ACTIVEBACKUP) ?
2067 (bond_is_active_slave(slave) ?
2068 "active " : "backup ") : "",
2069 slave->dev->name,
2070 bond->params.downdelay * bond->params.miimon);
2071 }
2072 /*FALLTHRU*/
2073 case BOND_LINK_FAIL:
2074 if (link_state) {
2075 /* recovered before downdelay expired */
2076 bond_propose_link_state(slave, BOND_LINK_UP);
2077 slave->last_link_up = jiffies;
2078 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2079 (bond->params.downdelay - slave->delay) *
2080 bond->params.miimon,
2081 slave->dev->name);
2082 commit++;
2083 continue;
2084 }
2085
2086 if (slave->delay <= 0) {
2087 slave->new_link = BOND_LINK_DOWN;
2088 commit++;
2089 continue;
2090 }
2091
2092 slave->delay--;
2093 break;
2094
2095 case BOND_LINK_DOWN:
2096 if (!link_state)
2097 continue;
2098
2099 bond_propose_link_state(slave, BOND_LINK_BACK);
2100 commit++;
2101 slave->delay = bond->params.updelay;
2102
2103 if (slave->delay) {
2104 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2105 slave->dev->name,
2106 ignore_updelay ? 0 :
2107 bond->params.updelay *
2108 bond->params.miimon);
2109 }
2110 /*FALLTHRU*/
2111 case BOND_LINK_BACK:
2112 if (!link_state) {
2113 bond_propose_link_state(slave, BOND_LINK_DOWN);
2114 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2115 (bond->params.updelay - slave->delay) *
2116 bond->params.miimon,
2117 slave->dev->name);
2118 commit++;
2119 continue;
2120 }
2121
2122 if (ignore_updelay)
2123 slave->delay = 0;
2124
2125 if (slave->delay <= 0) {
2126 slave->new_link = BOND_LINK_UP;
2127 commit++;
2128 ignore_updelay = false;
2129 continue;
2130 }
2131
2132 slave->delay--;
2133 break;
2134 }
2135 }
2136
2137 return commit;
2138}
2139
2140static void bond_miimon_commit(struct bonding *bond)
2141{
2142 struct list_head *iter;
2143 struct slave *slave, *primary;
2144
2145 bond_for_each_slave(bond, slave, iter) {
2146 switch (slave->new_link) {
2147 case BOND_LINK_NOCHANGE:
2148 continue;
2149
2150 case BOND_LINK_UP:
2151 if (bond_update_speed_duplex(slave) &&
2152 bond_needs_speed_duplex(bond)) {
2153 slave->link = BOND_LINK_DOWN;
2154 if (net_ratelimit())
2155 netdev_warn(bond->dev,
2156 "failed to get link speed/duplex for %s\n",
2157 slave->dev->name);
2158 continue;
2159 }
2160 bond_set_slave_link_state(slave, BOND_LINK_UP,
2161 BOND_SLAVE_NOTIFY_NOW);
2162 slave->last_link_up = jiffies;
2163
2164 primary = rtnl_dereference(bond->primary_slave);
2165 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2166 /* prevent it from being the active one */
2167 bond_set_backup_slave(slave);
2168 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2169 /* make it immediately active */
2170 bond_set_active_slave(slave);
2171 } else if (slave != primary) {
2172 /* prevent it from being the active one */
2173 bond_set_backup_slave(slave);
2174 }
2175
2176 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2177 slave->dev->name,
2178 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2179 slave->duplex ? "full" : "half");
2180
2181 /* notify ad that the link status has changed */
2182 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2183 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2184
2185 if (bond_is_lb(bond))
2186 bond_alb_handle_link_change(bond, slave,
2187 BOND_LINK_UP);
2188
2189 if (BOND_MODE(bond) == BOND_MODE_XOR)
2190 bond_update_slave_arr(bond, NULL);
2191
2192 if (!bond->curr_active_slave || slave == primary)
2193 goto do_failover;
2194
2195 continue;
2196
2197 case BOND_LINK_DOWN:
2198 if (slave->link_failure_count < UINT_MAX)
2199 slave->link_failure_count++;
2200
2201 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2202 BOND_SLAVE_NOTIFY_NOW);
2203
2204 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2205 BOND_MODE(bond) == BOND_MODE_8023AD)
2206 bond_set_slave_inactive_flags(slave,
2207 BOND_SLAVE_NOTIFY_NOW);
2208
2209 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2210 slave->dev->name);
2211
2212 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2213 bond_3ad_handle_link_change(slave,
2214 BOND_LINK_DOWN);
2215
2216 if (bond_is_lb(bond))
2217 bond_alb_handle_link_change(bond, slave,
2218 BOND_LINK_DOWN);
2219
2220 if (BOND_MODE(bond) == BOND_MODE_XOR)
2221 bond_update_slave_arr(bond, NULL);
2222
2223 if (slave == rcu_access_pointer(bond->curr_active_slave))
2224 goto do_failover;
2225
2226 continue;
2227
2228 default:
2229 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2230 slave->new_link, slave->dev->name);
2231 slave->new_link = BOND_LINK_NOCHANGE;
2232
2233 continue;
2234 }
2235
2236do_failover:
2237 block_netpoll_tx();
2238 bond_select_active_slave(bond);
2239 unblock_netpoll_tx();
2240 }
2241
2242 bond_set_carrier(bond);
2243}
2244
2245/* bond_mii_monitor
2246 *
2247 * Really a wrapper that splits the mii monitor into two phases: an
2248 * inspection, then (if inspection indicates something needs to be done)
2249 * an acquisition of appropriate locks followed by a commit phase to
2250 * implement whatever link state changes are indicated.
2251 */
2252static void bond_mii_monitor(struct work_struct *work)
2253{
2254 struct bonding *bond = container_of(work, struct bonding,
2255 mii_work.work);
2256 bool should_notify_peers = false;
2257 unsigned long delay;
2258 struct slave *slave;
2259 struct list_head *iter;
2260
2261 delay = msecs_to_jiffies(bond->params.miimon);
2262
2263 if (!bond_has_slaves(bond))
2264 goto re_arm;
2265
2266 rcu_read_lock();
2267
2268 should_notify_peers = bond_should_notify_peers(bond);
2269
2270 if (bond_miimon_inspect(bond)) {
2271 rcu_read_unlock();
2272
2273 /* Race avoidance with bond_close cancel of workqueue */
2274 if (!rtnl_trylock()) {
2275 delay = 1;
2276 should_notify_peers = false;
2277 goto re_arm;
2278 }
2279
2280 bond_for_each_slave(bond, slave, iter) {
2281 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2282 }
2283 bond_miimon_commit(bond);
2284
2285 rtnl_unlock(); /* might sleep, hold no other locks */
2286 } else
2287 rcu_read_unlock();
2288
2289re_arm:
2290 if (bond->params.miimon)
2291 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2292
2293 if (should_notify_peers) {
2294 if (!rtnl_trylock())
2295 return;
2296 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2297 rtnl_unlock();
2298 }
2299}
2300
2301static int bond_upper_dev_walk(struct net_device *upper, void *data)
2302{
2303 __be32 ip = *((__be32 *)data);
2304
2305 return ip == bond_confirm_addr(upper, 0, ip);
2306}
2307
2308static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2309{
2310 bool ret = false;
2311
2312 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2313 return true;
2314
2315 rcu_read_lock();
2316 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2317 ret = true;
2318 rcu_read_unlock();
2319
2320 return ret;
2321}
2322
2323/* We go to the (large) trouble of VLAN tagging ARP frames because
2324 * switches in VLAN mode (especially if ports are configured as
2325 * "native" to a VLAN) might not pass non-tagged frames.
2326 */
2327static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2328 __be32 dest_ip, __be32 src_ip,
2329 struct bond_vlan_tag *tags)
2330{
2331 struct sk_buff *skb;
2332 struct bond_vlan_tag *outer_tag = tags;
2333
2334 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2335 arp_op, slave_dev->name, &dest_ip, &src_ip);
2336
2337 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2338 NULL, slave_dev->dev_addr, NULL);
2339
2340 if (!skb) {
2341 net_err_ratelimited("ARP packet allocation failed\n");
2342 return;
2343 }
2344
2345 if (!tags || tags->vlan_proto == VLAN_N_VID)
2346 goto xmit;
2347
2348 tags++;
2349
2350 /* Go through all the tags backwards and add them to the packet */
2351 while (tags->vlan_proto != VLAN_N_VID) {
2352 if (!tags->vlan_id) {
2353 tags++;
2354 continue;
2355 }
2356
2357 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2358 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2359 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2360 tags->vlan_id);
2361 if (!skb) {
2362 net_err_ratelimited("failed to insert inner VLAN tag\n");
2363 return;
2364 }
2365
2366 tags++;
2367 }
2368 /* Set the outer tag */
2369 if (outer_tag->vlan_id) {
2370 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2371 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2372 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2373 outer_tag->vlan_id);
2374 }
2375
2376xmit:
2377 arp_xmit(skb);
2378}
2379
2380/* Validate the device path between the @start_dev and the @end_dev.
2381 * The path is valid if the @end_dev is reachable through device
2382 * stacking.
2383 * When the path is validated, collect any vlan information in the
2384 * path.
2385 */
2386struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2387 struct net_device *end_dev,
2388 int level)
2389{
2390 struct bond_vlan_tag *tags;
2391 struct net_device *upper;
2392 struct list_head *iter;
2393
2394 if (start_dev == end_dev) {
2395 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2396 if (!tags)
2397 return ERR_PTR(-ENOMEM);
2398 tags[level].vlan_proto = VLAN_N_VID;
2399 return tags;
2400 }
2401
2402 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2403 tags = bond_verify_device_path(upper, end_dev, level + 1);
2404 if (IS_ERR_OR_NULL(tags)) {
2405 if (IS_ERR(tags))
2406 return tags;
2407 continue;
2408 }
2409 if (is_vlan_dev(upper)) {
2410 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2411 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2412 }
2413
2414 return tags;
2415 }
2416
2417 return NULL;
2418}
2419
2420static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2421{
2422 struct rtable *rt;
2423 struct bond_vlan_tag *tags;
2424 __be32 *targets = bond->params.arp_targets, addr;
2425 int i;
2426
2427 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2428 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2429 tags = NULL;
2430
2431 /* Find out through which dev should the packet go */
2432 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2433 RTO_ONLINK, 0);
2434 if (IS_ERR(rt)) {
2435 /* there's no route to target - try to send arp
2436 * probe to generate any traffic (arp_validate=0)
2437 */
2438 if (bond->params.arp_validate)
2439 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2440 bond->dev->name,
2441 &targets[i]);
2442 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2443 0, tags);
2444 continue;
2445 }
2446
2447 /* bond device itself */
2448 if (rt->dst.dev == bond->dev)
2449 goto found;
2450
2451 rcu_read_lock();
2452 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2453 rcu_read_unlock();
2454
2455 if (!IS_ERR_OR_NULL(tags))
2456 goto found;
2457
2458 /* Not our device - skip */
2459 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2460 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2461
2462 ip_rt_put(rt);
2463 continue;
2464
2465found:
2466 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2467 ip_rt_put(rt);
2468 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2469 addr, tags);
2470 kfree(tags);
2471 }
2472}
2473
2474static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2475{
2476 int i;
2477
2478 if (!sip || !bond_has_this_ip(bond, tip)) {
2479 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2480 &sip, &tip);
2481 return;
2482 }
2483
2484 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2485 if (i == -1) {
2486 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2487 &sip);
2488 return;
2489 }
2490 slave->last_rx = jiffies;
2491 slave->target_last_arp_rx[i] = jiffies;
2492}
2493
2494int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2495 struct slave *slave)
2496{
2497 struct arphdr *arp = (struct arphdr *)skb->data;
2498 struct slave *curr_active_slave, *curr_arp_slave;
2499 unsigned char *arp_ptr;
2500 __be32 sip, tip;
2501 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2502 unsigned int alen;
2503
2504 if (!slave_do_arp_validate(bond, slave)) {
2505 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2506 !slave_do_arp_validate_only(bond))
2507 slave->last_rx = jiffies;
2508 return RX_HANDLER_ANOTHER;
2509 } else if (!is_arp) {
2510 return RX_HANDLER_ANOTHER;
2511 }
2512
2513 alen = arp_hdr_len(bond->dev);
2514
2515 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2516 skb->dev->name);
2517
2518 if (alen > skb_headlen(skb)) {
2519 arp = kmalloc(alen, GFP_ATOMIC);
2520 if (!arp)
2521 goto out_unlock;
2522 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2523 goto out_unlock;
2524 }
2525
2526 if (arp->ar_hln != bond->dev->addr_len ||
2527 skb->pkt_type == PACKET_OTHERHOST ||
2528 skb->pkt_type == PACKET_LOOPBACK ||
2529 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2530 arp->ar_pro != htons(ETH_P_IP) ||
2531 arp->ar_pln != 4)
2532 goto out_unlock;
2533
2534 arp_ptr = (unsigned char *)(arp + 1);
2535 arp_ptr += bond->dev->addr_len;
2536 memcpy(&sip, arp_ptr, 4);
2537 arp_ptr += 4 + bond->dev->addr_len;
2538 memcpy(&tip, arp_ptr, 4);
2539
2540 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2541 slave->dev->name, bond_slave_state(slave),
2542 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2543 &sip, &tip);
2544
2545 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2546 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2547
2548 /* We 'trust' the received ARP enough to validate it if:
2549 *
2550 * (a) the slave receiving the ARP is active (which includes the
2551 * current ARP slave, if any), or
2552 *
2553 * (b) the receiving slave isn't active, but there is a currently
2554 * active slave and it received valid arp reply(s) after it became
2555 * the currently active slave, or
2556 *
2557 * (c) there is an ARP slave that sent an ARP during the prior ARP
2558 * interval, and we receive an ARP reply on any slave. We accept
2559 * these because switch FDB update delays may deliver the ARP
2560 * reply to a slave other than the sender of the ARP request.
2561 *
2562 * Note: for (b), backup slaves are receiving the broadcast ARP
2563 * request, not a reply. This request passes from the sending
2564 * slave through the L2 switch(es) to the receiving slave. Since
2565 * this is checking the request, sip/tip are swapped for
2566 * validation.
2567 *
2568 * This is done to avoid endless looping when we can't reach the
2569 * arp_ip_target and fool ourselves with our own arp requests.
2570 */
2571 if (bond_is_active_slave(slave))
2572 bond_validate_arp(bond, slave, sip, tip);
2573 else if (curr_active_slave &&
2574 time_after(slave_last_rx(bond, curr_active_slave),
2575 curr_active_slave->last_link_up))
2576 bond_validate_arp(bond, slave, tip, sip);
2577 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2578 bond_time_in_interval(bond,
2579 dev_trans_start(curr_arp_slave->dev), 1))
2580 bond_validate_arp(bond, slave, sip, tip);
2581
2582out_unlock:
2583 if (arp != (struct arphdr *)skb->data)
2584 kfree(arp);
2585 return RX_HANDLER_ANOTHER;
2586}
2587
2588/* function to verify if we're in the arp_interval timeslice, returns true if
2589 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2590 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2591 */
2592static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2593 int mod)
2594{
2595 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2596
2597 return time_in_range(jiffies,
2598 last_act - delta_in_ticks,
2599 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2600}
2601
2602/* This function is called regularly to monitor each slave's link
2603 * ensuring that traffic is being sent and received when arp monitoring
2604 * is used in load-balancing mode. if the adapter has been dormant, then an
2605 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2606 * arp monitoring in active backup mode.
2607 */
2608static void bond_loadbalance_arp_mon(struct bonding *bond)
2609{
2610 struct slave *slave, *oldcurrent;
2611 struct list_head *iter;
2612 int do_failover = 0, slave_state_changed = 0;
2613
2614 if (!bond_has_slaves(bond))
2615 goto re_arm;
2616
2617 rcu_read_lock();
2618
2619 oldcurrent = rcu_dereference(bond->curr_active_slave);
2620 /* see if any of the previous devices are up now (i.e. they have
2621 * xmt and rcv traffic). the curr_active_slave does not come into
2622 * the picture unless it is null. also, slave->last_link_up is not
2623 * needed here because we send an arp on each slave and give a slave
2624 * as long as it needs to get the tx/rx within the delta.
2625 * TODO: what about up/down delay in arp mode? it wasn't here before
2626 * so it can wait
2627 */
2628 bond_for_each_slave_rcu(bond, slave, iter) {
2629 unsigned long trans_start = dev_trans_start(slave->dev);
2630
2631 slave->new_link = BOND_LINK_NOCHANGE;
2632
2633 if (slave->link != BOND_LINK_UP) {
2634 if (bond_time_in_interval(bond, trans_start, 1) &&
2635 bond_time_in_interval(bond, slave->last_rx, 1)) {
2636
2637 slave->new_link = BOND_LINK_UP;
2638 slave_state_changed = 1;
2639
2640 /* primary_slave has no meaning in round-robin
2641 * mode. the window of a slave being up and
2642 * curr_active_slave being null after enslaving
2643 * is closed.
2644 */
2645 if (!oldcurrent) {
2646 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2647 slave->dev->name);
2648 do_failover = 1;
2649 } else {
2650 netdev_info(bond->dev, "interface %s is now up\n",
2651 slave->dev->name);
2652 }
2653 }
2654 } else {
2655 /* slave->link == BOND_LINK_UP */
2656
2657 /* not all switches will respond to an arp request
2658 * when the source ip is 0, so don't take the link down
2659 * if we don't know our ip yet
2660 */
2661 if (!bond_time_in_interval(bond, trans_start, 2) ||
2662 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2663
2664 slave->new_link = BOND_LINK_DOWN;
2665 slave_state_changed = 1;
2666
2667 if (slave->link_failure_count < UINT_MAX)
2668 slave->link_failure_count++;
2669
2670 netdev_info(bond->dev, "interface %s is now down\n",
2671 slave->dev->name);
2672
2673 if (slave == oldcurrent)
2674 do_failover = 1;
2675 }
2676 }
2677
2678 /* note: if switch is in round-robin mode, all links
2679 * must tx arp to ensure all links rx an arp - otherwise
2680 * links may oscillate or not come up at all; if switch is
2681 * in something like xor mode, there is nothing we can
2682 * do - all replies will be rx'ed on same link causing slaves
2683 * to be unstable during low/no traffic periods
2684 */
2685 if (bond_slave_is_up(slave))
2686 bond_arp_send_all(bond, slave);
2687 }
2688
2689 rcu_read_unlock();
2690
2691 if (do_failover || slave_state_changed) {
2692 if (!rtnl_trylock())
2693 goto re_arm;
2694
2695 bond_for_each_slave(bond, slave, iter) {
2696 if (slave->new_link != BOND_LINK_NOCHANGE)
2697 slave->link = slave->new_link;
2698 }
2699
2700 if (slave_state_changed) {
2701 bond_slave_state_change(bond);
2702 if (BOND_MODE(bond) == BOND_MODE_XOR)
2703 bond_update_slave_arr(bond, NULL);
2704 }
2705 if (do_failover) {
2706 block_netpoll_tx();
2707 bond_select_active_slave(bond);
2708 unblock_netpoll_tx();
2709 }
2710 rtnl_unlock();
2711 }
2712
2713re_arm:
2714 if (bond->params.arp_interval)
2715 queue_delayed_work(bond->wq, &bond->arp_work,
2716 msecs_to_jiffies(bond->params.arp_interval));
2717}
2718
2719/* Called to inspect slaves for active-backup mode ARP monitor link state
2720 * changes. Sets new_link in slaves to specify what action should take
2721 * place for the slave. Returns 0 if no changes are found, >0 if changes
2722 * to link states must be committed.
2723 *
2724 * Called with rcu_read_lock held.
2725 */
2726static int bond_ab_arp_inspect(struct bonding *bond)
2727{
2728 unsigned long trans_start, last_rx;
2729 struct list_head *iter;
2730 struct slave *slave;
2731 int commit = 0;
2732
2733 bond_for_each_slave_rcu(bond, slave, iter) {
2734 slave->new_link = BOND_LINK_NOCHANGE;
2735 last_rx = slave_last_rx(bond, slave);
2736
2737 if (slave->link != BOND_LINK_UP) {
2738 if (bond_time_in_interval(bond, last_rx, 1)) {
2739 slave->new_link = BOND_LINK_UP;
2740 commit++;
2741 }
2742 continue;
2743 }
2744
2745 /* Give slaves 2*delta after being enslaved or made
2746 * active. This avoids bouncing, as the last receive
2747 * times need a full ARP monitor cycle to be updated.
2748 */
2749 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2750 continue;
2751
2752 /* Backup slave is down if:
2753 * - No current_arp_slave AND
2754 * - more than 3*delta since last receive AND
2755 * - the bond has an IP address
2756 *
2757 * Note: a non-null current_arp_slave indicates
2758 * the curr_active_slave went down and we are
2759 * searching for a new one; under this condition
2760 * we only take the curr_active_slave down - this
2761 * gives each slave a chance to tx/rx traffic
2762 * before being taken out
2763 */
2764 if (!bond_is_active_slave(slave) &&
2765 !rcu_access_pointer(bond->current_arp_slave) &&
2766 !bond_time_in_interval(bond, last_rx, 3)) {
2767 slave->new_link = BOND_LINK_DOWN;
2768 commit++;
2769 }
2770
2771 /* Active slave is down if:
2772 * - more than 2*delta since transmitting OR
2773 * - (more than 2*delta since receive AND
2774 * the bond has an IP address)
2775 */
2776 trans_start = dev_trans_start(slave->dev);
2777 if (bond_is_active_slave(slave) &&
2778 (!bond_time_in_interval(bond, trans_start, 2) ||
2779 !bond_time_in_interval(bond, last_rx, 2))) {
2780 slave->new_link = BOND_LINK_DOWN;
2781 commit++;
2782 }
2783 }
2784
2785 return commit;
2786}
2787
2788/* Called to commit link state changes noted by inspection step of
2789 * active-backup mode ARP monitor.
2790 *
2791 * Called with RTNL hold.
2792 */
2793static void bond_ab_arp_commit(struct bonding *bond)
2794{
2795 unsigned long trans_start;
2796 struct list_head *iter;
2797 struct slave *slave;
2798
2799 bond_for_each_slave(bond, slave, iter) {
2800 switch (slave->new_link) {
2801 case BOND_LINK_NOCHANGE:
2802 continue;
2803
2804 case BOND_LINK_UP:
2805 trans_start = dev_trans_start(slave->dev);
2806 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2807 (!rtnl_dereference(bond->curr_active_slave) &&
2808 bond_time_in_interval(bond, trans_start, 1))) {
2809 struct slave *current_arp_slave;
2810
2811 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2812 bond_set_slave_link_state(slave, BOND_LINK_UP,
2813 BOND_SLAVE_NOTIFY_NOW);
2814 if (current_arp_slave) {
2815 bond_set_slave_inactive_flags(
2816 current_arp_slave,
2817 BOND_SLAVE_NOTIFY_NOW);
2818 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2819 }
2820
2821 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2822 slave->dev->name);
2823
2824 if (!rtnl_dereference(bond->curr_active_slave) ||
2825 slave == rtnl_dereference(bond->primary_slave))
2826 goto do_failover;
2827
2828 }
2829
2830 continue;
2831
2832 case BOND_LINK_DOWN:
2833 if (slave->link_failure_count < UINT_MAX)
2834 slave->link_failure_count++;
2835
2836 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2837 BOND_SLAVE_NOTIFY_NOW);
2838 bond_set_slave_inactive_flags(slave,
2839 BOND_SLAVE_NOTIFY_NOW);
2840
2841 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2842 slave->dev->name);
2843
2844 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2845 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2846 goto do_failover;
2847 }
2848
2849 continue;
2850
2851 default:
2852 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2853 slave->new_link, slave->dev->name);
2854 continue;
2855 }
2856
2857do_failover:
2858 block_netpoll_tx();
2859 bond_select_active_slave(bond);
2860 unblock_netpoll_tx();
2861 }
2862
2863 bond_set_carrier(bond);
2864}
2865
2866/* Send ARP probes for active-backup mode ARP monitor.
2867 *
2868 * Called with rcu_read_lock held.
2869 */
2870static bool bond_ab_arp_probe(struct bonding *bond)
2871{
2872 struct slave *slave, *before = NULL, *new_slave = NULL,
2873 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2874 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2875 struct list_head *iter;
2876 bool found = false;
2877 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2878
2879 if (curr_arp_slave && curr_active_slave)
2880 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2881 curr_arp_slave->dev->name,
2882 curr_active_slave->dev->name);
2883
2884 if (curr_active_slave) {
2885 bond_arp_send_all(bond, curr_active_slave);
2886 return should_notify_rtnl;
2887 }
2888
2889 /* if we don't have a curr_active_slave, search for the next available
2890 * backup slave from the current_arp_slave and make it the candidate
2891 * for becoming the curr_active_slave
2892 */
2893
2894 if (!curr_arp_slave) {
2895 curr_arp_slave = bond_first_slave_rcu(bond);
2896 if (!curr_arp_slave)
2897 return should_notify_rtnl;
2898 }
2899
2900 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2901
2902 bond_for_each_slave_rcu(bond, slave, iter) {
2903 if (!found && !before && bond_slave_is_up(slave))
2904 before = slave;
2905
2906 if (found && !new_slave && bond_slave_is_up(slave))
2907 new_slave = slave;
2908 /* if the link state is up at this point, we
2909 * mark it down - this can happen if we have
2910 * simultaneous link failures and
2911 * reselect_active_interface doesn't make this
2912 * one the current slave so it is still marked
2913 * up when it is actually down
2914 */
2915 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2916 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2917 BOND_SLAVE_NOTIFY_LATER);
2918 if (slave->link_failure_count < UINT_MAX)
2919 slave->link_failure_count++;
2920
2921 bond_set_slave_inactive_flags(slave,
2922 BOND_SLAVE_NOTIFY_LATER);
2923
2924 netdev_info(bond->dev, "backup interface %s is now down\n",
2925 slave->dev->name);
2926 }
2927 if (slave == curr_arp_slave)
2928 found = true;
2929 }
2930
2931 if (!new_slave && before)
2932 new_slave = before;
2933
2934 if (!new_slave)
2935 goto check_state;
2936
2937 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2938 BOND_SLAVE_NOTIFY_LATER);
2939 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2940 bond_arp_send_all(bond, new_slave);
2941 new_slave->last_link_up = jiffies;
2942 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2943
2944check_state:
2945 bond_for_each_slave_rcu(bond, slave, iter) {
2946 if (slave->should_notify || slave->should_notify_link) {
2947 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2948 break;
2949 }
2950 }
2951 return should_notify_rtnl;
2952}
2953
2954static void bond_activebackup_arp_mon(struct bonding *bond)
2955{
2956 bool should_notify_peers = false;
2957 bool should_notify_rtnl = false;
2958 int delta_in_ticks;
2959
2960 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2961
2962 if (!bond_has_slaves(bond))
2963 goto re_arm;
2964
2965 rcu_read_lock();
2966
2967 should_notify_peers = bond_should_notify_peers(bond);
2968
2969 if (bond_ab_arp_inspect(bond)) {
2970 rcu_read_unlock();
2971
2972 /* Race avoidance with bond_close flush of workqueue */
2973 if (!rtnl_trylock()) {
2974 delta_in_ticks = 1;
2975 should_notify_peers = false;
2976 goto re_arm;
2977 }
2978
2979 bond_ab_arp_commit(bond);
2980
2981 rtnl_unlock();
2982 rcu_read_lock();
2983 }
2984
2985 should_notify_rtnl = bond_ab_arp_probe(bond);
2986 rcu_read_unlock();
2987
2988re_arm:
2989 if (bond->params.arp_interval)
2990 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2991
2992 if (should_notify_peers || should_notify_rtnl) {
2993 if (!rtnl_trylock())
2994 return;
2995
2996 if (should_notify_peers)
2997 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2998 bond->dev);
2999 if (should_notify_rtnl) {
3000 bond_slave_state_notify(bond);
3001 bond_slave_link_notify(bond);
3002 }
3003
3004 rtnl_unlock();
3005 }
3006}
3007
3008static void bond_arp_monitor(struct work_struct *work)
3009{
3010 struct bonding *bond = container_of(work, struct bonding,
3011 arp_work.work);
3012
3013 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3014 bond_activebackup_arp_mon(bond);
3015 else
3016 bond_loadbalance_arp_mon(bond);
3017}
3018
3019/*-------------------------- netdev event handling --------------------------*/
3020
3021/* Change device name */
3022static int bond_event_changename(struct bonding *bond)
3023{
3024 bond_remove_proc_entry(bond);
3025 bond_create_proc_entry(bond);
3026
3027 bond_debug_reregister(bond);
3028
3029 return NOTIFY_DONE;
3030}
3031
3032static int bond_master_netdev_event(unsigned long event,
3033 struct net_device *bond_dev)
3034{
3035 struct bonding *event_bond = netdev_priv(bond_dev);
3036
3037 switch (event) {
3038 case NETDEV_CHANGENAME:
3039 return bond_event_changename(event_bond);
3040 case NETDEV_UNREGISTER:
3041 bond_remove_proc_entry(event_bond);
3042 break;
3043 case NETDEV_REGISTER:
3044 bond_create_proc_entry(event_bond);
3045 break;
3046 case NETDEV_NOTIFY_PEERS:
3047 if (event_bond->send_peer_notif)
3048 event_bond->send_peer_notif--;
3049 break;
3050 default:
3051 break;
3052 }
3053
3054 return NOTIFY_DONE;
3055}
3056
3057static int bond_slave_netdev_event(unsigned long event,
3058 struct net_device *slave_dev)
3059{
3060 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3061 struct bonding *bond;
3062 struct net_device *bond_dev;
3063
3064 /* A netdev event can be generated while enslaving a device
3065 * before netdev_rx_handler_register is called in which case
3066 * slave will be NULL
3067 */
3068 if (!slave)
3069 return NOTIFY_DONE;
3070 bond_dev = slave->bond->dev;
3071 bond = slave->bond;
3072 primary = rtnl_dereference(bond->primary_slave);
3073
3074 switch (event) {
3075 case NETDEV_UNREGISTER:
3076 if (bond_dev->type != ARPHRD_ETHER)
3077 bond_release_and_destroy(bond_dev, slave_dev);
3078 else
3079 __bond_release_one(bond_dev, slave_dev, false, true);
3080 break;
3081 case NETDEV_UP:
3082 case NETDEV_CHANGE:
3083 /* For 802.3ad mode only:
3084 * Getting invalid Speed/Duplex values here will put slave
3085 * in weird state. So mark it as link-down for the time
3086 * being and let link-monitoring (miimon) set it right when
3087 * correct speeds/duplex are available.
3088 */
3089 if (bond_update_speed_duplex(slave) &&
3090 BOND_MODE(bond) == BOND_MODE_8023AD)
3091 slave->link = BOND_LINK_DOWN;
3092
3093 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3094 bond_3ad_adapter_speed_duplex_changed(slave);
3095 /* Fallthrough */
3096 case NETDEV_DOWN:
3097 /* Refresh slave-array if applicable!
3098 * If the setup does not use miimon or arpmon (mode-specific!),
3099 * then these events will not cause the slave-array to be
3100 * refreshed. This will cause xmit to use a slave that is not
3101 * usable. Avoid such situation by refeshing the array at these
3102 * events. If these (miimon/arpmon) parameters are configured
3103 * then array gets refreshed twice and that should be fine!
3104 */
3105 if (bond_mode_uses_xmit_hash(bond))
3106 bond_update_slave_arr(bond, NULL);
3107 break;
3108 case NETDEV_CHANGEMTU:
3109 /* TODO: Should slaves be allowed to
3110 * independently alter their MTU? For
3111 * an active-backup bond, slaves need
3112 * not be the same type of device, so
3113 * MTUs may vary. For other modes,
3114 * slaves arguably should have the
3115 * same MTUs. To do this, we'd need to
3116 * take over the slave's change_mtu
3117 * function for the duration of their
3118 * servitude.
3119 */
3120 break;
3121 case NETDEV_CHANGENAME:
3122 /* we don't care if we don't have primary set */
3123 if (!bond_uses_primary(bond) ||
3124 !bond->params.primary[0])
3125 break;
3126
3127 if (slave == primary) {
3128 /* slave's name changed - he's no longer primary */
3129 RCU_INIT_POINTER(bond->primary_slave, NULL);
3130 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3131 /* we have a new primary slave */
3132 rcu_assign_pointer(bond->primary_slave, slave);
3133 } else { /* we didn't change primary - exit */
3134 break;
3135 }
3136
3137 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3138 primary ? slave_dev->name : "none");
3139
3140 block_netpoll_tx();
3141 bond_select_active_slave(bond);
3142 unblock_netpoll_tx();
3143 break;
3144 case NETDEV_FEAT_CHANGE:
3145 bond_compute_features(bond);
3146 break;
3147 case NETDEV_RESEND_IGMP:
3148 /* Propagate to master device */
3149 call_netdevice_notifiers(event, slave->bond->dev);
3150 break;
3151 default:
3152 break;
3153 }
3154
3155 return NOTIFY_DONE;
3156}
3157
3158/* bond_netdev_event: handle netdev notifier chain events.
3159 *
3160 * This function receives events for the netdev chain. The caller (an
3161 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3162 * locks for us to safely manipulate the slave devices (RTNL lock,
3163 * dev_probe_lock).
3164 */
3165static int bond_netdev_event(struct notifier_block *this,
3166 unsigned long event, void *ptr)
3167{
3168 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3169
3170 netdev_dbg(event_dev, "event: %lx\n", event);
3171
3172 if (!(event_dev->priv_flags & IFF_BONDING))
3173 return NOTIFY_DONE;
3174
3175 if (event_dev->flags & IFF_MASTER) {
3176 netdev_dbg(event_dev, "IFF_MASTER\n");
3177 return bond_master_netdev_event(event, event_dev);
3178 }
3179
3180 if (event_dev->flags & IFF_SLAVE) {
3181 netdev_dbg(event_dev, "IFF_SLAVE\n");
3182 return bond_slave_netdev_event(event, event_dev);
3183 }
3184
3185 return NOTIFY_DONE;
3186}
3187
3188static struct notifier_block bond_netdev_notifier = {
3189 .notifier_call = bond_netdev_event,
3190};
3191
3192/*---------------------------- Hashing Policies -----------------------------*/
3193
3194/* L2 hash helper */
3195static inline u32 bond_eth_hash(struct sk_buff *skb)
3196{
3197 struct ethhdr *ep, hdr_tmp;
3198
3199 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3200 if (ep)
3201 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3202 return 0;
3203}
3204
3205/* Extract the appropriate headers based on bond's xmit policy */
3206static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3207 struct flow_keys *fk)
3208{
3209 const struct ipv6hdr *iph6;
3210 const struct iphdr *iph;
3211 int noff, proto = -1;
3212
3213 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3214 return skb_flow_dissect_flow_keys(skb, fk, 0);
3215
3216 fk->ports.ports = 0;
3217 noff = skb_network_offset(skb);
3218 if (skb->protocol == htons(ETH_P_IP)) {
3219 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3220 return false;
3221 iph = ip_hdr(skb);
3222 iph_to_flow_copy_v4addrs(fk, iph);
3223 noff += iph->ihl << 2;
3224 if (!ip_is_fragment(iph))
3225 proto = iph->protocol;
3226 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3227 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3228 return false;
3229 iph6 = ipv6_hdr(skb);
3230 iph_to_flow_copy_v6addrs(fk, iph6);
3231 noff += sizeof(*iph6);
3232 proto = iph6->nexthdr;
3233 } else {
3234 return false;
3235 }
3236 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3237 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3238
3239 return true;
3240}
3241
3242/**
3243 * bond_xmit_hash - generate a hash value based on the xmit policy
3244 * @bond: bonding device
3245 * @skb: buffer to use for headers
3246 *
3247 * This function will extract the necessary headers from the skb buffer and use
3248 * them to generate a hash based on the xmit_policy set in the bonding device
3249 */
3250u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3251{
3252 struct flow_keys flow;
3253 u32 hash;
3254
3255 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3256 skb->l4_hash)
3257 return skb->hash;
3258
3259 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3260 !bond_flow_dissect(bond, skb, &flow))
3261 return bond_eth_hash(skb);
3262
3263 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3264 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3265 hash = bond_eth_hash(skb);
3266 else
3267 hash = (__force u32)flow.ports.ports;
3268 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3269 (__force u32)flow_get_u32_src(&flow);
3270 hash ^= (hash >> 16);
3271 hash ^= (hash >> 8);
3272
3273 return hash >> 1;
3274}
3275
3276/*-------------------------- Device entry points ----------------------------*/
3277
3278void bond_work_init_all(struct bonding *bond)
3279{
3280 INIT_DELAYED_WORK(&bond->mcast_work,
3281 bond_resend_igmp_join_requests_delayed);
3282 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3283 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3284 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3285 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3286 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3287}
3288
3289static void bond_work_cancel_all(struct bonding *bond)
3290{
3291 cancel_delayed_work_sync(&bond->mii_work);
3292 cancel_delayed_work_sync(&bond->arp_work);
3293 cancel_delayed_work_sync(&bond->alb_work);
3294 cancel_delayed_work_sync(&bond->ad_work);
3295 cancel_delayed_work_sync(&bond->mcast_work);
3296 cancel_delayed_work_sync(&bond->slave_arr_work);
3297}
3298
3299static int bond_open(struct net_device *bond_dev)
3300{
3301 struct bonding *bond = netdev_priv(bond_dev);
3302 struct list_head *iter;
3303 struct slave *slave;
3304
3305 /* reset slave->backup and slave->inactive */
3306 if (bond_has_slaves(bond)) {
3307 bond_for_each_slave(bond, slave, iter) {
3308 if (bond_uses_primary(bond) &&
3309 slave != rcu_access_pointer(bond->curr_active_slave)) {
3310 bond_set_slave_inactive_flags(slave,
3311 BOND_SLAVE_NOTIFY_NOW);
3312 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3313 bond_set_slave_active_flags(slave,
3314 BOND_SLAVE_NOTIFY_NOW);
3315 }
3316 }
3317 }
3318
3319 if (bond_is_lb(bond)) {
3320 /* bond_alb_initialize must be called before the timer
3321 * is started.
3322 */
3323 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3324 return -ENOMEM;
3325 if (bond->params.tlb_dynamic_lb)
3326 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3327 }
3328
3329 if (bond->params.miimon) /* link check interval, in milliseconds. */
3330 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3331
3332 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3333 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3334 bond->recv_probe = bond_arp_rcv;
3335 }
3336
3337 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3338 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3339 /* register to receive LACPDUs */
3340 bond->recv_probe = bond_3ad_lacpdu_recv;
3341 bond_3ad_initiate_agg_selection(bond, 1);
3342 }
3343
3344 if (bond_mode_uses_xmit_hash(bond))
3345 bond_update_slave_arr(bond, NULL);
3346
3347 return 0;
3348}
3349
3350static int bond_close(struct net_device *bond_dev)
3351{
3352 struct bonding *bond = netdev_priv(bond_dev);
3353
3354 bond_work_cancel_all(bond);
3355 bond->send_peer_notif = 0;
3356 if (bond_is_lb(bond))
3357 bond_alb_deinitialize(bond);
3358 bond->recv_probe = NULL;
3359
3360 return 0;
3361}
3362
3363/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3364 * that some drivers can provide 32bit values only.
3365 */
3366static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3367 const struct rtnl_link_stats64 *_new,
3368 const struct rtnl_link_stats64 *_old)
3369{
3370 const u64 *new = (const u64 *)_new;
3371 const u64 *old = (const u64 *)_old;
3372 u64 *res = (u64 *)_res;
3373 int i;
3374
3375 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3376 u64 nv = new[i];
3377 u64 ov = old[i];
3378 s64 delta = nv - ov;
3379
3380 /* detects if this particular field is 32bit only */
3381 if (((nv | ov) >> 32) == 0)
3382 delta = (s64)(s32)((u32)nv - (u32)ov);
3383
3384 /* filter anomalies, some drivers reset their stats
3385 * at down/up events.
3386 */
3387 if (delta > 0)
3388 res[i] += delta;
3389 }
3390}
3391
3392static void bond_get_stats(struct net_device *bond_dev,
3393 struct rtnl_link_stats64 *stats)
3394{
3395 struct bonding *bond = netdev_priv(bond_dev);
3396 struct rtnl_link_stats64 temp;
3397 struct list_head *iter;
3398 struct slave *slave;
3399
3400 spin_lock(&bond->stats_lock);
3401 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3402
3403 rcu_read_lock();
3404 bond_for_each_slave_rcu(bond, slave, iter) {
3405 const struct rtnl_link_stats64 *new =
3406 dev_get_stats(slave->dev, &temp);
3407
3408 bond_fold_stats(stats, new, &slave->slave_stats);
3409
3410 /* save off the slave stats for the next run */
3411 memcpy(&slave->slave_stats, new, sizeof(*new));
3412 }
3413 rcu_read_unlock();
3414
3415 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3416 spin_unlock(&bond->stats_lock);
3417}
3418
3419static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3420{
3421 struct bonding *bond = netdev_priv(bond_dev);
3422 struct net_device *slave_dev = NULL;
3423 struct ifbond k_binfo;
3424 struct ifbond __user *u_binfo = NULL;
3425 struct ifslave k_sinfo;
3426 struct ifslave __user *u_sinfo = NULL;
3427 struct mii_ioctl_data *mii = NULL;
3428 struct bond_opt_value newval;
3429 struct net *net;
3430 int res = 0;
3431
3432 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3433
3434 switch (cmd) {
3435 case SIOCGMIIPHY:
3436 mii = if_mii(ifr);
3437 if (!mii)
3438 return -EINVAL;
3439
3440 mii->phy_id = 0;
3441 /* Fall Through */
3442 case SIOCGMIIREG:
3443 /* We do this again just in case we were called by SIOCGMIIREG
3444 * instead of SIOCGMIIPHY.
3445 */
3446 mii = if_mii(ifr);
3447 if (!mii)
3448 return -EINVAL;
3449
3450 if (mii->reg_num == 1) {
3451 mii->val_out = 0;
3452 if (netif_carrier_ok(bond->dev))
3453 mii->val_out = BMSR_LSTATUS;
3454 }
3455
3456 return 0;
3457 case BOND_INFO_QUERY_OLD:
3458 case SIOCBONDINFOQUERY:
3459 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3460
3461 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3462 return -EFAULT;
3463
3464 bond_info_query(bond_dev, &k_binfo);
3465 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3466 return -EFAULT;
3467
3468 return 0;
3469 case BOND_SLAVE_INFO_QUERY_OLD:
3470 case SIOCBONDSLAVEINFOQUERY:
3471 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3472
3473 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3474 return -EFAULT;
3475
3476 res = bond_slave_info_query(bond_dev, &k_sinfo);
3477 if (res == 0 &&
3478 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3479 return -EFAULT;
3480
3481 return res;
3482 default:
3483 break;
3484 }
3485
3486 net = dev_net(bond_dev);
3487
3488 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3489 return -EPERM;
3490
3491 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3492
3493 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3494
3495 if (!slave_dev)
3496 return -ENODEV;
3497
3498 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3499 switch (cmd) {
3500 case BOND_ENSLAVE_OLD:
3501 case SIOCBONDENSLAVE:
3502 res = bond_enslave(bond_dev, slave_dev, NULL);
3503 break;
3504 case BOND_RELEASE_OLD:
3505 case SIOCBONDRELEASE:
3506 res = bond_release(bond_dev, slave_dev);
3507 break;
3508 case BOND_SETHWADDR_OLD:
3509 case SIOCBONDSETHWADDR:
3510 bond_set_dev_addr(bond_dev, slave_dev);
3511 res = 0;
3512 break;
3513 case BOND_CHANGE_ACTIVE_OLD:
3514 case SIOCBONDCHANGEACTIVE:
3515 bond_opt_initstr(&newval, slave_dev->name);
3516 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3517 &newval);
3518 break;
3519 default:
3520 res = -EOPNOTSUPP;
3521 }
3522
3523 return res;
3524}
3525
3526static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3527{
3528 struct bonding *bond = netdev_priv(bond_dev);
3529
3530 if (change & IFF_PROMISC)
3531 bond_set_promiscuity(bond,
3532 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3533
3534 if (change & IFF_ALLMULTI)
3535 bond_set_allmulti(bond,
3536 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3537}
3538
3539static void bond_set_rx_mode(struct net_device *bond_dev)
3540{
3541 struct bonding *bond = netdev_priv(bond_dev);
3542 struct list_head *iter;
3543 struct slave *slave;
3544
3545 rcu_read_lock();
3546 if (bond_uses_primary(bond)) {
3547 slave = rcu_dereference(bond->curr_active_slave);
3548 if (slave) {
3549 dev_uc_sync(slave->dev, bond_dev);
3550 dev_mc_sync(slave->dev, bond_dev);
3551 }
3552 } else {
3553 bond_for_each_slave_rcu(bond, slave, iter) {
3554 dev_uc_sync_multiple(slave->dev, bond_dev);
3555 dev_mc_sync_multiple(slave->dev, bond_dev);
3556 }
3557 }
3558 rcu_read_unlock();
3559}
3560
3561static int bond_neigh_init(struct neighbour *n)
3562{
3563 struct bonding *bond = netdev_priv(n->dev);
3564 const struct net_device_ops *slave_ops;
3565 struct neigh_parms parms;
3566 struct slave *slave;
3567 int ret;
3568
3569 slave = bond_first_slave(bond);
3570 if (!slave)
3571 return 0;
3572 slave_ops = slave->dev->netdev_ops;
3573 if (!slave_ops->ndo_neigh_setup)
3574 return 0;
3575
3576 parms.neigh_setup = NULL;
3577 parms.neigh_cleanup = NULL;
3578 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3579 if (ret)
3580 return ret;
3581
3582 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3583 * after the last slave has been detached. Assumes that all slaves
3584 * utilize the same neigh_cleanup (true at this writing as only user
3585 * is ipoib).
3586 */
3587 n->parms->neigh_cleanup = parms.neigh_cleanup;
3588
3589 if (!parms.neigh_setup)
3590 return 0;
3591
3592 return parms.neigh_setup(n);
3593}
3594
3595/* The bonding ndo_neigh_setup is called at init time beofre any
3596 * slave exists. So we must declare proxy setup function which will
3597 * be used at run time to resolve the actual slave neigh param setup.
3598 *
3599 * It's also called by master devices (such as vlans) to setup their
3600 * underlying devices. In that case - do nothing, we're already set up from
3601 * our init.
3602 */
3603static int bond_neigh_setup(struct net_device *dev,
3604 struct neigh_parms *parms)
3605{
3606 /* modify only our neigh_parms */
3607 if (parms->dev == dev)
3608 parms->neigh_setup = bond_neigh_init;
3609
3610 return 0;
3611}
3612
3613/* Change the MTU of all of a master's slaves to match the master */
3614static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3615{
3616 struct bonding *bond = netdev_priv(bond_dev);
3617 struct slave *slave, *rollback_slave;
3618 struct list_head *iter;
3619 int res = 0;
3620
3621 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3622
3623 bond_for_each_slave(bond, slave, iter) {
3624 netdev_dbg(bond_dev, "s %p c_m %p\n",
3625 slave, slave->dev->netdev_ops->ndo_change_mtu);
3626
3627 res = dev_set_mtu(slave->dev, new_mtu);
3628
3629 if (res) {
3630 /* If we failed to set the slave's mtu to the new value
3631 * we must abort the operation even in ACTIVE_BACKUP
3632 * mode, because if we allow the backup slaves to have
3633 * different mtu values than the active slave we'll
3634 * need to change their mtu when doing a failover. That
3635 * means changing their mtu from timer context, which
3636 * is probably not a good idea.
3637 */
3638 netdev_dbg(bond_dev, "err %d %s\n", res,
3639 slave->dev->name);
3640 goto unwind;
3641 }
3642 }
3643
3644 bond_dev->mtu = new_mtu;
3645
3646 return 0;
3647
3648unwind:
3649 /* unwind from head to the slave that failed */
3650 bond_for_each_slave(bond, rollback_slave, iter) {
3651 int tmp_res;
3652
3653 if (rollback_slave == slave)
3654 break;
3655
3656 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3657 if (tmp_res) {
3658 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3659 tmp_res, rollback_slave->dev->name);
3660 }
3661 }
3662
3663 return res;
3664}
3665
3666/* Change HW address
3667 *
3668 * Note that many devices must be down to change the HW address, and
3669 * downing the master releases all slaves. We can make bonds full of
3670 * bonding devices to test this, however.
3671 */
3672static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3673{
3674 struct bonding *bond = netdev_priv(bond_dev);
3675 struct slave *slave, *rollback_slave;
3676 struct sockaddr_storage *ss = addr, tmp_ss;
3677 struct list_head *iter;
3678 int res = 0;
3679
3680 if (BOND_MODE(bond) == BOND_MODE_ALB)
3681 return bond_alb_set_mac_address(bond_dev, addr);
3682
3683
3684 netdev_dbg(bond_dev, "bond=%p\n", bond);
3685
3686 /* If fail_over_mac is enabled, do nothing and return success.
3687 * Returning an error causes ifenslave to fail.
3688 */
3689 if (bond->params.fail_over_mac &&
3690 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3691 return 0;
3692
3693 if (!is_valid_ether_addr(ss->__data))
3694 return -EADDRNOTAVAIL;
3695
3696 bond_for_each_slave(bond, slave, iter) {
3697 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3698 res = dev_set_mac_address(slave->dev, addr);
3699 if (res) {
3700 /* TODO: consider downing the slave
3701 * and retry ?
3702 * User should expect communications
3703 * breakage anyway until ARP finish
3704 * updating, so...
3705 */
3706 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3707 goto unwind;
3708 }
3709 }
3710
3711 /* success */
3712 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3713 return 0;
3714
3715unwind:
3716 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3717 tmp_ss.ss_family = bond_dev->type;
3718
3719 /* unwind from head to the slave that failed */
3720 bond_for_each_slave(bond, rollback_slave, iter) {
3721 int tmp_res;
3722
3723 if (rollback_slave == slave)
3724 break;
3725
3726 tmp_res = dev_set_mac_address(rollback_slave->dev,
3727 (struct sockaddr *)&tmp_ss);
3728 if (tmp_res) {
3729 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3730 tmp_res, rollback_slave->dev->name);
3731 }
3732 }
3733
3734 return res;
3735}
3736
3737/**
3738 * bond_xmit_slave_id - transmit skb through slave with slave_id
3739 * @bond: bonding device that is transmitting
3740 * @skb: buffer to transmit
3741 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3742 *
3743 * This function tries to transmit through slave with slave_id but in case
3744 * it fails, it tries to find the first available slave for transmission.
3745 * The skb is consumed in all cases, thus the function is void.
3746 */
3747static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3748{
3749 struct list_head *iter;
3750 struct slave *slave;
3751 int i = slave_id;
3752
3753 /* Here we start from the slave with slave_id */
3754 bond_for_each_slave_rcu(bond, slave, iter) {
3755 if (--i < 0) {
3756 if (bond_slave_can_tx(slave)) {
3757 bond_dev_queue_xmit(bond, skb, slave->dev);
3758 return;
3759 }
3760 }
3761 }
3762
3763 /* Here we start from the first slave up to slave_id */
3764 i = slave_id;
3765 bond_for_each_slave_rcu(bond, slave, iter) {
3766 if (--i < 0)
3767 break;
3768 if (bond_slave_can_tx(slave)) {
3769 bond_dev_queue_xmit(bond, skb, slave->dev);
3770 return;
3771 }
3772 }
3773 /* no slave that can tx has been found */
3774 bond_tx_drop(bond->dev, skb);
3775}
3776
3777/**
3778 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3779 * @bond: bonding device to use
3780 *
3781 * Based on the value of the bonding device's packets_per_slave parameter
3782 * this function generates a slave id, which is usually used as the next
3783 * slave to transmit through.
3784 */
3785static u32 bond_rr_gen_slave_id(struct bonding *bond)
3786{
3787 u32 slave_id;
3788 struct reciprocal_value reciprocal_packets_per_slave;
3789 int packets_per_slave = bond->params.packets_per_slave;
3790
3791 switch (packets_per_slave) {
3792 case 0:
3793 slave_id = prandom_u32();
3794 break;
3795 case 1:
3796 slave_id = bond->rr_tx_counter;
3797 break;
3798 default:
3799 reciprocal_packets_per_slave =
3800 bond->params.reciprocal_packets_per_slave;
3801 slave_id = reciprocal_divide(bond->rr_tx_counter,
3802 reciprocal_packets_per_slave);
3803 break;
3804 }
3805 bond->rr_tx_counter++;
3806
3807 return slave_id;
3808}
3809
3810static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3811{
3812 struct bonding *bond = netdev_priv(bond_dev);
3813 struct iphdr *iph = ip_hdr(skb);
3814 struct slave *slave;
3815 u32 slave_id;
3816
3817 /* Start with the curr_active_slave that joined the bond as the
3818 * default for sending IGMP traffic. For failover purposes one
3819 * needs to maintain some consistency for the interface that will
3820 * send the join/membership reports. The curr_active_slave found
3821 * will send all of this type of traffic.
3822 */
3823 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3824 slave = rcu_dereference(bond->curr_active_slave);
3825 if (slave)
3826 bond_dev_queue_xmit(bond, skb, slave->dev);
3827 else
3828 bond_xmit_slave_id(bond, skb, 0);
3829 } else {
3830 int slave_cnt = READ_ONCE(bond->slave_cnt);
3831
3832 if (likely(slave_cnt)) {
3833 slave_id = bond_rr_gen_slave_id(bond);
3834 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3835 } else {
3836 bond_tx_drop(bond_dev, skb);
3837 }
3838 }
3839
3840 return NETDEV_TX_OK;
3841}
3842
3843/* In active-backup mode, we know that bond->curr_active_slave is always valid if
3844 * the bond has a usable interface.
3845 */
3846static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3847{
3848 struct bonding *bond = netdev_priv(bond_dev);
3849 struct slave *slave;
3850
3851 slave = rcu_dereference(bond->curr_active_slave);
3852 if (slave)
3853 bond_dev_queue_xmit(bond, skb, slave->dev);
3854 else
3855 bond_tx_drop(bond_dev, skb);
3856
3857 return NETDEV_TX_OK;
3858}
3859
3860/* Use this to update slave_array when (a) it's not appropriate to update
3861 * slave_array right away (note that update_slave_array() may sleep)
3862 * and / or (b) RTNL is not held.
3863 */
3864void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3865{
3866 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3867}
3868
3869/* Slave array work handler. Holds only RTNL */
3870static void bond_slave_arr_handler(struct work_struct *work)
3871{
3872 struct bonding *bond = container_of(work, struct bonding,
3873 slave_arr_work.work);
3874 int ret;
3875
3876 if (!rtnl_trylock())
3877 goto err;
3878
3879 ret = bond_update_slave_arr(bond, NULL);
3880 rtnl_unlock();
3881 if (ret) {
3882 pr_warn_ratelimited("Failed to update slave array from WT\n");
3883 goto err;
3884 }
3885 return;
3886
3887err:
3888 bond_slave_arr_work_rearm(bond, 1);
3889}
3890
3891/* Build the usable slaves array in control path for modes that use xmit-hash
3892 * to determine the slave interface -
3893 * (a) BOND_MODE_8023AD
3894 * (b) BOND_MODE_XOR
3895 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3896 *
3897 * The caller is expected to hold RTNL only and NO other lock!
3898 */
3899int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3900{
3901 struct slave *slave;
3902 struct list_head *iter;
3903 struct bond_up_slave *new_arr, *old_arr;
3904 int agg_id = 0;
3905 int ret = 0;
3906
3907#ifdef CONFIG_LOCKDEP
3908 WARN_ON(lockdep_is_held(&bond->mode_lock));
3909#endif
3910
3911 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3912 GFP_KERNEL);
3913 if (!new_arr) {
3914 ret = -ENOMEM;
3915 pr_err("Failed to build slave-array.\n");
3916 goto out;
3917 }
3918 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3919 struct ad_info ad_info;
3920
3921 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3922 pr_debug("bond_3ad_get_active_agg_info failed\n");
3923 kfree_rcu(new_arr, rcu);
3924 /* No active aggragator means it's not safe to use
3925 * the previous array.
3926 */
3927 old_arr = rtnl_dereference(bond->slave_arr);
3928 if (old_arr) {
3929 RCU_INIT_POINTER(bond->slave_arr, NULL);
3930 kfree_rcu(old_arr, rcu);
3931 }
3932 goto out;
3933 }
3934 agg_id = ad_info.aggregator_id;
3935 }
3936 bond_for_each_slave(bond, slave, iter) {
3937 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3938 struct aggregator *agg;
3939
3940 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3941 if (!agg || agg->aggregator_identifier != agg_id)
3942 continue;
3943 }
3944 if (!bond_slave_can_tx(slave))
3945 continue;
3946 if (skipslave == slave)
3947 continue;
3948 new_arr->arr[new_arr->count++] = slave;
3949 }
3950
3951 old_arr = rtnl_dereference(bond->slave_arr);
3952 rcu_assign_pointer(bond->slave_arr, new_arr);
3953 if (old_arr)
3954 kfree_rcu(old_arr, rcu);
3955out:
3956 if (ret != 0 && skipslave) {
3957 int idx;
3958
3959 /* Rare situation where caller has asked to skip a specific
3960 * slave but allocation failed (most likely!). BTW this is
3961 * only possible when the call is initiated from
3962 * __bond_release_one(). In this situation; overwrite the
3963 * skipslave entry in the array with the last entry from the
3964 * array to avoid a situation where the xmit path may choose
3965 * this to-be-skipped slave to send a packet out.
3966 */
3967 old_arr = rtnl_dereference(bond->slave_arr);
3968 for (idx = 0; idx < old_arr->count; idx++) {
3969 if (skipslave == old_arr->arr[idx]) {
3970 old_arr->arr[idx] =
3971 old_arr->arr[old_arr->count-1];
3972 old_arr->count--;
3973 break;
3974 }
3975 }
3976 }
3977 return ret;
3978}
3979
3980/* Use this Xmit function for 3AD as well as XOR modes. The current
3981 * usable slave array is formed in the control path. The xmit function
3982 * just calculates hash and sends the packet out.
3983 */
3984static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3985{
3986 struct bonding *bond = netdev_priv(dev);
3987 struct slave *slave;
3988 struct bond_up_slave *slaves;
3989 unsigned int count;
3990
3991 slaves = rcu_dereference(bond->slave_arr);
3992 count = slaves ? READ_ONCE(slaves->count) : 0;
3993 if (likely(count)) {
3994 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3995 bond_dev_queue_xmit(bond, skb, slave->dev);
3996 } else {
3997 bond_tx_drop(dev, skb);
3998 }
3999
4000 return NETDEV_TX_OK;
4001}
4002
4003/* in broadcast mode, we send everything to all usable interfaces. */
4004static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4005{
4006 struct bonding *bond = netdev_priv(bond_dev);
4007 struct slave *slave = NULL;
4008 struct list_head *iter;
4009
4010 bond_for_each_slave_rcu(bond, slave, iter) {
4011 if (bond_is_last_slave(bond, slave))
4012 break;
4013 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4014 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4015
4016 if (!skb2) {
4017 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4018 bond_dev->name, __func__);
4019 continue;
4020 }
4021 bond_dev_queue_xmit(bond, skb2, slave->dev);
4022 }
4023 }
4024 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4025 bond_dev_queue_xmit(bond, skb, slave->dev);
4026 else
4027 bond_tx_drop(bond_dev, skb);
4028
4029 return NETDEV_TX_OK;
4030}
4031
4032/*------------------------- Device initialization ---------------------------*/
4033
4034/* Lookup the slave that corresponds to a qid */
4035static inline int bond_slave_override(struct bonding *bond,
4036 struct sk_buff *skb)
4037{
4038 struct slave *slave = NULL;
4039 struct list_head *iter;
4040
4041 if (!skb->queue_mapping)
4042 return 1;
4043
4044 /* Find out if any slaves have the same mapping as this skb. */
4045 bond_for_each_slave_rcu(bond, slave, iter) {
4046 if (slave->queue_id == skb->queue_mapping) {
4047 if (bond_slave_is_up(slave) &&
4048 slave->link == BOND_LINK_UP) {
4049 bond_dev_queue_xmit(bond, skb, slave->dev);
4050 return 0;
4051 }
4052 /* If the slave isn't UP, use default transmit policy. */
4053 break;
4054 }
4055 }
4056
4057 return 1;
4058}
4059
4060
4061static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4062 void *accel_priv, select_queue_fallback_t fallback)
4063{
4064 /* This helper function exists to help dev_pick_tx get the correct
4065 * destination queue. Using a helper function skips a call to
4066 * skb_tx_hash and will put the skbs in the queue we expect on their
4067 * way down to the bonding driver.
4068 */
4069 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4070
4071 /* Save the original txq to restore before passing to the driver */
4072 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4073
4074 if (unlikely(txq >= dev->real_num_tx_queues)) {
4075 do {
4076 txq -= dev->real_num_tx_queues;
4077 } while (txq >= dev->real_num_tx_queues);
4078 }
4079 return txq;
4080}
4081
4082static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4083{
4084 struct bonding *bond = netdev_priv(dev);
4085
4086 if (bond_should_override_tx_queue(bond) &&
4087 !bond_slave_override(bond, skb))
4088 return NETDEV_TX_OK;
4089
4090 switch (BOND_MODE(bond)) {
4091 case BOND_MODE_ROUNDROBIN:
4092 return bond_xmit_roundrobin(skb, dev);
4093 case BOND_MODE_ACTIVEBACKUP:
4094 return bond_xmit_activebackup(skb, dev);
4095 case BOND_MODE_8023AD:
4096 case BOND_MODE_XOR:
4097 return bond_3ad_xor_xmit(skb, dev);
4098 case BOND_MODE_BROADCAST:
4099 return bond_xmit_broadcast(skb, dev);
4100 case BOND_MODE_ALB:
4101 return bond_alb_xmit(skb, dev);
4102 case BOND_MODE_TLB:
4103 return bond_tlb_xmit(skb, dev);
4104 default:
4105 /* Should never happen, mode already checked */
4106 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4107 WARN_ON_ONCE(1);
4108 bond_tx_drop(dev, skb);
4109 return NETDEV_TX_OK;
4110 }
4111}
4112
4113static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4114{
4115 struct bonding *bond = netdev_priv(dev);
4116 netdev_tx_t ret = NETDEV_TX_OK;
4117
4118 /* If we risk deadlock from transmitting this in the
4119 * netpoll path, tell netpoll to queue the frame for later tx
4120 */
4121 if (unlikely(is_netpoll_tx_blocked(dev)))
4122 return NETDEV_TX_BUSY;
4123
4124 rcu_read_lock();
4125 if (bond_has_slaves(bond))
4126 ret = __bond_start_xmit(skb, dev);
4127 else
4128 bond_tx_drop(dev, skb);
4129 rcu_read_unlock();
4130
4131 return ret;
4132}
4133
4134static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4135 struct ethtool_link_ksettings *cmd)
4136{
4137 struct bonding *bond = netdev_priv(bond_dev);
4138 unsigned long speed = 0;
4139 struct list_head *iter;
4140 struct slave *slave;
4141
4142 cmd->base.duplex = DUPLEX_UNKNOWN;
4143 cmd->base.port = PORT_OTHER;
4144
4145 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4146 * do not need to check mode. Though link speed might not represent
4147 * the true receive or transmit bandwidth (not all modes are symmetric)
4148 * this is an accurate maximum.
4149 */
4150 bond_for_each_slave(bond, slave, iter) {
4151 if (bond_slave_can_tx(slave)) {
4152 if (slave->speed != SPEED_UNKNOWN)
4153 speed += slave->speed;
4154 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4155 slave->duplex != DUPLEX_UNKNOWN)
4156 cmd->base.duplex = slave->duplex;
4157 }
4158 }
4159 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4160
4161 return 0;
4162}
4163
4164static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4165 struct ethtool_drvinfo *drvinfo)
4166{
4167 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4168 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4169 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4170 BOND_ABI_VERSION);
4171}
4172
4173static const struct ethtool_ops bond_ethtool_ops = {
4174 .get_drvinfo = bond_ethtool_get_drvinfo,
4175 .get_link = ethtool_op_get_link,
4176 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4177};
4178
4179static const struct net_device_ops bond_netdev_ops = {
4180 .ndo_init = bond_init,
4181 .ndo_uninit = bond_uninit,
4182 .ndo_open = bond_open,
4183 .ndo_stop = bond_close,
4184 .ndo_start_xmit = bond_start_xmit,
4185 .ndo_select_queue = bond_select_queue,
4186 .ndo_get_stats64 = bond_get_stats,
4187 .ndo_do_ioctl = bond_do_ioctl,
4188 .ndo_change_rx_flags = bond_change_rx_flags,
4189 .ndo_set_rx_mode = bond_set_rx_mode,
4190 .ndo_change_mtu = bond_change_mtu,
4191 .ndo_set_mac_address = bond_set_mac_address,
4192 .ndo_neigh_setup = bond_neigh_setup,
4193 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4194 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4195#ifdef CONFIG_NET_POLL_CONTROLLER
4196 .ndo_netpoll_setup = bond_netpoll_setup,
4197 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4198 .ndo_poll_controller = bond_poll_controller,
4199#endif
4200 .ndo_add_slave = bond_enslave,
4201 .ndo_del_slave = bond_release,
4202 .ndo_fix_features = bond_fix_features,
4203 .ndo_features_check = passthru_features_check,
4204};
4205
4206static const struct device_type bond_type = {
4207 .name = "bond",
4208};
4209
4210static void bond_destructor(struct net_device *bond_dev)
4211{
4212 struct bonding *bond = netdev_priv(bond_dev);
4213 if (bond->wq)
4214 destroy_workqueue(bond->wq);
4215}
4216
4217void bond_setup(struct net_device *bond_dev)
4218{
4219 struct bonding *bond = netdev_priv(bond_dev);
4220
4221 spin_lock_init(&bond->mode_lock);
4222 spin_lock_init(&bond->stats_lock);
4223 bond->params = bonding_defaults;
4224
4225 /* Initialize pointers */
4226 bond->dev = bond_dev;
4227
4228 /* Initialize the device entry points */
4229 ether_setup(bond_dev);
4230 bond_dev->max_mtu = ETH_MAX_MTU;
4231 bond_dev->netdev_ops = &bond_netdev_ops;
4232 bond_dev->ethtool_ops = &bond_ethtool_ops;
4233
4234 bond_dev->needs_free_netdev = true;
4235 bond_dev->priv_destructor = bond_destructor;
4236
4237 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4238
4239 /* Initialize the device options */
4240 bond_dev->flags |= IFF_MASTER;
4241 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4242 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4243
4244 /* don't acquire bond device's netif_tx_lock when transmitting */
4245 bond_dev->features |= NETIF_F_LLTX;
4246
4247 /* By default, we declare the bond to be fully
4248 * VLAN hardware accelerated capable. Special
4249 * care is taken in the various xmit functions
4250 * when there are slaves that are not hw accel
4251 * capable
4252 */
4253
4254 /* Don't allow bond devices to change network namespaces. */
4255 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4256
4257 bond_dev->hw_features = BOND_VLAN_FEATURES |
4258 NETIF_F_HW_VLAN_CTAG_TX |
4259 NETIF_F_HW_VLAN_CTAG_RX |
4260 NETIF_F_HW_VLAN_CTAG_FILTER;
4261
4262 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4263 bond_dev->features |= bond_dev->hw_features;
4264}
4265
4266/* Destroy a bonding device.
4267 * Must be under rtnl_lock when this function is called.
4268 */
4269static void bond_uninit(struct net_device *bond_dev)
4270{
4271 struct bonding *bond = netdev_priv(bond_dev);
4272 struct list_head *iter;
4273 struct slave *slave;
4274 struct bond_up_slave *arr;
4275
4276 bond_netpoll_cleanup(bond_dev);
4277
4278 /* Release the bonded slaves */
4279 bond_for_each_slave(bond, slave, iter)
4280 __bond_release_one(bond_dev, slave->dev, true, true);
4281 netdev_info(bond_dev, "Released all slaves\n");
4282
4283 arr = rtnl_dereference(bond->slave_arr);
4284 if (arr) {
4285 RCU_INIT_POINTER(bond->slave_arr, NULL);
4286 kfree_rcu(arr, rcu);
4287 }
4288
4289 list_del(&bond->bond_list);
4290
4291 bond_debug_unregister(bond);
4292}
4293
4294/*------------------------- Module initialization ---------------------------*/
4295
4296static int bond_check_params(struct bond_params *params)
4297{
4298 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4299 struct bond_opt_value newval;
4300 const struct bond_opt_value *valptr;
4301 int arp_all_targets_value = 0;
4302 u16 ad_actor_sys_prio = 0;
4303 u16 ad_user_port_key = 0;
4304 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4305 int arp_ip_count;
4306 int bond_mode = BOND_MODE_ROUNDROBIN;
4307 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4308 int lacp_fast = 0;
4309 int tlb_dynamic_lb;
4310
4311 /* Convert string parameters. */
4312 if (mode) {
4313 bond_opt_initstr(&newval, mode);
4314 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4315 if (!valptr) {
4316 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4317 return -EINVAL;
4318 }
4319 bond_mode = valptr->value;
4320 }
4321
4322 if (xmit_hash_policy) {
4323 if ((bond_mode != BOND_MODE_XOR) &&
4324 (bond_mode != BOND_MODE_8023AD) &&
4325 (bond_mode != BOND_MODE_TLB)) {
4326 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4327 bond_mode_name(bond_mode));
4328 } else {
4329 bond_opt_initstr(&newval, xmit_hash_policy);
4330 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4331 &newval);
4332 if (!valptr) {
4333 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4334 xmit_hash_policy);
4335 return -EINVAL;
4336 }
4337 xmit_hashtype = valptr->value;
4338 }
4339 }
4340
4341 if (lacp_rate) {
4342 if (bond_mode != BOND_MODE_8023AD) {
4343 pr_info("lacp_rate param is irrelevant in mode %s\n",
4344 bond_mode_name(bond_mode));
4345 } else {
4346 bond_opt_initstr(&newval, lacp_rate);
4347 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4348 &newval);
4349 if (!valptr) {
4350 pr_err("Error: Invalid lacp rate \"%s\"\n",
4351 lacp_rate);
4352 return -EINVAL;
4353 }
4354 lacp_fast = valptr->value;
4355 }
4356 }
4357
4358 if (ad_select) {
4359 bond_opt_initstr(&newval, ad_select);
4360 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4361 &newval);
4362 if (!valptr) {
4363 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4364 return -EINVAL;
4365 }
4366 params->ad_select = valptr->value;
4367 if (bond_mode != BOND_MODE_8023AD)
4368 pr_warn("ad_select param only affects 802.3ad mode\n");
4369 } else {
4370 params->ad_select = BOND_AD_STABLE;
4371 }
4372
4373 if (max_bonds < 0) {
4374 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4375 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4376 max_bonds = BOND_DEFAULT_MAX_BONDS;
4377 }
4378
4379 if (miimon < 0) {
4380 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4381 miimon, INT_MAX);
4382 miimon = 0;
4383 }
4384
4385 if (updelay < 0) {
4386 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4387 updelay, INT_MAX);
4388 updelay = 0;
4389 }
4390
4391 if (downdelay < 0) {
4392 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4393 downdelay, INT_MAX);
4394 downdelay = 0;
4395 }
4396
4397 if ((use_carrier != 0) && (use_carrier != 1)) {
4398 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4399 use_carrier);
4400 use_carrier = 1;
4401 }
4402
4403 if (num_peer_notif < 0 || num_peer_notif > 255) {
4404 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4405 num_peer_notif);
4406 num_peer_notif = 1;
4407 }
4408
4409 /* reset values for 802.3ad/TLB/ALB */
4410 if (!bond_mode_uses_arp(bond_mode)) {
4411 if (!miimon) {
4412 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4413 pr_warn("Forcing miimon to 100msec\n");
4414 miimon = BOND_DEFAULT_MIIMON;
4415 }
4416 }
4417
4418 if (tx_queues < 1 || tx_queues > 255) {
4419 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4420 tx_queues, BOND_DEFAULT_TX_QUEUES);
4421 tx_queues = BOND_DEFAULT_TX_QUEUES;
4422 }
4423
4424 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4425 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4426 all_slaves_active);
4427 all_slaves_active = 0;
4428 }
4429
4430 if (resend_igmp < 0 || resend_igmp > 255) {
4431 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4432 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4433 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4434 }
4435
4436 bond_opt_initval(&newval, packets_per_slave);
4437 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4438 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4439 packets_per_slave, USHRT_MAX);
4440 packets_per_slave = 1;
4441 }
4442
4443 if (bond_mode == BOND_MODE_ALB) {
4444 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4445 updelay);
4446 }
4447
4448 if (!miimon) {
4449 if (updelay || downdelay) {
4450 /* just warn the user the up/down delay will have
4451 * no effect since miimon is zero...
4452 */
4453 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4454 updelay, downdelay);
4455 }
4456 } else {
4457 /* don't allow arp monitoring */
4458 if (arp_interval) {
4459 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4460 miimon, arp_interval);
4461 arp_interval = 0;
4462 }
4463
4464 if ((updelay % miimon) != 0) {
4465 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4466 updelay, miimon, (updelay / miimon) * miimon);
4467 }
4468
4469 updelay /= miimon;
4470
4471 if ((downdelay % miimon) != 0) {
4472 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4473 downdelay, miimon,
4474 (downdelay / miimon) * miimon);
4475 }
4476
4477 downdelay /= miimon;
4478 }
4479
4480 if (arp_interval < 0) {
4481 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4482 arp_interval, INT_MAX);
4483 arp_interval = 0;
4484 }
4485
4486 for (arp_ip_count = 0, i = 0;
4487 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4488 __be32 ip;
4489
4490 /* not a complete check, but good enough to catch mistakes */
4491 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4492 !bond_is_ip_target_ok(ip)) {
4493 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4494 arp_ip_target[i]);
4495 arp_interval = 0;
4496 } else {
4497 if (bond_get_targets_ip(arp_target, ip) == -1)
4498 arp_target[arp_ip_count++] = ip;
4499 else
4500 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4501 &ip);
4502 }
4503 }
4504
4505 if (arp_interval && !arp_ip_count) {
4506 /* don't allow arping if no arp_ip_target given... */
4507 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4508 arp_interval);
4509 arp_interval = 0;
4510 }
4511
4512 if (arp_validate) {
4513 if (!arp_interval) {
4514 pr_err("arp_validate requires arp_interval\n");
4515 return -EINVAL;
4516 }
4517
4518 bond_opt_initstr(&newval, arp_validate);
4519 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4520 &newval);
4521 if (!valptr) {
4522 pr_err("Error: invalid arp_validate \"%s\"\n",
4523 arp_validate);
4524 return -EINVAL;
4525 }
4526 arp_validate_value = valptr->value;
4527 } else {
4528 arp_validate_value = 0;
4529 }
4530
4531 if (arp_all_targets) {
4532 bond_opt_initstr(&newval, arp_all_targets);
4533 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4534 &newval);
4535 if (!valptr) {
4536 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4537 arp_all_targets);
4538 arp_all_targets_value = 0;
4539 } else {
4540 arp_all_targets_value = valptr->value;
4541 }
4542 }
4543
4544 if (miimon) {
4545 pr_info("MII link monitoring set to %d ms\n", miimon);
4546 } else if (arp_interval) {
4547 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4548 arp_validate_value);
4549 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4550 arp_interval, valptr->string, arp_ip_count);
4551
4552 for (i = 0; i < arp_ip_count; i++)
4553 pr_cont(" %s", arp_ip_target[i]);
4554
4555 pr_cont("\n");
4556
4557 } else if (max_bonds) {
4558 /* miimon and arp_interval not set, we need one so things
4559 * work as expected, see bonding.txt for details
4560 */
4561 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
4562 }
4563
4564 if (primary && !bond_mode_uses_primary(bond_mode)) {
4565 /* currently, using a primary only makes sense
4566 * in active backup, TLB or ALB modes
4567 */
4568 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4569 primary, bond_mode_name(bond_mode));
4570 primary = NULL;
4571 }
4572
4573 if (primary && primary_reselect) {
4574 bond_opt_initstr(&newval, primary_reselect);
4575 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4576 &newval);
4577 if (!valptr) {
4578 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4579 primary_reselect);
4580 return -EINVAL;
4581 }
4582 primary_reselect_value = valptr->value;
4583 } else {
4584 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4585 }
4586
4587 if (fail_over_mac) {
4588 bond_opt_initstr(&newval, fail_over_mac);
4589 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4590 &newval);
4591 if (!valptr) {
4592 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4593 fail_over_mac);
4594 return -EINVAL;
4595 }
4596 fail_over_mac_value = valptr->value;
4597 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4598 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4599 } else {
4600 fail_over_mac_value = BOND_FOM_NONE;
4601 }
4602
4603 bond_opt_initstr(&newval, "default");
4604 valptr = bond_opt_parse(
4605 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4606 &newval);
4607 if (!valptr) {
4608 pr_err("Error: No ad_actor_sys_prio default value");
4609 return -EINVAL;
4610 }
4611 ad_actor_sys_prio = valptr->value;
4612
4613 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4614 &newval);
4615 if (!valptr) {
4616 pr_err("Error: No ad_user_port_key default value");
4617 return -EINVAL;
4618 }
4619 ad_user_port_key = valptr->value;
4620
4621 bond_opt_initstr(&newval, "default");
4622 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4623 if (!valptr) {
4624 pr_err("Error: No tlb_dynamic_lb default value");
4625 return -EINVAL;
4626 }
4627 tlb_dynamic_lb = valptr->value;
4628
4629 if (lp_interval == 0) {
4630 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4631 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4632 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4633 }
4634
4635 /* fill params struct with the proper values */
4636 params->mode = bond_mode;
4637 params->xmit_policy = xmit_hashtype;
4638 params->miimon = miimon;
4639 params->num_peer_notif = num_peer_notif;
4640 params->arp_interval = arp_interval;
4641 params->arp_validate = arp_validate_value;
4642 params->arp_all_targets = arp_all_targets_value;
4643 params->updelay = updelay;
4644 params->downdelay = downdelay;
4645 params->use_carrier = use_carrier;
4646 params->lacp_fast = lacp_fast;
4647 params->primary[0] = 0;
4648 params->primary_reselect = primary_reselect_value;
4649 params->fail_over_mac = fail_over_mac_value;
4650 params->tx_queues = tx_queues;
4651 params->all_slaves_active = all_slaves_active;
4652 params->resend_igmp = resend_igmp;
4653 params->min_links = min_links;
4654 params->lp_interval = lp_interval;
4655 params->packets_per_slave = packets_per_slave;
4656 params->tlb_dynamic_lb = tlb_dynamic_lb;
4657 params->ad_actor_sys_prio = ad_actor_sys_prio;
4658 eth_zero_addr(params->ad_actor_system);
4659 params->ad_user_port_key = ad_user_port_key;
4660 if (packets_per_slave > 0) {
4661 params->reciprocal_packets_per_slave =
4662 reciprocal_value(packets_per_slave);
4663 } else {
4664 /* reciprocal_packets_per_slave is unused if
4665 * packets_per_slave is 0 or 1, just initialize it
4666 */
4667 params->reciprocal_packets_per_slave =
4668 (struct reciprocal_value) { 0 };
4669 }
4670
4671 if (primary) {
4672 strncpy(params->primary, primary, IFNAMSIZ);
4673 params->primary[IFNAMSIZ - 1] = 0;
4674 }
4675
4676 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4677
4678 return 0;
4679}
4680
4681/* Called from registration process */
4682static int bond_init(struct net_device *bond_dev)
4683{
4684 struct bonding *bond = netdev_priv(bond_dev);
4685 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4686
4687 netdev_dbg(bond_dev, "Begin bond_init\n");
4688
4689 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4690 if (!bond->wq)
4691 return -ENOMEM;
4692
4693 netdev_lockdep_set_classes(bond_dev);
4694
4695 list_add_tail(&bond->bond_list, &bn->dev_list);
4696
4697 bond_prepare_sysfs_group(bond);
4698
4699 bond_debug_register(bond);
4700
4701 /* Ensure valid dev_addr */
4702 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4703 bond_dev->addr_assign_type == NET_ADDR_PERM)
4704 eth_hw_addr_random(bond_dev);
4705
4706 return 0;
4707}
4708
4709unsigned int bond_get_num_tx_queues(void)
4710{
4711 return tx_queues;
4712}
4713
4714/* Create a new bond based on the specified name and bonding parameters.
4715 * If name is NULL, obtain a suitable "bond%d" name for us.
4716 * Caller must NOT hold rtnl_lock; we need to release it here before we
4717 * set up our sysfs entries.
4718 */
4719int bond_create(struct net *net, const char *name)
4720{
4721 struct net_device *bond_dev;
4722 struct bonding *bond;
4723 struct alb_bond_info *bond_info;
4724 int res;
4725
4726 rtnl_lock();
4727
4728 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4729 name ? name : "bond%d", NET_NAME_UNKNOWN,
4730 bond_setup, tx_queues);
4731 if (!bond_dev) {
4732 pr_err("%s: eek! can't alloc netdev!\n", name);
4733 rtnl_unlock();
4734 return -ENOMEM;
4735 }
4736
4737 /*
4738 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4739 * It is set to 0 by default which is wrong.
4740 */
4741 bond = netdev_priv(bond_dev);
4742 bond_info = &(BOND_ALB_INFO(bond));
4743 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4744
4745 dev_net_set(bond_dev, net);
4746 bond_dev->rtnl_link_ops = &bond_link_ops;
4747
4748 res = register_netdevice(bond_dev);
4749
4750 netif_carrier_off(bond_dev);
4751
4752 bond_work_init_all(bond);
4753
4754 rtnl_unlock();
4755 if (res < 0)
4756 free_netdev(bond_dev);
4757 return res;
4758}
4759
4760static int __net_init bond_net_init(struct net *net)
4761{
4762 struct bond_net *bn = net_generic(net, bond_net_id);
4763
4764 bn->net = net;
4765 INIT_LIST_HEAD(&bn->dev_list);
4766
4767 bond_create_proc_dir(bn);
4768 bond_create_sysfs(bn);
4769
4770 return 0;
4771}
4772
4773static void __net_exit bond_net_exit(struct net *net)
4774{
4775 struct bond_net *bn = net_generic(net, bond_net_id);
4776 struct bonding *bond, *tmp_bond;
4777 LIST_HEAD(list);
4778
4779 bond_destroy_sysfs(bn);
4780
4781 /* Kill off any bonds created after unregistering bond rtnl ops */
4782 rtnl_lock();
4783 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4784 unregister_netdevice_queue(bond->dev, &list);
4785 unregister_netdevice_many(&list);
4786 rtnl_unlock();
4787
4788 bond_destroy_proc_dir(bn);
4789}
4790
4791static struct pernet_operations bond_net_ops = {
4792 .init = bond_net_init,
4793 .exit = bond_net_exit,
4794 .id = &bond_net_id,
4795 .size = sizeof(struct bond_net),
4796};
4797
4798static int __init bonding_init(void)
4799{
4800 int i;
4801 int res;
4802
4803 pr_info("%s", bond_version);
4804
4805 res = bond_check_params(&bonding_defaults);
4806 if (res)
4807 goto out;
4808
4809 res = register_pernet_subsys(&bond_net_ops);
4810 if (res)
4811 goto out;
4812
4813 res = bond_netlink_init();
4814 if (res)
4815 goto err_link;
4816
4817 bond_create_debugfs();
4818
4819 for (i = 0; i < max_bonds; i++) {
4820 res = bond_create(&init_net, NULL);
4821 if (res)
4822 goto err;
4823 }
4824
4825 register_netdevice_notifier(&bond_netdev_notifier);
4826out:
4827 return res;
4828err:
4829 bond_destroy_debugfs();
4830 bond_netlink_fini();
4831err_link:
4832 unregister_pernet_subsys(&bond_net_ops);
4833 goto out;
4834
4835}
4836
4837static void __exit bonding_exit(void)
4838{
4839 unregister_netdevice_notifier(&bond_netdev_notifier);
4840
4841 bond_destroy_debugfs();
4842
4843 bond_netlink_fini();
4844 unregister_pernet_subsys(&bond_net_ops);
4845
4846#ifdef CONFIG_NET_POLL_CONTROLLER
4847 /* Make sure we don't have an imbalance on our netpoll blocking */
4848 WARN_ON(atomic_read(&netpoll_block_tx));
4849#endif
4850}
4851
4852module_init(bonding_init);
4853module_exit(bonding_exit);
4854MODULE_LICENSE("GPL");
4855MODULE_VERSION(DRV_VERSION);
4856MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4857MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
1/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
32 */
33
34#include <linux/kernel.h>
35#include <linux/module.h>
36#include <linux/types.h>
37#include <linux/fcntl.h>
38#include <linux/interrupt.h>
39#include <linux/ptrace.h>
40#include <linux/ioport.h>
41#include <linux/in.h>
42#include <net/ip.h>
43#include <linux/ip.h>
44#include <linux/tcp.h>
45#include <linux/udp.h>
46#include <linux/slab.h>
47#include <linux/string.h>
48#include <linux/init.h>
49#include <linux/timer.h>
50#include <linux/socket.h>
51#include <linux/ctype.h>
52#include <linux/inet.h>
53#include <linux/bitops.h>
54#include <linux/io.h>
55#include <asm/dma.h>
56#include <linux/uaccess.h>
57#include <linux/errno.h>
58#include <linux/netdevice.h>
59#include <linux/inetdevice.h>
60#include <linux/igmp.h>
61#include <linux/etherdevice.h>
62#include <linux/skbuff.h>
63#include <net/sock.h>
64#include <linux/rtnetlink.h>
65#include <linux/smp.h>
66#include <linux/if_ether.h>
67#include <net/arp.h>
68#include <linux/mii.h>
69#include <linux/ethtool.h>
70#include <linux/if_vlan.h>
71#include <linux/if_bonding.h>
72#include <linux/jiffies.h>
73#include <linux/preempt.h>
74#include <net/route.h>
75#include <net/net_namespace.h>
76#include <net/netns/generic.h>
77#include <net/pkt_sched.h>
78#include <linux/rculist.h>
79#include <net/flow_dissector.h>
80#include <net/switchdev.h>
81#include <net/bonding.h>
82#include <net/bond_3ad.h>
83#include <net/bond_alb.h>
84
85#include "bonding_priv.h"
86
87/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
90
91static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93static int num_peer_notif = 1;
94static int miimon;
95static int updelay;
96static int downdelay;
97static int use_carrier = 1;
98static char *mode;
99static char *primary;
100static char *primary_reselect;
101static char *lacp_rate;
102static int min_links;
103static char *ad_select;
104static char *xmit_hash_policy;
105static int arp_interval;
106static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107static char *arp_validate;
108static char *arp_all_targets;
109static char *fail_over_mac;
110static int all_slaves_active;
111static struct bond_params bonding_defaults;
112static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113static int packets_per_slave = 1;
114static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115
116module_param(max_bonds, int, 0);
117MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118module_param(tx_queues, int, 0);
119MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126module_param(miimon, int, 0);
127MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128module_param(updelay, int, 0);
129MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130module_param(downdelay, int, 0);
131MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
133module_param(use_carrier, int, 0);
134MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136module_param(mode, charp, 0);
137MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141module_param(primary, charp, 0);
142MODULE_PARM_DESC(primary, "Primary network device to use");
143module_param(primary_reselect, charp, 0);
144MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "once it comes up; "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
148 "better, "
149 "2 for only on active slave "
150 "failure");
151module_param(lacp_rate, charp, 0);
152MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154module_param(ad_select, charp, 0);
155MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
157 "2 for count");
158module_param(min_links, int, 0);
159MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160
161module_param(xmit_hash_policy, charp, 0);
162MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166module_param(arp_interval, int, 0);
167MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168module_param_array(arp_ip_target, charp, NULL, 0);
169MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170module_param(arp_validate, charp, 0);
171MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174module_param(arp_all_targets, charp, 0);
175MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176module_param(fail_over_mac, charp, 0);
177MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180module_param(all_slaves_active, int, 0);
181MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184module_param(resend_igmp, int, 0);
185MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
187module_param(packets_per_slave, int, 0);
188MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191module_param(lp_interval, uint, 0);
192MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
195
196/*----------------------------- Global variables ----------------------------*/
197
198#ifdef CONFIG_NET_POLL_CONTROLLER
199atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200#endif
201
202unsigned int bond_net_id __read_mostly;
203
204static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205static int arp_ip_count;
206static int bond_mode = BOND_MODE_ROUNDROBIN;
207static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208static int lacp_fast;
209
210/*-------------------------- Forward declarations ---------------------------*/
211
212static int bond_init(struct net_device *bond_dev);
213static void bond_uninit(struct net_device *bond_dev);
214static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
216static void bond_slave_arr_handler(struct work_struct *work);
217static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
218 int mod);
219
220/*---------------------------- General routines -----------------------------*/
221
222const char *bond_mode_name(int mode)
223{
224 static const char *names[] = {
225 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
226 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
227 [BOND_MODE_XOR] = "load balancing (xor)",
228 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
229 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
230 [BOND_MODE_TLB] = "transmit load balancing",
231 [BOND_MODE_ALB] = "adaptive load balancing",
232 };
233
234 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
235 return "unknown";
236
237 return names[mode];
238}
239
240/*---------------------------------- VLAN -----------------------------------*/
241
242/**
243 * bond_dev_queue_xmit - Prepare skb for xmit.
244 *
245 * @bond: bond device that got this skb for tx.
246 * @skb: hw accel VLAN tagged skb to transmit
247 * @slave_dev: slave that is supposed to xmit this skbuff
248 */
249void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
250 struct net_device *slave_dev)
251{
252 skb->dev = slave_dev;
253
254 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
255 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
256 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
257
258 if (unlikely(netpoll_tx_running(bond->dev)))
259 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
260 else
261 dev_queue_xmit(skb);
262}
263
264/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
265 * We don't protect the slave list iteration with a lock because:
266 * a. This operation is performed in IOCTL context,
267 * b. The operation is protected by the RTNL semaphore in the 8021q code,
268 * c. Holding a lock with BH disabled while directly calling a base driver
269 * entry point is generally a BAD idea.
270 *
271 * The design of synchronization/protection for this operation in the 8021q
272 * module is good for one or more VLAN devices over a single physical device
273 * and cannot be extended for a teaming solution like bonding, so there is a
274 * potential race condition here where a net device from the vlan group might
275 * be referenced (either by a base driver or the 8021q code) while it is being
276 * removed from the system. However, it turns out we're not making matters
277 * worse, and if it works for regular VLAN usage it will work here too.
278*/
279
280/**
281 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
282 * @bond_dev: bonding net device that got called
283 * @vid: vlan id being added
284 */
285static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
286 __be16 proto, u16 vid)
287{
288 struct bonding *bond = netdev_priv(bond_dev);
289 struct slave *slave, *rollback_slave;
290 struct list_head *iter;
291 int res;
292
293 bond_for_each_slave(bond, slave, iter) {
294 res = vlan_vid_add(slave->dev, proto, vid);
295 if (res)
296 goto unwind;
297 }
298
299 return 0;
300
301unwind:
302 /* unwind to the slave that failed */
303 bond_for_each_slave(bond, rollback_slave, iter) {
304 if (rollback_slave == slave)
305 break;
306
307 vlan_vid_del(rollback_slave->dev, proto, vid);
308 }
309
310 return res;
311}
312
313/**
314 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
315 * @bond_dev: bonding net device that got called
316 * @vid: vlan id being removed
317 */
318static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
319 __be16 proto, u16 vid)
320{
321 struct bonding *bond = netdev_priv(bond_dev);
322 struct list_head *iter;
323 struct slave *slave;
324
325 bond_for_each_slave(bond, slave, iter)
326 vlan_vid_del(slave->dev, proto, vid);
327
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vid);
330
331 return 0;
332}
333
334/*------------------------------- Link status -------------------------------*/
335
336/* Set the carrier state for the master according to the state of its
337 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
338 * do special 802.3ad magic.
339 *
340 * Returns zero if carrier state does not change, nonzero if it does.
341 */
342int bond_set_carrier(struct bonding *bond)
343{
344 struct list_head *iter;
345 struct slave *slave;
346
347 if (!bond_has_slaves(bond))
348 goto down;
349
350 if (BOND_MODE(bond) == BOND_MODE_8023AD)
351 return bond_3ad_set_carrier(bond);
352
353 bond_for_each_slave(bond, slave, iter) {
354 if (slave->link == BOND_LINK_UP) {
355 if (!netif_carrier_ok(bond->dev)) {
356 netif_carrier_on(bond->dev);
357 return 1;
358 }
359 return 0;
360 }
361 }
362
363down:
364 if (netif_carrier_ok(bond->dev)) {
365 netif_carrier_off(bond->dev);
366 return 1;
367 }
368 return 0;
369}
370
371/* Get link speed and duplex from the slave's base driver
372 * using ethtool. If for some reason the call fails or the
373 * values are invalid, set speed and duplex to -1,
374 * and return.
375 */
376static void bond_update_speed_duplex(struct slave *slave)
377{
378 struct net_device *slave_dev = slave->dev;
379 struct ethtool_link_ksettings ecmd;
380 int res;
381
382 slave->speed = SPEED_UNKNOWN;
383 slave->duplex = DUPLEX_UNKNOWN;
384
385 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
386 if (res < 0)
387 return;
388
389 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
390 return;
391
392 switch (ecmd.base.duplex) {
393 case DUPLEX_FULL:
394 case DUPLEX_HALF:
395 break;
396 default:
397 return;
398 }
399
400 slave->speed = ecmd.base.speed;
401 slave->duplex = ecmd.base.duplex;
402
403 return;
404}
405
406const char *bond_slave_link_status(s8 link)
407{
408 switch (link) {
409 case BOND_LINK_UP:
410 return "up";
411 case BOND_LINK_FAIL:
412 return "going down";
413 case BOND_LINK_DOWN:
414 return "down";
415 case BOND_LINK_BACK:
416 return "going back";
417 default:
418 return "unknown";
419 }
420}
421
422/* if <dev> supports MII link status reporting, check its link status.
423 *
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
426 *
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
429 * down.
430 *
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
436 */
437static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
439{
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
442 struct ifreq ifr;
443 struct mii_ioctl_data *mii;
444
445 if (!reporting && !netif_running(slave_dev))
446 return 0;
447
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
450
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 BMSR_LSTATUS : 0;
455
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
458 if (ioctl) {
459 /* TODO: set pointer to correct ioctl on a per team member
460 * bases to make this more efficient. that is, once
461 * we determine the correct ioctl, we will always
462 * call it and not the others for that team
463 * member.
464 */
465
466 /* We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
468 * support that.
469 */
470
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
473 mii = if_mii(&ifr);
474 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
478 }
479 }
480
481 /* If reporting, report that either there's no dev->do_ioctl,
482 * or both SIOCGMIIREG and get_link failed (meaning that we
483 * cannot report link status). If not reporting, pretend
484 * we're ok.
485 */
486 return reporting ? -1 : BMSR_LSTATUS;
487}
488
489/*----------------------------- Multicast list ------------------------------*/
490
491/* Push the promiscuity flag down to appropriate slaves */
492static int bond_set_promiscuity(struct bonding *bond, int inc)
493{
494 struct list_head *iter;
495 int err = 0;
496
497 if (bond_uses_primary(bond)) {
498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
499
500 if (curr_active)
501 err = dev_set_promiscuity(curr_active->dev, inc);
502 } else {
503 struct slave *slave;
504
505 bond_for_each_slave(bond, slave, iter) {
506 err = dev_set_promiscuity(slave->dev, inc);
507 if (err)
508 return err;
509 }
510 }
511 return err;
512}
513
514/* Push the allmulti flag down to all slaves */
515static int bond_set_allmulti(struct bonding *bond, int inc)
516{
517 struct list_head *iter;
518 int err = 0;
519
520 if (bond_uses_primary(bond)) {
521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
522
523 if (curr_active)
524 err = dev_set_allmulti(curr_active->dev, inc);
525 } else {
526 struct slave *slave;
527
528 bond_for_each_slave(bond, slave, iter) {
529 err = dev_set_allmulti(slave->dev, inc);
530 if (err)
531 return err;
532 }
533 }
534 return err;
535}
536
537/* Retrieve the list of registered multicast addresses for the bonding
538 * device and retransmit an IGMP JOIN request to the current active
539 * slave.
540 */
541static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
542{
543 struct bonding *bond = container_of(work, struct bonding,
544 mcast_work.work);
545
546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
548 return;
549 }
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
551
552 if (bond->igmp_retrans > 1) {
553 bond->igmp_retrans--;
554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
555 }
556 rtnl_unlock();
557}
558
559/* Flush bond's hardware addresses from slave */
560static void bond_hw_addr_flush(struct net_device *bond_dev,
561 struct net_device *slave_dev)
562{
563 struct bonding *bond = netdev_priv(bond_dev);
564
565 dev_uc_unsync(slave_dev, bond_dev);
566 dev_mc_unsync(slave_dev, bond_dev);
567
568 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
571
572 dev_mc_del(slave_dev, lacpdu_multicast);
573 }
574}
575
576/*--------------------------- Active slave change ---------------------------*/
577
578/* Update the hardware address list and promisc/allmulti for the new and
579 * old active slaves (if any). Modes that are not using primary keep all
580 * slaves up date at all times; only the modes that use primary need to call
581 * this function to swap these settings during a failover.
582 */
583static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 struct slave *old_active)
585{
586 if (old_active) {
587 if (bond->dev->flags & IFF_PROMISC)
588 dev_set_promiscuity(old_active->dev, -1);
589
590 if (bond->dev->flags & IFF_ALLMULTI)
591 dev_set_allmulti(old_active->dev, -1);
592
593 bond_hw_addr_flush(bond->dev, old_active->dev);
594 }
595
596 if (new_active) {
597 /* FIXME: Signal errors upstream. */
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(new_active->dev, 1);
600
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(new_active->dev, 1);
603
604 netif_addr_lock_bh(bond->dev);
605 dev_uc_sync(new_active->dev, bond->dev);
606 dev_mc_sync(new_active->dev, bond->dev);
607 netif_addr_unlock_bh(bond->dev);
608 }
609}
610
611/**
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
615 *
616 * Should be called with RTNL held.
617 */
618static void bond_set_dev_addr(struct net_device *bond_dev,
619 struct net_device *slave_dev)
620{
621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
622 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
626}
627
628static struct slave *bond_get_old_active(struct bonding *bond,
629 struct slave *new_active)
630{
631 struct slave *slave;
632 struct list_head *iter;
633
634 bond_for_each_slave(bond, slave, iter) {
635 if (slave == new_active)
636 continue;
637
638 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
639 return slave;
640 }
641
642 return NULL;
643}
644
645/* bond_do_fail_over_mac
646 *
647 * Perform special MAC address swapping for fail_over_mac settings
648 *
649 * Called with RTNL
650 */
651static void bond_do_fail_over_mac(struct bonding *bond,
652 struct slave *new_active,
653 struct slave *old_active)
654{
655 u8 tmp_mac[ETH_ALEN];
656 struct sockaddr saddr;
657 int rv;
658
659 switch (bond->params.fail_over_mac) {
660 case BOND_FOM_ACTIVE:
661 if (new_active)
662 bond_set_dev_addr(bond->dev, new_active->dev);
663 break;
664 case BOND_FOM_FOLLOW:
665 /* if new_active && old_active, swap them
666 * if just old_active, do nothing (going to no active slave)
667 * if just new_active, set new_active to bond's MAC
668 */
669 if (!new_active)
670 return;
671
672 if (!old_active)
673 old_active = bond_get_old_active(bond, new_active);
674
675 if (old_active) {
676 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
677 ether_addr_copy(saddr.sa_data,
678 old_active->dev->dev_addr);
679 saddr.sa_family = new_active->dev->type;
680 } else {
681 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
682 saddr.sa_family = bond->dev->type;
683 }
684
685 rv = dev_set_mac_address(new_active->dev, &saddr);
686 if (rv) {
687 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
688 -rv, new_active->dev->name);
689 goto out;
690 }
691
692 if (!old_active)
693 goto out;
694
695 ether_addr_copy(saddr.sa_data, tmp_mac);
696 saddr.sa_family = old_active->dev->type;
697
698 rv = dev_set_mac_address(old_active->dev, &saddr);
699 if (rv)
700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
702out:
703 break;
704 default:
705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
707 break;
708 }
709
710}
711
712static struct slave *bond_choose_primary_or_current(struct bonding *bond)
713{
714 struct slave *prim = rtnl_dereference(bond->primary_slave);
715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
716
717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
719 return NULL;
720 return curr;
721 }
722
723 if (bond->force_primary) {
724 bond->force_primary = false;
725 return prim;
726 }
727
728 if (!curr || curr->link != BOND_LINK_UP)
729 return prim;
730
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
734 return prim;
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
737 return curr;
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
739 return curr;
740 return prim;
741 case BOND_PRI_RESELECT_FAILURE:
742 return curr;
743 default:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
746 return curr;
747 }
748}
749
750/**
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
753 */
754static struct slave *bond_find_best_slave(struct bonding *bond)
755{
756 struct slave *slave, *bestslave = NULL;
757 struct list_head *iter;
758 int mintime = bond->params.updelay;
759
760 slave = bond_choose_primary_or_current(bond);
761 if (slave)
762 return slave;
763
764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
766 return slave;
767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 slave->delay < mintime) {
769 mintime = slave->delay;
770 bestslave = slave;
771 }
772 }
773
774 return bestslave;
775}
776
777static bool bond_should_notify_peers(struct bonding *bond)
778{
779 struct slave *slave;
780
781 rcu_read_lock();
782 slave = rcu_dereference(bond->curr_active_slave);
783 rcu_read_unlock();
784
785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
787
788 if (!slave || !bond->send_peer_notif ||
789 !netif_carrier_ok(bond->dev) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
791 return false;
792
793 return true;
794}
795
796/**
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
800 *
801 * Set the new slave to the bond's settings and unset them on the old
802 * curr_active_slave.
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
804 *
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
808 *
809 * Caller must hold RTNL.
810 */
811void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
812{
813 struct slave *old_active;
814
815 ASSERT_RTNL();
816
817 old_active = rtnl_dereference(bond->curr_active_slave);
818
819 if (old_active == new_active)
820 return;
821
822 if (new_active) {
823 new_active->last_link_up = jiffies;
824
825 if (new_active->link == BOND_LINK_BACK) {
826 if (bond_uses_primary(bond)) {
827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
830 }
831
832 new_active->delay = 0;
833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
835
836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
838
839 if (bond_is_lb(bond))
840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
841 } else {
842 if (bond_uses_primary(bond)) {
843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
845 }
846 }
847 }
848
849 if (bond_uses_primary(bond))
850 bond_hw_addr_swap(bond, new_active, old_active);
851
852 if (bond_is_lb(bond)) {
853 bond_alb_handle_active_change(bond, new_active);
854 if (old_active)
855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
857 if (new_active)
858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
860 } else {
861 rcu_assign_pointer(bond->curr_active_slave, new_active);
862 }
863
864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
865 if (old_active)
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
868
869 if (new_active) {
870 bool should_notify_peers = false;
871
872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
874
875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
877 old_active);
878
879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
884 }
885
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 if (should_notify_peers)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
889 bond->dev);
890 }
891 }
892
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
897 */
898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 ((bond_uses_primary(bond) && new_active) ||
900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 bond->igmp_retrans = bond->params.resend_igmp;
902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
903 }
904}
905
906/**
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
909 *
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
914 *
915 * Caller must hold RTNL.
916 */
917void bond_select_active_slave(struct bonding *bond)
918{
919 struct slave *best_slave;
920 int rv;
921
922 ASSERT_RTNL();
923
924 best_slave = bond_find_best_slave(bond);
925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 bond_change_active_slave(bond, best_slave);
927 rv = bond_set_carrier(bond);
928 if (!rv)
929 return;
930
931 if (netif_carrier_ok(bond->dev))
932 netdev_info(bond->dev, "first active interface up!\n");
933 else
934 netdev_info(bond->dev, "now running without any active interface!\n");
935 }
936}
937
938#ifdef CONFIG_NET_POLL_CONTROLLER
939static inline int slave_enable_netpoll(struct slave *slave)
940{
941 struct netpoll *np;
942 int err = 0;
943
944 np = kzalloc(sizeof(*np), GFP_KERNEL);
945 err = -ENOMEM;
946 if (!np)
947 goto out;
948
949 err = __netpoll_setup(np, slave->dev);
950 if (err) {
951 kfree(np);
952 goto out;
953 }
954 slave->np = np;
955out:
956 return err;
957}
958static inline void slave_disable_netpoll(struct slave *slave)
959{
960 struct netpoll *np = slave->np;
961
962 if (!np)
963 return;
964
965 slave->np = NULL;
966 __netpoll_free_async(np);
967}
968
969static void bond_poll_controller(struct net_device *bond_dev)
970{
971 struct bonding *bond = netdev_priv(bond_dev);
972 struct slave *slave = NULL;
973 struct list_head *iter;
974 struct ad_info ad_info;
975 struct netpoll_info *ni;
976 const struct net_device_ops *ops;
977
978 if (BOND_MODE(bond) == BOND_MODE_8023AD)
979 if (bond_3ad_get_active_agg_info(bond, &ad_info))
980 return;
981
982 bond_for_each_slave_rcu(bond, slave, iter) {
983 ops = slave->dev->netdev_ops;
984 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
985 continue;
986
987 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
988 struct aggregator *agg =
989 SLAVE_AD_INFO(slave)->port.aggregator;
990
991 if (agg &&
992 agg->aggregator_identifier != ad_info.aggregator_id)
993 continue;
994 }
995
996 ni = rcu_dereference_bh(slave->dev->npinfo);
997 if (down_trylock(&ni->dev_lock))
998 continue;
999 ops->ndo_poll_controller(slave->dev);
1000 up(&ni->dev_lock);
1001 }
1002}
1003
1004static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005{
1006 struct bonding *bond = netdev_priv(bond_dev);
1007 struct list_head *iter;
1008 struct slave *slave;
1009
1010 bond_for_each_slave(bond, slave, iter)
1011 if (bond_slave_is_up(slave))
1012 slave_disable_netpoll(slave);
1013}
1014
1015static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016{
1017 struct bonding *bond = netdev_priv(dev);
1018 struct list_head *iter;
1019 struct slave *slave;
1020 int err = 0;
1021
1022 bond_for_each_slave(bond, slave, iter) {
1023 err = slave_enable_netpoll(slave);
1024 if (err) {
1025 bond_netpoll_cleanup(dev);
1026 break;
1027 }
1028 }
1029 return err;
1030}
1031#else
1032static inline int slave_enable_netpoll(struct slave *slave)
1033{
1034 return 0;
1035}
1036static inline void slave_disable_netpoll(struct slave *slave)
1037{
1038}
1039static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040{
1041}
1042#endif
1043
1044/*---------------------------------- IOCTL ----------------------------------*/
1045
1046static netdev_features_t bond_fix_features(struct net_device *dev,
1047 netdev_features_t features)
1048{
1049 struct bonding *bond = netdev_priv(dev);
1050 struct list_head *iter;
1051 netdev_features_t mask;
1052 struct slave *slave;
1053
1054 mask = features;
1055
1056 features &= ~NETIF_F_ONE_FOR_ALL;
1057 features |= NETIF_F_ALL_FOR_ALL;
1058
1059 bond_for_each_slave(bond, slave, iter) {
1060 features = netdev_increment_features(features,
1061 slave->dev->features,
1062 mask);
1063 }
1064 features = netdev_add_tso_features(features, mask);
1065
1066 return features;
1067}
1068
1069#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071 NETIF_F_HIGHDMA | NETIF_F_LRO)
1072
1073#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075
1076static void bond_compute_features(struct bonding *bond)
1077{
1078 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079 IFF_XMIT_DST_RELEASE_PERM;
1080 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081 netdev_features_t enc_features = BOND_ENC_FEATURES;
1082 struct net_device *bond_dev = bond->dev;
1083 struct list_head *iter;
1084 struct slave *slave;
1085 unsigned short max_hard_header_len = ETH_HLEN;
1086 unsigned int gso_max_size = GSO_MAX_SIZE;
1087 u16 gso_max_segs = GSO_MAX_SEGS;
1088
1089 if (!bond_has_slaves(bond))
1090 goto done;
1091 vlan_features &= NETIF_F_ALL_FOR_ALL;
1092
1093 bond_for_each_slave(bond, slave, iter) {
1094 vlan_features = netdev_increment_features(vlan_features,
1095 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096
1097 enc_features = netdev_increment_features(enc_features,
1098 slave->dev->hw_enc_features,
1099 BOND_ENC_FEATURES);
1100 dst_release_flag &= slave->dev->priv_flags;
1101 if (slave->dev->hard_header_len > max_hard_header_len)
1102 max_hard_header_len = slave->dev->hard_header_len;
1103
1104 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106 }
1107
1108done:
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 bond_dev->hard_header_len = max_hard_header_len;
1112 bond_dev->gso_max_segs = gso_max_segs;
1113 netif_set_gso_max_size(bond_dev, gso_max_size);
1114
1115 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119
1120 netdev_change_features(bond_dev);
1121}
1122
1123static void bond_setup_by_slave(struct net_device *bond_dev,
1124 struct net_device *slave_dev)
1125{
1126 bond_dev->header_ops = slave_dev->header_ops;
1127
1128 bond_dev->type = slave_dev->type;
1129 bond_dev->hard_header_len = slave_dev->hard_header_len;
1130 bond_dev->addr_len = slave_dev->addr_len;
1131
1132 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133 slave_dev->addr_len);
1134}
1135
1136/* On bonding slaves other than the currently active slave, suppress
1137 * duplicates except for alb non-mcast/bcast.
1138 */
1139static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140 struct slave *slave,
1141 struct bonding *bond)
1142{
1143 if (bond_is_slave_inactive(slave)) {
1144 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145 skb->pkt_type != PACKET_BROADCAST &&
1146 skb->pkt_type != PACKET_MULTICAST)
1147 return false;
1148 return true;
1149 }
1150 return false;
1151}
1152
1153static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154{
1155 struct sk_buff *skb = *pskb;
1156 struct slave *slave;
1157 struct bonding *bond;
1158 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159 struct slave *);
1160 int ret = RX_HANDLER_ANOTHER;
1161
1162 skb = skb_share_check(skb, GFP_ATOMIC);
1163 if (unlikely(!skb))
1164 return RX_HANDLER_CONSUMED;
1165
1166 *pskb = skb;
1167
1168 slave = bond_slave_get_rcu(skb->dev);
1169 bond = slave->bond;
1170
1171 recv_probe = ACCESS_ONCE(bond->recv_probe);
1172 if (recv_probe) {
1173 ret = recv_probe(skb, bond, slave);
1174 if (ret == RX_HANDLER_CONSUMED) {
1175 consume_skb(skb);
1176 return ret;
1177 }
1178 }
1179
1180 if (bond_should_deliver_exact_match(skb, slave, bond))
1181 return RX_HANDLER_EXACT;
1182
1183 skb->dev = bond->dev;
1184
1185 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187 skb->pkt_type == PACKET_HOST) {
1188
1189 if (unlikely(skb_cow_head(skb,
1190 skb->data - skb_mac_header(skb)))) {
1191 kfree_skb(skb);
1192 return RX_HANDLER_CONSUMED;
1193 }
1194 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1195 }
1196
1197 return ret;
1198}
1199
1200static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1201{
1202 switch (BOND_MODE(bond)) {
1203 case BOND_MODE_ROUNDROBIN:
1204 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205 case BOND_MODE_ACTIVEBACKUP:
1206 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207 case BOND_MODE_BROADCAST:
1208 return NETDEV_LAG_TX_TYPE_BROADCAST;
1209 case BOND_MODE_XOR:
1210 case BOND_MODE_8023AD:
1211 return NETDEV_LAG_TX_TYPE_HASH;
1212 default:
1213 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1214 }
1215}
1216
1217static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1218{
1219 struct netdev_lag_upper_info lag_upper_info;
1220 int err;
1221
1222 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1224 &lag_upper_info);
1225 if (err)
1226 return err;
1227 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1228 return 0;
1229}
1230
1231static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232{
1233 netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 slave->dev->flags &= ~IFF_SLAVE;
1235 rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1236}
1237
1238static struct slave *bond_alloc_slave(struct bonding *bond)
1239{
1240 struct slave *slave = NULL;
1241
1242 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1243 if (!slave)
1244 return NULL;
1245
1246 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1248 GFP_KERNEL);
1249 if (!SLAVE_AD_INFO(slave)) {
1250 kfree(slave);
1251 return NULL;
1252 }
1253 }
1254 return slave;
1255}
1256
1257static void bond_free_slave(struct slave *slave)
1258{
1259 struct bonding *bond = bond_get_bond_by_slave(slave);
1260
1261 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262 kfree(SLAVE_AD_INFO(slave));
1263
1264 kfree(slave);
1265}
1266
1267static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1268{
1269 info->bond_mode = BOND_MODE(bond);
1270 info->miimon = bond->params.miimon;
1271 info->num_slaves = bond->slave_cnt;
1272}
1273
1274static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1275{
1276 strcpy(info->slave_name, slave->dev->name);
1277 info->link = slave->link;
1278 info->state = bond_slave_state(slave);
1279 info->link_failure_count = slave->link_failure_count;
1280}
1281
1282static void bond_netdev_notify(struct net_device *dev,
1283 struct netdev_bonding_info *info)
1284{
1285 rtnl_lock();
1286 netdev_bonding_info_change(dev, info);
1287 rtnl_unlock();
1288}
1289
1290static void bond_netdev_notify_work(struct work_struct *_work)
1291{
1292 struct netdev_notify_work *w =
1293 container_of(_work, struct netdev_notify_work, work.work);
1294
1295 bond_netdev_notify(w->dev, &w->bonding_info);
1296 dev_put(w->dev);
1297 kfree(w);
1298}
1299
1300void bond_queue_slave_event(struct slave *slave)
1301{
1302 struct bonding *bond = slave->bond;
1303 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1304
1305 if (!nnw)
1306 return;
1307
1308 dev_hold(slave->dev);
1309 nnw->dev = slave->dev;
1310 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1313
1314 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1315}
1316
1317void bond_lower_state_changed(struct slave *slave)
1318{
1319 struct netdev_lag_lower_state_info info;
1320
1321 info.link_up = slave->link == BOND_LINK_UP ||
1322 slave->link == BOND_LINK_FAIL;
1323 info.tx_enabled = bond_is_active_slave(slave);
1324 netdev_lower_state_changed(slave->dev, &info);
1325}
1326
1327/* enslave device <slave> to bond device <master> */
1328int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1329{
1330 struct bonding *bond = netdev_priv(bond_dev);
1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 struct slave *new_slave = NULL, *prev_slave;
1333 struct sockaddr addr;
1334 int link_reporting;
1335 int res = 0, i;
1336
1337 if (!bond->params.use_carrier &&
1338 slave_dev->ethtool_ops->get_link == NULL &&
1339 slave_ops->ndo_do_ioctl == NULL) {
1340 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341 slave_dev->name);
1342 }
1343
1344 /* already in-use? */
1345 if (netdev_is_rx_handler_busy(slave_dev)) {
1346 netdev_err(bond_dev,
1347 "Error: Device is in use and cannot be enslaved\n");
1348 return -EBUSY;
1349 }
1350
1351 if (bond_dev == slave_dev) {
1352 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1353 return -EPERM;
1354 }
1355
1356 /* vlan challenged mutual exclusion */
1357 /* no need to lock since we're protected by rtnl_lock */
1358 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1359 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1360 slave_dev->name);
1361 if (vlan_uses_dev(bond_dev)) {
1362 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1363 slave_dev->name, bond_dev->name);
1364 return -EPERM;
1365 } else {
1366 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1367 slave_dev->name, slave_dev->name,
1368 bond_dev->name);
1369 }
1370 } else {
1371 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1372 slave_dev->name);
1373 }
1374
1375 /* Old ifenslave binaries are no longer supported. These can
1376 * be identified with moderate accuracy by the state of the slave:
1377 * the current ifenslave will set the interface down prior to
1378 * enslaving it; the old ifenslave will not.
1379 */
1380 if (slave_dev->flags & IFF_UP) {
1381 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1382 slave_dev->name);
1383 return -EPERM;
1384 }
1385
1386 /* set bonding device ether type by slave - bonding netdevices are
1387 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1388 * there is a need to override some of the type dependent attribs/funcs.
1389 *
1390 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1391 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1392 */
1393 if (!bond_has_slaves(bond)) {
1394 if (bond_dev->type != slave_dev->type) {
1395 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1396 bond_dev->type, slave_dev->type);
1397
1398 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1399 bond_dev);
1400 res = notifier_to_errno(res);
1401 if (res) {
1402 netdev_err(bond_dev, "refused to change device type\n");
1403 return -EBUSY;
1404 }
1405
1406 /* Flush unicast and multicast addresses */
1407 dev_uc_flush(bond_dev);
1408 dev_mc_flush(bond_dev);
1409
1410 if (slave_dev->type != ARPHRD_ETHER)
1411 bond_setup_by_slave(bond_dev, slave_dev);
1412 else {
1413 ether_setup(bond_dev);
1414 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1415 }
1416
1417 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1418 bond_dev);
1419 }
1420 } else if (bond_dev->type != slave_dev->type) {
1421 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1422 slave_dev->name, slave_dev->type, bond_dev->type);
1423 return -EINVAL;
1424 }
1425
1426 if (slave_dev->type == ARPHRD_INFINIBAND &&
1427 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1428 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1429 slave_dev->type);
1430 res = -EOPNOTSUPP;
1431 goto err_undo_flags;
1432 }
1433
1434 if (!slave_ops->ndo_set_mac_address ||
1435 slave_dev->type == ARPHRD_INFINIBAND) {
1436 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1437 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1438 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1439 if (!bond_has_slaves(bond)) {
1440 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1441 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1442 } else {
1443 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1444 res = -EOPNOTSUPP;
1445 goto err_undo_flags;
1446 }
1447 }
1448 }
1449
1450 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1451
1452 /* If this is the first slave, then we need to set the master's hardware
1453 * address to be the same as the slave's.
1454 */
1455 if (!bond_has_slaves(bond) &&
1456 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1457 bond_set_dev_addr(bond->dev, slave_dev);
1458
1459 new_slave = bond_alloc_slave(bond);
1460 if (!new_slave) {
1461 res = -ENOMEM;
1462 goto err_undo_flags;
1463 }
1464
1465 new_slave->bond = bond;
1466 new_slave->dev = slave_dev;
1467 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1468 * is set via sysfs or module option if desired.
1469 */
1470 new_slave->queue_id = 0;
1471
1472 /* Save slave's original mtu and then set it to match the bond */
1473 new_slave->original_mtu = slave_dev->mtu;
1474 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1475 if (res) {
1476 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1477 goto err_free;
1478 }
1479
1480 /* Save slave's original ("permanent") mac address for modes
1481 * that need it, and for restoring it upon release, and then
1482 * set it to the master's address
1483 */
1484 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1485
1486 if (!bond->params.fail_over_mac ||
1487 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1488 /* Set slave to master's mac address. The application already
1489 * set the master's mac address to that of the first slave
1490 */
1491 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1492 addr.sa_family = slave_dev->type;
1493 res = dev_set_mac_address(slave_dev, &addr);
1494 if (res) {
1495 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1496 goto err_restore_mtu;
1497 }
1498 }
1499
1500 /* set slave flag before open to prevent IPv6 addrconf */
1501 slave_dev->flags |= IFF_SLAVE;
1502
1503 /* open the slave since the application closed it */
1504 res = dev_open(slave_dev);
1505 if (res) {
1506 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1507 goto err_restore_mac;
1508 }
1509
1510 slave_dev->priv_flags |= IFF_BONDING;
1511 /* initialize slave stats */
1512 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1513
1514 if (bond_is_lb(bond)) {
1515 /* bond_alb_init_slave() must be called before all other stages since
1516 * it might fail and we do not want to have to undo everything
1517 */
1518 res = bond_alb_init_slave(bond, new_slave);
1519 if (res)
1520 goto err_close;
1521 }
1522
1523 /* If the mode uses primary, then the following is handled by
1524 * bond_change_active_slave().
1525 */
1526 if (!bond_uses_primary(bond)) {
1527 /* set promiscuity level to new slave */
1528 if (bond_dev->flags & IFF_PROMISC) {
1529 res = dev_set_promiscuity(slave_dev, 1);
1530 if (res)
1531 goto err_close;
1532 }
1533
1534 /* set allmulti level to new slave */
1535 if (bond_dev->flags & IFF_ALLMULTI) {
1536 res = dev_set_allmulti(slave_dev, 1);
1537 if (res)
1538 goto err_close;
1539 }
1540
1541 netif_addr_lock_bh(bond_dev);
1542
1543 dev_mc_sync_multiple(slave_dev, bond_dev);
1544 dev_uc_sync_multiple(slave_dev, bond_dev);
1545
1546 netif_addr_unlock_bh(bond_dev);
1547 }
1548
1549 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1550 /* add lacpdu mc addr to mc list */
1551 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1552
1553 dev_mc_add(slave_dev, lacpdu_multicast);
1554 }
1555
1556 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1557 if (res) {
1558 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1559 slave_dev->name);
1560 goto err_close;
1561 }
1562
1563 prev_slave = bond_last_slave(bond);
1564
1565 new_slave->delay = 0;
1566 new_slave->link_failure_count = 0;
1567
1568 bond_update_speed_duplex(new_slave);
1569
1570 new_slave->last_rx = jiffies -
1571 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1572 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1573 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1574
1575 if (bond->params.miimon && !bond->params.use_carrier) {
1576 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1577
1578 if ((link_reporting == -1) && !bond->params.arp_interval) {
1579 /* miimon is set but a bonded network driver
1580 * does not support ETHTOOL/MII and
1581 * arp_interval is not set. Note: if
1582 * use_carrier is enabled, we will never go
1583 * here (because netif_carrier is always
1584 * supported); thus, we don't need to change
1585 * the messages for netif_carrier.
1586 */
1587 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1588 slave_dev->name);
1589 } else if (link_reporting == -1) {
1590 /* unable get link status using mii/ethtool */
1591 netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1592 slave_dev->name);
1593 }
1594 }
1595
1596 /* check for initial state */
1597 new_slave->link = BOND_LINK_NOCHANGE;
1598 if (bond->params.miimon) {
1599 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1600 if (bond->params.updelay) {
1601 bond_set_slave_link_state(new_slave,
1602 BOND_LINK_BACK,
1603 BOND_SLAVE_NOTIFY_NOW);
1604 new_slave->delay = bond->params.updelay;
1605 } else {
1606 bond_set_slave_link_state(new_slave,
1607 BOND_LINK_UP,
1608 BOND_SLAVE_NOTIFY_NOW);
1609 }
1610 } else {
1611 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1612 BOND_SLAVE_NOTIFY_NOW);
1613 }
1614 } else if (bond->params.arp_interval) {
1615 bond_set_slave_link_state(new_slave,
1616 (netif_carrier_ok(slave_dev) ?
1617 BOND_LINK_UP : BOND_LINK_DOWN),
1618 BOND_SLAVE_NOTIFY_NOW);
1619 } else {
1620 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1621 BOND_SLAVE_NOTIFY_NOW);
1622 }
1623
1624 if (new_slave->link != BOND_LINK_DOWN)
1625 new_slave->last_link_up = jiffies;
1626 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1627 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1628 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1629
1630 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1631 /* if there is a primary slave, remember it */
1632 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1633 rcu_assign_pointer(bond->primary_slave, new_slave);
1634 bond->force_primary = true;
1635 }
1636 }
1637
1638 switch (BOND_MODE(bond)) {
1639 case BOND_MODE_ACTIVEBACKUP:
1640 bond_set_slave_inactive_flags(new_slave,
1641 BOND_SLAVE_NOTIFY_NOW);
1642 break;
1643 case BOND_MODE_8023AD:
1644 /* in 802.3ad mode, the internal mechanism
1645 * will activate the slaves in the selected
1646 * aggregator
1647 */
1648 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1649 /* if this is the first slave */
1650 if (!prev_slave) {
1651 SLAVE_AD_INFO(new_slave)->id = 1;
1652 /* Initialize AD with the number of times that the AD timer is called in 1 second
1653 * can be called only after the mac address of the bond is set
1654 */
1655 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1656 } else {
1657 SLAVE_AD_INFO(new_slave)->id =
1658 SLAVE_AD_INFO(prev_slave)->id + 1;
1659 }
1660
1661 bond_3ad_bind_slave(new_slave);
1662 break;
1663 case BOND_MODE_TLB:
1664 case BOND_MODE_ALB:
1665 bond_set_active_slave(new_slave);
1666 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1667 break;
1668 default:
1669 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1670
1671 /* always active in trunk mode */
1672 bond_set_active_slave(new_slave);
1673
1674 /* In trunking mode there is little meaning to curr_active_slave
1675 * anyway (it holds no special properties of the bond device),
1676 * so we can change it without calling change_active_interface()
1677 */
1678 if (!rcu_access_pointer(bond->curr_active_slave) &&
1679 new_slave->link == BOND_LINK_UP)
1680 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1681
1682 break;
1683 } /* switch(bond_mode) */
1684
1685#ifdef CONFIG_NET_POLL_CONTROLLER
1686 slave_dev->npinfo = bond->dev->npinfo;
1687 if (slave_dev->npinfo) {
1688 if (slave_enable_netpoll(new_slave)) {
1689 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1690 res = -EBUSY;
1691 goto err_detach;
1692 }
1693 }
1694#endif
1695
1696 if (!(bond_dev->features & NETIF_F_LRO))
1697 dev_disable_lro(slave_dev);
1698
1699 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1700 new_slave);
1701 if (res) {
1702 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1703 goto err_detach;
1704 }
1705
1706 res = bond_master_upper_dev_link(bond, new_slave);
1707 if (res) {
1708 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1709 goto err_unregister;
1710 }
1711
1712 res = bond_sysfs_slave_add(new_slave);
1713 if (res) {
1714 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1715 goto err_upper_unlink;
1716 }
1717
1718 bond->slave_cnt++;
1719 bond_compute_features(bond);
1720 bond_set_carrier(bond);
1721
1722 if (bond_uses_primary(bond)) {
1723 block_netpoll_tx();
1724 bond_select_active_slave(bond);
1725 unblock_netpoll_tx();
1726 }
1727
1728 if (bond_mode_uses_xmit_hash(bond))
1729 bond_update_slave_arr(bond, NULL);
1730
1731 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1732 slave_dev->name,
1733 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1734 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1735
1736 /* enslave is successful */
1737 bond_queue_slave_event(new_slave);
1738 return 0;
1739
1740/* Undo stages on error */
1741err_upper_unlink:
1742 bond_upper_dev_unlink(bond, new_slave);
1743
1744err_unregister:
1745 netdev_rx_handler_unregister(slave_dev);
1746
1747err_detach:
1748 if (!bond_uses_primary(bond))
1749 bond_hw_addr_flush(bond_dev, slave_dev);
1750
1751 vlan_vids_del_by_dev(slave_dev, bond_dev);
1752 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1753 RCU_INIT_POINTER(bond->primary_slave, NULL);
1754 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1755 block_netpoll_tx();
1756 bond_change_active_slave(bond, NULL);
1757 bond_select_active_slave(bond);
1758 unblock_netpoll_tx();
1759 }
1760 /* either primary_slave or curr_active_slave might've changed */
1761 synchronize_rcu();
1762 slave_disable_netpoll(new_slave);
1763
1764err_close:
1765 slave_dev->priv_flags &= ~IFF_BONDING;
1766 dev_close(slave_dev);
1767
1768err_restore_mac:
1769 slave_dev->flags &= ~IFF_SLAVE;
1770 if (!bond->params.fail_over_mac ||
1771 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1772 /* XXX TODO - fom follow mode needs to change master's
1773 * MAC if this slave's MAC is in use by the bond, or at
1774 * least print a warning.
1775 */
1776 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1777 addr.sa_family = slave_dev->type;
1778 dev_set_mac_address(slave_dev, &addr);
1779 }
1780
1781err_restore_mtu:
1782 dev_set_mtu(slave_dev, new_slave->original_mtu);
1783
1784err_free:
1785 bond_free_slave(new_slave);
1786
1787err_undo_flags:
1788 /* Enslave of first slave has failed and we need to fix master's mac */
1789 if (!bond_has_slaves(bond)) {
1790 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1791 slave_dev->dev_addr))
1792 eth_hw_addr_random(bond_dev);
1793 if (bond_dev->type != ARPHRD_ETHER) {
1794 dev_close(bond_dev);
1795 ether_setup(bond_dev);
1796 bond_dev->flags |= IFF_MASTER;
1797 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1798 }
1799 }
1800
1801 return res;
1802}
1803
1804/* Try to release the slave device <slave> from the bond device <master>
1805 * It is legal to access curr_active_slave without a lock because all the function
1806 * is RTNL-locked. If "all" is true it means that the function is being called
1807 * while destroying a bond interface and all slaves are being released.
1808 *
1809 * The rules for slave state should be:
1810 * for Active/Backup:
1811 * Active stays on all backups go down
1812 * for Bonded connections:
1813 * The first up interface should be left on and all others downed.
1814 */
1815static int __bond_release_one(struct net_device *bond_dev,
1816 struct net_device *slave_dev,
1817 bool all)
1818{
1819 struct bonding *bond = netdev_priv(bond_dev);
1820 struct slave *slave, *oldcurrent;
1821 struct sockaddr addr;
1822 int old_flags = bond_dev->flags;
1823 netdev_features_t old_features = bond_dev->features;
1824
1825 /* slave is not a slave or master is not master of this slave */
1826 if (!(slave_dev->flags & IFF_SLAVE) ||
1827 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1828 netdev_dbg(bond_dev, "cannot release %s\n",
1829 slave_dev->name);
1830 return -EINVAL;
1831 }
1832
1833 block_netpoll_tx();
1834
1835 slave = bond_get_slave_by_dev(bond, slave_dev);
1836 if (!slave) {
1837 /* not a slave of this bond */
1838 netdev_info(bond_dev, "%s not enslaved\n",
1839 slave_dev->name);
1840 unblock_netpoll_tx();
1841 return -EINVAL;
1842 }
1843
1844 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1845
1846 bond_sysfs_slave_del(slave);
1847
1848 /* recompute stats just before removing the slave */
1849 bond_get_stats(bond->dev, &bond->bond_stats);
1850
1851 bond_upper_dev_unlink(bond, slave);
1852 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1853 * for this slave anymore.
1854 */
1855 netdev_rx_handler_unregister(slave_dev);
1856
1857 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1858 bond_3ad_unbind_slave(slave);
1859
1860 if (bond_mode_uses_xmit_hash(bond))
1861 bond_update_slave_arr(bond, slave);
1862
1863 netdev_info(bond_dev, "Releasing %s interface %s\n",
1864 bond_is_active_slave(slave) ? "active" : "backup",
1865 slave_dev->name);
1866
1867 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1868
1869 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1870
1871 if (!all && (!bond->params.fail_over_mac ||
1872 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1873 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1874 bond_has_slaves(bond))
1875 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1876 slave_dev->name, slave->perm_hwaddr,
1877 bond_dev->name, slave_dev->name);
1878 }
1879
1880 if (rtnl_dereference(bond->primary_slave) == slave)
1881 RCU_INIT_POINTER(bond->primary_slave, NULL);
1882
1883 if (oldcurrent == slave)
1884 bond_change_active_slave(bond, NULL);
1885
1886 if (bond_is_lb(bond)) {
1887 /* Must be called only after the slave has been
1888 * detached from the list and the curr_active_slave
1889 * has been cleared (if our_slave == old_current),
1890 * but before a new active slave is selected.
1891 */
1892 bond_alb_deinit_slave(bond, slave);
1893 }
1894
1895 if (all) {
1896 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1897 } else if (oldcurrent == slave) {
1898 /* Note that we hold RTNL over this sequence, so there
1899 * is no concern that another slave add/remove event
1900 * will interfere.
1901 */
1902 bond_select_active_slave(bond);
1903 }
1904
1905 if (!bond_has_slaves(bond)) {
1906 bond_set_carrier(bond);
1907 eth_hw_addr_random(bond_dev);
1908 }
1909
1910 unblock_netpoll_tx();
1911 synchronize_rcu();
1912 bond->slave_cnt--;
1913
1914 if (!bond_has_slaves(bond)) {
1915 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1916 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1917 }
1918
1919 bond_compute_features(bond);
1920 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1921 (old_features & NETIF_F_VLAN_CHALLENGED))
1922 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1923 slave_dev->name, bond_dev->name);
1924
1925 vlan_vids_del_by_dev(slave_dev, bond_dev);
1926
1927 /* If the mode uses primary, then this case was handled above by
1928 * bond_change_active_slave(..., NULL)
1929 */
1930 if (!bond_uses_primary(bond)) {
1931 /* unset promiscuity level from slave
1932 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1933 * of the IFF_PROMISC flag in the bond_dev, but we need the
1934 * value of that flag before that change, as that was the value
1935 * when this slave was attached, so we cache at the start of the
1936 * function and use it here. Same goes for ALLMULTI below
1937 */
1938 if (old_flags & IFF_PROMISC)
1939 dev_set_promiscuity(slave_dev, -1);
1940
1941 /* unset allmulti level from slave */
1942 if (old_flags & IFF_ALLMULTI)
1943 dev_set_allmulti(slave_dev, -1);
1944
1945 bond_hw_addr_flush(bond_dev, slave_dev);
1946 }
1947
1948 slave_disable_netpoll(slave);
1949
1950 /* close slave before restoring its mac address */
1951 dev_close(slave_dev);
1952
1953 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1954 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1955 /* restore original ("permanent") mac address */
1956 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1957 addr.sa_family = slave_dev->type;
1958 dev_set_mac_address(slave_dev, &addr);
1959 }
1960
1961 dev_set_mtu(slave_dev, slave->original_mtu);
1962
1963 slave_dev->priv_flags &= ~IFF_BONDING;
1964
1965 bond_free_slave(slave);
1966
1967 return 0;
1968}
1969
1970/* A wrapper used because of ndo_del_link */
1971int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1972{
1973 return __bond_release_one(bond_dev, slave_dev, false);
1974}
1975
1976/* First release a slave and then destroy the bond if no more slaves are left.
1977 * Must be under rtnl_lock when this function is called.
1978 */
1979static int bond_release_and_destroy(struct net_device *bond_dev,
1980 struct net_device *slave_dev)
1981{
1982 struct bonding *bond = netdev_priv(bond_dev);
1983 int ret;
1984
1985 ret = bond_release(bond_dev, slave_dev);
1986 if (ret == 0 && !bond_has_slaves(bond)) {
1987 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1988 netdev_info(bond_dev, "Destroying bond %s\n",
1989 bond_dev->name);
1990 bond_remove_proc_entry(bond);
1991 unregister_netdevice(bond_dev);
1992 }
1993 return ret;
1994}
1995
1996static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1997{
1998 struct bonding *bond = netdev_priv(bond_dev);
1999 bond_fill_ifbond(bond, info);
2000 return 0;
2001}
2002
2003static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2004{
2005 struct bonding *bond = netdev_priv(bond_dev);
2006 struct list_head *iter;
2007 int i = 0, res = -ENODEV;
2008 struct slave *slave;
2009
2010 bond_for_each_slave(bond, slave, iter) {
2011 if (i++ == (int)info->slave_id) {
2012 res = 0;
2013 bond_fill_ifslave(slave, info);
2014 break;
2015 }
2016 }
2017
2018 return res;
2019}
2020
2021/*-------------------------------- Monitoring -------------------------------*/
2022
2023/* called with rcu_read_lock() */
2024static int bond_miimon_inspect(struct bonding *bond)
2025{
2026 int link_state, commit = 0;
2027 struct list_head *iter;
2028 struct slave *slave;
2029 bool ignore_updelay;
2030
2031 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2032
2033 bond_for_each_slave_rcu(bond, slave, iter) {
2034 slave->new_link = BOND_LINK_NOCHANGE;
2035
2036 link_state = bond_check_dev_link(bond, slave->dev, 0);
2037
2038 switch (slave->link) {
2039 case BOND_LINK_UP:
2040 if (link_state)
2041 continue;
2042
2043 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2044 BOND_SLAVE_NOTIFY_LATER);
2045 slave->delay = bond->params.downdelay;
2046 if (slave->delay) {
2047 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2048 (BOND_MODE(bond) ==
2049 BOND_MODE_ACTIVEBACKUP) ?
2050 (bond_is_active_slave(slave) ?
2051 "active " : "backup ") : "",
2052 slave->dev->name,
2053 bond->params.downdelay * bond->params.miimon);
2054 }
2055 /*FALLTHRU*/
2056 case BOND_LINK_FAIL:
2057 if (link_state) {
2058 /* recovered before downdelay expired */
2059 bond_set_slave_link_state(slave, BOND_LINK_UP,
2060 BOND_SLAVE_NOTIFY_LATER);
2061 slave->last_link_up = jiffies;
2062 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2063 (bond->params.downdelay - slave->delay) *
2064 bond->params.miimon,
2065 slave->dev->name);
2066 continue;
2067 }
2068
2069 if (slave->delay <= 0) {
2070 slave->new_link = BOND_LINK_DOWN;
2071 commit++;
2072 continue;
2073 }
2074
2075 slave->delay--;
2076 break;
2077
2078 case BOND_LINK_DOWN:
2079 if (!link_state)
2080 continue;
2081
2082 bond_set_slave_link_state(slave, BOND_LINK_BACK,
2083 BOND_SLAVE_NOTIFY_LATER);
2084 slave->delay = bond->params.updelay;
2085
2086 if (slave->delay) {
2087 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2088 slave->dev->name,
2089 ignore_updelay ? 0 :
2090 bond->params.updelay *
2091 bond->params.miimon);
2092 }
2093 /*FALLTHRU*/
2094 case BOND_LINK_BACK:
2095 if (!link_state) {
2096 bond_set_slave_link_state(slave,
2097 BOND_LINK_DOWN,
2098 BOND_SLAVE_NOTIFY_LATER);
2099 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2100 (bond->params.updelay - slave->delay) *
2101 bond->params.miimon,
2102 slave->dev->name);
2103
2104 continue;
2105 }
2106
2107 if (ignore_updelay)
2108 slave->delay = 0;
2109
2110 if (slave->delay <= 0) {
2111 slave->new_link = BOND_LINK_UP;
2112 commit++;
2113 ignore_updelay = false;
2114 continue;
2115 }
2116
2117 slave->delay--;
2118 break;
2119 }
2120 }
2121
2122 return commit;
2123}
2124
2125static void bond_miimon_commit(struct bonding *bond)
2126{
2127 struct list_head *iter;
2128 struct slave *slave, *primary;
2129
2130 bond_for_each_slave(bond, slave, iter) {
2131 switch (slave->new_link) {
2132 case BOND_LINK_NOCHANGE:
2133 continue;
2134
2135 case BOND_LINK_UP:
2136 bond_update_speed_duplex(slave);
2137 bond_set_slave_link_state(slave, BOND_LINK_UP,
2138 BOND_SLAVE_NOTIFY_NOW);
2139 slave->last_link_up = jiffies;
2140
2141 primary = rtnl_dereference(bond->primary_slave);
2142 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2143 /* prevent it from being the active one */
2144 bond_set_backup_slave(slave);
2145 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2146 /* make it immediately active */
2147 bond_set_active_slave(slave);
2148 } else if (slave != primary) {
2149 /* prevent it from being the active one */
2150 bond_set_backup_slave(slave);
2151 }
2152
2153 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2154 slave->dev->name,
2155 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2156 slave->duplex ? "full" : "half");
2157
2158 /* notify ad that the link status has changed */
2159 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2160 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2161
2162 if (bond_is_lb(bond))
2163 bond_alb_handle_link_change(bond, slave,
2164 BOND_LINK_UP);
2165
2166 if (BOND_MODE(bond) == BOND_MODE_XOR)
2167 bond_update_slave_arr(bond, NULL);
2168
2169 if (!bond->curr_active_slave || slave == primary)
2170 goto do_failover;
2171
2172 continue;
2173
2174 case BOND_LINK_DOWN:
2175 if (slave->link_failure_count < UINT_MAX)
2176 slave->link_failure_count++;
2177
2178 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2179 BOND_SLAVE_NOTIFY_NOW);
2180
2181 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2182 BOND_MODE(bond) == BOND_MODE_8023AD)
2183 bond_set_slave_inactive_flags(slave,
2184 BOND_SLAVE_NOTIFY_NOW);
2185
2186 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2187 slave->dev->name);
2188
2189 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2190 bond_3ad_handle_link_change(slave,
2191 BOND_LINK_DOWN);
2192
2193 if (bond_is_lb(bond))
2194 bond_alb_handle_link_change(bond, slave,
2195 BOND_LINK_DOWN);
2196
2197 if (BOND_MODE(bond) == BOND_MODE_XOR)
2198 bond_update_slave_arr(bond, NULL);
2199
2200 if (slave == rcu_access_pointer(bond->curr_active_slave))
2201 goto do_failover;
2202
2203 continue;
2204
2205 default:
2206 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2207 slave->new_link, slave->dev->name);
2208 slave->new_link = BOND_LINK_NOCHANGE;
2209
2210 continue;
2211 }
2212
2213do_failover:
2214 block_netpoll_tx();
2215 bond_select_active_slave(bond);
2216 unblock_netpoll_tx();
2217 }
2218
2219 bond_set_carrier(bond);
2220}
2221
2222/* bond_mii_monitor
2223 *
2224 * Really a wrapper that splits the mii monitor into two phases: an
2225 * inspection, then (if inspection indicates something needs to be done)
2226 * an acquisition of appropriate locks followed by a commit phase to
2227 * implement whatever link state changes are indicated.
2228 */
2229static void bond_mii_monitor(struct work_struct *work)
2230{
2231 struct bonding *bond = container_of(work, struct bonding,
2232 mii_work.work);
2233 bool should_notify_peers = false;
2234 unsigned long delay;
2235
2236 delay = msecs_to_jiffies(bond->params.miimon);
2237
2238 if (!bond_has_slaves(bond))
2239 goto re_arm;
2240
2241 rcu_read_lock();
2242
2243 should_notify_peers = bond_should_notify_peers(bond);
2244
2245 if (bond_miimon_inspect(bond)) {
2246 rcu_read_unlock();
2247
2248 /* Race avoidance with bond_close cancel of workqueue */
2249 if (!rtnl_trylock()) {
2250 delay = 1;
2251 should_notify_peers = false;
2252 goto re_arm;
2253 }
2254
2255 bond_miimon_commit(bond);
2256
2257 rtnl_unlock(); /* might sleep, hold no other locks */
2258 } else
2259 rcu_read_unlock();
2260
2261re_arm:
2262 if (bond->params.miimon)
2263 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2264
2265 if (should_notify_peers) {
2266 if (!rtnl_trylock())
2267 return;
2268 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2269 rtnl_unlock();
2270 }
2271}
2272
2273static int bond_upper_dev_walk(struct net_device *upper, void *data)
2274{
2275 __be32 ip = *((__be32 *)data);
2276
2277 return ip == bond_confirm_addr(upper, 0, ip);
2278}
2279
2280static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2281{
2282 bool ret = false;
2283
2284 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2285 return true;
2286
2287 rcu_read_lock();
2288 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2289 ret = true;
2290 rcu_read_unlock();
2291
2292 return ret;
2293}
2294
2295/* We go to the (large) trouble of VLAN tagging ARP frames because
2296 * switches in VLAN mode (especially if ports are configured as
2297 * "native" to a VLAN) might not pass non-tagged frames.
2298 */
2299static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2300 __be32 dest_ip, __be32 src_ip,
2301 struct bond_vlan_tag *tags)
2302{
2303 struct sk_buff *skb;
2304 struct bond_vlan_tag *outer_tag = tags;
2305
2306 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2307 arp_op, slave_dev->name, &dest_ip, &src_ip);
2308
2309 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2310 NULL, slave_dev->dev_addr, NULL);
2311
2312 if (!skb) {
2313 net_err_ratelimited("ARP packet allocation failed\n");
2314 return;
2315 }
2316
2317 if (!tags || tags->vlan_proto == VLAN_N_VID)
2318 goto xmit;
2319
2320 tags++;
2321
2322 /* Go through all the tags backwards and add them to the packet */
2323 while (tags->vlan_proto != VLAN_N_VID) {
2324 if (!tags->vlan_id) {
2325 tags++;
2326 continue;
2327 }
2328
2329 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2330 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2331 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2332 tags->vlan_id);
2333 if (!skb) {
2334 net_err_ratelimited("failed to insert inner VLAN tag\n");
2335 return;
2336 }
2337
2338 tags++;
2339 }
2340 /* Set the outer tag */
2341 if (outer_tag->vlan_id) {
2342 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2343 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2344 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2345 outer_tag->vlan_id);
2346 }
2347
2348xmit:
2349 arp_xmit(skb);
2350}
2351
2352/* Validate the device path between the @start_dev and the @end_dev.
2353 * The path is valid if the @end_dev is reachable through device
2354 * stacking.
2355 * When the path is validated, collect any vlan information in the
2356 * path.
2357 */
2358struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2359 struct net_device *end_dev,
2360 int level)
2361{
2362 struct bond_vlan_tag *tags;
2363 struct net_device *upper;
2364 struct list_head *iter;
2365
2366 if (start_dev == end_dev) {
2367 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2368 if (!tags)
2369 return ERR_PTR(-ENOMEM);
2370 tags[level].vlan_proto = VLAN_N_VID;
2371 return tags;
2372 }
2373
2374 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2375 tags = bond_verify_device_path(upper, end_dev, level + 1);
2376 if (IS_ERR_OR_NULL(tags)) {
2377 if (IS_ERR(tags))
2378 return tags;
2379 continue;
2380 }
2381 if (is_vlan_dev(upper)) {
2382 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2383 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2384 }
2385
2386 return tags;
2387 }
2388
2389 return NULL;
2390}
2391
2392static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2393{
2394 struct rtable *rt;
2395 struct bond_vlan_tag *tags;
2396 __be32 *targets = bond->params.arp_targets, addr;
2397 int i;
2398
2399 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2400 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2401 tags = NULL;
2402
2403 /* Find out through which dev should the packet go */
2404 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2405 RTO_ONLINK, 0);
2406 if (IS_ERR(rt)) {
2407 /* there's no route to target - try to send arp
2408 * probe to generate any traffic (arp_validate=0)
2409 */
2410 if (bond->params.arp_validate)
2411 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2412 bond->dev->name,
2413 &targets[i]);
2414 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2415 0, tags);
2416 continue;
2417 }
2418
2419 /* bond device itself */
2420 if (rt->dst.dev == bond->dev)
2421 goto found;
2422
2423 rcu_read_lock();
2424 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2425 rcu_read_unlock();
2426
2427 if (!IS_ERR_OR_NULL(tags))
2428 goto found;
2429
2430 /* Not our device - skip */
2431 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2432 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2433
2434 ip_rt_put(rt);
2435 continue;
2436
2437found:
2438 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2439 ip_rt_put(rt);
2440 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2441 addr, tags);
2442 kfree(tags);
2443 }
2444}
2445
2446static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2447{
2448 int i;
2449
2450 if (!sip || !bond_has_this_ip(bond, tip)) {
2451 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2452 &sip, &tip);
2453 return;
2454 }
2455
2456 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2457 if (i == -1) {
2458 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2459 &sip);
2460 return;
2461 }
2462 slave->last_rx = jiffies;
2463 slave->target_last_arp_rx[i] = jiffies;
2464}
2465
2466int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2467 struct slave *slave)
2468{
2469 struct arphdr *arp = (struct arphdr *)skb->data;
2470 struct slave *curr_active_slave, *curr_arp_slave;
2471 unsigned char *arp_ptr;
2472 __be32 sip, tip;
2473 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2474
2475 if (!slave_do_arp_validate(bond, slave)) {
2476 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2477 !slave_do_arp_validate_only(bond))
2478 slave->last_rx = jiffies;
2479 return RX_HANDLER_ANOTHER;
2480 } else if (!is_arp) {
2481 return RX_HANDLER_ANOTHER;
2482 }
2483
2484 alen = arp_hdr_len(bond->dev);
2485
2486 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2487 skb->dev->name);
2488
2489 if (alen > skb_headlen(skb)) {
2490 arp = kmalloc(alen, GFP_ATOMIC);
2491 if (!arp)
2492 goto out_unlock;
2493 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2494 goto out_unlock;
2495 }
2496
2497 if (arp->ar_hln != bond->dev->addr_len ||
2498 skb->pkt_type == PACKET_OTHERHOST ||
2499 skb->pkt_type == PACKET_LOOPBACK ||
2500 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2501 arp->ar_pro != htons(ETH_P_IP) ||
2502 arp->ar_pln != 4)
2503 goto out_unlock;
2504
2505 arp_ptr = (unsigned char *)(arp + 1);
2506 arp_ptr += bond->dev->addr_len;
2507 memcpy(&sip, arp_ptr, 4);
2508 arp_ptr += 4 + bond->dev->addr_len;
2509 memcpy(&tip, arp_ptr, 4);
2510
2511 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2512 slave->dev->name, bond_slave_state(slave),
2513 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2514 &sip, &tip);
2515
2516 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2517 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2518
2519 /* We 'trust' the received ARP enough to validate it if:
2520 *
2521 * (a) the slave receiving the ARP is active (which includes the
2522 * current ARP slave, if any), or
2523 *
2524 * (b) the receiving slave isn't active, but there is a currently
2525 * active slave and it received valid arp reply(s) after it became
2526 * the currently active slave, or
2527 *
2528 * (c) there is an ARP slave that sent an ARP during the prior ARP
2529 * interval, and we receive an ARP reply on any slave. We accept
2530 * these because switch FDB update delays may deliver the ARP
2531 * reply to a slave other than the sender of the ARP request.
2532 *
2533 * Note: for (b), backup slaves are receiving the broadcast ARP
2534 * request, not a reply. This request passes from the sending
2535 * slave through the L2 switch(es) to the receiving slave. Since
2536 * this is checking the request, sip/tip are swapped for
2537 * validation.
2538 *
2539 * This is done to avoid endless looping when we can't reach the
2540 * arp_ip_target and fool ourselves with our own arp requests.
2541 */
2542 if (bond_is_active_slave(slave))
2543 bond_validate_arp(bond, slave, sip, tip);
2544 else if (curr_active_slave &&
2545 time_after(slave_last_rx(bond, curr_active_slave),
2546 curr_active_slave->last_link_up))
2547 bond_validate_arp(bond, slave, tip, sip);
2548 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2549 bond_time_in_interval(bond,
2550 dev_trans_start(curr_arp_slave->dev), 1))
2551 bond_validate_arp(bond, slave, sip, tip);
2552
2553out_unlock:
2554 if (arp != (struct arphdr *)skb->data)
2555 kfree(arp);
2556 return RX_HANDLER_ANOTHER;
2557}
2558
2559/* function to verify if we're in the arp_interval timeslice, returns true if
2560 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2561 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2562 */
2563static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2564 int mod)
2565{
2566 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2567
2568 return time_in_range(jiffies,
2569 last_act - delta_in_ticks,
2570 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2571}
2572
2573/* This function is called regularly to monitor each slave's link
2574 * ensuring that traffic is being sent and received when arp monitoring
2575 * is used in load-balancing mode. if the adapter has been dormant, then an
2576 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2577 * arp monitoring in active backup mode.
2578 */
2579static void bond_loadbalance_arp_mon(struct work_struct *work)
2580{
2581 struct bonding *bond = container_of(work, struct bonding,
2582 arp_work.work);
2583 struct slave *slave, *oldcurrent;
2584 struct list_head *iter;
2585 int do_failover = 0, slave_state_changed = 0;
2586
2587 if (!bond_has_slaves(bond))
2588 goto re_arm;
2589
2590 rcu_read_lock();
2591
2592 oldcurrent = rcu_dereference(bond->curr_active_slave);
2593 /* see if any of the previous devices are up now (i.e. they have
2594 * xmt and rcv traffic). the curr_active_slave does not come into
2595 * the picture unless it is null. also, slave->last_link_up is not
2596 * needed here because we send an arp on each slave and give a slave
2597 * as long as it needs to get the tx/rx within the delta.
2598 * TODO: what about up/down delay in arp mode? it wasn't here before
2599 * so it can wait
2600 */
2601 bond_for_each_slave_rcu(bond, slave, iter) {
2602 unsigned long trans_start = dev_trans_start(slave->dev);
2603
2604 if (slave->link != BOND_LINK_UP) {
2605 if (bond_time_in_interval(bond, trans_start, 1) &&
2606 bond_time_in_interval(bond, slave->last_rx, 1)) {
2607
2608 slave->link = BOND_LINK_UP;
2609 slave_state_changed = 1;
2610
2611 /* primary_slave has no meaning in round-robin
2612 * mode. the window of a slave being up and
2613 * curr_active_slave being null after enslaving
2614 * is closed.
2615 */
2616 if (!oldcurrent) {
2617 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2618 slave->dev->name);
2619 do_failover = 1;
2620 } else {
2621 netdev_info(bond->dev, "interface %s is now up\n",
2622 slave->dev->name);
2623 }
2624 }
2625 } else {
2626 /* slave->link == BOND_LINK_UP */
2627
2628 /* not all switches will respond to an arp request
2629 * when the source ip is 0, so don't take the link down
2630 * if we don't know our ip yet
2631 */
2632 if (!bond_time_in_interval(bond, trans_start, 2) ||
2633 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2634
2635 slave->link = BOND_LINK_DOWN;
2636 slave_state_changed = 1;
2637
2638 if (slave->link_failure_count < UINT_MAX)
2639 slave->link_failure_count++;
2640
2641 netdev_info(bond->dev, "interface %s is now down\n",
2642 slave->dev->name);
2643
2644 if (slave == oldcurrent)
2645 do_failover = 1;
2646 }
2647 }
2648
2649 /* note: if switch is in round-robin mode, all links
2650 * must tx arp to ensure all links rx an arp - otherwise
2651 * links may oscillate or not come up at all; if switch is
2652 * in something like xor mode, there is nothing we can
2653 * do - all replies will be rx'ed on same link causing slaves
2654 * to be unstable during low/no traffic periods
2655 */
2656 if (bond_slave_is_up(slave))
2657 bond_arp_send_all(bond, slave);
2658 }
2659
2660 rcu_read_unlock();
2661
2662 if (do_failover || slave_state_changed) {
2663 if (!rtnl_trylock())
2664 goto re_arm;
2665
2666 if (slave_state_changed) {
2667 bond_slave_state_change(bond);
2668 if (BOND_MODE(bond) == BOND_MODE_XOR)
2669 bond_update_slave_arr(bond, NULL);
2670 }
2671 if (do_failover) {
2672 block_netpoll_tx();
2673 bond_select_active_slave(bond);
2674 unblock_netpoll_tx();
2675 }
2676 rtnl_unlock();
2677 }
2678
2679re_arm:
2680 if (bond->params.arp_interval)
2681 queue_delayed_work(bond->wq, &bond->arp_work,
2682 msecs_to_jiffies(bond->params.arp_interval));
2683}
2684
2685/* Called to inspect slaves for active-backup mode ARP monitor link state
2686 * changes. Sets new_link in slaves to specify what action should take
2687 * place for the slave. Returns 0 if no changes are found, >0 if changes
2688 * to link states must be committed.
2689 *
2690 * Called with rcu_read_lock held.
2691 */
2692static int bond_ab_arp_inspect(struct bonding *bond)
2693{
2694 unsigned long trans_start, last_rx;
2695 struct list_head *iter;
2696 struct slave *slave;
2697 int commit = 0;
2698
2699 bond_for_each_slave_rcu(bond, slave, iter) {
2700 slave->new_link = BOND_LINK_NOCHANGE;
2701 last_rx = slave_last_rx(bond, slave);
2702
2703 if (slave->link != BOND_LINK_UP) {
2704 if (bond_time_in_interval(bond, last_rx, 1)) {
2705 slave->new_link = BOND_LINK_UP;
2706 commit++;
2707 }
2708 continue;
2709 }
2710
2711 /* Give slaves 2*delta after being enslaved or made
2712 * active. This avoids bouncing, as the last receive
2713 * times need a full ARP monitor cycle to be updated.
2714 */
2715 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2716 continue;
2717
2718 /* Backup slave is down if:
2719 * - No current_arp_slave AND
2720 * - more than 3*delta since last receive AND
2721 * - the bond has an IP address
2722 *
2723 * Note: a non-null current_arp_slave indicates
2724 * the curr_active_slave went down and we are
2725 * searching for a new one; under this condition
2726 * we only take the curr_active_slave down - this
2727 * gives each slave a chance to tx/rx traffic
2728 * before being taken out
2729 */
2730 if (!bond_is_active_slave(slave) &&
2731 !rcu_access_pointer(bond->current_arp_slave) &&
2732 !bond_time_in_interval(bond, last_rx, 3)) {
2733 slave->new_link = BOND_LINK_DOWN;
2734 commit++;
2735 }
2736
2737 /* Active slave is down if:
2738 * - more than 2*delta since transmitting OR
2739 * - (more than 2*delta since receive AND
2740 * the bond has an IP address)
2741 */
2742 trans_start = dev_trans_start(slave->dev);
2743 if (bond_is_active_slave(slave) &&
2744 (!bond_time_in_interval(bond, trans_start, 2) ||
2745 !bond_time_in_interval(bond, last_rx, 2))) {
2746 slave->new_link = BOND_LINK_DOWN;
2747 commit++;
2748 }
2749 }
2750
2751 return commit;
2752}
2753
2754/* Called to commit link state changes noted by inspection step of
2755 * active-backup mode ARP monitor.
2756 *
2757 * Called with RTNL hold.
2758 */
2759static void bond_ab_arp_commit(struct bonding *bond)
2760{
2761 unsigned long trans_start;
2762 struct list_head *iter;
2763 struct slave *slave;
2764
2765 bond_for_each_slave(bond, slave, iter) {
2766 switch (slave->new_link) {
2767 case BOND_LINK_NOCHANGE:
2768 continue;
2769
2770 case BOND_LINK_UP:
2771 trans_start = dev_trans_start(slave->dev);
2772 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2773 (!rtnl_dereference(bond->curr_active_slave) &&
2774 bond_time_in_interval(bond, trans_start, 1))) {
2775 struct slave *current_arp_slave;
2776
2777 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2778 bond_set_slave_link_state(slave, BOND_LINK_UP,
2779 BOND_SLAVE_NOTIFY_NOW);
2780 if (current_arp_slave) {
2781 bond_set_slave_inactive_flags(
2782 current_arp_slave,
2783 BOND_SLAVE_NOTIFY_NOW);
2784 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2785 }
2786
2787 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2788 slave->dev->name);
2789
2790 if (!rtnl_dereference(bond->curr_active_slave) ||
2791 slave == rtnl_dereference(bond->primary_slave))
2792 goto do_failover;
2793
2794 }
2795
2796 continue;
2797
2798 case BOND_LINK_DOWN:
2799 if (slave->link_failure_count < UINT_MAX)
2800 slave->link_failure_count++;
2801
2802 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2803 BOND_SLAVE_NOTIFY_NOW);
2804 bond_set_slave_inactive_flags(slave,
2805 BOND_SLAVE_NOTIFY_NOW);
2806
2807 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2808 slave->dev->name);
2809
2810 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2811 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2812 goto do_failover;
2813 }
2814
2815 continue;
2816
2817 default:
2818 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2819 slave->new_link, slave->dev->name);
2820 continue;
2821 }
2822
2823do_failover:
2824 block_netpoll_tx();
2825 bond_select_active_slave(bond);
2826 unblock_netpoll_tx();
2827 }
2828
2829 bond_set_carrier(bond);
2830}
2831
2832/* Send ARP probes for active-backup mode ARP monitor.
2833 *
2834 * Called with rcu_read_lock held.
2835 */
2836static bool bond_ab_arp_probe(struct bonding *bond)
2837{
2838 struct slave *slave, *before = NULL, *new_slave = NULL,
2839 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2840 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2841 struct list_head *iter;
2842 bool found = false;
2843 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2844
2845 if (curr_arp_slave && curr_active_slave)
2846 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2847 curr_arp_slave->dev->name,
2848 curr_active_slave->dev->name);
2849
2850 if (curr_active_slave) {
2851 bond_arp_send_all(bond, curr_active_slave);
2852 return should_notify_rtnl;
2853 }
2854
2855 /* if we don't have a curr_active_slave, search for the next available
2856 * backup slave from the current_arp_slave and make it the candidate
2857 * for becoming the curr_active_slave
2858 */
2859
2860 if (!curr_arp_slave) {
2861 curr_arp_slave = bond_first_slave_rcu(bond);
2862 if (!curr_arp_slave)
2863 return should_notify_rtnl;
2864 }
2865
2866 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2867
2868 bond_for_each_slave_rcu(bond, slave, iter) {
2869 if (!found && !before && bond_slave_is_up(slave))
2870 before = slave;
2871
2872 if (found && !new_slave && bond_slave_is_up(slave))
2873 new_slave = slave;
2874 /* if the link state is up at this point, we
2875 * mark it down - this can happen if we have
2876 * simultaneous link failures and
2877 * reselect_active_interface doesn't make this
2878 * one the current slave so it is still marked
2879 * up when it is actually down
2880 */
2881 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2882 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2883 BOND_SLAVE_NOTIFY_LATER);
2884 if (slave->link_failure_count < UINT_MAX)
2885 slave->link_failure_count++;
2886
2887 bond_set_slave_inactive_flags(slave,
2888 BOND_SLAVE_NOTIFY_LATER);
2889
2890 netdev_info(bond->dev, "backup interface %s is now down\n",
2891 slave->dev->name);
2892 }
2893 if (slave == curr_arp_slave)
2894 found = true;
2895 }
2896
2897 if (!new_slave && before)
2898 new_slave = before;
2899
2900 if (!new_slave)
2901 goto check_state;
2902
2903 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2904 BOND_SLAVE_NOTIFY_LATER);
2905 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2906 bond_arp_send_all(bond, new_slave);
2907 new_slave->last_link_up = jiffies;
2908 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2909
2910check_state:
2911 bond_for_each_slave_rcu(bond, slave, iter) {
2912 if (slave->should_notify || slave->should_notify_link) {
2913 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2914 break;
2915 }
2916 }
2917 return should_notify_rtnl;
2918}
2919
2920static void bond_activebackup_arp_mon(struct work_struct *work)
2921{
2922 struct bonding *bond = container_of(work, struct bonding,
2923 arp_work.work);
2924 bool should_notify_peers = false;
2925 bool should_notify_rtnl = false;
2926 int delta_in_ticks;
2927
2928 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2929
2930 if (!bond_has_slaves(bond))
2931 goto re_arm;
2932
2933 rcu_read_lock();
2934
2935 should_notify_peers = bond_should_notify_peers(bond);
2936
2937 if (bond_ab_arp_inspect(bond)) {
2938 rcu_read_unlock();
2939
2940 /* Race avoidance with bond_close flush of workqueue */
2941 if (!rtnl_trylock()) {
2942 delta_in_ticks = 1;
2943 should_notify_peers = false;
2944 goto re_arm;
2945 }
2946
2947 bond_ab_arp_commit(bond);
2948
2949 rtnl_unlock();
2950 rcu_read_lock();
2951 }
2952
2953 should_notify_rtnl = bond_ab_arp_probe(bond);
2954 rcu_read_unlock();
2955
2956re_arm:
2957 if (bond->params.arp_interval)
2958 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2959
2960 if (should_notify_peers || should_notify_rtnl) {
2961 if (!rtnl_trylock())
2962 return;
2963
2964 if (should_notify_peers)
2965 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2966 bond->dev);
2967 if (should_notify_rtnl) {
2968 bond_slave_state_notify(bond);
2969 bond_slave_link_notify(bond);
2970 }
2971
2972 rtnl_unlock();
2973 }
2974}
2975
2976/*-------------------------- netdev event handling --------------------------*/
2977
2978/* Change device name */
2979static int bond_event_changename(struct bonding *bond)
2980{
2981 bond_remove_proc_entry(bond);
2982 bond_create_proc_entry(bond);
2983
2984 bond_debug_reregister(bond);
2985
2986 return NOTIFY_DONE;
2987}
2988
2989static int bond_master_netdev_event(unsigned long event,
2990 struct net_device *bond_dev)
2991{
2992 struct bonding *event_bond = netdev_priv(bond_dev);
2993
2994 switch (event) {
2995 case NETDEV_CHANGENAME:
2996 return bond_event_changename(event_bond);
2997 case NETDEV_UNREGISTER:
2998 bond_remove_proc_entry(event_bond);
2999 break;
3000 case NETDEV_REGISTER:
3001 bond_create_proc_entry(event_bond);
3002 break;
3003 case NETDEV_NOTIFY_PEERS:
3004 if (event_bond->send_peer_notif)
3005 event_bond->send_peer_notif--;
3006 break;
3007 default:
3008 break;
3009 }
3010
3011 return NOTIFY_DONE;
3012}
3013
3014static int bond_slave_netdev_event(unsigned long event,
3015 struct net_device *slave_dev)
3016{
3017 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3018 struct bonding *bond;
3019 struct net_device *bond_dev;
3020
3021 /* A netdev event can be generated while enslaving a device
3022 * before netdev_rx_handler_register is called in which case
3023 * slave will be NULL
3024 */
3025 if (!slave)
3026 return NOTIFY_DONE;
3027 bond_dev = slave->bond->dev;
3028 bond = slave->bond;
3029 primary = rtnl_dereference(bond->primary_slave);
3030
3031 switch (event) {
3032 case NETDEV_UNREGISTER:
3033 if (bond_dev->type != ARPHRD_ETHER)
3034 bond_release_and_destroy(bond_dev, slave_dev);
3035 else
3036 bond_release(bond_dev, slave_dev);
3037 break;
3038 case NETDEV_UP:
3039 case NETDEV_CHANGE:
3040 bond_update_speed_duplex(slave);
3041 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3042 bond_3ad_adapter_speed_duplex_changed(slave);
3043 /* Fallthrough */
3044 case NETDEV_DOWN:
3045 /* Refresh slave-array if applicable!
3046 * If the setup does not use miimon or arpmon (mode-specific!),
3047 * then these events will not cause the slave-array to be
3048 * refreshed. This will cause xmit to use a slave that is not
3049 * usable. Avoid such situation by refeshing the array at these
3050 * events. If these (miimon/arpmon) parameters are configured
3051 * then array gets refreshed twice and that should be fine!
3052 */
3053 if (bond_mode_uses_xmit_hash(bond))
3054 bond_update_slave_arr(bond, NULL);
3055 break;
3056 case NETDEV_CHANGEMTU:
3057 /* TODO: Should slaves be allowed to
3058 * independently alter their MTU? For
3059 * an active-backup bond, slaves need
3060 * not be the same type of device, so
3061 * MTUs may vary. For other modes,
3062 * slaves arguably should have the
3063 * same MTUs. To do this, we'd need to
3064 * take over the slave's change_mtu
3065 * function for the duration of their
3066 * servitude.
3067 */
3068 break;
3069 case NETDEV_CHANGENAME:
3070 /* we don't care if we don't have primary set */
3071 if (!bond_uses_primary(bond) ||
3072 !bond->params.primary[0])
3073 break;
3074
3075 if (slave == primary) {
3076 /* slave's name changed - he's no longer primary */
3077 RCU_INIT_POINTER(bond->primary_slave, NULL);
3078 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3079 /* we have a new primary slave */
3080 rcu_assign_pointer(bond->primary_slave, slave);
3081 } else { /* we didn't change primary - exit */
3082 break;
3083 }
3084
3085 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3086 primary ? slave_dev->name : "none");
3087
3088 block_netpoll_tx();
3089 bond_select_active_slave(bond);
3090 unblock_netpoll_tx();
3091 break;
3092 case NETDEV_FEAT_CHANGE:
3093 bond_compute_features(bond);
3094 break;
3095 case NETDEV_RESEND_IGMP:
3096 /* Propagate to master device */
3097 call_netdevice_notifiers(event, slave->bond->dev);
3098 break;
3099 default:
3100 break;
3101 }
3102
3103 return NOTIFY_DONE;
3104}
3105
3106/* bond_netdev_event: handle netdev notifier chain events.
3107 *
3108 * This function receives events for the netdev chain. The caller (an
3109 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3110 * locks for us to safely manipulate the slave devices (RTNL lock,
3111 * dev_probe_lock).
3112 */
3113static int bond_netdev_event(struct notifier_block *this,
3114 unsigned long event, void *ptr)
3115{
3116 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3117
3118 netdev_dbg(event_dev, "event: %lx\n", event);
3119
3120 if (!(event_dev->priv_flags & IFF_BONDING))
3121 return NOTIFY_DONE;
3122
3123 if (event_dev->flags & IFF_MASTER) {
3124 netdev_dbg(event_dev, "IFF_MASTER\n");
3125 return bond_master_netdev_event(event, event_dev);
3126 }
3127
3128 if (event_dev->flags & IFF_SLAVE) {
3129 netdev_dbg(event_dev, "IFF_SLAVE\n");
3130 return bond_slave_netdev_event(event, event_dev);
3131 }
3132
3133 return NOTIFY_DONE;
3134}
3135
3136static struct notifier_block bond_netdev_notifier = {
3137 .notifier_call = bond_netdev_event,
3138};
3139
3140/*---------------------------- Hashing Policies -----------------------------*/
3141
3142/* L2 hash helper */
3143static inline u32 bond_eth_hash(struct sk_buff *skb)
3144{
3145 struct ethhdr *ep, hdr_tmp;
3146
3147 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3148 if (ep)
3149 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3150 return 0;
3151}
3152
3153/* Extract the appropriate headers based on bond's xmit policy */
3154static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3155 struct flow_keys *fk)
3156{
3157 const struct ipv6hdr *iph6;
3158 const struct iphdr *iph;
3159 int noff, proto = -1;
3160
3161 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3162 return skb_flow_dissect_flow_keys(skb, fk, 0);
3163
3164 fk->ports.ports = 0;
3165 noff = skb_network_offset(skb);
3166 if (skb->protocol == htons(ETH_P_IP)) {
3167 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3168 return false;
3169 iph = ip_hdr(skb);
3170 iph_to_flow_copy_v4addrs(fk, iph);
3171 noff += iph->ihl << 2;
3172 if (!ip_is_fragment(iph))
3173 proto = iph->protocol;
3174 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3175 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3176 return false;
3177 iph6 = ipv6_hdr(skb);
3178 iph_to_flow_copy_v6addrs(fk, iph6);
3179 noff += sizeof(*iph6);
3180 proto = iph6->nexthdr;
3181 } else {
3182 return false;
3183 }
3184 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3185 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3186
3187 return true;
3188}
3189
3190/**
3191 * bond_xmit_hash - generate a hash value based on the xmit policy
3192 * @bond: bonding device
3193 * @skb: buffer to use for headers
3194 *
3195 * This function will extract the necessary headers from the skb buffer and use
3196 * them to generate a hash based on the xmit_policy set in the bonding device
3197 */
3198u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3199{
3200 struct flow_keys flow;
3201 u32 hash;
3202
3203 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3204 skb->l4_hash)
3205 return skb->hash;
3206
3207 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3208 !bond_flow_dissect(bond, skb, &flow))
3209 return bond_eth_hash(skb);
3210
3211 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3212 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3213 hash = bond_eth_hash(skb);
3214 else
3215 hash = (__force u32)flow.ports.ports;
3216 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3217 (__force u32)flow_get_u32_src(&flow);
3218 hash ^= (hash >> 16);
3219 hash ^= (hash >> 8);
3220
3221 return hash;
3222}
3223
3224/*-------------------------- Device entry points ----------------------------*/
3225
3226static void bond_work_init_all(struct bonding *bond)
3227{
3228 INIT_DELAYED_WORK(&bond->mcast_work,
3229 bond_resend_igmp_join_requests_delayed);
3230 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3231 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3232 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3233 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3234 else
3235 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3236 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3237 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3238}
3239
3240static void bond_work_cancel_all(struct bonding *bond)
3241{
3242 cancel_delayed_work_sync(&bond->mii_work);
3243 cancel_delayed_work_sync(&bond->arp_work);
3244 cancel_delayed_work_sync(&bond->alb_work);
3245 cancel_delayed_work_sync(&bond->ad_work);
3246 cancel_delayed_work_sync(&bond->mcast_work);
3247 cancel_delayed_work_sync(&bond->slave_arr_work);
3248}
3249
3250static int bond_open(struct net_device *bond_dev)
3251{
3252 struct bonding *bond = netdev_priv(bond_dev);
3253 struct list_head *iter;
3254 struct slave *slave;
3255
3256 /* reset slave->backup and slave->inactive */
3257 if (bond_has_slaves(bond)) {
3258 bond_for_each_slave(bond, slave, iter) {
3259 if (bond_uses_primary(bond) &&
3260 slave != rcu_access_pointer(bond->curr_active_slave)) {
3261 bond_set_slave_inactive_flags(slave,
3262 BOND_SLAVE_NOTIFY_NOW);
3263 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3264 bond_set_slave_active_flags(slave,
3265 BOND_SLAVE_NOTIFY_NOW);
3266 }
3267 }
3268 }
3269
3270 bond_work_init_all(bond);
3271
3272 if (bond_is_lb(bond)) {
3273 /* bond_alb_initialize must be called before the timer
3274 * is started.
3275 */
3276 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3277 return -ENOMEM;
3278 if (bond->params.tlb_dynamic_lb)
3279 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3280 }
3281
3282 if (bond->params.miimon) /* link check interval, in milliseconds. */
3283 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3284
3285 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3286 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3287 bond->recv_probe = bond_arp_rcv;
3288 }
3289
3290 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3291 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3292 /* register to receive LACPDUs */
3293 bond->recv_probe = bond_3ad_lacpdu_recv;
3294 bond_3ad_initiate_agg_selection(bond, 1);
3295 }
3296
3297 if (bond_mode_uses_xmit_hash(bond))
3298 bond_update_slave_arr(bond, NULL);
3299
3300 return 0;
3301}
3302
3303static int bond_close(struct net_device *bond_dev)
3304{
3305 struct bonding *bond = netdev_priv(bond_dev);
3306
3307 bond_work_cancel_all(bond);
3308 bond->send_peer_notif = 0;
3309 if (bond_is_lb(bond))
3310 bond_alb_deinitialize(bond);
3311 bond->recv_probe = NULL;
3312
3313 return 0;
3314}
3315
3316/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3317 * that some drivers can provide 32bit values only.
3318 */
3319static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3320 const struct rtnl_link_stats64 *_new,
3321 const struct rtnl_link_stats64 *_old)
3322{
3323 const u64 *new = (const u64 *)_new;
3324 const u64 *old = (const u64 *)_old;
3325 u64 *res = (u64 *)_res;
3326 int i;
3327
3328 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3329 u64 nv = new[i];
3330 u64 ov = old[i];
3331
3332 /* detects if this particular field is 32bit only */
3333 if (((nv | ov) >> 32) == 0)
3334 res[i] += (u32)nv - (u32)ov;
3335 else
3336 res[i] += nv - ov;
3337 }
3338}
3339
3340static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3341 struct rtnl_link_stats64 *stats)
3342{
3343 struct bonding *bond = netdev_priv(bond_dev);
3344 struct rtnl_link_stats64 temp;
3345 struct list_head *iter;
3346 struct slave *slave;
3347
3348 spin_lock(&bond->stats_lock);
3349 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3350
3351 rcu_read_lock();
3352 bond_for_each_slave_rcu(bond, slave, iter) {
3353 const struct rtnl_link_stats64 *new =
3354 dev_get_stats(slave->dev, &temp);
3355
3356 bond_fold_stats(stats, new, &slave->slave_stats);
3357
3358 /* save off the slave stats for the next run */
3359 memcpy(&slave->slave_stats, new, sizeof(*new));
3360 }
3361 rcu_read_unlock();
3362
3363 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3364 spin_unlock(&bond->stats_lock);
3365
3366 return stats;
3367}
3368
3369static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3370{
3371 struct bonding *bond = netdev_priv(bond_dev);
3372 struct net_device *slave_dev = NULL;
3373 struct ifbond k_binfo;
3374 struct ifbond __user *u_binfo = NULL;
3375 struct ifslave k_sinfo;
3376 struct ifslave __user *u_sinfo = NULL;
3377 struct mii_ioctl_data *mii = NULL;
3378 struct bond_opt_value newval;
3379 struct net *net;
3380 int res = 0;
3381
3382 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3383
3384 switch (cmd) {
3385 case SIOCGMIIPHY:
3386 mii = if_mii(ifr);
3387 if (!mii)
3388 return -EINVAL;
3389
3390 mii->phy_id = 0;
3391 /* Fall Through */
3392 case SIOCGMIIREG:
3393 /* We do this again just in case we were called by SIOCGMIIREG
3394 * instead of SIOCGMIIPHY.
3395 */
3396 mii = if_mii(ifr);
3397 if (!mii)
3398 return -EINVAL;
3399
3400 if (mii->reg_num == 1) {
3401 mii->val_out = 0;
3402 if (netif_carrier_ok(bond->dev))
3403 mii->val_out = BMSR_LSTATUS;
3404 }
3405
3406 return 0;
3407 case BOND_INFO_QUERY_OLD:
3408 case SIOCBONDINFOQUERY:
3409 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3410
3411 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3412 return -EFAULT;
3413
3414 res = bond_info_query(bond_dev, &k_binfo);
3415 if (res == 0 &&
3416 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3417 return -EFAULT;
3418
3419 return res;
3420 case BOND_SLAVE_INFO_QUERY_OLD:
3421 case SIOCBONDSLAVEINFOQUERY:
3422 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3423
3424 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3425 return -EFAULT;
3426
3427 res = bond_slave_info_query(bond_dev, &k_sinfo);
3428 if (res == 0 &&
3429 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3430 return -EFAULT;
3431
3432 return res;
3433 default:
3434 break;
3435 }
3436
3437 net = dev_net(bond_dev);
3438
3439 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3440 return -EPERM;
3441
3442 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3443
3444 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3445
3446 if (!slave_dev)
3447 return -ENODEV;
3448
3449 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3450 switch (cmd) {
3451 case BOND_ENSLAVE_OLD:
3452 case SIOCBONDENSLAVE:
3453 res = bond_enslave(bond_dev, slave_dev);
3454 break;
3455 case BOND_RELEASE_OLD:
3456 case SIOCBONDRELEASE:
3457 res = bond_release(bond_dev, slave_dev);
3458 break;
3459 case BOND_SETHWADDR_OLD:
3460 case SIOCBONDSETHWADDR:
3461 bond_set_dev_addr(bond_dev, slave_dev);
3462 res = 0;
3463 break;
3464 case BOND_CHANGE_ACTIVE_OLD:
3465 case SIOCBONDCHANGEACTIVE:
3466 bond_opt_initstr(&newval, slave_dev->name);
3467 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3468 break;
3469 default:
3470 res = -EOPNOTSUPP;
3471 }
3472
3473 return res;
3474}
3475
3476static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3477{
3478 struct bonding *bond = netdev_priv(bond_dev);
3479
3480 if (change & IFF_PROMISC)
3481 bond_set_promiscuity(bond,
3482 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3483
3484 if (change & IFF_ALLMULTI)
3485 bond_set_allmulti(bond,
3486 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3487}
3488
3489static void bond_set_rx_mode(struct net_device *bond_dev)
3490{
3491 struct bonding *bond = netdev_priv(bond_dev);
3492 struct list_head *iter;
3493 struct slave *slave;
3494
3495 rcu_read_lock();
3496 if (bond_uses_primary(bond)) {
3497 slave = rcu_dereference(bond->curr_active_slave);
3498 if (slave) {
3499 dev_uc_sync(slave->dev, bond_dev);
3500 dev_mc_sync(slave->dev, bond_dev);
3501 }
3502 } else {
3503 bond_for_each_slave_rcu(bond, slave, iter) {
3504 dev_uc_sync_multiple(slave->dev, bond_dev);
3505 dev_mc_sync_multiple(slave->dev, bond_dev);
3506 }
3507 }
3508 rcu_read_unlock();
3509}
3510
3511static int bond_neigh_init(struct neighbour *n)
3512{
3513 struct bonding *bond = netdev_priv(n->dev);
3514 const struct net_device_ops *slave_ops;
3515 struct neigh_parms parms;
3516 struct slave *slave;
3517 int ret;
3518
3519 slave = bond_first_slave(bond);
3520 if (!slave)
3521 return 0;
3522 slave_ops = slave->dev->netdev_ops;
3523 if (!slave_ops->ndo_neigh_setup)
3524 return 0;
3525
3526 parms.neigh_setup = NULL;
3527 parms.neigh_cleanup = NULL;
3528 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3529 if (ret)
3530 return ret;
3531
3532 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3533 * after the last slave has been detached. Assumes that all slaves
3534 * utilize the same neigh_cleanup (true at this writing as only user
3535 * is ipoib).
3536 */
3537 n->parms->neigh_cleanup = parms.neigh_cleanup;
3538
3539 if (!parms.neigh_setup)
3540 return 0;
3541
3542 return parms.neigh_setup(n);
3543}
3544
3545/* The bonding ndo_neigh_setup is called at init time beofre any
3546 * slave exists. So we must declare proxy setup function which will
3547 * be used at run time to resolve the actual slave neigh param setup.
3548 *
3549 * It's also called by master devices (such as vlans) to setup their
3550 * underlying devices. In that case - do nothing, we're already set up from
3551 * our init.
3552 */
3553static int bond_neigh_setup(struct net_device *dev,
3554 struct neigh_parms *parms)
3555{
3556 /* modify only our neigh_parms */
3557 if (parms->dev == dev)
3558 parms->neigh_setup = bond_neigh_init;
3559
3560 return 0;
3561}
3562
3563/* Change the MTU of all of a master's slaves to match the master */
3564static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3565{
3566 struct bonding *bond = netdev_priv(bond_dev);
3567 struct slave *slave, *rollback_slave;
3568 struct list_head *iter;
3569 int res = 0;
3570
3571 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3572
3573 bond_for_each_slave(bond, slave, iter) {
3574 netdev_dbg(bond_dev, "s %p c_m %p\n",
3575 slave, slave->dev->netdev_ops->ndo_change_mtu);
3576
3577 res = dev_set_mtu(slave->dev, new_mtu);
3578
3579 if (res) {
3580 /* If we failed to set the slave's mtu to the new value
3581 * we must abort the operation even in ACTIVE_BACKUP
3582 * mode, because if we allow the backup slaves to have
3583 * different mtu values than the active slave we'll
3584 * need to change their mtu when doing a failover. That
3585 * means changing their mtu from timer context, which
3586 * is probably not a good idea.
3587 */
3588 netdev_dbg(bond_dev, "err %d %s\n", res,
3589 slave->dev->name);
3590 goto unwind;
3591 }
3592 }
3593
3594 bond_dev->mtu = new_mtu;
3595
3596 return 0;
3597
3598unwind:
3599 /* unwind from head to the slave that failed */
3600 bond_for_each_slave(bond, rollback_slave, iter) {
3601 int tmp_res;
3602
3603 if (rollback_slave == slave)
3604 break;
3605
3606 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3607 if (tmp_res) {
3608 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3609 tmp_res, rollback_slave->dev->name);
3610 }
3611 }
3612
3613 return res;
3614}
3615
3616/* Change HW address
3617 *
3618 * Note that many devices must be down to change the HW address, and
3619 * downing the master releases all slaves. We can make bonds full of
3620 * bonding devices to test this, however.
3621 */
3622static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3623{
3624 struct bonding *bond = netdev_priv(bond_dev);
3625 struct slave *slave, *rollback_slave;
3626 struct sockaddr *sa = addr, tmp_sa;
3627 struct list_head *iter;
3628 int res = 0;
3629
3630 if (BOND_MODE(bond) == BOND_MODE_ALB)
3631 return bond_alb_set_mac_address(bond_dev, addr);
3632
3633
3634 netdev_dbg(bond_dev, "bond=%p\n", bond);
3635
3636 /* If fail_over_mac is enabled, do nothing and return success.
3637 * Returning an error causes ifenslave to fail.
3638 */
3639 if (bond->params.fail_over_mac &&
3640 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3641 return 0;
3642
3643 if (!is_valid_ether_addr(sa->sa_data))
3644 return -EADDRNOTAVAIL;
3645
3646 bond_for_each_slave(bond, slave, iter) {
3647 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3648 res = dev_set_mac_address(slave->dev, addr);
3649 if (res) {
3650 /* TODO: consider downing the slave
3651 * and retry ?
3652 * User should expect communications
3653 * breakage anyway until ARP finish
3654 * updating, so...
3655 */
3656 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3657 goto unwind;
3658 }
3659 }
3660
3661 /* success */
3662 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3663 return 0;
3664
3665unwind:
3666 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3667 tmp_sa.sa_family = bond_dev->type;
3668
3669 /* unwind from head to the slave that failed */
3670 bond_for_each_slave(bond, rollback_slave, iter) {
3671 int tmp_res;
3672
3673 if (rollback_slave == slave)
3674 break;
3675
3676 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3677 if (tmp_res) {
3678 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3679 tmp_res, rollback_slave->dev->name);
3680 }
3681 }
3682
3683 return res;
3684}
3685
3686/**
3687 * bond_xmit_slave_id - transmit skb through slave with slave_id
3688 * @bond: bonding device that is transmitting
3689 * @skb: buffer to transmit
3690 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3691 *
3692 * This function tries to transmit through slave with slave_id but in case
3693 * it fails, it tries to find the first available slave for transmission.
3694 * The skb is consumed in all cases, thus the function is void.
3695 */
3696static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3697{
3698 struct list_head *iter;
3699 struct slave *slave;
3700 int i = slave_id;
3701
3702 /* Here we start from the slave with slave_id */
3703 bond_for_each_slave_rcu(bond, slave, iter) {
3704 if (--i < 0) {
3705 if (bond_slave_can_tx(slave)) {
3706 bond_dev_queue_xmit(bond, skb, slave->dev);
3707 return;
3708 }
3709 }
3710 }
3711
3712 /* Here we start from the first slave up to slave_id */
3713 i = slave_id;
3714 bond_for_each_slave_rcu(bond, slave, iter) {
3715 if (--i < 0)
3716 break;
3717 if (bond_slave_can_tx(slave)) {
3718 bond_dev_queue_xmit(bond, skb, slave->dev);
3719 return;
3720 }
3721 }
3722 /* no slave that can tx has been found */
3723 bond_tx_drop(bond->dev, skb);
3724}
3725
3726/**
3727 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3728 * @bond: bonding device to use
3729 *
3730 * Based on the value of the bonding device's packets_per_slave parameter
3731 * this function generates a slave id, which is usually used as the next
3732 * slave to transmit through.
3733 */
3734static u32 bond_rr_gen_slave_id(struct bonding *bond)
3735{
3736 u32 slave_id;
3737 struct reciprocal_value reciprocal_packets_per_slave;
3738 int packets_per_slave = bond->params.packets_per_slave;
3739
3740 switch (packets_per_slave) {
3741 case 0:
3742 slave_id = prandom_u32();
3743 break;
3744 case 1:
3745 slave_id = bond->rr_tx_counter;
3746 break;
3747 default:
3748 reciprocal_packets_per_slave =
3749 bond->params.reciprocal_packets_per_slave;
3750 slave_id = reciprocal_divide(bond->rr_tx_counter,
3751 reciprocal_packets_per_slave);
3752 break;
3753 }
3754 bond->rr_tx_counter++;
3755
3756 return slave_id;
3757}
3758
3759static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3760{
3761 struct bonding *bond = netdev_priv(bond_dev);
3762 struct iphdr *iph = ip_hdr(skb);
3763 struct slave *slave;
3764 u32 slave_id;
3765
3766 /* Start with the curr_active_slave that joined the bond as the
3767 * default for sending IGMP traffic. For failover purposes one
3768 * needs to maintain some consistency for the interface that will
3769 * send the join/membership reports. The curr_active_slave found
3770 * will send all of this type of traffic.
3771 */
3772 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3773 slave = rcu_dereference(bond->curr_active_slave);
3774 if (slave)
3775 bond_dev_queue_xmit(bond, skb, slave->dev);
3776 else
3777 bond_xmit_slave_id(bond, skb, 0);
3778 } else {
3779 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3780
3781 if (likely(slave_cnt)) {
3782 slave_id = bond_rr_gen_slave_id(bond);
3783 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3784 } else {
3785 bond_tx_drop(bond_dev, skb);
3786 }
3787 }
3788
3789 return NETDEV_TX_OK;
3790}
3791
3792/* In active-backup mode, we know that bond->curr_active_slave is always valid if
3793 * the bond has a usable interface.
3794 */
3795static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3796{
3797 struct bonding *bond = netdev_priv(bond_dev);
3798 struct slave *slave;
3799
3800 slave = rcu_dereference(bond->curr_active_slave);
3801 if (slave)
3802 bond_dev_queue_xmit(bond, skb, slave->dev);
3803 else
3804 bond_tx_drop(bond_dev, skb);
3805
3806 return NETDEV_TX_OK;
3807}
3808
3809/* Use this to update slave_array when (a) it's not appropriate to update
3810 * slave_array right away (note that update_slave_array() may sleep)
3811 * and / or (b) RTNL is not held.
3812 */
3813void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3814{
3815 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3816}
3817
3818/* Slave array work handler. Holds only RTNL */
3819static void bond_slave_arr_handler(struct work_struct *work)
3820{
3821 struct bonding *bond = container_of(work, struct bonding,
3822 slave_arr_work.work);
3823 int ret;
3824
3825 if (!rtnl_trylock())
3826 goto err;
3827
3828 ret = bond_update_slave_arr(bond, NULL);
3829 rtnl_unlock();
3830 if (ret) {
3831 pr_warn_ratelimited("Failed to update slave array from WT\n");
3832 goto err;
3833 }
3834 return;
3835
3836err:
3837 bond_slave_arr_work_rearm(bond, 1);
3838}
3839
3840/* Build the usable slaves array in control path for modes that use xmit-hash
3841 * to determine the slave interface -
3842 * (a) BOND_MODE_8023AD
3843 * (b) BOND_MODE_XOR
3844 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3845 *
3846 * The caller is expected to hold RTNL only and NO other lock!
3847 */
3848int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3849{
3850 struct slave *slave;
3851 struct list_head *iter;
3852 struct bond_up_slave *new_arr, *old_arr;
3853 int agg_id = 0;
3854 int ret = 0;
3855
3856#ifdef CONFIG_LOCKDEP
3857 WARN_ON(lockdep_is_held(&bond->mode_lock));
3858#endif
3859
3860 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3861 GFP_KERNEL);
3862 if (!new_arr) {
3863 ret = -ENOMEM;
3864 pr_err("Failed to build slave-array.\n");
3865 goto out;
3866 }
3867 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3868 struct ad_info ad_info;
3869
3870 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3871 pr_debug("bond_3ad_get_active_agg_info failed\n");
3872 kfree_rcu(new_arr, rcu);
3873 /* No active aggragator means it's not safe to use
3874 * the previous array.
3875 */
3876 old_arr = rtnl_dereference(bond->slave_arr);
3877 if (old_arr) {
3878 RCU_INIT_POINTER(bond->slave_arr, NULL);
3879 kfree_rcu(old_arr, rcu);
3880 }
3881 goto out;
3882 }
3883 agg_id = ad_info.aggregator_id;
3884 }
3885 bond_for_each_slave(bond, slave, iter) {
3886 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3887 struct aggregator *agg;
3888
3889 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3890 if (!agg || agg->aggregator_identifier != agg_id)
3891 continue;
3892 }
3893 if (!bond_slave_can_tx(slave))
3894 continue;
3895 if (skipslave == slave)
3896 continue;
3897 new_arr->arr[new_arr->count++] = slave;
3898 }
3899
3900 old_arr = rtnl_dereference(bond->slave_arr);
3901 rcu_assign_pointer(bond->slave_arr, new_arr);
3902 if (old_arr)
3903 kfree_rcu(old_arr, rcu);
3904out:
3905 if (ret != 0 && skipslave) {
3906 int idx;
3907
3908 /* Rare situation where caller has asked to skip a specific
3909 * slave but allocation failed (most likely!). BTW this is
3910 * only possible when the call is initiated from
3911 * __bond_release_one(). In this situation; overwrite the
3912 * skipslave entry in the array with the last entry from the
3913 * array to avoid a situation where the xmit path may choose
3914 * this to-be-skipped slave to send a packet out.
3915 */
3916 old_arr = rtnl_dereference(bond->slave_arr);
3917 for (idx = 0; idx < old_arr->count; idx++) {
3918 if (skipslave == old_arr->arr[idx]) {
3919 old_arr->arr[idx] =
3920 old_arr->arr[old_arr->count-1];
3921 old_arr->count--;
3922 break;
3923 }
3924 }
3925 }
3926 return ret;
3927}
3928
3929/* Use this Xmit function for 3AD as well as XOR modes. The current
3930 * usable slave array is formed in the control path. The xmit function
3931 * just calculates hash and sends the packet out.
3932 */
3933static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3934{
3935 struct bonding *bond = netdev_priv(dev);
3936 struct slave *slave;
3937 struct bond_up_slave *slaves;
3938 unsigned int count;
3939
3940 slaves = rcu_dereference(bond->slave_arr);
3941 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3942 if (likely(count)) {
3943 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3944 bond_dev_queue_xmit(bond, skb, slave->dev);
3945 } else {
3946 bond_tx_drop(dev, skb);
3947 }
3948
3949 return NETDEV_TX_OK;
3950}
3951
3952/* in broadcast mode, we send everything to all usable interfaces. */
3953static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3954{
3955 struct bonding *bond = netdev_priv(bond_dev);
3956 struct slave *slave = NULL;
3957 struct list_head *iter;
3958
3959 bond_for_each_slave_rcu(bond, slave, iter) {
3960 if (bond_is_last_slave(bond, slave))
3961 break;
3962 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3963 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3964
3965 if (!skb2) {
3966 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3967 bond_dev->name, __func__);
3968 continue;
3969 }
3970 bond_dev_queue_xmit(bond, skb2, slave->dev);
3971 }
3972 }
3973 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3974 bond_dev_queue_xmit(bond, skb, slave->dev);
3975 else
3976 bond_tx_drop(bond_dev, skb);
3977
3978 return NETDEV_TX_OK;
3979}
3980
3981/*------------------------- Device initialization ---------------------------*/
3982
3983/* Lookup the slave that corresponds to a qid */
3984static inline int bond_slave_override(struct bonding *bond,
3985 struct sk_buff *skb)
3986{
3987 struct slave *slave = NULL;
3988 struct list_head *iter;
3989
3990 if (!skb->queue_mapping)
3991 return 1;
3992
3993 /* Find out if any slaves have the same mapping as this skb. */
3994 bond_for_each_slave_rcu(bond, slave, iter) {
3995 if (slave->queue_id == skb->queue_mapping) {
3996 if (bond_slave_is_up(slave) &&
3997 slave->link == BOND_LINK_UP) {
3998 bond_dev_queue_xmit(bond, skb, slave->dev);
3999 return 0;
4000 }
4001 /* If the slave isn't UP, use default transmit policy. */
4002 break;
4003 }
4004 }
4005
4006 return 1;
4007}
4008
4009
4010static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4011 void *accel_priv, select_queue_fallback_t fallback)
4012{
4013 /* This helper function exists to help dev_pick_tx get the correct
4014 * destination queue. Using a helper function skips a call to
4015 * skb_tx_hash and will put the skbs in the queue we expect on their
4016 * way down to the bonding driver.
4017 */
4018 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4019
4020 /* Save the original txq to restore before passing to the driver */
4021 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4022
4023 if (unlikely(txq >= dev->real_num_tx_queues)) {
4024 do {
4025 txq -= dev->real_num_tx_queues;
4026 } while (txq >= dev->real_num_tx_queues);
4027 }
4028 return txq;
4029}
4030
4031static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4032{
4033 struct bonding *bond = netdev_priv(dev);
4034
4035 if (bond_should_override_tx_queue(bond) &&
4036 !bond_slave_override(bond, skb))
4037 return NETDEV_TX_OK;
4038
4039 switch (BOND_MODE(bond)) {
4040 case BOND_MODE_ROUNDROBIN:
4041 return bond_xmit_roundrobin(skb, dev);
4042 case BOND_MODE_ACTIVEBACKUP:
4043 return bond_xmit_activebackup(skb, dev);
4044 case BOND_MODE_8023AD:
4045 case BOND_MODE_XOR:
4046 return bond_3ad_xor_xmit(skb, dev);
4047 case BOND_MODE_BROADCAST:
4048 return bond_xmit_broadcast(skb, dev);
4049 case BOND_MODE_ALB:
4050 return bond_alb_xmit(skb, dev);
4051 case BOND_MODE_TLB:
4052 return bond_tlb_xmit(skb, dev);
4053 default:
4054 /* Should never happen, mode already checked */
4055 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4056 WARN_ON_ONCE(1);
4057 bond_tx_drop(dev, skb);
4058 return NETDEV_TX_OK;
4059 }
4060}
4061
4062static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4063{
4064 struct bonding *bond = netdev_priv(dev);
4065 netdev_tx_t ret = NETDEV_TX_OK;
4066
4067 /* If we risk deadlock from transmitting this in the
4068 * netpoll path, tell netpoll to queue the frame for later tx
4069 */
4070 if (unlikely(is_netpoll_tx_blocked(dev)))
4071 return NETDEV_TX_BUSY;
4072
4073 rcu_read_lock();
4074 if (bond_has_slaves(bond))
4075 ret = __bond_start_xmit(skb, dev);
4076 else
4077 bond_tx_drop(dev, skb);
4078 rcu_read_unlock();
4079
4080 return ret;
4081}
4082
4083static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4084 struct ethtool_link_ksettings *cmd)
4085{
4086 struct bonding *bond = netdev_priv(bond_dev);
4087 unsigned long speed = 0;
4088 struct list_head *iter;
4089 struct slave *slave;
4090
4091 cmd->base.duplex = DUPLEX_UNKNOWN;
4092 cmd->base.port = PORT_OTHER;
4093
4094 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4095 * do not need to check mode. Though link speed might not represent
4096 * the true receive or transmit bandwidth (not all modes are symmetric)
4097 * this is an accurate maximum.
4098 */
4099 bond_for_each_slave(bond, slave, iter) {
4100 if (bond_slave_can_tx(slave)) {
4101 if (slave->speed != SPEED_UNKNOWN)
4102 speed += slave->speed;
4103 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4104 slave->duplex != DUPLEX_UNKNOWN)
4105 cmd->base.duplex = slave->duplex;
4106 }
4107 }
4108 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4109
4110 return 0;
4111}
4112
4113static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4114 struct ethtool_drvinfo *drvinfo)
4115{
4116 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4117 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4118 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4119 BOND_ABI_VERSION);
4120}
4121
4122static const struct ethtool_ops bond_ethtool_ops = {
4123 .get_drvinfo = bond_ethtool_get_drvinfo,
4124 .get_link = ethtool_op_get_link,
4125 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4126};
4127
4128static const struct net_device_ops bond_netdev_ops = {
4129 .ndo_init = bond_init,
4130 .ndo_uninit = bond_uninit,
4131 .ndo_open = bond_open,
4132 .ndo_stop = bond_close,
4133 .ndo_start_xmit = bond_start_xmit,
4134 .ndo_select_queue = bond_select_queue,
4135 .ndo_get_stats64 = bond_get_stats,
4136 .ndo_do_ioctl = bond_do_ioctl,
4137 .ndo_change_rx_flags = bond_change_rx_flags,
4138 .ndo_set_rx_mode = bond_set_rx_mode,
4139 .ndo_change_mtu = bond_change_mtu,
4140 .ndo_set_mac_address = bond_set_mac_address,
4141 .ndo_neigh_setup = bond_neigh_setup,
4142 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4143 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4144#ifdef CONFIG_NET_POLL_CONTROLLER
4145 .ndo_netpoll_setup = bond_netpoll_setup,
4146 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4147 .ndo_poll_controller = bond_poll_controller,
4148#endif
4149 .ndo_add_slave = bond_enslave,
4150 .ndo_del_slave = bond_release,
4151 .ndo_fix_features = bond_fix_features,
4152 .ndo_neigh_construct = netdev_default_l2upper_neigh_construct,
4153 .ndo_neigh_destroy = netdev_default_l2upper_neigh_destroy,
4154 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
4155 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
4156 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
4157 .ndo_fdb_add = switchdev_port_fdb_add,
4158 .ndo_fdb_del = switchdev_port_fdb_del,
4159 .ndo_fdb_dump = switchdev_port_fdb_dump,
4160 .ndo_features_check = passthru_features_check,
4161};
4162
4163static const struct device_type bond_type = {
4164 .name = "bond",
4165};
4166
4167static void bond_destructor(struct net_device *bond_dev)
4168{
4169 struct bonding *bond = netdev_priv(bond_dev);
4170 if (bond->wq)
4171 destroy_workqueue(bond->wq);
4172 free_netdev(bond_dev);
4173}
4174
4175void bond_setup(struct net_device *bond_dev)
4176{
4177 struct bonding *bond = netdev_priv(bond_dev);
4178
4179 spin_lock_init(&bond->mode_lock);
4180 spin_lock_init(&bond->stats_lock);
4181 bond->params = bonding_defaults;
4182
4183 /* Initialize pointers */
4184 bond->dev = bond_dev;
4185
4186 /* Initialize the device entry points */
4187 ether_setup(bond_dev);
4188 bond_dev->max_mtu = ETH_MAX_MTU;
4189 bond_dev->netdev_ops = &bond_netdev_ops;
4190 bond_dev->ethtool_ops = &bond_ethtool_ops;
4191
4192 bond_dev->destructor = bond_destructor;
4193
4194 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4195
4196 /* Initialize the device options */
4197 bond_dev->flags |= IFF_MASTER;
4198 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4199 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4200
4201 /* don't acquire bond device's netif_tx_lock when transmitting */
4202 bond_dev->features |= NETIF_F_LLTX;
4203
4204 /* By default, we declare the bond to be fully
4205 * VLAN hardware accelerated capable. Special
4206 * care is taken in the various xmit functions
4207 * when there are slaves that are not hw accel
4208 * capable
4209 */
4210
4211 /* Don't allow bond devices to change network namespaces. */
4212 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4213
4214 bond_dev->hw_features = BOND_VLAN_FEATURES |
4215 NETIF_F_HW_VLAN_CTAG_TX |
4216 NETIF_F_HW_VLAN_CTAG_RX |
4217 NETIF_F_HW_VLAN_CTAG_FILTER;
4218
4219 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4220 bond_dev->features |= bond_dev->hw_features;
4221}
4222
4223/* Destroy a bonding device.
4224 * Must be under rtnl_lock when this function is called.
4225 */
4226static void bond_uninit(struct net_device *bond_dev)
4227{
4228 struct bonding *bond = netdev_priv(bond_dev);
4229 struct list_head *iter;
4230 struct slave *slave;
4231 struct bond_up_slave *arr;
4232
4233 bond_netpoll_cleanup(bond_dev);
4234
4235 /* Release the bonded slaves */
4236 bond_for_each_slave(bond, slave, iter)
4237 __bond_release_one(bond_dev, slave->dev, true);
4238 netdev_info(bond_dev, "Released all slaves\n");
4239
4240 arr = rtnl_dereference(bond->slave_arr);
4241 if (arr) {
4242 RCU_INIT_POINTER(bond->slave_arr, NULL);
4243 kfree_rcu(arr, rcu);
4244 }
4245
4246 list_del(&bond->bond_list);
4247
4248 bond_debug_unregister(bond);
4249}
4250
4251/*------------------------- Module initialization ---------------------------*/
4252
4253static int bond_check_params(struct bond_params *params)
4254{
4255 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4256 struct bond_opt_value newval;
4257 const struct bond_opt_value *valptr;
4258 int arp_all_targets_value;
4259 u16 ad_actor_sys_prio = 0;
4260 u16 ad_user_port_key = 0;
4261
4262 /* Convert string parameters. */
4263 if (mode) {
4264 bond_opt_initstr(&newval, mode);
4265 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4266 if (!valptr) {
4267 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4268 return -EINVAL;
4269 }
4270 bond_mode = valptr->value;
4271 }
4272
4273 if (xmit_hash_policy) {
4274 if ((bond_mode != BOND_MODE_XOR) &&
4275 (bond_mode != BOND_MODE_8023AD) &&
4276 (bond_mode != BOND_MODE_TLB)) {
4277 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4278 bond_mode_name(bond_mode));
4279 } else {
4280 bond_opt_initstr(&newval, xmit_hash_policy);
4281 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4282 &newval);
4283 if (!valptr) {
4284 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4285 xmit_hash_policy);
4286 return -EINVAL;
4287 }
4288 xmit_hashtype = valptr->value;
4289 }
4290 }
4291
4292 if (lacp_rate) {
4293 if (bond_mode != BOND_MODE_8023AD) {
4294 pr_info("lacp_rate param is irrelevant in mode %s\n",
4295 bond_mode_name(bond_mode));
4296 } else {
4297 bond_opt_initstr(&newval, lacp_rate);
4298 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4299 &newval);
4300 if (!valptr) {
4301 pr_err("Error: Invalid lacp rate \"%s\"\n",
4302 lacp_rate);
4303 return -EINVAL;
4304 }
4305 lacp_fast = valptr->value;
4306 }
4307 }
4308
4309 if (ad_select) {
4310 bond_opt_initstr(&newval, ad_select);
4311 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4312 &newval);
4313 if (!valptr) {
4314 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4315 return -EINVAL;
4316 }
4317 params->ad_select = valptr->value;
4318 if (bond_mode != BOND_MODE_8023AD)
4319 pr_warn("ad_select param only affects 802.3ad mode\n");
4320 } else {
4321 params->ad_select = BOND_AD_STABLE;
4322 }
4323
4324 if (max_bonds < 0) {
4325 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4326 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4327 max_bonds = BOND_DEFAULT_MAX_BONDS;
4328 }
4329
4330 if (miimon < 0) {
4331 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4332 miimon, INT_MAX);
4333 miimon = 0;
4334 }
4335
4336 if (updelay < 0) {
4337 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4338 updelay, INT_MAX);
4339 updelay = 0;
4340 }
4341
4342 if (downdelay < 0) {
4343 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4344 downdelay, INT_MAX);
4345 downdelay = 0;
4346 }
4347
4348 if ((use_carrier != 0) && (use_carrier != 1)) {
4349 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4350 use_carrier);
4351 use_carrier = 1;
4352 }
4353
4354 if (num_peer_notif < 0 || num_peer_notif > 255) {
4355 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4356 num_peer_notif);
4357 num_peer_notif = 1;
4358 }
4359
4360 /* reset values for 802.3ad/TLB/ALB */
4361 if (!bond_mode_uses_arp(bond_mode)) {
4362 if (!miimon) {
4363 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");
4364 pr_warn("Forcing miimon to 100msec\n");
4365 miimon = BOND_DEFAULT_MIIMON;
4366 }
4367 }
4368
4369 if (tx_queues < 1 || tx_queues > 255) {
4370 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4371 tx_queues, BOND_DEFAULT_TX_QUEUES);
4372 tx_queues = BOND_DEFAULT_TX_QUEUES;
4373 }
4374
4375 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4376 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4377 all_slaves_active);
4378 all_slaves_active = 0;
4379 }
4380
4381 if (resend_igmp < 0 || resend_igmp > 255) {
4382 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4383 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4384 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4385 }
4386
4387 bond_opt_initval(&newval, packets_per_slave);
4388 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4389 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4390 packets_per_slave, USHRT_MAX);
4391 packets_per_slave = 1;
4392 }
4393
4394 if (bond_mode == BOND_MODE_ALB) {
4395 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",
4396 updelay);
4397 }
4398
4399 if (!miimon) {
4400 if (updelay || downdelay) {
4401 /* just warn the user the up/down delay will have
4402 * no effect since miimon is zero...
4403 */
4404 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",
4405 updelay, downdelay);
4406 }
4407 } else {
4408 /* don't allow arp monitoring */
4409 if (arp_interval) {
4410 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4411 miimon, arp_interval);
4412 arp_interval = 0;
4413 }
4414
4415 if ((updelay % miimon) != 0) {
4416 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4417 updelay, miimon, (updelay / miimon) * miimon);
4418 }
4419
4420 updelay /= miimon;
4421
4422 if ((downdelay % miimon) != 0) {
4423 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4424 downdelay, miimon,
4425 (downdelay / miimon) * miimon);
4426 }
4427
4428 downdelay /= miimon;
4429 }
4430
4431 if (arp_interval < 0) {
4432 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4433 arp_interval, INT_MAX);
4434 arp_interval = 0;
4435 }
4436
4437 for (arp_ip_count = 0, i = 0;
4438 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4439 __be32 ip;
4440
4441 /* not a complete check, but good enough to catch mistakes */
4442 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4443 !bond_is_ip_target_ok(ip)) {
4444 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4445 arp_ip_target[i]);
4446 arp_interval = 0;
4447 } else {
4448 if (bond_get_targets_ip(arp_target, ip) == -1)
4449 arp_target[arp_ip_count++] = ip;
4450 else
4451 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4452 &ip);
4453 }
4454 }
4455
4456 if (arp_interval && !arp_ip_count) {
4457 /* don't allow arping if no arp_ip_target given... */
4458 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4459 arp_interval);
4460 arp_interval = 0;
4461 }
4462
4463 if (arp_validate) {
4464 if (!arp_interval) {
4465 pr_err("arp_validate requires arp_interval\n");
4466 return -EINVAL;
4467 }
4468
4469 bond_opt_initstr(&newval, arp_validate);
4470 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4471 &newval);
4472 if (!valptr) {
4473 pr_err("Error: invalid arp_validate \"%s\"\n",
4474 arp_validate);
4475 return -EINVAL;
4476 }
4477 arp_validate_value = valptr->value;
4478 } else {
4479 arp_validate_value = 0;
4480 }
4481
4482 arp_all_targets_value = 0;
4483 if (arp_all_targets) {
4484 bond_opt_initstr(&newval, arp_all_targets);
4485 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4486 &newval);
4487 if (!valptr) {
4488 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4489 arp_all_targets);
4490 arp_all_targets_value = 0;
4491 } else {
4492 arp_all_targets_value = valptr->value;
4493 }
4494 }
4495
4496 if (miimon) {
4497 pr_info("MII link monitoring set to %d ms\n", miimon);
4498 } else if (arp_interval) {
4499 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4500 arp_validate_value);
4501 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4502 arp_interval, valptr->string, arp_ip_count);
4503
4504 for (i = 0; i < arp_ip_count; i++)
4505 pr_cont(" %s", arp_ip_target[i]);
4506
4507 pr_cont("\n");
4508
4509 } else if (max_bonds) {
4510 /* miimon and arp_interval not set, we need one so things
4511 * work as expected, see bonding.txt for details
4512 */
4513 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");
4514 }
4515
4516 if (primary && !bond_mode_uses_primary(bond_mode)) {
4517 /* currently, using a primary only makes sense
4518 * in active backup, TLB or ALB modes
4519 */
4520 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4521 primary, bond_mode_name(bond_mode));
4522 primary = NULL;
4523 }
4524
4525 if (primary && primary_reselect) {
4526 bond_opt_initstr(&newval, primary_reselect);
4527 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4528 &newval);
4529 if (!valptr) {
4530 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4531 primary_reselect);
4532 return -EINVAL;
4533 }
4534 primary_reselect_value = valptr->value;
4535 } else {
4536 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4537 }
4538
4539 if (fail_over_mac) {
4540 bond_opt_initstr(&newval, fail_over_mac);
4541 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4542 &newval);
4543 if (!valptr) {
4544 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4545 fail_over_mac);
4546 return -EINVAL;
4547 }
4548 fail_over_mac_value = valptr->value;
4549 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4550 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4551 } else {
4552 fail_over_mac_value = BOND_FOM_NONE;
4553 }
4554
4555 bond_opt_initstr(&newval, "default");
4556 valptr = bond_opt_parse(
4557 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4558 &newval);
4559 if (!valptr) {
4560 pr_err("Error: No ad_actor_sys_prio default value");
4561 return -EINVAL;
4562 }
4563 ad_actor_sys_prio = valptr->value;
4564
4565 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4566 &newval);
4567 if (!valptr) {
4568 pr_err("Error: No ad_user_port_key default value");
4569 return -EINVAL;
4570 }
4571 ad_user_port_key = valptr->value;
4572
4573 if (lp_interval == 0) {
4574 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4575 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4576 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4577 }
4578
4579 /* fill params struct with the proper values */
4580 params->mode = bond_mode;
4581 params->xmit_policy = xmit_hashtype;
4582 params->miimon = miimon;
4583 params->num_peer_notif = num_peer_notif;
4584 params->arp_interval = arp_interval;
4585 params->arp_validate = arp_validate_value;
4586 params->arp_all_targets = arp_all_targets_value;
4587 params->updelay = updelay;
4588 params->downdelay = downdelay;
4589 params->use_carrier = use_carrier;
4590 params->lacp_fast = lacp_fast;
4591 params->primary[0] = 0;
4592 params->primary_reselect = primary_reselect_value;
4593 params->fail_over_mac = fail_over_mac_value;
4594 params->tx_queues = tx_queues;
4595 params->all_slaves_active = all_slaves_active;
4596 params->resend_igmp = resend_igmp;
4597 params->min_links = min_links;
4598 params->lp_interval = lp_interval;
4599 params->packets_per_slave = packets_per_slave;
4600 params->tlb_dynamic_lb = 1; /* Default value */
4601 params->ad_actor_sys_prio = ad_actor_sys_prio;
4602 eth_zero_addr(params->ad_actor_system);
4603 params->ad_user_port_key = ad_user_port_key;
4604 if (packets_per_slave > 0) {
4605 params->reciprocal_packets_per_slave =
4606 reciprocal_value(packets_per_slave);
4607 } else {
4608 /* reciprocal_packets_per_slave is unused if
4609 * packets_per_slave is 0 or 1, just initialize it
4610 */
4611 params->reciprocal_packets_per_slave =
4612 (struct reciprocal_value) { 0 };
4613 }
4614
4615 if (primary) {
4616 strncpy(params->primary, primary, IFNAMSIZ);
4617 params->primary[IFNAMSIZ - 1] = 0;
4618 }
4619
4620 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4621
4622 return 0;
4623}
4624
4625/* Called from registration process */
4626static int bond_init(struct net_device *bond_dev)
4627{
4628 struct bonding *bond = netdev_priv(bond_dev);
4629 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4630
4631 netdev_dbg(bond_dev, "Begin bond_init\n");
4632
4633 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4634 if (!bond->wq)
4635 return -ENOMEM;
4636
4637 netdev_lockdep_set_classes(bond_dev);
4638
4639 list_add_tail(&bond->bond_list, &bn->dev_list);
4640
4641 bond_prepare_sysfs_group(bond);
4642
4643 bond_debug_register(bond);
4644
4645 /* Ensure valid dev_addr */
4646 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4647 bond_dev->addr_assign_type == NET_ADDR_PERM)
4648 eth_hw_addr_random(bond_dev);
4649
4650 return 0;
4651}
4652
4653unsigned int bond_get_num_tx_queues(void)
4654{
4655 return tx_queues;
4656}
4657
4658/* Create a new bond based on the specified name and bonding parameters.
4659 * If name is NULL, obtain a suitable "bond%d" name for us.
4660 * Caller must NOT hold rtnl_lock; we need to release it here before we
4661 * set up our sysfs entries.
4662 */
4663int bond_create(struct net *net, const char *name)
4664{
4665 struct net_device *bond_dev;
4666 struct bonding *bond;
4667 struct alb_bond_info *bond_info;
4668 int res;
4669
4670 rtnl_lock();
4671
4672 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4673 name ? name : "bond%d", NET_NAME_UNKNOWN,
4674 bond_setup, tx_queues);
4675 if (!bond_dev) {
4676 pr_err("%s: eek! can't alloc netdev!\n", name);
4677 rtnl_unlock();
4678 return -ENOMEM;
4679 }
4680
4681 /*
4682 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4683 * It is set to 0 by default which is wrong.
4684 */
4685 bond = netdev_priv(bond_dev);
4686 bond_info = &(BOND_ALB_INFO(bond));
4687 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4688
4689 dev_net_set(bond_dev, net);
4690 bond_dev->rtnl_link_ops = &bond_link_ops;
4691
4692 res = register_netdevice(bond_dev);
4693
4694 netif_carrier_off(bond_dev);
4695
4696 rtnl_unlock();
4697 if (res < 0)
4698 bond_destructor(bond_dev);
4699 return res;
4700}
4701
4702static int __net_init bond_net_init(struct net *net)
4703{
4704 struct bond_net *bn = net_generic(net, bond_net_id);
4705
4706 bn->net = net;
4707 INIT_LIST_HEAD(&bn->dev_list);
4708
4709 bond_create_proc_dir(bn);
4710 bond_create_sysfs(bn);
4711
4712 return 0;
4713}
4714
4715static void __net_exit bond_net_exit(struct net *net)
4716{
4717 struct bond_net *bn = net_generic(net, bond_net_id);
4718 struct bonding *bond, *tmp_bond;
4719 LIST_HEAD(list);
4720
4721 bond_destroy_sysfs(bn);
4722
4723 /* Kill off any bonds created after unregistering bond rtnl ops */
4724 rtnl_lock();
4725 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4726 unregister_netdevice_queue(bond->dev, &list);
4727 unregister_netdevice_many(&list);
4728 rtnl_unlock();
4729
4730 bond_destroy_proc_dir(bn);
4731}
4732
4733static struct pernet_operations bond_net_ops = {
4734 .init = bond_net_init,
4735 .exit = bond_net_exit,
4736 .id = &bond_net_id,
4737 .size = sizeof(struct bond_net),
4738};
4739
4740static int __init bonding_init(void)
4741{
4742 int i;
4743 int res;
4744
4745 pr_info("%s", bond_version);
4746
4747 res = bond_check_params(&bonding_defaults);
4748 if (res)
4749 goto out;
4750
4751 res = register_pernet_subsys(&bond_net_ops);
4752 if (res)
4753 goto out;
4754
4755 res = bond_netlink_init();
4756 if (res)
4757 goto err_link;
4758
4759 bond_create_debugfs();
4760
4761 for (i = 0; i < max_bonds; i++) {
4762 res = bond_create(&init_net, NULL);
4763 if (res)
4764 goto err;
4765 }
4766
4767 register_netdevice_notifier(&bond_netdev_notifier);
4768out:
4769 return res;
4770err:
4771 bond_destroy_debugfs();
4772 bond_netlink_fini();
4773err_link:
4774 unregister_pernet_subsys(&bond_net_ops);
4775 goto out;
4776
4777}
4778
4779static void __exit bonding_exit(void)
4780{
4781 unregister_netdevice_notifier(&bond_netdev_notifier);
4782
4783 bond_destroy_debugfs();
4784
4785 bond_netlink_fini();
4786 unregister_pernet_subsys(&bond_net_ops);
4787
4788#ifdef CONFIG_NET_POLL_CONTROLLER
4789 /* Make sure we don't have an imbalance on our netpoll blocking */
4790 WARN_ON(atomic_read(&netpoll_block_tx));
4791#endif
4792}
4793
4794module_init(bonding_init);
4795module_exit(bonding_exit);
4796MODULE_LICENSE("GPL");
4797MODULE_VERSION(DRV_VERSION);
4798MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4799MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");