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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36#include <linux/kernel.h>
37#include <linux/module.h>
38#include <linux/types.h>
39#include <linux/fcntl.h>
40#include <linux/interrupt.h>
41#include <linux/ptrace.h>
42#include <linux/ioport.h>
43#include <linux/in.h>
44#include <net/ip.h>
45#include <linux/ip.h>
46#include <linux/tcp.h>
47#include <linux/udp.h>
48#include <linux/slab.h>
49#include <linux/string.h>
50#include <linux/init.h>
51#include <linux/timer.h>
52#include <linux/socket.h>
53#include <linux/ctype.h>
54#include <linux/inet.h>
55#include <linux/bitops.h>
56#include <linux/io.h>
57#include <asm/dma.h>
58#include <linux/uaccess.h>
59#include <linux/errno.h>
60#include <linux/netdevice.h>
61#include <linux/inetdevice.h>
62#include <linux/igmp.h>
63#include <linux/etherdevice.h>
64#include <linux/skbuff.h>
65#include <net/sock.h>
66#include <linux/rtnetlink.h>
67#include <linux/smp.h>
68#include <linux/if_ether.h>
69#include <net/arp.h>
70#include <linux/mii.h>
71#include <linux/ethtool.h>
72#include <linux/if_vlan.h>
73#include <linux/if_bonding.h>
74#include <linux/jiffies.h>
75#include <linux/preempt.h>
76#include <net/route.h>
77#include <net/net_namespace.h>
78#include <net/netns/generic.h>
79#include <net/pkt_sched.h>
80#include <linux/rculist.h>
81#include <net/flow_keys.h>
82#include "bonding.h"
83#include "bond_3ad.h"
84#include "bond_alb.h"
85
86/*---------------------------- Module parameters ----------------------------*/
87
88/* monitor all links that often (in milliseconds). <=0 disables monitoring */
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92static int num_peer_notif = 1;
93static int miimon;
94static int updelay;
95static int downdelay;
96static int use_carrier = 1;
97static char *mode;
98static char *primary;
99static char *primary_reselect;
100static char *lacp_rate;
101static int min_links;
102static char *ad_select;
103static char *xmit_hash_policy;
104static int arp_interval;
105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *arp_all_targets;
108static char *fail_over_mac;
109static int all_slaves_active;
110static struct bond_params bonding_defaults;
111static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112static int packets_per_slave = 1;
113static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
114
115module_param(max_bonds, int, 0);
116MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117module_param(tx_queues, int, 0);
118MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125module_param(miimon, int, 0);
126MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127module_param(updelay, int, 0);
128MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129module_param(downdelay, int, 0);
130MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 "in milliseconds");
132module_param(use_carrier, int, 0);
133MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135module_param(mode, charp, 0);
136MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140module_param(primary, charp, 0);
141MODULE_PARM_DESC(primary, "Primary network device to use");
142module_param(primary_reselect, charp, 0);
143MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "once it comes up; "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
147 "better, "
148 "2 for only on active slave "
149 "failure");
150module_param(lacp_rate, charp, 0);
151MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153module_param(ad_select, charp, 0);
154MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
156 "2 for count");
157module_param(min_links, int, 0);
158MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159
160module_param(xmit_hash_policy, charp, 0);
161MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165module_param(arp_interval, int, 0);
166MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167module_param_array(arp_ip_target, charp, NULL, 0);
168MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169module_param(arp_validate, charp, 0);
170MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173module_param(arp_all_targets, charp, 0);
174MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175module_param(fail_over_mac, charp, 0);
176MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179module_param(all_slaves_active, int, 0);
180MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183module_param(resend_igmp, int, 0);
184MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 "link failure");
186module_param(packets_per_slave, int, 0);
187MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190module_param(lp_interval, uint, 0);
191MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
194
195/*----------------------------- Global variables ----------------------------*/
196
197#ifdef CONFIG_NET_POLL_CONTROLLER
198atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199#endif
200
201int bond_net_id __read_mostly;
202
203static __be32 arp_target[BOND_MAX_ARP_TARGETS];
204static int arp_ip_count;
205static int bond_mode = BOND_MODE_ROUNDROBIN;
206static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
207static int lacp_fast;
208
209/*-------------------------- Forward declarations ---------------------------*/
210
211static int bond_init(struct net_device *bond_dev);
212static void bond_uninit(struct net_device *bond_dev);
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/*
259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
265 *
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
273*/
274
275/**
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
279 */
280static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
282{
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
286 int res;
287
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
290 if (res)
291 goto unwind;
292 }
293
294 return 0;
295
296unwind:
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
300 break;
301
302 vlan_vid_del(rollback_slave->dev, proto, vid);
303 }
304
305 return res;
306}
307
308/**
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
312 */
313static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
315{
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
318 struct slave *slave;
319
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
322
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
325
326 return 0;
327}
328
329/*------------------------------- Link status -------------------------------*/
330
331/*
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
335 *
336 * Returns zero if carrier state does not change, nonzero if it does.
337 */
338static int bond_set_carrier(struct bonding *bond)
339{
340 struct list_head *iter;
341 struct slave *slave;
342
343 if (!bond_has_slaves(bond))
344 goto down;
345
346 if (bond->params.mode == BOND_MODE_8023AD)
347 return bond_3ad_set_carrier(bond);
348
349 bond_for_each_slave(bond, slave, iter) {
350 if (slave->link == BOND_LINK_UP) {
351 if (!netif_carrier_ok(bond->dev)) {
352 netif_carrier_on(bond->dev);
353 return 1;
354 }
355 return 0;
356 }
357 }
358
359down:
360 if (netif_carrier_ok(bond->dev)) {
361 netif_carrier_off(bond->dev);
362 return 1;
363 }
364 return 0;
365}
366
367/*
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
370 * values are invalid, set speed and duplex to -1,
371 * and return.
372 */
373static void bond_update_speed_duplex(struct slave *slave)
374{
375 struct net_device *slave_dev = slave->dev;
376 struct ethtool_cmd ecmd;
377 u32 slave_speed;
378 int res;
379
380 slave->speed = SPEED_UNKNOWN;
381 slave->duplex = DUPLEX_UNKNOWN;
382
383 res = __ethtool_get_settings(slave_dev, &ecmd);
384 if (res < 0)
385 return;
386
387 slave_speed = ethtool_cmd_speed(&ecmd);
388 if (slave_speed == 0 || slave_speed == ((__u32) -1))
389 return;
390
391 switch (ecmd.duplex) {
392 case DUPLEX_FULL:
393 case DUPLEX_HALF:
394 break;
395 default:
396 return;
397 }
398
399 slave->speed = slave_speed;
400 slave->duplex = ecmd.duplex;
401
402 return;
403}
404
405const char *bond_slave_link_status(s8 link)
406{
407 switch (link) {
408 case BOND_LINK_UP:
409 return "up";
410 case BOND_LINK_FAIL:
411 return "going down";
412 case BOND_LINK_DOWN:
413 return "down";
414 case BOND_LINK_BACK:
415 return "going back";
416 default:
417 return "unknown";
418 }
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 /*
482 * If reporting, report that either there's no dev->do_ioctl,
483 * or both SIOCGMIIREG and get_link failed (meaning that we
484 * cannot report link status). If not reporting, pretend
485 * we're ok.
486 */
487 return reporting ? -1 : BMSR_LSTATUS;
488}
489
490/*----------------------------- Multicast list ------------------------------*/
491
492/*
493 * Push the promiscuity flag down to appropriate slaves
494 */
495static int bond_set_promiscuity(struct bonding *bond, int inc)
496{
497 struct list_head *iter;
498 int err = 0;
499
500 if (USES_PRIMARY(bond->params.mode)) {
501 /* write lock already acquired */
502 if (bond->curr_active_slave) {
503 err = dev_set_promiscuity(bond->curr_active_slave->dev,
504 inc);
505 }
506 } else {
507 struct slave *slave;
508
509 bond_for_each_slave(bond, slave, iter) {
510 err = dev_set_promiscuity(slave->dev, inc);
511 if (err)
512 return err;
513 }
514 }
515 return err;
516}
517
518/*
519 * Push the allmulti flag down to all slaves
520 */
521static int bond_set_allmulti(struct bonding *bond, int inc)
522{
523 struct list_head *iter;
524 int err = 0;
525
526 if (USES_PRIMARY(bond->params.mode)) {
527 /* write lock already acquired */
528 if (bond->curr_active_slave) {
529 err = dev_set_allmulti(bond->curr_active_slave->dev,
530 inc);
531 }
532 } else {
533 struct slave *slave;
534
535 bond_for_each_slave(bond, slave, iter) {
536 err = dev_set_allmulti(slave->dev, inc);
537 if (err)
538 return err;
539 }
540 }
541 return err;
542}
543
544/*
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
547 * slave.
548 */
549static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
550{
551 struct bonding *bond = container_of(work, struct bonding,
552 mcast_work.work);
553
554 if (!rtnl_trylock()) {
555 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
556 return;
557 }
558 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
559
560 if (bond->igmp_retrans > 1) {
561 bond->igmp_retrans--;
562 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
563 }
564 rtnl_unlock();
565}
566
567/* Flush bond's hardware addresses from slave
568 */
569static void bond_hw_addr_flush(struct net_device *bond_dev,
570 struct net_device *slave_dev)
571{
572 struct bonding *bond = netdev_priv(bond_dev);
573
574 dev_uc_unsync(slave_dev, bond_dev);
575 dev_mc_unsync(slave_dev, bond_dev);
576
577 if (bond->params.mode == BOND_MODE_8023AD) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
580
581 dev_mc_del(slave_dev, lacpdu_multicast);
582 }
583}
584
585/*--------------------------- Active slave change ---------------------------*/
586
587/* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
589 * slaves up date at all times; only the USES_PRIMARY modes need to call
590 * this function to swap these settings during a failover.
591 */
592static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
593 struct slave *old_active)
594{
595 ASSERT_RTNL();
596
597 if (old_active) {
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(old_active->dev, -1);
600
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(old_active->dev, -1);
603
604 bond_hw_addr_flush(bond->dev, old_active->dev);
605 }
606
607 if (new_active) {
608 /* FIXME: Signal errors upstream. */
609 if (bond->dev->flags & IFF_PROMISC)
610 dev_set_promiscuity(new_active->dev, 1);
611
612 if (bond->dev->flags & IFF_ALLMULTI)
613 dev_set_allmulti(new_active->dev, 1);
614
615 netif_addr_lock_bh(bond->dev);
616 dev_uc_sync(new_active->dev, bond->dev);
617 dev_mc_sync(new_active->dev, bond->dev);
618 netif_addr_unlock_bh(bond->dev);
619 }
620}
621
622/**
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
626 *
627 * Should be called with RTNL held.
628 */
629static void bond_set_dev_addr(struct net_device *bond_dev,
630 struct net_device *slave_dev)
631{
632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev, slave_dev, slave_dev->addr_len);
634 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
635 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
637}
638
639/*
640 * bond_do_fail_over_mac
641 *
642 * Perform special MAC address swapping for fail_over_mac settings
643 *
644 * Called with RTNL, curr_slave_lock for write_bh.
645 */
646static void bond_do_fail_over_mac(struct bonding *bond,
647 struct slave *new_active,
648 struct slave *old_active)
649 __releases(&bond->curr_slave_lock)
650 __acquires(&bond->curr_slave_lock)
651{
652 u8 tmp_mac[ETH_ALEN];
653 struct sockaddr saddr;
654 int rv;
655
656 switch (bond->params.fail_over_mac) {
657 case BOND_FOM_ACTIVE:
658 if (new_active) {
659 write_unlock_bh(&bond->curr_slave_lock);
660 bond_set_dev_addr(bond->dev, new_active->dev);
661 write_lock_bh(&bond->curr_slave_lock);
662 }
663 break;
664 case BOND_FOM_FOLLOW:
665 /*
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
669 */
670 if (!new_active)
671 return;
672
673 write_unlock_bh(&bond->curr_slave_lock);
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 pr_err("%s: Error %d setting MAC of slave %s\n",
688 bond->dev->name, -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 pr_err("%s: Error %d setting MAC of slave %s\n",
701 bond->dev->name, -rv, new_active->dev->name);
702out:
703 write_lock_bh(&bond->curr_slave_lock);
704 break;
705 default:
706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
707 bond->dev->name, bond->params.fail_over_mac);
708 break;
709 }
710
711}
712
713static bool bond_should_change_active(struct bonding *bond)
714{
715 struct slave *prim = bond->primary_slave;
716 struct slave *curr = bond->curr_active_slave;
717
718 if (!prim || !curr || curr->link != BOND_LINK_UP)
719 return true;
720 if (bond->force_primary) {
721 bond->force_primary = false;
722 return true;
723 }
724 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
725 (prim->speed < curr->speed ||
726 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
727 return false;
728 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
729 return false;
730 return true;
731}
732
733/**
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
736 */
737static struct slave *bond_find_best_slave(struct bonding *bond)
738{
739 struct slave *slave, *bestslave = NULL;
740 struct list_head *iter;
741 int mintime = bond->params.updelay;
742
743 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
744 bond_should_change_active(bond))
745 return bond->primary_slave;
746
747 bond_for_each_slave(bond, slave, iter) {
748 if (slave->link == BOND_LINK_UP)
749 return slave;
750 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
751 slave->delay < mintime) {
752 mintime = slave->delay;
753 bestslave = slave;
754 }
755 }
756
757 return bestslave;
758}
759
760static bool bond_should_notify_peers(struct bonding *bond)
761{
762 struct slave *slave;
763
764 rcu_read_lock();
765 slave = rcu_dereference(bond->curr_active_slave);
766 rcu_read_unlock();
767
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond->dev->name, slave ? slave->dev->name : "NULL");
770
771 if (!slave || !bond->send_peer_notif ||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
773 return false;
774
775 return true;
776}
777
778/**
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
782 *
783 * Set the new slave to the bond's settings and unset them on the old
784 * curr_active_slave.
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
786 *
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
790 *
791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
792 */
793void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
794{
795 struct slave *old_active = bond->curr_active_slave;
796
797 if (old_active == new_active)
798 return;
799
800 if (new_active) {
801 new_active->last_link_up = jiffies;
802
803 if (new_active->link == BOND_LINK_BACK) {
804 if (USES_PRIMARY(bond->params.mode)) {
805 pr_info("%s: making interface %s the new active one %d ms earlier\n",
806 bond->dev->name, new_active->dev->name,
807 (bond->params.updelay - new_active->delay) * bond->params.miimon);
808 }
809
810 new_active->delay = 0;
811 new_active->link = BOND_LINK_UP;
812
813 if (bond->params.mode == BOND_MODE_8023AD)
814 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
815
816 if (bond_is_lb(bond))
817 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
818 } else {
819 if (USES_PRIMARY(bond->params.mode)) {
820 pr_info("%s: making interface %s the new active one\n",
821 bond->dev->name, new_active->dev->name);
822 }
823 }
824 }
825
826 if (USES_PRIMARY(bond->params.mode))
827 bond_hw_addr_swap(bond, new_active, old_active);
828
829 if (bond_is_lb(bond)) {
830 bond_alb_handle_active_change(bond, new_active);
831 if (old_active)
832 bond_set_slave_inactive_flags(old_active,
833 BOND_SLAVE_NOTIFY_NOW);
834 if (new_active)
835 bond_set_slave_active_flags(new_active,
836 BOND_SLAVE_NOTIFY_NOW);
837 } else {
838 rcu_assign_pointer(bond->curr_active_slave, new_active);
839 }
840
841 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
842 if (old_active)
843 bond_set_slave_inactive_flags(old_active,
844 BOND_SLAVE_NOTIFY_NOW);
845
846 if (new_active) {
847 bool should_notify_peers = false;
848
849 bond_set_slave_active_flags(new_active,
850 BOND_SLAVE_NOTIFY_NOW);
851
852 if (bond->params.fail_over_mac)
853 bond_do_fail_over_mac(bond, new_active,
854 old_active);
855
856 if (netif_running(bond->dev)) {
857 bond->send_peer_notif =
858 bond->params.num_peer_notif;
859 should_notify_peers =
860 bond_should_notify_peers(bond);
861 }
862
863 write_unlock_bh(&bond->curr_slave_lock);
864
865 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
866 if (should_notify_peers)
867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
868 bond->dev);
869
870 write_lock_bh(&bond->curr_slave_lock);
871 }
872 }
873
874 /* resend IGMP joins since active slave has changed or
875 * all were sent on curr_active_slave.
876 * resend only if bond is brought up with the affected
877 * bonding modes and the retransmission is enabled */
878 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
879 ((USES_PRIMARY(bond->params.mode) && new_active) ||
880 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
881 bond->igmp_retrans = bond->params.resend_igmp;
882 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
883 }
884}
885
886/**
887 * bond_select_active_slave - select a new active slave, if needed
888 * @bond: our bonding struct
889 *
890 * This functions should be called when one of the following occurs:
891 * - The old curr_active_slave has been released or lost its link.
892 * - The primary_slave has got its link back.
893 * - A slave has got its link back and there's no old curr_active_slave.
894 *
895 * Caller must hold curr_slave_lock for write_bh.
896 */
897void bond_select_active_slave(struct bonding *bond)
898{
899 struct slave *best_slave;
900 int rv;
901
902 best_slave = bond_find_best_slave(bond);
903 if (best_slave != bond->curr_active_slave) {
904 bond_change_active_slave(bond, best_slave);
905 rv = bond_set_carrier(bond);
906 if (!rv)
907 return;
908
909 if (netif_carrier_ok(bond->dev)) {
910 pr_info("%s: first active interface up!\n",
911 bond->dev->name);
912 } else {
913 pr_info("%s: now running without any active interface!\n",
914 bond->dev->name);
915 }
916 }
917}
918
919#ifdef CONFIG_NET_POLL_CONTROLLER
920static inline int slave_enable_netpoll(struct slave *slave)
921{
922 struct netpoll *np;
923 int err = 0;
924
925 np = kzalloc(sizeof(*np), GFP_KERNEL);
926 err = -ENOMEM;
927 if (!np)
928 goto out;
929
930 err = __netpoll_setup(np, slave->dev);
931 if (err) {
932 kfree(np);
933 goto out;
934 }
935 slave->np = np;
936out:
937 return err;
938}
939static inline void slave_disable_netpoll(struct slave *slave)
940{
941 struct netpoll *np = slave->np;
942
943 if (!np)
944 return;
945
946 slave->np = NULL;
947 __netpoll_free_async(np);
948}
949
950static void bond_poll_controller(struct net_device *bond_dev)
951{
952}
953
954static void bond_netpoll_cleanup(struct net_device *bond_dev)
955{
956 struct bonding *bond = netdev_priv(bond_dev);
957 struct list_head *iter;
958 struct slave *slave;
959
960 bond_for_each_slave(bond, slave, iter)
961 if (IS_UP(slave->dev))
962 slave_disable_netpoll(slave);
963}
964
965static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
966{
967 struct bonding *bond = netdev_priv(dev);
968 struct list_head *iter;
969 struct slave *slave;
970 int err = 0;
971
972 bond_for_each_slave(bond, slave, iter) {
973 err = slave_enable_netpoll(slave);
974 if (err) {
975 bond_netpoll_cleanup(dev);
976 break;
977 }
978 }
979 return err;
980}
981#else
982static inline int slave_enable_netpoll(struct slave *slave)
983{
984 return 0;
985}
986static inline void slave_disable_netpoll(struct slave *slave)
987{
988}
989static void bond_netpoll_cleanup(struct net_device *bond_dev)
990{
991}
992#endif
993
994/*---------------------------------- IOCTL ----------------------------------*/
995
996static netdev_features_t bond_fix_features(struct net_device *dev,
997 netdev_features_t features)
998{
999 struct bonding *bond = netdev_priv(dev);
1000 struct list_head *iter;
1001 netdev_features_t mask;
1002 struct slave *slave;
1003
1004 if (!bond_has_slaves(bond)) {
1005 /* Disable adding VLANs to empty bond. But why? --mq */
1006 features |= NETIF_F_VLAN_CHALLENGED;
1007 return features;
1008 }
1009
1010 mask = features;
1011 features &= ~NETIF_F_ONE_FOR_ALL;
1012 features |= NETIF_F_ALL_FOR_ALL;
1013
1014 bond_for_each_slave(bond, slave, iter) {
1015 features = netdev_increment_features(features,
1016 slave->dev->features,
1017 mask);
1018 }
1019 features = netdev_add_tso_features(features, mask);
1020
1021 return features;
1022}
1023
1024#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1025 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1026 NETIF_F_HIGHDMA | NETIF_F_LRO)
1027
1028static void bond_compute_features(struct bonding *bond)
1029{
1030 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1031 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1032 struct net_device *bond_dev = bond->dev;
1033 struct list_head *iter;
1034 struct slave *slave;
1035 unsigned short max_hard_header_len = ETH_HLEN;
1036 unsigned int gso_max_size = GSO_MAX_SIZE;
1037 u16 gso_max_segs = GSO_MAX_SEGS;
1038
1039 if (!bond_has_slaves(bond))
1040 goto done;
1041
1042 bond_for_each_slave(bond, slave, iter) {
1043 vlan_features = netdev_increment_features(vlan_features,
1044 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1045
1046 dst_release_flag &= slave->dev->priv_flags;
1047 if (slave->dev->hard_header_len > max_hard_header_len)
1048 max_hard_header_len = slave->dev->hard_header_len;
1049
1050 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1051 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1052 }
1053
1054done:
1055 bond_dev->vlan_features = vlan_features;
1056 bond_dev->hard_header_len = max_hard_header_len;
1057 bond_dev->gso_max_segs = gso_max_segs;
1058 netif_set_gso_max_size(bond_dev, gso_max_size);
1059
1060 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1061 bond_dev->priv_flags = flags | dst_release_flag;
1062
1063 netdev_change_features(bond_dev);
1064}
1065
1066static void bond_setup_by_slave(struct net_device *bond_dev,
1067 struct net_device *slave_dev)
1068{
1069 bond_dev->header_ops = slave_dev->header_ops;
1070
1071 bond_dev->type = slave_dev->type;
1072 bond_dev->hard_header_len = slave_dev->hard_header_len;
1073 bond_dev->addr_len = slave_dev->addr_len;
1074
1075 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1076 slave_dev->addr_len);
1077}
1078
1079/* On bonding slaves other than the currently active slave, suppress
1080 * duplicates except for alb non-mcast/bcast.
1081 */
1082static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1083 struct slave *slave,
1084 struct bonding *bond)
1085{
1086 if (bond_is_slave_inactive(slave)) {
1087 if (bond->params.mode == BOND_MODE_ALB &&
1088 skb->pkt_type != PACKET_BROADCAST &&
1089 skb->pkt_type != PACKET_MULTICAST)
1090 return false;
1091 return true;
1092 }
1093 return false;
1094}
1095
1096static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1097{
1098 struct sk_buff *skb = *pskb;
1099 struct slave *slave;
1100 struct bonding *bond;
1101 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1102 struct slave *);
1103 int ret = RX_HANDLER_ANOTHER;
1104
1105 skb = skb_share_check(skb, GFP_ATOMIC);
1106 if (unlikely(!skb))
1107 return RX_HANDLER_CONSUMED;
1108
1109 *pskb = skb;
1110
1111 slave = bond_slave_get_rcu(skb->dev);
1112 bond = slave->bond;
1113
1114 recv_probe = ACCESS_ONCE(bond->recv_probe);
1115 if (recv_probe) {
1116 ret = recv_probe(skb, bond, slave);
1117 if (ret == RX_HANDLER_CONSUMED) {
1118 consume_skb(skb);
1119 return ret;
1120 }
1121 }
1122
1123 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1124 return RX_HANDLER_EXACT;
1125 }
1126
1127 skb->dev = bond->dev;
1128
1129 if (bond->params.mode == BOND_MODE_ALB &&
1130 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1131 skb->pkt_type == PACKET_HOST) {
1132
1133 if (unlikely(skb_cow_head(skb,
1134 skb->data - skb_mac_header(skb)))) {
1135 kfree_skb(skb);
1136 return RX_HANDLER_CONSUMED;
1137 }
1138 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1139 }
1140
1141 return ret;
1142}
1143
1144static int bond_master_upper_dev_link(struct net_device *bond_dev,
1145 struct net_device *slave_dev,
1146 struct slave *slave)
1147{
1148 int err;
1149
1150 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1151 if (err)
1152 return err;
1153 slave_dev->flags |= IFF_SLAVE;
1154 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1155 return 0;
1156}
1157
1158static void bond_upper_dev_unlink(struct net_device *bond_dev,
1159 struct net_device *slave_dev)
1160{
1161 netdev_upper_dev_unlink(slave_dev, bond_dev);
1162 slave_dev->flags &= ~IFF_SLAVE;
1163 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1164}
1165
1166/* enslave device <slave> to bond device <master> */
1167int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1168{
1169 struct bonding *bond = netdev_priv(bond_dev);
1170 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1171 struct slave *new_slave = NULL, *prev_slave;
1172 struct sockaddr addr;
1173 int link_reporting;
1174 int res = 0, i;
1175
1176 if (!bond->params.use_carrier &&
1177 slave_dev->ethtool_ops->get_link == NULL &&
1178 slave_ops->ndo_do_ioctl == NULL) {
1179 pr_warn("%s: Warning: no link monitoring support for %s\n",
1180 bond_dev->name, slave_dev->name);
1181 }
1182
1183 /* already enslaved */
1184 if (slave_dev->flags & IFF_SLAVE) {
1185 pr_debug("Error: Device was already enslaved\n");
1186 return -EBUSY;
1187 }
1188
1189 if (bond_dev == slave_dev) {
1190 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1191 return -EPERM;
1192 }
1193
1194 /* vlan challenged mutual exclusion */
1195 /* no need to lock since we're protected by rtnl_lock */
1196 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1197 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1198 if (vlan_uses_dev(bond_dev)) {
1199 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1200 bond_dev->name, slave_dev->name, bond_dev->name);
1201 return -EPERM;
1202 } else {
1203 pr_warn("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1204 bond_dev->name, slave_dev->name,
1205 slave_dev->name, bond_dev->name);
1206 }
1207 } else {
1208 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1209 }
1210
1211 /*
1212 * Old ifenslave binaries are no longer supported. These can
1213 * be identified with moderate accuracy by the state of the slave:
1214 * the current ifenslave will set the interface down prior to
1215 * enslaving it; the old ifenslave will not.
1216 */
1217 if ((slave_dev->flags & IFF_UP)) {
1218 pr_err("%s is up - this may be due to an out of date ifenslave\n",
1219 slave_dev->name);
1220 res = -EPERM;
1221 goto err_undo_flags;
1222 }
1223
1224 /* set bonding device ether type by slave - bonding netdevices are
1225 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1226 * there is a need to override some of the type dependent attribs/funcs.
1227 *
1228 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1229 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1230 */
1231 if (!bond_has_slaves(bond)) {
1232 if (bond_dev->type != slave_dev->type) {
1233 pr_debug("%s: change device type from %d to %d\n",
1234 bond_dev->name,
1235 bond_dev->type, slave_dev->type);
1236
1237 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1238 bond_dev);
1239 res = notifier_to_errno(res);
1240 if (res) {
1241 pr_err("%s: refused to change device type\n",
1242 bond_dev->name);
1243 res = -EBUSY;
1244 goto err_undo_flags;
1245 }
1246
1247 /* Flush unicast and multicast addresses */
1248 dev_uc_flush(bond_dev);
1249 dev_mc_flush(bond_dev);
1250
1251 if (slave_dev->type != ARPHRD_ETHER)
1252 bond_setup_by_slave(bond_dev, slave_dev);
1253 else {
1254 ether_setup(bond_dev);
1255 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1256 }
1257
1258 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1259 bond_dev);
1260 }
1261 } else if (bond_dev->type != slave_dev->type) {
1262 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1263 slave_dev->name, slave_dev->type, bond_dev->type);
1264 res = -EINVAL;
1265 goto err_undo_flags;
1266 }
1267
1268 if (slave_ops->ndo_set_mac_address == NULL) {
1269 if (!bond_has_slaves(bond)) {
1270 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address\n",
1271 bond_dev->name);
1272 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1273 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1274 pr_warn("%s: Setting fail_over_mac to active for active-backup mode\n",
1275 bond_dev->name);
1276 }
1277 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1278 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n",
1279 bond_dev->name);
1280 res = -EOPNOTSUPP;
1281 goto err_undo_flags;
1282 }
1283 }
1284
1285 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1286
1287 /* If this is the first slave, then we need to set the master's hardware
1288 * address to be the same as the slave's. */
1289 if (!bond_has_slaves(bond) &&
1290 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1291 bond_set_dev_addr(bond->dev, slave_dev);
1292
1293 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1294 if (!new_slave) {
1295 res = -ENOMEM;
1296 goto err_undo_flags;
1297 }
1298 /*
1299 * Set the new_slave's queue_id to be zero. Queue ID mapping
1300 * is set via sysfs or module option if desired.
1301 */
1302 new_slave->queue_id = 0;
1303
1304 /* Save slave's original mtu and then set it to match the bond */
1305 new_slave->original_mtu = slave_dev->mtu;
1306 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1307 if (res) {
1308 pr_debug("Error %d calling dev_set_mtu\n", res);
1309 goto err_free;
1310 }
1311
1312 /*
1313 * Save slave's original ("permanent") mac address for modes
1314 * that need it, and for restoring it upon release, and then
1315 * set it to the master's address
1316 */
1317 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1318
1319 if (!bond->params.fail_over_mac ||
1320 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1321 /*
1322 * Set slave to master's mac address. The application already
1323 * set the master's mac address to that of the first slave
1324 */
1325 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1326 addr.sa_family = slave_dev->type;
1327 res = dev_set_mac_address(slave_dev, &addr);
1328 if (res) {
1329 pr_debug("Error %d calling set_mac_address\n", res);
1330 goto err_restore_mtu;
1331 }
1332 }
1333
1334 /* open the slave since the application closed it */
1335 res = dev_open(slave_dev);
1336 if (res) {
1337 pr_debug("Opening slave %s failed\n", slave_dev->name);
1338 goto err_restore_mac;
1339 }
1340
1341 new_slave->bond = bond;
1342 new_slave->dev = slave_dev;
1343 slave_dev->priv_flags |= IFF_BONDING;
1344
1345 if (bond_is_lb(bond)) {
1346 /* bond_alb_init_slave() must be called before all other stages since
1347 * it might fail and we do not want to have to undo everything
1348 */
1349 res = bond_alb_init_slave(bond, new_slave);
1350 if (res)
1351 goto err_close;
1352 }
1353
1354 /* If the mode USES_PRIMARY, then the following is handled by
1355 * bond_change_active_slave().
1356 */
1357 if (!USES_PRIMARY(bond->params.mode)) {
1358 /* set promiscuity level to new slave */
1359 if (bond_dev->flags & IFF_PROMISC) {
1360 res = dev_set_promiscuity(slave_dev, 1);
1361 if (res)
1362 goto err_close;
1363 }
1364
1365 /* set allmulti level to new slave */
1366 if (bond_dev->flags & IFF_ALLMULTI) {
1367 res = dev_set_allmulti(slave_dev, 1);
1368 if (res)
1369 goto err_close;
1370 }
1371
1372 netif_addr_lock_bh(bond_dev);
1373
1374 dev_mc_sync_multiple(slave_dev, bond_dev);
1375 dev_uc_sync_multiple(slave_dev, bond_dev);
1376
1377 netif_addr_unlock_bh(bond_dev);
1378 }
1379
1380 if (bond->params.mode == BOND_MODE_8023AD) {
1381 /* add lacpdu mc addr to mc list */
1382 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1383
1384 dev_mc_add(slave_dev, lacpdu_multicast);
1385 }
1386
1387 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1388 if (res) {
1389 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1390 bond_dev->name, slave_dev->name);
1391 goto err_close;
1392 }
1393
1394 prev_slave = bond_last_slave(bond);
1395
1396 new_slave->delay = 0;
1397 new_slave->link_failure_count = 0;
1398
1399 bond_update_speed_duplex(new_slave);
1400
1401 new_slave->last_rx = jiffies -
1402 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1403 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1404 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1405
1406 if (bond->params.miimon && !bond->params.use_carrier) {
1407 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1408
1409 if ((link_reporting == -1) && !bond->params.arp_interval) {
1410 /*
1411 * miimon is set but a bonded network driver
1412 * does not support ETHTOOL/MII and
1413 * arp_interval is not set. Note: if
1414 * use_carrier is enabled, we will never go
1415 * here (because netif_carrier is always
1416 * supported); thus, we don't need to change
1417 * the messages for netif_carrier.
1418 */
1419 pr_warn("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1420 bond_dev->name, slave_dev->name);
1421 } else if (link_reporting == -1) {
1422 /* unable get link status using mii/ethtool */
1423 pr_warn("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
1424 bond_dev->name, slave_dev->name);
1425 }
1426 }
1427
1428 /* check for initial state */
1429 if (bond->params.miimon) {
1430 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1431 if (bond->params.updelay) {
1432 new_slave->link = BOND_LINK_BACK;
1433 new_slave->delay = bond->params.updelay;
1434 } else {
1435 new_slave->link = BOND_LINK_UP;
1436 }
1437 } else {
1438 new_slave->link = BOND_LINK_DOWN;
1439 }
1440 } else if (bond->params.arp_interval) {
1441 new_slave->link = (netif_carrier_ok(slave_dev) ?
1442 BOND_LINK_UP : BOND_LINK_DOWN);
1443 } else {
1444 new_slave->link = BOND_LINK_UP;
1445 }
1446
1447 if (new_slave->link != BOND_LINK_DOWN)
1448 new_slave->last_link_up = jiffies;
1449 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1450 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1451 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1452
1453 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1454 /* if there is a primary slave, remember it */
1455 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1456 bond->primary_slave = new_slave;
1457 bond->force_primary = true;
1458 }
1459 }
1460
1461 switch (bond->params.mode) {
1462 case BOND_MODE_ACTIVEBACKUP:
1463 bond_set_slave_inactive_flags(new_slave,
1464 BOND_SLAVE_NOTIFY_NOW);
1465 break;
1466 case BOND_MODE_8023AD:
1467 /* in 802.3ad mode, the internal mechanism
1468 * will activate the slaves in the selected
1469 * aggregator
1470 */
1471 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1472 /* if this is the first slave */
1473 if (!prev_slave) {
1474 SLAVE_AD_INFO(new_slave).id = 1;
1475 /* Initialize AD with the number of times that the AD timer is called in 1 second
1476 * can be called only after the mac address of the bond is set
1477 */
1478 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1479 } else {
1480 SLAVE_AD_INFO(new_slave).id =
1481 SLAVE_AD_INFO(prev_slave).id + 1;
1482 }
1483
1484 bond_3ad_bind_slave(new_slave);
1485 break;
1486 case BOND_MODE_TLB:
1487 case BOND_MODE_ALB:
1488 bond_set_active_slave(new_slave);
1489 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1490 break;
1491 default:
1492 pr_debug("This slave is always active in trunk mode\n");
1493
1494 /* always active in trunk mode */
1495 bond_set_active_slave(new_slave);
1496
1497 /* In trunking mode there is little meaning to curr_active_slave
1498 * anyway (it holds no special properties of the bond device),
1499 * so we can change it without calling change_active_interface()
1500 */
1501 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1502 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1503
1504 break;
1505 } /* switch(bond_mode) */
1506
1507#ifdef CONFIG_NET_POLL_CONTROLLER
1508 slave_dev->npinfo = bond->dev->npinfo;
1509 if (slave_dev->npinfo) {
1510 if (slave_enable_netpoll(new_slave)) {
1511 pr_info("Error, %s: master_dev is using netpoll, but new slave device does not support netpoll\n",
1512 bond_dev->name);
1513 res = -EBUSY;
1514 goto err_detach;
1515 }
1516 }
1517#endif
1518
1519 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1520 new_slave);
1521 if (res) {
1522 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1523 goto err_detach;
1524 }
1525
1526 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1527 if (res) {
1528 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1529 goto err_unregister;
1530 }
1531
1532 res = bond_sysfs_slave_add(new_slave);
1533 if (res) {
1534 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1535 goto err_upper_unlink;
1536 }
1537
1538 bond->slave_cnt++;
1539 bond_compute_features(bond);
1540 bond_set_carrier(bond);
1541
1542 if (USES_PRIMARY(bond->params.mode)) {
1543 block_netpoll_tx();
1544 write_lock_bh(&bond->curr_slave_lock);
1545 bond_select_active_slave(bond);
1546 write_unlock_bh(&bond->curr_slave_lock);
1547 unblock_netpoll_tx();
1548 }
1549
1550 pr_info("%s: Enslaving %s as %s interface with %s link\n",
1551 bond_dev->name, slave_dev->name,
1552 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1553 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1554
1555 /* enslave is successful */
1556 return 0;
1557
1558/* Undo stages on error */
1559err_upper_unlink:
1560 bond_upper_dev_unlink(bond_dev, slave_dev);
1561
1562err_unregister:
1563 netdev_rx_handler_unregister(slave_dev);
1564
1565err_detach:
1566 if (!USES_PRIMARY(bond->params.mode))
1567 bond_hw_addr_flush(bond_dev, slave_dev);
1568
1569 vlan_vids_del_by_dev(slave_dev, bond_dev);
1570 if (bond->primary_slave == new_slave)
1571 bond->primary_slave = NULL;
1572 if (bond->curr_active_slave == new_slave) {
1573 block_netpoll_tx();
1574 write_lock_bh(&bond->curr_slave_lock);
1575 bond_change_active_slave(bond, NULL);
1576 bond_select_active_slave(bond);
1577 write_unlock_bh(&bond->curr_slave_lock);
1578 unblock_netpoll_tx();
1579 }
1580 slave_disable_netpoll(new_slave);
1581
1582err_close:
1583 slave_dev->priv_flags &= ~IFF_BONDING;
1584 dev_close(slave_dev);
1585
1586err_restore_mac:
1587 if (!bond->params.fail_over_mac ||
1588 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1589 /* XXX TODO - fom follow mode needs to change master's
1590 * MAC if this slave's MAC is in use by the bond, or at
1591 * least print a warning.
1592 */
1593 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1594 addr.sa_family = slave_dev->type;
1595 dev_set_mac_address(slave_dev, &addr);
1596 }
1597
1598err_restore_mtu:
1599 dev_set_mtu(slave_dev, new_slave->original_mtu);
1600
1601err_free:
1602 kfree(new_slave);
1603
1604err_undo_flags:
1605 /* Enslave of first slave has failed and we need to fix master's mac */
1606 if (!bond_has_slaves(bond) &&
1607 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1608 eth_hw_addr_random(bond_dev);
1609
1610 return res;
1611}
1612
1613/*
1614 * Try to release the slave device <slave> from the bond device <master>
1615 * It is legal to access curr_active_slave without a lock because all the function
1616 * is write-locked. If "all" is true it means that the function is being called
1617 * while destroying a bond interface and all slaves are being released.
1618 *
1619 * The rules for slave state should be:
1620 * for Active/Backup:
1621 * Active stays on all backups go down
1622 * for Bonded connections:
1623 * The first up interface should be left on and all others downed.
1624 */
1625static int __bond_release_one(struct net_device *bond_dev,
1626 struct net_device *slave_dev,
1627 bool all)
1628{
1629 struct bonding *bond = netdev_priv(bond_dev);
1630 struct slave *slave, *oldcurrent;
1631 struct sockaddr addr;
1632 int old_flags = bond_dev->flags;
1633 netdev_features_t old_features = bond_dev->features;
1634
1635 /* slave is not a slave or master is not master of this slave */
1636 if (!(slave_dev->flags & IFF_SLAVE) ||
1637 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1638 pr_err("%s: Error: cannot release %s\n",
1639 bond_dev->name, slave_dev->name);
1640 return -EINVAL;
1641 }
1642
1643 block_netpoll_tx();
1644
1645 slave = bond_get_slave_by_dev(bond, slave_dev);
1646 if (!slave) {
1647 /* not a slave of this bond */
1648 pr_info("%s: %s not enslaved\n",
1649 bond_dev->name, slave_dev->name);
1650 unblock_netpoll_tx();
1651 return -EINVAL;
1652 }
1653
1654 bond_sysfs_slave_del(slave);
1655
1656 bond_upper_dev_unlink(bond_dev, slave_dev);
1657 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1658 * for this slave anymore.
1659 */
1660 netdev_rx_handler_unregister(slave_dev);
1661 write_lock_bh(&bond->lock);
1662
1663 /* Inform AD package of unbinding of slave. */
1664 if (bond->params.mode == BOND_MODE_8023AD)
1665 bond_3ad_unbind_slave(slave);
1666
1667 write_unlock_bh(&bond->lock);
1668
1669 pr_info("%s: Releasing %s interface %s\n",
1670 bond_dev->name,
1671 bond_is_active_slave(slave) ? "active" : "backup",
1672 slave_dev->name);
1673
1674 oldcurrent = bond->curr_active_slave;
1675
1676 bond->current_arp_slave = NULL;
1677
1678 if (!all && (!bond->params.fail_over_mac ||
1679 bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1680 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1681 bond_has_slaves(bond))
1682 pr_warn("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1683 bond_dev->name, slave_dev->name,
1684 slave->perm_hwaddr,
1685 bond_dev->name, slave_dev->name);
1686 }
1687
1688 if (bond->primary_slave == slave)
1689 bond->primary_slave = NULL;
1690
1691 if (oldcurrent == slave) {
1692 write_lock_bh(&bond->curr_slave_lock);
1693 bond_change_active_slave(bond, NULL);
1694 write_unlock_bh(&bond->curr_slave_lock);
1695 }
1696
1697 if (bond_is_lb(bond)) {
1698 /* Must be called only after the slave has been
1699 * detached from the list and the curr_active_slave
1700 * has been cleared (if our_slave == old_current),
1701 * but before a new active slave is selected.
1702 */
1703 bond_alb_deinit_slave(bond, slave);
1704 }
1705
1706 if (all) {
1707 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1708 } else if (oldcurrent == slave) {
1709 /*
1710 * Note that we hold RTNL over this sequence, so there
1711 * is no concern that another slave add/remove event
1712 * will interfere.
1713 */
1714 write_lock_bh(&bond->curr_slave_lock);
1715
1716 bond_select_active_slave(bond);
1717
1718 write_unlock_bh(&bond->curr_slave_lock);
1719 }
1720
1721 if (!bond_has_slaves(bond)) {
1722 bond_set_carrier(bond);
1723 eth_hw_addr_random(bond_dev);
1724
1725 if (vlan_uses_dev(bond_dev)) {
1726 pr_warn("%s: Warning: clearing HW address of %s while it still has VLANs\n",
1727 bond_dev->name, bond_dev->name);
1728 pr_warn("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs\n",
1729 bond_dev->name);
1730 }
1731 }
1732
1733 unblock_netpoll_tx();
1734 synchronize_rcu();
1735 bond->slave_cnt--;
1736
1737 if (!bond_has_slaves(bond)) {
1738 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1739 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1740 }
1741
1742 bond_compute_features(bond);
1743 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1744 (old_features & NETIF_F_VLAN_CHALLENGED))
1745 pr_info("%s: last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1746 bond_dev->name, slave_dev->name, bond_dev->name);
1747
1748 /* must do this from outside any spinlocks */
1749 vlan_vids_del_by_dev(slave_dev, bond_dev);
1750
1751 /* If the mode USES_PRIMARY, then this cases was handled above by
1752 * bond_change_active_slave(..., NULL)
1753 */
1754 if (!USES_PRIMARY(bond->params.mode)) {
1755 /* unset promiscuity level from slave
1756 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1757 * of the IFF_PROMISC flag in the bond_dev, but we need the
1758 * value of that flag before that change, as that was the value
1759 * when this slave was attached, so we cache at the start of the
1760 * function and use it here. Same goes for ALLMULTI below
1761 */
1762 if (old_flags & IFF_PROMISC)
1763 dev_set_promiscuity(slave_dev, -1);
1764
1765 /* unset allmulti level from slave */
1766 if (old_flags & IFF_ALLMULTI)
1767 dev_set_allmulti(slave_dev, -1);
1768
1769 bond_hw_addr_flush(bond_dev, slave_dev);
1770 }
1771
1772 slave_disable_netpoll(slave);
1773
1774 /* close slave before restoring its mac address */
1775 dev_close(slave_dev);
1776
1777 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1778 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1779 /* restore original ("permanent") mac address */
1780 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1781 addr.sa_family = slave_dev->type;
1782 dev_set_mac_address(slave_dev, &addr);
1783 }
1784
1785 dev_set_mtu(slave_dev, slave->original_mtu);
1786
1787 slave_dev->priv_flags &= ~IFF_BONDING;
1788
1789 kfree(slave);
1790
1791 return 0; /* deletion OK */
1792}
1793
1794/* A wrapper used because of ndo_del_link */
1795int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1796{
1797 return __bond_release_one(bond_dev, slave_dev, false);
1798}
1799
1800/*
1801* First release a slave and then destroy the bond if no more slaves are left.
1802* Must be under rtnl_lock when this function is called.
1803*/
1804static int bond_release_and_destroy(struct net_device *bond_dev,
1805 struct net_device *slave_dev)
1806{
1807 struct bonding *bond = netdev_priv(bond_dev);
1808 int ret;
1809
1810 ret = bond_release(bond_dev, slave_dev);
1811 if (ret == 0 && !bond_has_slaves(bond)) {
1812 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1813 pr_info("%s: Destroying bond %s\n",
1814 bond_dev->name, bond_dev->name);
1815 unregister_netdevice(bond_dev);
1816 }
1817 return ret;
1818}
1819
1820static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1821{
1822 struct bonding *bond = netdev_priv(bond_dev);
1823
1824 info->bond_mode = bond->params.mode;
1825 info->miimon = bond->params.miimon;
1826
1827 info->num_slaves = bond->slave_cnt;
1828
1829 return 0;
1830}
1831
1832static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1833{
1834 struct bonding *bond = netdev_priv(bond_dev);
1835 struct list_head *iter;
1836 int i = 0, res = -ENODEV;
1837 struct slave *slave;
1838
1839 bond_for_each_slave(bond, slave, iter) {
1840 if (i++ == (int)info->slave_id) {
1841 res = 0;
1842 strcpy(info->slave_name, slave->dev->name);
1843 info->link = slave->link;
1844 info->state = bond_slave_state(slave);
1845 info->link_failure_count = slave->link_failure_count;
1846 break;
1847 }
1848 }
1849
1850 return res;
1851}
1852
1853/*-------------------------------- Monitoring -------------------------------*/
1854
1855
1856static int bond_miimon_inspect(struct bonding *bond)
1857{
1858 int link_state, commit = 0;
1859 struct list_head *iter;
1860 struct slave *slave;
1861 bool ignore_updelay;
1862
1863 ignore_updelay = !bond->curr_active_slave ? true : false;
1864
1865 bond_for_each_slave_rcu(bond, slave, iter) {
1866 slave->new_link = BOND_LINK_NOCHANGE;
1867
1868 link_state = bond_check_dev_link(bond, slave->dev, 0);
1869
1870 switch (slave->link) {
1871 case BOND_LINK_UP:
1872 if (link_state)
1873 continue;
1874
1875 slave->link = BOND_LINK_FAIL;
1876 slave->delay = bond->params.downdelay;
1877 if (slave->delay) {
1878 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms\n",
1879 bond->dev->name,
1880 (bond->params.mode ==
1881 BOND_MODE_ACTIVEBACKUP) ?
1882 (bond_is_active_slave(slave) ?
1883 "active " : "backup ") : "",
1884 slave->dev->name,
1885 bond->params.downdelay * bond->params.miimon);
1886 }
1887 /*FALLTHRU*/
1888 case BOND_LINK_FAIL:
1889 if (link_state) {
1890 /*
1891 * recovered before downdelay expired
1892 */
1893 slave->link = BOND_LINK_UP;
1894 slave->last_link_up = jiffies;
1895 pr_info("%s: link status up again after %d ms for interface %s\n",
1896 bond->dev->name,
1897 (bond->params.downdelay - slave->delay) *
1898 bond->params.miimon,
1899 slave->dev->name);
1900 continue;
1901 }
1902
1903 if (slave->delay <= 0) {
1904 slave->new_link = BOND_LINK_DOWN;
1905 commit++;
1906 continue;
1907 }
1908
1909 slave->delay--;
1910 break;
1911
1912 case BOND_LINK_DOWN:
1913 if (!link_state)
1914 continue;
1915
1916 slave->link = BOND_LINK_BACK;
1917 slave->delay = bond->params.updelay;
1918
1919 if (slave->delay) {
1920 pr_info("%s: link status up for interface %s, enabling it in %d ms\n",
1921 bond->dev->name, slave->dev->name,
1922 ignore_updelay ? 0 :
1923 bond->params.updelay *
1924 bond->params.miimon);
1925 }
1926 /*FALLTHRU*/
1927 case BOND_LINK_BACK:
1928 if (!link_state) {
1929 slave->link = BOND_LINK_DOWN;
1930 pr_info("%s: link status down again after %d ms for interface %s\n",
1931 bond->dev->name,
1932 (bond->params.updelay - slave->delay) *
1933 bond->params.miimon,
1934 slave->dev->name);
1935
1936 continue;
1937 }
1938
1939 if (ignore_updelay)
1940 slave->delay = 0;
1941
1942 if (slave->delay <= 0) {
1943 slave->new_link = BOND_LINK_UP;
1944 commit++;
1945 ignore_updelay = false;
1946 continue;
1947 }
1948
1949 slave->delay--;
1950 break;
1951 }
1952 }
1953
1954 return commit;
1955}
1956
1957static void bond_miimon_commit(struct bonding *bond)
1958{
1959 struct list_head *iter;
1960 struct slave *slave;
1961
1962 bond_for_each_slave(bond, slave, iter) {
1963 switch (slave->new_link) {
1964 case BOND_LINK_NOCHANGE:
1965 continue;
1966
1967 case BOND_LINK_UP:
1968 slave->link = BOND_LINK_UP;
1969 slave->last_link_up = jiffies;
1970
1971 if (bond->params.mode == BOND_MODE_8023AD) {
1972 /* prevent it from being the active one */
1973 bond_set_backup_slave(slave);
1974 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1975 /* make it immediately active */
1976 bond_set_active_slave(slave);
1977 } else if (slave != bond->primary_slave) {
1978 /* prevent it from being the active one */
1979 bond_set_backup_slave(slave);
1980 }
1981
1982 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex\n",
1983 bond->dev->name, slave->dev->name,
1984 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1985 slave->duplex ? "full" : "half");
1986
1987 /* notify ad that the link status has changed */
1988 if (bond->params.mode == BOND_MODE_8023AD)
1989 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1990
1991 if (bond_is_lb(bond))
1992 bond_alb_handle_link_change(bond, slave,
1993 BOND_LINK_UP);
1994
1995 if (!bond->curr_active_slave ||
1996 (slave == bond->primary_slave))
1997 goto do_failover;
1998
1999 continue;
2000
2001 case BOND_LINK_DOWN:
2002 if (slave->link_failure_count < UINT_MAX)
2003 slave->link_failure_count++;
2004
2005 slave->link = BOND_LINK_DOWN;
2006
2007 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2008 bond->params.mode == BOND_MODE_8023AD)
2009 bond_set_slave_inactive_flags(slave,
2010 BOND_SLAVE_NOTIFY_NOW);
2011
2012 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2013 bond->dev->name, slave->dev->name);
2014
2015 if (bond->params.mode == BOND_MODE_8023AD)
2016 bond_3ad_handle_link_change(slave,
2017 BOND_LINK_DOWN);
2018
2019 if (bond_is_lb(bond))
2020 bond_alb_handle_link_change(bond, slave,
2021 BOND_LINK_DOWN);
2022
2023 if (slave == bond->curr_active_slave)
2024 goto do_failover;
2025
2026 continue;
2027
2028 default:
2029 pr_err("%s: invalid new link %d on slave %s\n",
2030 bond->dev->name, slave->new_link,
2031 slave->dev->name);
2032 slave->new_link = BOND_LINK_NOCHANGE;
2033
2034 continue;
2035 }
2036
2037do_failover:
2038 ASSERT_RTNL();
2039 block_netpoll_tx();
2040 write_lock_bh(&bond->curr_slave_lock);
2041 bond_select_active_slave(bond);
2042 write_unlock_bh(&bond->curr_slave_lock);
2043 unblock_netpoll_tx();
2044 }
2045
2046 bond_set_carrier(bond);
2047}
2048
2049/*
2050 * bond_mii_monitor
2051 *
2052 * Really a wrapper that splits the mii monitor into two phases: an
2053 * inspection, then (if inspection indicates something needs to be done)
2054 * an acquisition of appropriate locks followed by a commit phase to
2055 * implement whatever link state changes are indicated.
2056 */
2057static void bond_mii_monitor(struct work_struct *work)
2058{
2059 struct bonding *bond = container_of(work, struct bonding,
2060 mii_work.work);
2061 bool should_notify_peers = false;
2062 unsigned long delay;
2063
2064 delay = msecs_to_jiffies(bond->params.miimon);
2065
2066 if (!bond_has_slaves(bond))
2067 goto re_arm;
2068
2069 rcu_read_lock();
2070
2071 should_notify_peers = bond_should_notify_peers(bond);
2072
2073 if (bond_miimon_inspect(bond)) {
2074 rcu_read_unlock();
2075
2076 /* Race avoidance with bond_close cancel of workqueue */
2077 if (!rtnl_trylock()) {
2078 delay = 1;
2079 should_notify_peers = false;
2080 goto re_arm;
2081 }
2082
2083 bond_miimon_commit(bond);
2084
2085 rtnl_unlock(); /* might sleep, hold no other locks */
2086 } else
2087 rcu_read_unlock();
2088
2089re_arm:
2090 if (bond->params.miimon)
2091 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2092
2093 if (should_notify_peers) {
2094 if (!rtnl_trylock())
2095 return;
2096 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2097 rtnl_unlock();
2098 }
2099}
2100
2101static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2102{
2103 struct net_device *upper;
2104 struct list_head *iter;
2105 bool ret = false;
2106
2107 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2108 return true;
2109
2110 rcu_read_lock();
2111 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2112 if (ip == bond_confirm_addr(upper, 0, ip)) {
2113 ret = true;
2114 break;
2115 }
2116 }
2117 rcu_read_unlock();
2118
2119 return ret;
2120}
2121
2122/*
2123 * We go to the (large) trouble of VLAN tagging ARP frames because
2124 * switches in VLAN mode (especially if ports are configured as
2125 * "native" to a VLAN) might not pass non-tagged frames.
2126 */
2127static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2128 __be32 dest_ip, __be32 src_ip,
2129 struct bond_vlan_tag *tags)
2130{
2131 struct sk_buff *skb;
2132 int i;
2133
2134 pr_debug("arp %d on slave %s: dst %pI4 src %pI4\n",
2135 arp_op, slave_dev->name, &dest_ip, &src_ip);
2136
2137 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2138 NULL, slave_dev->dev_addr, NULL);
2139
2140 if (!skb) {
2141 net_err_ratelimited("ARP packet allocation failed\n");
2142 return;
2143 }
2144
2145 /* Go through all the tags backwards and add them to the packet */
2146 for (i = BOND_MAX_VLAN_ENCAP - 1; i > 0; i--) {
2147 if (!tags[i].vlan_id)
2148 continue;
2149
2150 pr_debug("inner tag: proto %X vid %X\n",
2151 ntohs(tags[i].vlan_proto), tags[i].vlan_id);
2152 skb = __vlan_put_tag(skb, tags[i].vlan_proto,
2153 tags[i].vlan_id);
2154 if (!skb) {
2155 net_err_ratelimited("failed to insert inner VLAN tag\n");
2156 return;
2157 }
2158 }
2159 /* Set the outer tag */
2160 if (tags[0].vlan_id) {
2161 pr_debug("outer tag: proto %X vid %X\n",
2162 ntohs(tags[0].vlan_proto), tags[0].vlan_id);
2163 skb = vlan_put_tag(skb, tags[0].vlan_proto, tags[0].vlan_id);
2164 if (!skb) {
2165 net_err_ratelimited("failed to insert outer VLAN tag\n");
2166 return;
2167 }
2168 }
2169 arp_xmit(skb);
2170}
2171
2172/* Validate the device path between the @start_dev and the @end_dev.
2173 * The path is valid if the @end_dev is reachable through device
2174 * stacking.
2175 * When the path is validated, collect any vlan information in the
2176 * path.
2177 */
2178static bool bond_verify_device_path(struct net_device *start_dev,
2179 struct net_device *end_dev,
2180 struct bond_vlan_tag *tags)
2181{
2182 struct net_device *upper;
2183 struct list_head *iter;
2184 int idx;
2185
2186 if (start_dev == end_dev)
2187 return true;
2188
2189 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2190 if (bond_verify_device_path(upper, end_dev, tags)) {
2191 if (is_vlan_dev(upper)) {
2192 idx = vlan_get_encap_level(upper);
2193 if (idx >= BOND_MAX_VLAN_ENCAP)
2194 return false;
2195
2196 tags[idx].vlan_proto =
2197 vlan_dev_vlan_proto(upper);
2198 tags[idx].vlan_id = vlan_dev_vlan_id(upper);
2199 }
2200 return true;
2201 }
2202 }
2203
2204 return false;
2205}
2206
2207static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2208{
2209 struct rtable *rt;
2210 struct bond_vlan_tag tags[BOND_MAX_VLAN_ENCAP];
2211 __be32 *targets = bond->params.arp_targets, addr;
2212 int i;
2213 bool ret;
2214
2215 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2216 pr_debug("basa: target %pI4\n", &targets[i]);
2217 memset(tags, 0, sizeof(tags));
2218
2219 /* Find out through which dev should the packet go */
2220 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2221 RTO_ONLINK, 0);
2222 if (IS_ERR(rt)) {
2223 /* there's no route to target - try to send arp
2224 * probe to generate any traffic (arp_validate=0)
2225 */
2226 if (bond->params.arp_validate)
2227 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2228 bond->dev->name,
2229 &targets[i]);
2230 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2231 0, tags);
2232 continue;
2233 }
2234
2235 /* bond device itself */
2236 if (rt->dst.dev == bond->dev)
2237 goto found;
2238
2239 rcu_read_lock();
2240 ret = bond_verify_device_path(bond->dev, rt->dst.dev, tags);
2241 rcu_read_unlock();
2242
2243 if (ret)
2244 goto found;
2245
2246 /* Not our device - skip */
2247 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2248 bond->dev->name, &targets[i],
2249 rt->dst.dev ? rt->dst.dev->name : "NULL");
2250
2251 ip_rt_put(rt);
2252 continue;
2253
2254found:
2255 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2256 ip_rt_put(rt);
2257 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2258 addr, tags);
2259 }
2260}
2261
2262static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2263{
2264 int i;
2265
2266 if (!sip || !bond_has_this_ip(bond, tip)) {
2267 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2268 return;
2269 }
2270
2271 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2272 if (i == -1) {
2273 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2274 return;
2275 }
2276 slave->last_rx = jiffies;
2277 slave->target_last_arp_rx[i] = jiffies;
2278}
2279
2280int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2281 struct slave *slave)
2282{
2283 struct arphdr *arp = (struct arphdr *)skb->data;
2284 struct slave *curr_active_slave;
2285 unsigned char *arp_ptr;
2286 __be32 sip, tip;
2287 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2288
2289 if (!slave_do_arp_validate(bond, slave)) {
2290 if ((slave_do_arp_validate_only(bond, slave) && is_arp) ||
2291 !slave_do_arp_validate_only(bond, slave))
2292 slave->last_rx = jiffies;
2293 return RX_HANDLER_ANOTHER;
2294 } else if (!is_arp) {
2295 return RX_HANDLER_ANOTHER;
2296 }
2297
2298 alen = arp_hdr_len(bond->dev);
2299
2300 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2301 bond->dev->name, skb->dev->name);
2302
2303 if (alen > skb_headlen(skb)) {
2304 arp = kmalloc(alen, GFP_ATOMIC);
2305 if (!arp)
2306 goto out_unlock;
2307 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2308 goto out_unlock;
2309 }
2310
2311 if (arp->ar_hln != bond->dev->addr_len ||
2312 skb->pkt_type == PACKET_OTHERHOST ||
2313 skb->pkt_type == PACKET_LOOPBACK ||
2314 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2315 arp->ar_pro != htons(ETH_P_IP) ||
2316 arp->ar_pln != 4)
2317 goto out_unlock;
2318
2319 arp_ptr = (unsigned char *)(arp + 1);
2320 arp_ptr += bond->dev->addr_len;
2321 memcpy(&sip, arp_ptr, 4);
2322 arp_ptr += 4 + bond->dev->addr_len;
2323 memcpy(&tip, arp_ptr, 4);
2324
2325 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2326 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2327 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2328 &sip, &tip);
2329
2330 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2331
2332 /*
2333 * Backup slaves won't see the ARP reply, but do come through
2334 * here for each ARP probe (so we swap the sip/tip to validate
2335 * the probe). In a "redundant switch, common router" type of
2336 * configuration, the ARP probe will (hopefully) travel from
2337 * the active, through one switch, the router, then the other
2338 * switch before reaching the backup.
2339 *
2340 * We 'trust' the arp requests if there is an active slave and
2341 * it received valid arp reply(s) after it became active. This
2342 * is done to avoid endless looping when we can't reach the
2343 * arp_ip_target and fool ourselves with our own arp requests.
2344 */
2345
2346 if (bond_is_active_slave(slave))
2347 bond_validate_arp(bond, slave, sip, tip);
2348 else if (curr_active_slave &&
2349 time_after(slave_last_rx(bond, curr_active_slave),
2350 curr_active_slave->last_link_up))
2351 bond_validate_arp(bond, slave, tip, sip);
2352
2353out_unlock:
2354 if (arp != (struct arphdr *)skb->data)
2355 kfree(arp);
2356 return RX_HANDLER_ANOTHER;
2357}
2358
2359/* function to verify if we're in the arp_interval timeslice, returns true if
2360 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2361 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2362 */
2363static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2364 int mod)
2365{
2366 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2367
2368 return time_in_range(jiffies,
2369 last_act - delta_in_ticks,
2370 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2371}
2372
2373/*
2374 * this function is called regularly to monitor each slave's link
2375 * ensuring that traffic is being sent and received when arp monitoring
2376 * is used in load-balancing mode. if the adapter has been dormant, then an
2377 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2378 * arp monitoring in active backup mode.
2379 */
2380static void bond_loadbalance_arp_mon(struct work_struct *work)
2381{
2382 struct bonding *bond = container_of(work, struct bonding,
2383 arp_work.work);
2384 struct slave *slave, *oldcurrent;
2385 struct list_head *iter;
2386 int do_failover = 0, slave_state_changed = 0;
2387
2388 if (!bond_has_slaves(bond))
2389 goto re_arm;
2390
2391 rcu_read_lock();
2392
2393 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2394 /* see if any of the previous devices are up now (i.e. they have
2395 * xmt and rcv traffic). the curr_active_slave does not come into
2396 * the picture unless it is null. also, slave->last_link_up is not
2397 * needed here because we send an arp on each slave and give a slave
2398 * as long as it needs to get the tx/rx within the delta.
2399 * TODO: what about up/down delay in arp mode? it wasn't here before
2400 * so it can wait
2401 */
2402 bond_for_each_slave_rcu(bond, slave, iter) {
2403 unsigned long trans_start = dev_trans_start(slave->dev);
2404
2405 if (slave->link != BOND_LINK_UP) {
2406 if (bond_time_in_interval(bond, trans_start, 1) &&
2407 bond_time_in_interval(bond, slave->last_rx, 1)) {
2408
2409 slave->link = BOND_LINK_UP;
2410 slave_state_changed = 1;
2411
2412 /* primary_slave has no meaning in round-robin
2413 * mode. the window of a slave being up and
2414 * curr_active_slave being null after enslaving
2415 * is closed.
2416 */
2417 if (!oldcurrent) {
2418 pr_info("%s: link status definitely up for interface %s\n",
2419 bond->dev->name,
2420 slave->dev->name);
2421 do_failover = 1;
2422 } else {
2423 pr_info("%s: interface %s is now up\n",
2424 bond->dev->name,
2425 slave->dev->name);
2426 }
2427 }
2428 } else {
2429 /* slave->link == BOND_LINK_UP */
2430
2431 /* not all switches will respond to an arp request
2432 * when the source ip is 0, so don't take the link down
2433 * if we don't know our ip yet
2434 */
2435 if (!bond_time_in_interval(bond, trans_start, 2) ||
2436 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2437
2438 slave->link = BOND_LINK_DOWN;
2439 slave_state_changed = 1;
2440
2441 if (slave->link_failure_count < UINT_MAX)
2442 slave->link_failure_count++;
2443
2444 pr_info("%s: interface %s is now down\n",
2445 bond->dev->name, slave->dev->name);
2446
2447 if (slave == oldcurrent)
2448 do_failover = 1;
2449 }
2450 }
2451
2452 /* note: if switch is in round-robin mode, all links
2453 * must tx arp to ensure all links rx an arp - otherwise
2454 * links may oscillate or not come up at all; if switch is
2455 * in something like xor mode, there is nothing we can
2456 * do - all replies will be rx'ed on same link causing slaves
2457 * to be unstable during low/no traffic periods
2458 */
2459 if (IS_UP(slave->dev))
2460 bond_arp_send_all(bond, slave);
2461 }
2462
2463 rcu_read_unlock();
2464
2465 if (do_failover || slave_state_changed) {
2466 if (!rtnl_trylock())
2467 goto re_arm;
2468
2469 if (slave_state_changed) {
2470 bond_slave_state_change(bond);
2471 } else if (do_failover) {
2472 /* the bond_select_active_slave must hold RTNL
2473 * and curr_slave_lock for write.
2474 */
2475 block_netpoll_tx();
2476 write_lock_bh(&bond->curr_slave_lock);
2477
2478 bond_select_active_slave(bond);
2479
2480 write_unlock_bh(&bond->curr_slave_lock);
2481 unblock_netpoll_tx();
2482 }
2483 rtnl_unlock();
2484 }
2485
2486re_arm:
2487 if (bond->params.arp_interval)
2488 queue_delayed_work(bond->wq, &bond->arp_work,
2489 msecs_to_jiffies(bond->params.arp_interval));
2490}
2491
2492/*
2493 * Called to inspect slaves for active-backup mode ARP monitor link state
2494 * changes. Sets new_link in slaves to specify what action should take
2495 * place for the slave. Returns 0 if no changes are found, >0 if changes
2496 * to link states must be committed.
2497 *
2498 * Called with rcu_read_lock hold.
2499 */
2500static int bond_ab_arp_inspect(struct bonding *bond)
2501{
2502 unsigned long trans_start, last_rx;
2503 struct list_head *iter;
2504 struct slave *slave;
2505 int commit = 0;
2506
2507 bond_for_each_slave_rcu(bond, slave, iter) {
2508 slave->new_link = BOND_LINK_NOCHANGE;
2509 last_rx = slave_last_rx(bond, slave);
2510
2511 if (slave->link != BOND_LINK_UP) {
2512 if (bond_time_in_interval(bond, last_rx, 1)) {
2513 slave->new_link = BOND_LINK_UP;
2514 commit++;
2515 }
2516 continue;
2517 }
2518
2519 /*
2520 * Give slaves 2*delta after being enslaved or made
2521 * active. This avoids bouncing, as the last receive
2522 * times need a full ARP monitor cycle to be updated.
2523 */
2524 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2525 continue;
2526
2527 /*
2528 * Backup slave is down if:
2529 * - No current_arp_slave AND
2530 * - more than 3*delta since last receive AND
2531 * - the bond has an IP address
2532 *
2533 * Note: a non-null current_arp_slave indicates
2534 * the curr_active_slave went down and we are
2535 * searching for a new one; under this condition
2536 * we only take the curr_active_slave down - this
2537 * gives each slave a chance to tx/rx traffic
2538 * before being taken out
2539 */
2540 if (!bond_is_active_slave(slave) &&
2541 !bond->current_arp_slave &&
2542 !bond_time_in_interval(bond, last_rx, 3)) {
2543 slave->new_link = BOND_LINK_DOWN;
2544 commit++;
2545 }
2546
2547 /*
2548 * Active slave is down if:
2549 * - more than 2*delta since transmitting OR
2550 * - (more than 2*delta since receive AND
2551 * the bond has an IP address)
2552 */
2553 trans_start = dev_trans_start(slave->dev);
2554 if (bond_is_active_slave(slave) &&
2555 (!bond_time_in_interval(bond, trans_start, 2) ||
2556 !bond_time_in_interval(bond, last_rx, 2))) {
2557 slave->new_link = BOND_LINK_DOWN;
2558 commit++;
2559 }
2560 }
2561
2562 return commit;
2563}
2564
2565/*
2566 * Called to commit link state changes noted by inspection step of
2567 * active-backup mode ARP monitor.
2568 *
2569 * Called with RTNL hold.
2570 */
2571static void bond_ab_arp_commit(struct bonding *bond)
2572{
2573 unsigned long trans_start;
2574 struct list_head *iter;
2575 struct slave *slave;
2576
2577 bond_for_each_slave(bond, slave, iter) {
2578 switch (slave->new_link) {
2579 case BOND_LINK_NOCHANGE:
2580 continue;
2581
2582 case BOND_LINK_UP:
2583 trans_start = dev_trans_start(slave->dev);
2584 if (bond->curr_active_slave != slave ||
2585 (!bond->curr_active_slave &&
2586 bond_time_in_interval(bond, trans_start, 1))) {
2587 slave->link = BOND_LINK_UP;
2588 if (bond->current_arp_slave) {
2589 bond_set_slave_inactive_flags(
2590 bond->current_arp_slave,
2591 BOND_SLAVE_NOTIFY_NOW);
2592 bond->current_arp_slave = NULL;
2593 }
2594
2595 pr_info("%s: link status definitely up for interface %s\n",
2596 bond->dev->name, slave->dev->name);
2597
2598 if (!bond->curr_active_slave ||
2599 (slave == bond->primary_slave))
2600 goto do_failover;
2601
2602 }
2603
2604 continue;
2605
2606 case BOND_LINK_DOWN:
2607 if (slave->link_failure_count < UINT_MAX)
2608 slave->link_failure_count++;
2609
2610 slave->link = BOND_LINK_DOWN;
2611 bond_set_slave_inactive_flags(slave,
2612 BOND_SLAVE_NOTIFY_NOW);
2613
2614 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2615 bond->dev->name, slave->dev->name);
2616
2617 if (slave == bond->curr_active_slave) {
2618 bond->current_arp_slave = NULL;
2619 goto do_failover;
2620 }
2621
2622 continue;
2623
2624 default:
2625 pr_err("%s: impossible: new_link %d on slave %s\n",
2626 bond->dev->name, slave->new_link,
2627 slave->dev->name);
2628 continue;
2629 }
2630
2631do_failover:
2632 ASSERT_RTNL();
2633 block_netpoll_tx();
2634 write_lock_bh(&bond->curr_slave_lock);
2635 bond_select_active_slave(bond);
2636 write_unlock_bh(&bond->curr_slave_lock);
2637 unblock_netpoll_tx();
2638 }
2639
2640 bond_set_carrier(bond);
2641}
2642
2643/*
2644 * Send ARP probes for active-backup mode ARP monitor.
2645 *
2646 * Called with rcu_read_lock hold.
2647 */
2648static bool bond_ab_arp_probe(struct bonding *bond)
2649{
2650 struct slave *slave, *before = NULL, *new_slave = NULL,
2651 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2652 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2653 struct list_head *iter;
2654 bool found = false;
2655 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2656
2657 if (curr_arp_slave && curr_active_slave)
2658 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2659 curr_arp_slave->dev->name,
2660 curr_active_slave->dev->name);
2661
2662 if (curr_active_slave) {
2663 bond_arp_send_all(bond, curr_active_slave);
2664 return should_notify_rtnl;
2665 }
2666
2667 /* if we don't have a curr_active_slave, search for the next available
2668 * backup slave from the current_arp_slave and make it the candidate
2669 * for becoming the curr_active_slave
2670 */
2671
2672 if (!curr_arp_slave) {
2673 curr_arp_slave = bond_first_slave_rcu(bond);
2674 if (!curr_arp_slave)
2675 return should_notify_rtnl;
2676 }
2677
2678 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2679
2680 bond_for_each_slave_rcu(bond, slave, iter) {
2681 if (!found && !before && IS_UP(slave->dev))
2682 before = slave;
2683
2684 if (found && !new_slave && IS_UP(slave->dev))
2685 new_slave = slave;
2686 /* if the link state is up at this point, we
2687 * mark it down - this can happen if we have
2688 * simultaneous link failures and
2689 * reselect_active_interface doesn't make this
2690 * one the current slave so it is still marked
2691 * up when it is actually down
2692 */
2693 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2694 slave->link = BOND_LINK_DOWN;
2695 if (slave->link_failure_count < UINT_MAX)
2696 slave->link_failure_count++;
2697
2698 bond_set_slave_inactive_flags(slave,
2699 BOND_SLAVE_NOTIFY_LATER);
2700
2701 pr_info("%s: backup interface %s is now down\n",
2702 bond->dev->name, slave->dev->name);
2703 }
2704 if (slave == curr_arp_slave)
2705 found = true;
2706 }
2707
2708 if (!new_slave && before)
2709 new_slave = before;
2710
2711 if (!new_slave)
2712 goto check_state;
2713
2714 new_slave->link = BOND_LINK_BACK;
2715 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2716 bond_arp_send_all(bond, new_slave);
2717 new_slave->last_link_up = jiffies;
2718 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2719
2720check_state:
2721 bond_for_each_slave_rcu(bond, slave, iter) {
2722 if (slave->should_notify) {
2723 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2724 break;
2725 }
2726 }
2727 return should_notify_rtnl;
2728}
2729
2730static void bond_activebackup_arp_mon(struct work_struct *work)
2731{
2732 struct bonding *bond = container_of(work, struct bonding,
2733 arp_work.work);
2734 bool should_notify_peers = false;
2735 bool should_notify_rtnl = false;
2736 int delta_in_ticks;
2737
2738 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2739
2740 if (!bond_has_slaves(bond))
2741 goto re_arm;
2742
2743 rcu_read_lock();
2744
2745 should_notify_peers = bond_should_notify_peers(bond);
2746
2747 if (bond_ab_arp_inspect(bond)) {
2748 rcu_read_unlock();
2749
2750 /* Race avoidance with bond_close flush of workqueue */
2751 if (!rtnl_trylock()) {
2752 delta_in_ticks = 1;
2753 should_notify_peers = false;
2754 goto re_arm;
2755 }
2756
2757 bond_ab_arp_commit(bond);
2758
2759 rtnl_unlock();
2760 rcu_read_lock();
2761 }
2762
2763 should_notify_rtnl = bond_ab_arp_probe(bond);
2764 rcu_read_unlock();
2765
2766re_arm:
2767 if (bond->params.arp_interval)
2768 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2769
2770 if (should_notify_peers || should_notify_rtnl) {
2771 if (!rtnl_trylock())
2772 return;
2773
2774 if (should_notify_peers)
2775 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2776 bond->dev);
2777 if (should_notify_rtnl)
2778 bond_slave_state_notify(bond);
2779
2780 rtnl_unlock();
2781 }
2782}
2783
2784/*-------------------------- netdev event handling --------------------------*/
2785
2786/*
2787 * Change device name
2788 */
2789static int bond_event_changename(struct bonding *bond)
2790{
2791 bond_remove_proc_entry(bond);
2792 bond_create_proc_entry(bond);
2793
2794 bond_debug_reregister(bond);
2795
2796 return NOTIFY_DONE;
2797}
2798
2799static int bond_master_netdev_event(unsigned long event,
2800 struct net_device *bond_dev)
2801{
2802 struct bonding *event_bond = netdev_priv(bond_dev);
2803
2804 switch (event) {
2805 case NETDEV_CHANGENAME:
2806 return bond_event_changename(event_bond);
2807 case NETDEV_UNREGISTER:
2808 bond_remove_proc_entry(event_bond);
2809 break;
2810 case NETDEV_REGISTER:
2811 bond_create_proc_entry(event_bond);
2812 break;
2813 case NETDEV_NOTIFY_PEERS:
2814 if (event_bond->send_peer_notif)
2815 event_bond->send_peer_notif--;
2816 break;
2817 default:
2818 break;
2819 }
2820
2821 return NOTIFY_DONE;
2822}
2823
2824static int bond_slave_netdev_event(unsigned long event,
2825 struct net_device *slave_dev)
2826{
2827 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2828 struct bonding *bond;
2829 struct net_device *bond_dev;
2830 u32 old_speed;
2831 u8 old_duplex;
2832
2833 /* A netdev event can be generated while enslaving a device
2834 * before netdev_rx_handler_register is called in which case
2835 * slave will be NULL
2836 */
2837 if (!slave)
2838 return NOTIFY_DONE;
2839 bond_dev = slave->bond->dev;
2840 bond = slave->bond;
2841
2842 switch (event) {
2843 case NETDEV_UNREGISTER:
2844 if (bond_dev->type != ARPHRD_ETHER)
2845 bond_release_and_destroy(bond_dev, slave_dev);
2846 else
2847 bond_release(bond_dev, slave_dev);
2848 break;
2849 case NETDEV_UP:
2850 case NETDEV_CHANGE:
2851 old_speed = slave->speed;
2852 old_duplex = slave->duplex;
2853
2854 bond_update_speed_duplex(slave);
2855
2856 if (bond->params.mode == BOND_MODE_8023AD) {
2857 if (old_speed != slave->speed)
2858 bond_3ad_adapter_speed_changed(slave);
2859 if (old_duplex != slave->duplex)
2860 bond_3ad_adapter_duplex_changed(slave);
2861 }
2862 break;
2863 case NETDEV_DOWN:
2864 /*
2865 * ... Or is it this?
2866 */
2867 break;
2868 case NETDEV_CHANGEMTU:
2869 /*
2870 * TODO: Should slaves be allowed to
2871 * independently alter their MTU? For
2872 * an active-backup bond, slaves need
2873 * not be the same type of device, so
2874 * MTUs may vary. For other modes,
2875 * slaves arguably should have the
2876 * same MTUs. To do this, we'd need to
2877 * take over the slave's change_mtu
2878 * function for the duration of their
2879 * servitude.
2880 */
2881 break;
2882 case NETDEV_CHANGENAME:
2883 /* we don't care if we don't have primary set */
2884 if (!USES_PRIMARY(bond->params.mode) ||
2885 !bond->params.primary[0])
2886 break;
2887
2888 if (slave == bond->primary_slave) {
2889 /* slave's name changed - he's no longer primary */
2890 bond->primary_slave = NULL;
2891 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2892 /* we have a new primary slave */
2893 bond->primary_slave = slave;
2894 } else { /* we didn't change primary - exit */
2895 break;
2896 }
2897
2898 pr_info("%s: Primary slave changed to %s, reselecting active slave\n",
2899 bond->dev->name,
2900 bond->primary_slave ? slave_dev->name : "none");
2901
2902 block_netpoll_tx();
2903 write_lock_bh(&bond->curr_slave_lock);
2904 bond_select_active_slave(bond);
2905 write_unlock_bh(&bond->curr_slave_lock);
2906 unblock_netpoll_tx();
2907 break;
2908 case NETDEV_FEAT_CHANGE:
2909 bond_compute_features(bond);
2910 break;
2911 case NETDEV_RESEND_IGMP:
2912 /* Propagate to master device */
2913 call_netdevice_notifiers(event, slave->bond->dev);
2914 break;
2915 default:
2916 break;
2917 }
2918
2919 return NOTIFY_DONE;
2920}
2921
2922/*
2923 * bond_netdev_event: handle netdev notifier chain events.
2924 *
2925 * This function receives events for the netdev chain. The caller (an
2926 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2927 * locks for us to safely manipulate the slave devices (RTNL lock,
2928 * dev_probe_lock).
2929 */
2930static int bond_netdev_event(struct notifier_block *this,
2931 unsigned long event, void *ptr)
2932{
2933 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2934
2935 pr_debug("event_dev: %s, event: %lx\n",
2936 event_dev ? event_dev->name : "None", event);
2937
2938 if (!(event_dev->priv_flags & IFF_BONDING))
2939 return NOTIFY_DONE;
2940
2941 if (event_dev->flags & IFF_MASTER) {
2942 pr_debug("IFF_MASTER\n");
2943 return bond_master_netdev_event(event, event_dev);
2944 }
2945
2946 if (event_dev->flags & IFF_SLAVE) {
2947 pr_debug("IFF_SLAVE\n");
2948 return bond_slave_netdev_event(event, event_dev);
2949 }
2950
2951 return NOTIFY_DONE;
2952}
2953
2954static struct notifier_block bond_netdev_notifier = {
2955 .notifier_call = bond_netdev_event,
2956};
2957
2958/*---------------------------- Hashing Policies -----------------------------*/
2959
2960/* L2 hash helper */
2961static inline u32 bond_eth_hash(struct sk_buff *skb)
2962{
2963 struct ethhdr *data = (struct ethhdr *)skb->data;
2964
2965 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2966 return data->h_dest[5] ^ data->h_source[5];
2967
2968 return 0;
2969}
2970
2971/* Extract the appropriate headers based on bond's xmit policy */
2972static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2973 struct flow_keys *fk)
2974{
2975 const struct ipv6hdr *iph6;
2976 const struct iphdr *iph;
2977 int noff, proto = -1;
2978
2979 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2980 return skb_flow_dissect(skb, fk);
2981
2982 fk->ports = 0;
2983 noff = skb_network_offset(skb);
2984 if (skb->protocol == htons(ETH_P_IP)) {
2985 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
2986 return false;
2987 iph = ip_hdr(skb);
2988 fk->src = iph->saddr;
2989 fk->dst = iph->daddr;
2990 noff += iph->ihl << 2;
2991 if (!ip_is_fragment(iph))
2992 proto = iph->protocol;
2993 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2994 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
2995 return false;
2996 iph6 = ipv6_hdr(skb);
2997 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2998 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2999 noff += sizeof(*iph6);
3000 proto = iph6->nexthdr;
3001 } else {
3002 return false;
3003 }
3004 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3005 fk->ports = skb_flow_get_ports(skb, noff, proto);
3006
3007 return true;
3008}
3009
3010/**
3011 * bond_xmit_hash - generate a hash value based on the xmit policy
3012 * @bond: bonding device
3013 * @skb: buffer to use for headers
3014 * @count: modulo value
3015 *
3016 * This function will extract the necessary headers from the skb buffer and use
3017 * them to generate a hash based on the xmit_policy set in the bonding device
3018 * which will be reduced modulo count before returning.
3019 */
3020int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3021{
3022 struct flow_keys flow;
3023 u32 hash;
3024
3025 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3026 !bond_flow_dissect(bond, skb, &flow))
3027 return bond_eth_hash(skb) % count;
3028
3029 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3030 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3031 hash = bond_eth_hash(skb);
3032 else
3033 hash = (__force u32)flow.ports;
3034 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3035 hash ^= (hash >> 16);
3036 hash ^= (hash >> 8);
3037
3038 return hash % count;
3039}
3040
3041/*-------------------------- Device entry points ----------------------------*/
3042
3043static void bond_work_init_all(struct bonding *bond)
3044{
3045 INIT_DELAYED_WORK(&bond->mcast_work,
3046 bond_resend_igmp_join_requests_delayed);
3047 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3048 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3049 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3050 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3051 else
3052 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3053 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3054}
3055
3056static void bond_work_cancel_all(struct bonding *bond)
3057{
3058 cancel_delayed_work_sync(&bond->mii_work);
3059 cancel_delayed_work_sync(&bond->arp_work);
3060 cancel_delayed_work_sync(&bond->alb_work);
3061 cancel_delayed_work_sync(&bond->ad_work);
3062 cancel_delayed_work_sync(&bond->mcast_work);
3063}
3064
3065static int bond_open(struct net_device *bond_dev)
3066{
3067 struct bonding *bond = netdev_priv(bond_dev);
3068 struct list_head *iter;
3069 struct slave *slave;
3070
3071 /* reset slave->backup and slave->inactive */
3072 read_lock(&bond->lock);
3073 if (bond_has_slaves(bond)) {
3074 read_lock(&bond->curr_slave_lock);
3075 bond_for_each_slave(bond, slave, iter) {
3076 if (USES_PRIMARY(bond->params.mode)
3077 && (slave != bond->curr_active_slave)) {
3078 bond_set_slave_inactive_flags(slave,
3079 BOND_SLAVE_NOTIFY_NOW);
3080 } else {
3081 bond_set_slave_active_flags(slave,
3082 BOND_SLAVE_NOTIFY_NOW);
3083 }
3084 }
3085 read_unlock(&bond->curr_slave_lock);
3086 }
3087 read_unlock(&bond->lock);
3088
3089 bond_work_init_all(bond);
3090
3091 if (bond_is_lb(bond)) {
3092 /* bond_alb_initialize must be called before the timer
3093 * is started.
3094 */
3095 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3096 return -ENOMEM;
3097 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3098 }
3099
3100 if (bond->params.miimon) /* link check interval, in milliseconds. */
3101 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3102
3103 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3104 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3105 bond->recv_probe = bond_arp_rcv;
3106 }
3107
3108 if (bond->params.mode == BOND_MODE_8023AD) {
3109 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3110 /* register to receive LACPDUs */
3111 bond->recv_probe = bond_3ad_lacpdu_recv;
3112 bond_3ad_initiate_agg_selection(bond, 1);
3113 }
3114
3115 return 0;
3116}
3117
3118static int bond_close(struct net_device *bond_dev)
3119{
3120 struct bonding *bond = netdev_priv(bond_dev);
3121
3122 bond_work_cancel_all(bond);
3123 bond->send_peer_notif = 0;
3124 if (bond_is_lb(bond))
3125 bond_alb_deinitialize(bond);
3126 bond->recv_probe = NULL;
3127
3128 return 0;
3129}
3130
3131static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3132 struct rtnl_link_stats64 *stats)
3133{
3134 struct bonding *bond = netdev_priv(bond_dev);
3135 struct rtnl_link_stats64 temp;
3136 struct list_head *iter;
3137 struct slave *slave;
3138
3139 memset(stats, 0, sizeof(*stats));
3140
3141 read_lock_bh(&bond->lock);
3142 bond_for_each_slave(bond, slave, iter) {
3143 const struct rtnl_link_stats64 *sstats =
3144 dev_get_stats(slave->dev, &temp);
3145
3146 stats->rx_packets += sstats->rx_packets;
3147 stats->rx_bytes += sstats->rx_bytes;
3148 stats->rx_errors += sstats->rx_errors;
3149 stats->rx_dropped += sstats->rx_dropped;
3150
3151 stats->tx_packets += sstats->tx_packets;
3152 stats->tx_bytes += sstats->tx_bytes;
3153 stats->tx_errors += sstats->tx_errors;
3154 stats->tx_dropped += sstats->tx_dropped;
3155
3156 stats->multicast += sstats->multicast;
3157 stats->collisions += sstats->collisions;
3158
3159 stats->rx_length_errors += sstats->rx_length_errors;
3160 stats->rx_over_errors += sstats->rx_over_errors;
3161 stats->rx_crc_errors += sstats->rx_crc_errors;
3162 stats->rx_frame_errors += sstats->rx_frame_errors;
3163 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3164 stats->rx_missed_errors += sstats->rx_missed_errors;
3165
3166 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3167 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3168 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3169 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3170 stats->tx_window_errors += sstats->tx_window_errors;
3171 }
3172 read_unlock_bh(&bond->lock);
3173
3174 return stats;
3175}
3176
3177static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3178{
3179 struct bonding *bond = netdev_priv(bond_dev);
3180 struct net_device *slave_dev = NULL;
3181 struct ifbond k_binfo;
3182 struct ifbond __user *u_binfo = NULL;
3183 struct ifslave k_sinfo;
3184 struct ifslave __user *u_sinfo = NULL;
3185 struct mii_ioctl_data *mii = NULL;
3186 struct bond_opt_value newval;
3187 struct net *net;
3188 int res = 0;
3189
3190 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3191
3192 switch (cmd) {
3193 case SIOCGMIIPHY:
3194 mii = if_mii(ifr);
3195 if (!mii)
3196 return -EINVAL;
3197
3198 mii->phy_id = 0;
3199 /* Fall Through */
3200 case SIOCGMIIREG:
3201 /*
3202 * We do this again just in case we were called by SIOCGMIIREG
3203 * instead of SIOCGMIIPHY.
3204 */
3205 mii = if_mii(ifr);
3206 if (!mii)
3207 return -EINVAL;
3208
3209
3210 if (mii->reg_num == 1) {
3211 mii->val_out = 0;
3212 read_lock(&bond->lock);
3213 read_lock(&bond->curr_slave_lock);
3214 if (netif_carrier_ok(bond->dev))
3215 mii->val_out = BMSR_LSTATUS;
3216
3217 read_unlock(&bond->curr_slave_lock);
3218 read_unlock(&bond->lock);
3219 }
3220
3221 return 0;
3222 case BOND_INFO_QUERY_OLD:
3223 case SIOCBONDINFOQUERY:
3224 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3225
3226 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3227 return -EFAULT;
3228
3229 res = bond_info_query(bond_dev, &k_binfo);
3230 if (res == 0 &&
3231 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3232 return -EFAULT;
3233
3234 return res;
3235 case BOND_SLAVE_INFO_QUERY_OLD:
3236 case SIOCBONDSLAVEINFOQUERY:
3237 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3238
3239 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3240 return -EFAULT;
3241
3242 res = bond_slave_info_query(bond_dev, &k_sinfo);
3243 if (res == 0 &&
3244 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3245 return -EFAULT;
3246
3247 return res;
3248 default:
3249 /* Go on */
3250 break;
3251 }
3252
3253 net = dev_net(bond_dev);
3254
3255 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3256 return -EPERM;
3257
3258 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3259
3260 pr_debug("slave_dev=%p:\n", slave_dev);
3261
3262 if (!slave_dev)
3263 return -ENODEV;
3264
3265 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3266 switch (cmd) {
3267 case BOND_ENSLAVE_OLD:
3268 case SIOCBONDENSLAVE:
3269 res = bond_enslave(bond_dev, slave_dev);
3270 break;
3271 case BOND_RELEASE_OLD:
3272 case SIOCBONDRELEASE:
3273 res = bond_release(bond_dev, slave_dev);
3274 break;
3275 case BOND_SETHWADDR_OLD:
3276 case SIOCBONDSETHWADDR:
3277 bond_set_dev_addr(bond_dev, slave_dev);
3278 res = 0;
3279 break;
3280 case BOND_CHANGE_ACTIVE_OLD:
3281 case SIOCBONDCHANGEACTIVE:
3282 bond_opt_initstr(&newval, slave_dev->name);
3283 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3284 break;
3285 default:
3286 res = -EOPNOTSUPP;
3287 }
3288
3289 return res;
3290}
3291
3292static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3293{
3294 struct bonding *bond = netdev_priv(bond_dev);
3295
3296 if (change & IFF_PROMISC)
3297 bond_set_promiscuity(bond,
3298 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3299
3300 if (change & IFF_ALLMULTI)
3301 bond_set_allmulti(bond,
3302 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3303}
3304
3305static void bond_set_rx_mode(struct net_device *bond_dev)
3306{
3307 struct bonding *bond = netdev_priv(bond_dev);
3308 struct list_head *iter;
3309 struct slave *slave;
3310
3311
3312 rcu_read_lock();
3313 if (USES_PRIMARY(bond->params.mode)) {
3314 slave = rcu_dereference(bond->curr_active_slave);
3315 if (slave) {
3316 dev_uc_sync(slave->dev, bond_dev);
3317 dev_mc_sync(slave->dev, bond_dev);
3318 }
3319 } else {
3320 bond_for_each_slave_rcu(bond, slave, iter) {
3321 dev_uc_sync_multiple(slave->dev, bond_dev);
3322 dev_mc_sync_multiple(slave->dev, bond_dev);
3323 }
3324 }
3325 rcu_read_unlock();
3326}
3327
3328static int bond_neigh_init(struct neighbour *n)
3329{
3330 struct bonding *bond = netdev_priv(n->dev);
3331 const struct net_device_ops *slave_ops;
3332 struct neigh_parms parms;
3333 struct slave *slave;
3334 int ret;
3335
3336 slave = bond_first_slave(bond);
3337 if (!slave)
3338 return 0;
3339 slave_ops = slave->dev->netdev_ops;
3340 if (!slave_ops->ndo_neigh_setup)
3341 return 0;
3342
3343 parms.neigh_setup = NULL;
3344 parms.neigh_cleanup = NULL;
3345 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3346 if (ret)
3347 return ret;
3348
3349 /*
3350 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3351 * after the last slave has been detached. Assumes that all slaves
3352 * utilize the same neigh_cleanup (true at this writing as only user
3353 * is ipoib).
3354 */
3355 n->parms->neigh_cleanup = parms.neigh_cleanup;
3356
3357 if (!parms.neigh_setup)
3358 return 0;
3359
3360 return parms.neigh_setup(n);
3361}
3362
3363/*
3364 * The bonding ndo_neigh_setup is called at init time beofre any
3365 * slave exists. So we must declare proxy setup function which will
3366 * be used at run time to resolve the actual slave neigh param setup.
3367 *
3368 * It's also called by master devices (such as vlans) to setup their
3369 * underlying devices. In that case - do nothing, we're already set up from
3370 * our init.
3371 */
3372static int bond_neigh_setup(struct net_device *dev,
3373 struct neigh_parms *parms)
3374{
3375 /* modify only our neigh_parms */
3376 if (parms->dev == dev)
3377 parms->neigh_setup = bond_neigh_init;
3378
3379 return 0;
3380}
3381
3382/*
3383 * Change the MTU of all of a master's slaves to match the master
3384 */
3385static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3386{
3387 struct bonding *bond = netdev_priv(bond_dev);
3388 struct slave *slave, *rollback_slave;
3389 struct list_head *iter;
3390 int res = 0;
3391
3392 pr_debug("bond=%p, name=%s, new_mtu=%d\n",
3393 bond, bond_dev ? bond_dev->name : "None", new_mtu);
3394
3395 /* Can't hold bond->lock with bh disabled here since
3396 * some base drivers panic. On the other hand we can't
3397 * hold bond->lock without bh disabled because we'll
3398 * deadlock. The only solution is to rely on the fact
3399 * that we're under rtnl_lock here, and the slaves
3400 * list won't change. This doesn't solve the problem
3401 * of setting the slave's MTU while it is
3402 * transmitting, but the assumption is that the base
3403 * driver can handle that.
3404 *
3405 * TODO: figure out a way to safely iterate the slaves
3406 * list, but without holding a lock around the actual
3407 * call to the base driver.
3408 */
3409
3410 bond_for_each_slave(bond, slave, iter) {
3411 pr_debug("s %p c_m %p\n",
3412 slave, slave->dev->netdev_ops->ndo_change_mtu);
3413
3414 res = dev_set_mtu(slave->dev, new_mtu);
3415
3416 if (res) {
3417 /* If we failed to set the slave's mtu to the new value
3418 * we must abort the operation even in ACTIVE_BACKUP
3419 * mode, because if we allow the backup slaves to have
3420 * different mtu values than the active slave we'll
3421 * need to change their mtu when doing a failover. That
3422 * means changing their mtu from timer context, which
3423 * is probably not a good idea.
3424 */
3425 pr_debug("err %d %s\n", res, slave->dev->name);
3426 goto unwind;
3427 }
3428 }
3429
3430 bond_dev->mtu = new_mtu;
3431
3432 return 0;
3433
3434unwind:
3435 /* unwind from head to the slave that failed */
3436 bond_for_each_slave(bond, rollback_slave, iter) {
3437 int tmp_res;
3438
3439 if (rollback_slave == slave)
3440 break;
3441
3442 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3443 if (tmp_res) {
3444 pr_debug("unwind err %d dev %s\n",
3445 tmp_res, rollback_slave->dev->name);
3446 }
3447 }
3448
3449 return res;
3450}
3451
3452/*
3453 * Change HW address
3454 *
3455 * Note that many devices must be down to change the HW address, and
3456 * downing the master releases all slaves. We can make bonds full of
3457 * bonding devices to test this, however.
3458 */
3459static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3460{
3461 struct bonding *bond = netdev_priv(bond_dev);
3462 struct slave *slave, *rollback_slave;
3463 struct sockaddr *sa = addr, tmp_sa;
3464 struct list_head *iter;
3465 int res = 0;
3466
3467 if (bond->params.mode == BOND_MODE_ALB)
3468 return bond_alb_set_mac_address(bond_dev, addr);
3469
3470
3471 pr_debug("bond=%p, name=%s\n",
3472 bond, bond_dev ? bond_dev->name : "None");
3473
3474 /* If fail_over_mac is enabled, do nothing and return success.
3475 * Returning an error causes ifenslave to fail.
3476 */
3477 if (bond->params.fail_over_mac &&
3478 bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3479 return 0;
3480
3481 if (!is_valid_ether_addr(sa->sa_data))
3482 return -EADDRNOTAVAIL;
3483
3484 /* Can't hold bond->lock with bh disabled here since
3485 * some base drivers panic. On the other hand we can't
3486 * hold bond->lock without bh disabled because we'll
3487 * deadlock. The only solution is to rely on the fact
3488 * that we're under rtnl_lock here, and the slaves
3489 * list won't change. This doesn't solve the problem
3490 * of setting the slave's hw address while it is
3491 * transmitting, but the assumption is that the base
3492 * driver can handle that.
3493 *
3494 * TODO: figure out a way to safely iterate the slaves
3495 * list, but without holding a lock around the actual
3496 * call to the base driver.
3497 */
3498
3499 bond_for_each_slave(bond, slave, iter) {
3500 pr_debug("slave %p %s\n", slave, slave->dev->name);
3501 res = dev_set_mac_address(slave->dev, addr);
3502 if (res) {
3503 /* TODO: consider downing the slave
3504 * and retry ?
3505 * User should expect communications
3506 * breakage anyway until ARP finish
3507 * updating, so...
3508 */
3509 pr_debug("err %d %s\n", res, slave->dev->name);
3510 goto unwind;
3511 }
3512 }
3513
3514 /* success */
3515 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3516 return 0;
3517
3518unwind:
3519 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3520 tmp_sa.sa_family = bond_dev->type;
3521
3522 /* unwind from head to the slave that failed */
3523 bond_for_each_slave(bond, rollback_slave, iter) {
3524 int tmp_res;
3525
3526 if (rollback_slave == slave)
3527 break;
3528
3529 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3530 if (tmp_res) {
3531 pr_debug("unwind err %d dev %s\n",
3532 tmp_res, rollback_slave->dev->name);
3533 }
3534 }
3535
3536 return res;
3537}
3538
3539/**
3540 * bond_xmit_slave_id - transmit skb through slave with slave_id
3541 * @bond: bonding device that is transmitting
3542 * @skb: buffer to transmit
3543 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3544 *
3545 * This function tries to transmit through slave with slave_id but in case
3546 * it fails, it tries to find the first available slave for transmission.
3547 * The skb is consumed in all cases, thus the function is void.
3548 */
3549static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3550{
3551 struct list_head *iter;
3552 struct slave *slave;
3553 int i = slave_id;
3554
3555 /* Here we start from the slave with slave_id */
3556 bond_for_each_slave_rcu(bond, slave, iter) {
3557 if (--i < 0) {
3558 if (slave_can_tx(slave)) {
3559 bond_dev_queue_xmit(bond, skb, slave->dev);
3560 return;
3561 }
3562 }
3563 }
3564
3565 /* Here we start from the first slave up to slave_id */
3566 i = slave_id;
3567 bond_for_each_slave_rcu(bond, slave, iter) {
3568 if (--i < 0)
3569 break;
3570 if (slave_can_tx(slave)) {
3571 bond_dev_queue_xmit(bond, skb, slave->dev);
3572 return;
3573 }
3574 }
3575 /* no slave that can tx has been found */
3576 dev_kfree_skb_any(skb);
3577}
3578
3579/**
3580 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3581 * @bond: bonding device to use
3582 *
3583 * Based on the value of the bonding device's packets_per_slave parameter
3584 * this function generates a slave id, which is usually used as the next
3585 * slave to transmit through.
3586 */
3587static u32 bond_rr_gen_slave_id(struct bonding *bond)
3588{
3589 u32 slave_id;
3590 struct reciprocal_value reciprocal_packets_per_slave;
3591 int packets_per_slave = bond->params.packets_per_slave;
3592
3593 switch (packets_per_slave) {
3594 case 0:
3595 slave_id = prandom_u32();
3596 break;
3597 case 1:
3598 slave_id = bond->rr_tx_counter;
3599 break;
3600 default:
3601 reciprocal_packets_per_slave =
3602 bond->params.reciprocal_packets_per_slave;
3603 slave_id = reciprocal_divide(bond->rr_tx_counter,
3604 reciprocal_packets_per_slave);
3605 break;
3606 }
3607 bond->rr_tx_counter++;
3608
3609 return slave_id;
3610}
3611
3612static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3613{
3614 struct bonding *bond = netdev_priv(bond_dev);
3615 struct iphdr *iph = ip_hdr(skb);
3616 struct slave *slave;
3617 u32 slave_id;
3618
3619 /* Start with the curr_active_slave that joined the bond as the
3620 * default for sending IGMP traffic. For failover purposes one
3621 * needs to maintain some consistency for the interface that will
3622 * send the join/membership reports. The curr_active_slave found
3623 * will send all of this type of traffic.
3624 */
3625 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3626 slave = rcu_dereference(bond->curr_active_slave);
3627 if (slave && slave_can_tx(slave))
3628 bond_dev_queue_xmit(bond, skb, slave->dev);
3629 else
3630 bond_xmit_slave_id(bond, skb, 0);
3631 } else {
3632 slave_id = bond_rr_gen_slave_id(bond);
3633 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3634 }
3635
3636 return NETDEV_TX_OK;
3637}
3638
3639/*
3640 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3641 * the bond has a usable interface.
3642 */
3643static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3644{
3645 struct bonding *bond = netdev_priv(bond_dev);
3646 struct slave *slave;
3647
3648 slave = rcu_dereference(bond->curr_active_slave);
3649 if (slave)
3650 bond_dev_queue_xmit(bond, skb, slave->dev);
3651 else
3652 dev_kfree_skb_any(skb);
3653
3654 return NETDEV_TX_OK;
3655}
3656
3657/* In bond_xmit_xor() , we determine the output device by using a pre-
3658 * determined xmit_hash_policy(), If the selected device is not enabled,
3659 * find the next active slave.
3660 */
3661static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3662{
3663 struct bonding *bond = netdev_priv(bond_dev);
3664
3665 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3666
3667 return NETDEV_TX_OK;
3668}
3669
3670/* in broadcast mode, we send everything to all usable interfaces. */
3671static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3672{
3673 struct bonding *bond = netdev_priv(bond_dev);
3674 struct slave *slave = NULL;
3675 struct list_head *iter;
3676
3677 bond_for_each_slave_rcu(bond, slave, iter) {
3678 if (bond_is_last_slave(bond, slave))
3679 break;
3680 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3681 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3682
3683 if (!skb2) {
3684 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3685 bond_dev->name, __func__);
3686 continue;
3687 }
3688 /* bond_dev_queue_xmit always returns 0 */
3689 bond_dev_queue_xmit(bond, skb2, slave->dev);
3690 }
3691 }
3692 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3693 bond_dev_queue_xmit(bond, skb, slave->dev);
3694 else
3695 dev_kfree_skb_any(skb);
3696
3697 return NETDEV_TX_OK;
3698}
3699
3700/*------------------------- Device initialization ---------------------------*/
3701
3702/*
3703 * Lookup the slave that corresponds to a qid
3704 */
3705static inline int bond_slave_override(struct bonding *bond,
3706 struct sk_buff *skb)
3707{
3708 struct slave *slave = NULL;
3709 struct list_head *iter;
3710
3711 if (!skb->queue_mapping)
3712 return 1;
3713
3714 /* Find out if any slaves have the same mapping as this skb. */
3715 bond_for_each_slave_rcu(bond, slave, iter) {
3716 if (slave->queue_id == skb->queue_mapping) {
3717 if (slave_can_tx(slave)) {
3718 bond_dev_queue_xmit(bond, skb, slave->dev);
3719 return 0;
3720 }
3721 /* If the slave isn't UP, use default transmit policy. */
3722 break;
3723 }
3724 }
3725
3726 return 1;
3727}
3728
3729
3730static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3731 void *accel_priv, select_queue_fallback_t fallback)
3732{
3733 /*
3734 * This helper function exists to help dev_pick_tx get the correct
3735 * destination queue. Using a helper function skips a call to
3736 * skb_tx_hash and will put the skbs in the queue we expect on their
3737 * way down to the bonding driver.
3738 */
3739 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3740
3741 /*
3742 * Save the original txq to restore before passing to the driver
3743 */
3744 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3745
3746 if (unlikely(txq >= dev->real_num_tx_queues)) {
3747 do {
3748 txq -= dev->real_num_tx_queues;
3749 } while (txq >= dev->real_num_tx_queues);
3750 }
3751 return txq;
3752}
3753
3754static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3755{
3756 struct bonding *bond = netdev_priv(dev);
3757
3758 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3759 if (!bond_slave_override(bond, skb))
3760 return NETDEV_TX_OK;
3761 }
3762
3763 switch (bond->params.mode) {
3764 case BOND_MODE_ROUNDROBIN:
3765 return bond_xmit_roundrobin(skb, dev);
3766 case BOND_MODE_ACTIVEBACKUP:
3767 return bond_xmit_activebackup(skb, dev);
3768 case BOND_MODE_XOR:
3769 return bond_xmit_xor(skb, dev);
3770 case BOND_MODE_BROADCAST:
3771 return bond_xmit_broadcast(skb, dev);
3772 case BOND_MODE_8023AD:
3773 return bond_3ad_xmit_xor(skb, dev);
3774 case BOND_MODE_ALB:
3775 case BOND_MODE_TLB:
3776 return bond_alb_xmit(skb, dev);
3777 default:
3778 /* Should never happen, mode already checked */
3779 pr_err("%s: Error: Unknown bonding mode %d\n",
3780 dev->name, bond->params.mode);
3781 WARN_ON_ONCE(1);
3782 dev_kfree_skb_any(skb);
3783 return NETDEV_TX_OK;
3784 }
3785}
3786
3787static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3788{
3789 struct bonding *bond = netdev_priv(dev);
3790 netdev_tx_t ret = NETDEV_TX_OK;
3791
3792 /*
3793 * If we risk deadlock from transmitting this in the
3794 * netpoll path, tell netpoll to queue the frame for later tx
3795 */
3796 if (unlikely(is_netpoll_tx_blocked(dev)))
3797 return NETDEV_TX_BUSY;
3798
3799 rcu_read_lock();
3800 if (bond_has_slaves(bond))
3801 ret = __bond_start_xmit(skb, dev);
3802 else
3803 dev_kfree_skb_any(skb);
3804 rcu_read_unlock();
3805
3806 return ret;
3807}
3808
3809static int bond_ethtool_get_settings(struct net_device *bond_dev,
3810 struct ethtool_cmd *ecmd)
3811{
3812 struct bonding *bond = netdev_priv(bond_dev);
3813 unsigned long speed = 0;
3814 struct list_head *iter;
3815 struct slave *slave;
3816
3817 ecmd->duplex = DUPLEX_UNKNOWN;
3818 ecmd->port = PORT_OTHER;
3819
3820 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3821 * do not need to check mode. Though link speed might not represent
3822 * the true receive or transmit bandwidth (not all modes are symmetric)
3823 * this is an accurate maximum.
3824 */
3825 read_lock(&bond->lock);
3826 bond_for_each_slave(bond, slave, iter) {
3827 if (SLAVE_IS_OK(slave)) {
3828 if (slave->speed != SPEED_UNKNOWN)
3829 speed += slave->speed;
3830 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3831 slave->duplex != DUPLEX_UNKNOWN)
3832 ecmd->duplex = slave->duplex;
3833 }
3834 }
3835 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3836 read_unlock(&bond->lock);
3837
3838 return 0;
3839}
3840
3841static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3842 struct ethtool_drvinfo *drvinfo)
3843{
3844 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3845 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3846 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3847 BOND_ABI_VERSION);
3848}
3849
3850static const struct ethtool_ops bond_ethtool_ops = {
3851 .get_drvinfo = bond_ethtool_get_drvinfo,
3852 .get_settings = bond_ethtool_get_settings,
3853 .get_link = ethtool_op_get_link,
3854};
3855
3856static const struct net_device_ops bond_netdev_ops = {
3857 .ndo_init = bond_init,
3858 .ndo_uninit = bond_uninit,
3859 .ndo_open = bond_open,
3860 .ndo_stop = bond_close,
3861 .ndo_start_xmit = bond_start_xmit,
3862 .ndo_select_queue = bond_select_queue,
3863 .ndo_get_stats64 = bond_get_stats,
3864 .ndo_do_ioctl = bond_do_ioctl,
3865 .ndo_change_rx_flags = bond_change_rx_flags,
3866 .ndo_set_rx_mode = bond_set_rx_mode,
3867 .ndo_change_mtu = bond_change_mtu,
3868 .ndo_set_mac_address = bond_set_mac_address,
3869 .ndo_neigh_setup = bond_neigh_setup,
3870 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3871 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3872#ifdef CONFIG_NET_POLL_CONTROLLER
3873 .ndo_netpoll_setup = bond_netpoll_setup,
3874 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3875 .ndo_poll_controller = bond_poll_controller,
3876#endif
3877 .ndo_add_slave = bond_enslave,
3878 .ndo_del_slave = bond_release,
3879 .ndo_fix_features = bond_fix_features,
3880};
3881
3882static const struct device_type bond_type = {
3883 .name = "bond",
3884};
3885
3886static void bond_destructor(struct net_device *bond_dev)
3887{
3888 struct bonding *bond = netdev_priv(bond_dev);
3889 if (bond->wq)
3890 destroy_workqueue(bond->wq);
3891 free_netdev(bond_dev);
3892}
3893
3894void bond_setup(struct net_device *bond_dev)
3895{
3896 struct bonding *bond = netdev_priv(bond_dev);
3897
3898 /* initialize rwlocks */
3899 rwlock_init(&bond->lock);
3900 rwlock_init(&bond->curr_slave_lock);
3901 bond->params = bonding_defaults;
3902
3903 /* Initialize pointers */
3904 bond->dev = bond_dev;
3905
3906 /* Initialize the device entry points */
3907 ether_setup(bond_dev);
3908 bond_dev->netdev_ops = &bond_netdev_ops;
3909 bond_dev->ethtool_ops = &bond_ethtool_ops;
3910
3911 bond_dev->destructor = bond_destructor;
3912
3913 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3914
3915 /* Initialize the device options */
3916 bond_dev->tx_queue_len = 0;
3917 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3918 bond_dev->priv_flags |= IFF_BONDING;
3919 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3920
3921 /* At first, we block adding VLANs. That's the only way to
3922 * prevent problems that occur when adding VLANs over an
3923 * empty bond. The block will be removed once non-challenged
3924 * slaves are enslaved.
3925 */
3926 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3927
3928 /* don't acquire bond device's netif_tx_lock when
3929 * transmitting */
3930 bond_dev->features |= NETIF_F_LLTX;
3931
3932 /* By default, we declare the bond to be fully
3933 * VLAN hardware accelerated capable. Special
3934 * care is taken in the various xmit functions
3935 * when there are slaves that are not hw accel
3936 * capable
3937 */
3938
3939 /* Don't allow bond devices to change network namespaces. */
3940 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3941
3942 bond_dev->hw_features = BOND_VLAN_FEATURES |
3943 NETIF_F_HW_VLAN_CTAG_TX |
3944 NETIF_F_HW_VLAN_CTAG_RX |
3945 NETIF_F_HW_VLAN_CTAG_FILTER;
3946
3947 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3948 bond_dev->features |= bond_dev->hw_features;
3949}
3950
3951/*
3952* Destroy a bonding device.
3953* Must be under rtnl_lock when this function is called.
3954*/
3955static void bond_uninit(struct net_device *bond_dev)
3956{
3957 struct bonding *bond = netdev_priv(bond_dev);
3958 struct list_head *iter;
3959 struct slave *slave;
3960
3961 bond_netpoll_cleanup(bond_dev);
3962
3963 /* Release the bonded slaves */
3964 bond_for_each_slave(bond, slave, iter)
3965 __bond_release_one(bond_dev, slave->dev, true);
3966 pr_info("%s: Released all slaves\n", bond_dev->name);
3967
3968 list_del(&bond->bond_list);
3969
3970 bond_debug_unregister(bond);
3971}
3972
3973/*------------------------- Module initialization ---------------------------*/
3974
3975static int bond_check_params(struct bond_params *params)
3976{
3977 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3978 struct bond_opt_value newval;
3979 const struct bond_opt_value *valptr;
3980 int arp_all_targets_value;
3981
3982 /*
3983 * Convert string parameters.
3984 */
3985 if (mode) {
3986 bond_opt_initstr(&newval, mode);
3987 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
3988 if (!valptr) {
3989 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
3990 return -EINVAL;
3991 }
3992 bond_mode = valptr->value;
3993 }
3994
3995 if (xmit_hash_policy) {
3996 if ((bond_mode != BOND_MODE_XOR) &&
3997 (bond_mode != BOND_MODE_8023AD)) {
3998 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
3999 bond_mode_name(bond_mode));
4000 } else {
4001 bond_opt_initstr(&newval, xmit_hash_policy);
4002 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4003 &newval);
4004 if (!valptr) {
4005 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4006 xmit_hash_policy);
4007 return -EINVAL;
4008 }
4009 xmit_hashtype = valptr->value;
4010 }
4011 }
4012
4013 if (lacp_rate) {
4014 if (bond_mode != BOND_MODE_8023AD) {
4015 pr_info("lacp_rate param is irrelevant in mode %s\n",
4016 bond_mode_name(bond_mode));
4017 } else {
4018 bond_opt_initstr(&newval, lacp_rate);
4019 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4020 &newval);
4021 if (!valptr) {
4022 pr_err("Error: Invalid lacp rate \"%s\"\n",
4023 lacp_rate);
4024 return -EINVAL;
4025 }
4026 lacp_fast = valptr->value;
4027 }
4028 }
4029
4030 if (ad_select) {
4031 bond_opt_initstr(&newval, lacp_rate);
4032 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4033 &newval);
4034 if (!valptr) {
4035 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4036 return -EINVAL;
4037 }
4038 params->ad_select = valptr->value;
4039 if (bond_mode != BOND_MODE_8023AD)
4040 pr_warn("ad_select param only affects 802.3ad mode\n");
4041 } else {
4042 params->ad_select = BOND_AD_STABLE;
4043 }
4044
4045 if (max_bonds < 0) {
4046 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4047 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4048 max_bonds = BOND_DEFAULT_MAX_BONDS;
4049 }
4050
4051 if (miimon < 0) {
4052 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4053 miimon, INT_MAX);
4054 miimon = 0;
4055 }
4056
4057 if (updelay < 0) {
4058 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4059 updelay, INT_MAX);
4060 updelay = 0;
4061 }
4062
4063 if (downdelay < 0) {
4064 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4065 downdelay, INT_MAX);
4066 downdelay = 0;
4067 }
4068
4069 if ((use_carrier != 0) && (use_carrier != 1)) {
4070 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4071 use_carrier);
4072 use_carrier = 1;
4073 }
4074
4075 if (num_peer_notif < 0 || num_peer_notif > 255) {
4076 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4077 num_peer_notif);
4078 num_peer_notif = 1;
4079 }
4080
4081 /* reset values for 802.3ad/TLB/ALB */
4082 if (BOND_NO_USES_ARP(bond_mode)) {
4083 if (!miimon) {
4084 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");
4085 pr_warn("Forcing miimon to 100msec\n");
4086 miimon = BOND_DEFAULT_MIIMON;
4087 }
4088 }
4089
4090 if (tx_queues < 1 || tx_queues > 255) {
4091 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4092 tx_queues, BOND_DEFAULT_TX_QUEUES);
4093 tx_queues = BOND_DEFAULT_TX_QUEUES;
4094 }
4095
4096 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4097 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4098 all_slaves_active);
4099 all_slaves_active = 0;
4100 }
4101
4102 if (resend_igmp < 0 || resend_igmp > 255) {
4103 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4104 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4105 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4106 }
4107
4108 bond_opt_initval(&newval, packets_per_slave);
4109 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4110 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4111 packets_per_slave, USHRT_MAX);
4112 packets_per_slave = 1;
4113 }
4114
4115 if (bond_mode == BOND_MODE_ALB) {
4116 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",
4117 updelay);
4118 }
4119
4120 if (!miimon) {
4121 if (updelay || downdelay) {
4122 /* just warn the user the up/down delay will have
4123 * no effect since miimon is zero...
4124 */
4125 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",
4126 updelay, downdelay);
4127 }
4128 } else {
4129 /* don't allow arp monitoring */
4130 if (arp_interval) {
4131 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4132 miimon, arp_interval);
4133 arp_interval = 0;
4134 }
4135
4136 if ((updelay % miimon) != 0) {
4137 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4138 updelay, miimon, (updelay / miimon) * miimon);
4139 }
4140
4141 updelay /= miimon;
4142
4143 if ((downdelay % miimon) != 0) {
4144 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4145 downdelay, miimon,
4146 (downdelay / miimon) * miimon);
4147 }
4148
4149 downdelay /= miimon;
4150 }
4151
4152 if (arp_interval < 0) {
4153 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4154 arp_interval, INT_MAX);
4155 arp_interval = 0;
4156 }
4157
4158 for (arp_ip_count = 0, i = 0;
4159 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4160 /* not complete check, but should be good enough to
4161 catch mistakes */
4162 __be32 ip;
4163 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4164 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4165 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4166 arp_ip_target[i]);
4167 arp_interval = 0;
4168 } else {
4169 if (bond_get_targets_ip(arp_target, ip) == -1)
4170 arp_target[arp_ip_count++] = ip;
4171 else
4172 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4173 &ip);
4174 }
4175 }
4176
4177 if (arp_interval && !arp_ip_count) {
4178 /* don't allow arping if no arp_ip_target given... */
4179 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4180 arp_interval);
4181 arp_interval = 0;
4182 }
4183
4184 if (arp_validate) {
4185 if (!arp_interval) {
4186 pr_err("arp_validate requires arp_interval\n");
4187 return -EINVAL;
4188 }
4189
4190 bond_opt_initstr(&newval, arp_validate);
4191 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4192 &newval);
4193 if (!valptr) {
4194 pr_err("Error: invalid arp_validate \"%s\"\n",
4195 arp_validate);
4196 return -EINVAL;
4197 }
4198 arp_validate_value = valptr->value;
4199 } else {
4200 arp_validate_value = 0;
4201 }
4202
4203 arp_all_targets_value = 0;
4204 if (arp_all_targets) {
4205 bond_opt_initstr(&newval, arp_all_targets);
4206 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4207 &newval);
4208 if (!valptr) {
4209 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4210 arp_all_targets);
4211 arp_all_targets_value = 0;
4212 } else {
4213 arp_all_targets_value = valptr->value;
4214 }
4215 }
4216
4217 if (miimon) {
4218 pr_info("MII link monitoring set to %d ms\n", miimon);
4219 } else if (arp_interval) {
4220 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4221 arp_validate_value);
4222 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4223 arp_interval, valptr->string, arp_ip_count);
4224
4225 for (i = 0; i < arp_ip_count; i++)
4226 pr_cont(" %s", arp_ip_target[i]);
4227
4228 pr_cont("\n");
4229
4230 } else if (max_bonds) {
4231 /* miimon and arp_interval not set, we need one so things
4232 * work as expected, see bonding.txt for details
4233 */
4234 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");
4235 }
4236
4237 if (primary && !USES_PRIMARY(bond_mode)) {
4238 /* currently, using a primary only makes sense
4239 * in active backup, TLB or ALB modes
4240 */
4241 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4242 primary, bond_mode_name(bond_mode));
4243 primary = NULL;
4244 }
4245
4246 if (primary && primary_reselect) {
4247 bond_opt_initstr(&newval, primary_reselect);
4248 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4249 &newval);
4250 if (!valptr) {
4251 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4252 primary_reselect);
4253 return -EINVAL;
4254 }
4255 primary_reselect_value = valptr->value;
4256 } else {
4257 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4258 }
4259
4260 if (fail_over_mac) {
4261 bond_opt_initstr(&newval, fail_over_mac);
4262 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4263 &newval);
4264 if (!valptr) {
4265 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4266 fail_over_mac);
4267 return -EINVAL;
4268 }
4269 fail_over_mac_value = valptr->value;
4270 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4271 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4272 } else {
4273 fail_over_mac_value = BOND_FOM_NONE;
4274 }
4275
4276 if (lp_interval == 0) {
4277 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4278 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4279 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4280 }
4281
4282 /* fill params struct with the proper values */
4283 params->mode = bond_mode;
4284 params->xmit_policy = xmit_hashtype;
4285 params->miimon = miimon;
4286 params->num_peer_notif = num_peer_notif;
4287 params->arp_interval = arp_interval;
4288 params->arp_validate = arp_validate_value;
4289 params->arp_all_targets = arp_all_targets_value;
4290 params->updelay = updelay;
4291 params->downdelay = downdelay;
4292 params->use_carrier = use_carrier;
4293 params->lacp_fast = lacp_fast;
4294 params->primary[0] = 0;
4295 params->primary_reselect = primary_reselect_value;
4296 params->fail_over_mac = fail_over_mac_value;
4297 params->tx_queues = tx_queues;
4298 params->all_slaves_active = all_slaves_active;
4299 params->resend_igmp = resend_igmp;
4300 params->min_links = min_links;
4301 params->lp_interval = lp_interval;
4302 params->packets_per_slave = packets_per_slave;
4303 if (packets_per_slave > 0) {
4304 params->reciprocal_packets_per_slave =
4305 reciprocal_value(packets_per_slave);
4306 } else {
4307 /* reciprocal_packets_per_slave is unused if
4308 * packets_per_slave is 0 or 1, just initialize it
4309 */
4310 params->reciprocal_packets_per_slave =
4311 (struct reciprocal_value) { 0 };
4312 }
4313
4314 if (primary) {
4315 strncpy(params->primary, primary, IFNAMSIZ);
4316 params->primary[IFNAMSIZ - 1] = 0;
4317 }
4318
4319 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4320
4321 return 0;
4322}
4323
4324static struct lock_class_key bonding_netdev_xmit_lock_key;
4325static struct lock_class_key bonding_netdev_addr_lock_key;
4326static struct lock_class_key bonding_tx_busylock_key;
4327
4328static void bond_set_lockdep_class_one(struct net_device *dev,
4329 struct netdev_queue *txq,
4330 void *_unused)
4331{
4332 lockdep_set_class(&txq->_xmit_lock,
4333 &bonding_netdev_xmit_lock_key);
4334}
4335
4336static void bond_set_lockdep_class(struct net_device *dev)
4337{
4338 lockdep_set_class(&dev->addr_list_lock,
4339 &bonding_netdev_addr_lock_key);
4340 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4341 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4342}
4343
4344/*
4345 * Called from registration process
4346 */
4347static int bond_init(struct net_device *bond_dev)
4348{
4349 struct bonding *bond = netdev_priv(bond_dev);
4350 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4351 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4352
4353 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4354
4355 /*
4356 * Initialize locks that may be required during
4357 * en/deslave operations. All of the bond_open work
4358 * (of which this is part) should really be moved to
4359 * a phase prior to dev_open
4360 */
4361 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4362 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4363
4364 bond->wq = create_singlethread_workqueue(bond_dev->name);
4365 if (!bond->wq)
4366 return -ENOMEM;
4367
4368 bond_set_lockdep_class(bond_dev);
4369
4370 list_add_tail(&bond->bond_list, &bn->dev_list);
4371
4372 bond_prepare_sysfs_group(bond);
4373
4374 bond_debug_register(bond);
4375
4376 /* Ensure valid dev_addr */
4377 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4378 bond_dev->addr_assign_type == NET_ADDR_PERM)
4379 eth_hw_addr_random(bond_dev);
4380
4381 return 0;
4382}
4383
4384unsigned int bond_get_num_tx_queues(void)
4385{
4386 return tx_queues;
4387}
4388
4389/* Create a new bond based on the specified name and bonding parameters.
4390 * If name is NULL, obtain a suitable "bond%d" name for us.
4391 * Caller must NOT hold rtnl_lock; we need to release it here before we
4392 * set up our sysfs entries.
4393 */
4394int bond_create(struct net *net, const char *name)
4395{
4396 struct net_device *bond_dev;
4397 int res;
4398
4399 rtnl_lock();
4400
4401 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4402 name ? name : "bond%d",
4403 bond_setup, tx_queues);
4404 if (!bond_dev) {
4405 pr_err("%s: eek! can't alloc netdev!\n", name);
4406 rtnl_unlock();
4407 return -ENOMEM;
4408 }
4409
4410 dev_net_set(bond_dev, net);
4411 bond_dev->rtnl_link_ops = &bond_link_ops;
4412
4413 res = register_netdevice(bond_dev);
4414
4415 netif_carrier_off(bond_dev);
4416
4417 rtnl_unlock();
4418 if (res < 0)
4419 bond_destructor(bond_dev);
4420 return res;
4421}
4422
4423static int __net_init bond_net_init(struct net *net)
4424{
4425 struct bond_net *bn = net_generic(net, bond_net_id);
4426
4427 bn->net = net;
4428 INIT_LIST_HEAD(&bn->dev_list);
4429
4430 bond_create_proc_dir(bn);
4431 bond_create_sysfs(bn);
4432
4433 return 0;
4434}
4435
4436static void __net_exit bond_net_exit(struct net *net)
4437{
4438 struct bond_net *bn = net_generic(net, bond_net_id);
4439 struct bonding *bond, *tmp_bond;
4440 LIST_HEAD(list);
4441
4442 bond_destroy_sysfs(bn);
4443 bond_destroy_proc_dir(bn);
4444
4445 /* Kill off any bonds created after unregistering bond rtnl ops */
4446 rtnl_lock();
4447 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4448 unregister_netdevice_queue(bond->dev, &list);
4449 unregister_netdevice_many(&list);
4450 rtnl_unlock();
4451}
4452
4453static struct pernet_operations bond_net_ops = {
4454 .init = bond_net_init,
4455 .exit = bond_net_exit,
4456 .id = &bond_net_id,
4457 .size = sizeof(struct bond_net),
4458};
4459
4460static int __init bonding_init(void)
4461{
4462 int i;
4463 int res;
4464
4465 pr_info("%s", bond_version);
4466
4467 res = bond_check_params(&bonding_defaults);
4468 if (res)
4469 goto out;
4470
4471 res = register_pernet_subsys(&bond_net_ops);
4472 if (res)
4473 goto out;
4474
4475 res = bond_netlink_init();
4476 if (res)
4477 goto err_link;
4478
4479 bond_create_debugfs();
4480
4481 for (i = 0; i < max_bonds; i++) {
4482 res = bond_create(&init_net, NULL);
4483 if (res)
4484 goto err;
4485 }
4486
4487 register_netdevice_notifier(&bond_netdev_notifier);
4488out:
4489 return res;
4490err:
4491 bond_destroy_debugfs();
4492 bond_netlink_fini();
4493err_link:
4494 unregister_pernet_subsys(&bond_net_ops);
4495 goto out;
4496
4497}
4498
4499static void __exit bonding_exit(void)
4500{
4501 unregister_netdevice_notifier(&bond_netdev_notifier);
4502
4503 bond_destroy_debugfs();
4504
4505 bond_netlink_fini();
4506 unregister_pernet_subsys(&bond_net_ops);
4507
4508#ifdef CONFIG_NET_POLL_CONTROLLER
4509 /*
4510 * Make sure we don't have an imbalance on our netpoll blocking
4511 */
4512 WARN_ON(atomic_read(&netpoll_block_tx));
4513#endif
4514}
4515
4516module_init(bonding_init);
4517module_exit(bonding_exit);
4518MODULE_LICENSE("GPL");
4519MODULE_VERSION(DRV_VERSION);
4520MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4521MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");