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#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");
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/system.h>
58#include <asm/dma.h>
59#include <linux/uaccess.h>
60#include <linux/errno.h>
61#include <linux/netdevice.h>
62#include <linux/inetdevice.h>
63#include <linux/igmp.h>
64#include <linux/etherdevice.h>
65#include <linux/skbuff.h>
66#include <net/sock.h>
67#include <linux/rtnetlink.h>
68#include <linux/smp.h>
69#include <linux/if_ether.h>
70#include <net/arp.h>
71#include <linux/mii.h>
72#include <linux/ethtool.h>
73#include <linux/if_vlan.h>
74#include <linux/if_bonding.h>
75#include <linux/jiffies.h>
76#include <linux/preempt.h>
77#include <net/route.h>
78#include <net/net_namespace.h>
79#include <net/netns/generic.h>
80#include "bonding.h"
81#include "bond_3ad.h"
82#include "bond_alb.h"
83
84/*---------------------------- Module parameters ----------------------------*/
85
86/* monitor all links that often (in milliseconds). <=0 disables monitoring */
87#define BOND_LINK_MON_INTERV 0
88#define BOND_LINK_ARP_INTERV 0
89
90static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92static int num_peer_notif = 1;
93static int miimon = BOND_LINK_MON_INTERV;
94static int updelay;
95static int downdelay;
96static int use_carrier = 1;
97static char *mode;
98static char *primary;
99static char *primary_reselect;
100static char *lacp_rate;
101static int min_links;
102static char *ad_select;
103static char *xmit_hash_policy;
104static int arp_interval = BOND_LINK_ARP_INTERV;
105static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106static char *arp_validate;
107static char *fail_over_mac;
108static int all_slaves_active = 0;
109static struct bond_params bonding_defaults;
110static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111
112module_param(max_bonds, int, 0);
113MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114module_param(tx_queues, int, 0);
115MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122module_param(miimon, int, 0);
123MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124module_param(updelay, int, 0);
125MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126module_param(downdelay, int, 0);
127MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 "in milliseconds");
129module_param(use_carrier, int, 0);
130MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132module_param(mode, charp, 0);
133MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137module_param(primary, charp, 0);
138MODULE_PARM_DESC(primary, "Primary network device to use");
139module_param(primary_reselect, charp, 0);
140MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "once it comes up; "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
144 "better, "
145 "2 for only on active slave "
146 "failure");
147module_param(lacp_rate, charp, 0);
148MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150module_param(ad_select, charp, 0);
151MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
153 "2 for count");
154module_param(min_links, int, 0);
155MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156
157module_param(xmit_hash_policy, charp, 0);
158MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
160 "2 for layer 2+3");
161module_param(arp_interval, int, 0);
162MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163module_param_array(arp_ip_target, charp, NULL, 0);
164MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165module_param(arp_validate, charp, 0);
166MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169module_param(fail_over_mac, charp, 0);
170MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173module_param(all_slaves_active, int, 0);
174MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177module_param(resend_igmp, int, 0);
178MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179 "link failure");
180
181/*----------------------------- Global variables ----------------------------*/
182
183#ifdef CONFIG_NET_POLL_CONTROLLER
184atomic_t netpoll_block_tx = ATOMIC_INIT(0);
185#endif
186
187int bond_net_id __read_mostly;
188
189static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190static int arp_ip_count;
191static int bond_mode = BOND_MODE_ROUNDROBIN;
192static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193static int lacp_fast;
194
195const struct bond_parm_tbl bond_lacp_tbl[] = {
196{ "slow", AD_LACP_SLOW},
197{ "fast", AD_LACP_FAST},
198{ NULL, -1},
199};
200
201const struct bond_parm_tbl bond_mode_tbl[] = {
202{ "balance-rr", BOND_MODE_ROUNDROBIN},
203{ "active-backup", BOND_MODE_ACTIVEBACKUP},
204{ "balance-xor", BOND_MODE_XOR},
205{ "broadcast", BOND_MODE_BROADCAST},
206{ "802.3ad", BOND_MODE_8023AD},
207{ "balance-tlb", BOND_MODE_TLB},
208{ "balance-alb", BOND_MODE_ALB},
209{ NULL, -1},
210};
211
212const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213{ "layer2", BOND_XMIT_POLICY_LAYER2},
214{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
215{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
216{ NULL, -1},
217};
218
219const struct bond_parm_tbl arp_validate_tbl[] = {
220{ "none", BOND_ARP_VALIDATE_NONE},
221{ "active", BOND_ARP_VALIDATE_ACTIVE},
222{ "backup", BOND_ARP_VALIDATE_BACKUP},
223{ "all", BOND_ARP_VALIDATE_ALL},
224{ NULL, -1},
225};
226
227const struct bond_parm_tbl fail_over_mac_tbl[] = {
228{ "none", BOND_FOM_NONE},
229{ "active", BOND_FOM_ACTIVE},
230{ "follow", BOND_FOM_FOLLOW},
231{ NULL, -1},
232};
233
234const struct bond_parm_tbl pri_reselect_tbl[] = {
235{ "always", BOND_PRI_RESELECT_ALWAYS},
236{ "better", BOND_PRI_RESELECT_BETTER},
237{ "failure", BOND_PRI_RESELECT_FAILURE},
238{ NULL, -1},
239};
240
241struct bond_parm_tbl ad_select_tbl[] = {
242{ "stable", BOND_AD_STABLE},
243{ "bandwidth", BOND_AD_BANDWIDTH},
244{ "count", BOND_AD_COUNT},
245{ NULL, -1},
246};
247
248/*-------------------------- Forward declarations ---------------------------*/
249
250static int bond_init(struct net_device *bond_dev);
251static void bond_uninit(struct net_device *bond_dev);
252
253/*---------------------------- General routines -----------------------------*/
254
255const char *bond_mode_name(int mode)
256{
257 static const char *names[] = {
258 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB] = "transmit load balancing",
264 [BOND_MODE_ALB] = "adaptive load balancing",
265 };
266
267 if (mode < 0 || mode > BOND_MODE_ALB)
268 return "unknown";
269
270 return names[mode];
271}
272
273/*---------------------------------- VLAN -----------------------------------*/
274
275/**
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
279 *
280 * Returns -ENOMEM if allocation failed.
281 */
282static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283{
284 struct vlan_entry *vlan;
285
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond ? bond->dev->name : "None"), vlan_id);
288
289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
290 if (!vlan)
291 return -ENOMEM;
292
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
295
296 write_lock_bh(&bond->lock);
297
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299
300 write_unlock_bh(&bond->lock);
301
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
303
304 return 0;
305}
306
307/**
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
311 *
312 * returns -ENODEV if @vlan_id was not found in @bond.
313 */
314static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315{
316 struct vlan_entry *vlan;
317 int res = -ENODEV;
318
319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
320
321 block_netpoll_tx();
322 write_lock_bh(&bond->lock);
323
324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
327
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vlan_id);
330
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
333
334 kfree(vlan);
335
336 res = 0;
337 goto out;
338 }
339 }
340
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
343
344out:
345 write_unlock_bh(&bond->lock);
346 unblock_netpoll_tx();
347 return res;
348}
349
350/**
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
354 *
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
357 *
358 * Caller must hold bond->lock
359 */
360struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361{
362 struct vlan_entry *next, *last;
363
364 if (list_empty(&bond->vlan_list))
365 return NULL;
366
367 if (!curr) {
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
370 } else {
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
373 if (last == curr) {
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
376 } else {
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
379 }
380 }
381
382 return next;
383}
384
385#define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
386
387/**
388 * bond_dev_queue_xmit - Prepare skb for xmit.
389 *
390 * @bond: bond device that got this skb for tx.
391 * @skb: hw accel VLAN tagged skb to transmit
392 * @slave_dev: slave that is supposed to xmit this skbuff
393 */
394int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
395 struct net_device *slave_dev)
396{
397 skb->dev = slave_dev;
398 skb->priority = 1;
399
400 skb->queue_mapping = bond_queue_mapping(skb);
401
402 if (unlikely(netpoll_tx_running(slave_dev)))
403 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
404 else
405 dev_queue_xmit(skb);
406
407 return 0;
408}
409
410/*
411 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412 * We don't protect the slave list iteration with a lock because:
413 * a. This operation is performed in IOCTL context,
414 * b. The operation is protected by the RTNL semaphore in the 8021q code,
415 * c. Holding a lock with BH disabled while directly calling a base driver
416 * entry point is generally a BAD idea.
417 *
418 * The design of synchronization/protection for this operation in the 8021q
419 * module is good for one or more VLAN devices over a single physical device
420 * and cannot be extended for a teaming solution like bonding, so there is a
421 * potential race condition here where a net device from the vlan group might
422 * be referenced (either by a base driver or the 8021q code) while it is being
423 * removed from the system. However, it turns out we're not making matters
424 * worse, and if it works for regular VLAN usage it will work here too.
425*/
426
427/**
428 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429 * @bond_dev: bonding net device that got called
430 * @vid: vlan id being added
431 */
432static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433{
434 struct bonding *bond = netdev_priv(bond_dev);
435 struct slave *slave;
436 int i, res;
437
438 bond_for_each_slave(bond, slave, i) {
439 struct net_device *slave_dev = slave->dev;
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441
442 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
443 slave_ops->ndo_vlan_rx_add_vid) {
444 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
445 }
446 }
447
448 res = bond_add_vlan(bond, vid);
449 if (res) {
450 pr_err("%s: Error: Failed to add vlan id %d\n",
451 bond_dev->name, vid);
452 }
453}
454
455/**
456 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being removed
459 */
460static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
461{
462 struct bonding *bond = netdev_priv(bond_dev);
463 struct slave *slave;
464 int i, res;
465
466 bond_for_each_slave(bond, slave, i) {
467 struct net_device *slave_dev = slave->dev;
468 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
469
470 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
471 slave_ops->ndo_vlan_rx_kill_vid) {
472 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
473 }
474 }
475
476 res = bond_del_vlan(bond, vid);
477 if (res) {
478 pr_err("%s: Error: Failed to remove vlan id %d\n",
479 bond_dev->name, vid);
480 }
481}
482
483static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
484{
485 struct vlan_entry *vlan;
486 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
487
488 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
489 !(slave_ops->ndo_vlan_rx_add_vid))
490 return;
491
492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
493 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
494}
495
496static void bond_del_vlans_from_slave(struct bonding *bond,
497 struct net_device *slave_dev)
498{
499 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
500 struct vlan_entry *vlan;
501
502 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
503 !(slave_ops->ndo_vlan_rx_kill_vid))
504 return;
505
506 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
507 if (!vlan->vlan_id)
508 continue;
509 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
510 }
511}
512
513/*------------------------------- Link status -------------------------------*/
514
515/*
516 * Set the carrier state for the master according to the state of its
517 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
518 * do special 802.3ad magic.
519 *
520 * Returns zero if carrier state does not change, nonzero if it does.
521 */
522static int bond_set_carrier(struct bonding *bond)
523{
524 struct slave *slave;
525 int i;
526
527 if (bond->slave_cnt == 0)
528 goto down;
529
530 if (bond->params.mode == BOND_MODE_8023AD)
531 return bond_3ad_set_carrier(bond);
532
533 bond_for_each_slave(bond, slave, i) {
534 if (slave->link == BOND_LINK_UP) {
535 if (!netif_carrier_ok(bond->dev)) {
536 netif_carrier_on(bond->dev);
537 return 1;
538 }
539 return 0;
540 }
541 }
542
543down:
544 if (netif_carrier_ok(bond->dev)) {
545 netif_carrier_off(bond->dev);
546 return 1;
547 }
548 return 0;
549}
550
551/*
552 * Get link speed and duplex from the slave's base driver
553 * using ethtool. If for some reason the call fails or the
554 * values are invalid, fake speed and duplex to 100/Full
555 * and return error.
556 */
557static int bond_update_speed_duplex(struct slave *slave)
558{
559 struct net_device *slave_dev = slave->dev;
560 struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
561 u32 slave_speed;
562 int res;
563
564 /* Fake speed and duplex */
565 slave->speed = SPEED_100;
566 slave->duplex = DUPLEX_FULL;
567
568 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
569 return -1;
570
571 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
572 if (res < 0)
573 return -1;
574
575 slave_speed = ethtool_cmd_speed(&etool);
576 if (slave_speed == 0 || slave_speed == ((__u32) -1))
577 return -1;
578
579 switch (etool.duplex) {
580 case DUPLEX_FULL:
581 case DUPLEX_HALF:
582 break;
583 default:
584 return -1;
585 }
586
587 slave->speed = slave_speed;
588 slave->duplex = etool.duplex;
589
590 return 0;
591}
592
593/*
594 * if <dev> supports MII link status reporting, check its link status.
595 *
596 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
597 * depending upon the setting of the use_carrier parameter.
598 *
599 * Return either BMSR_LSTATUS, meaning that the link is up (or we
600 * can't tell and just pretend it is), or 0, meaning that the link is
601 * down.
602 *
603 * If reporting is non-zero, instead of faking link up, return -1 if
604 * both ETHTOOL and MII ioctls fail (meaning the device does not
605 * support them). If use_carrier is set, return whatever it says.
606 * It'd be nice if there was a good way to tell if a driver supports
607 * netif_carrier, but there really isn't.
608 */
609static int bond_check_dev_link(struct bonding *bond,
610 struct net_device *slave_dev, int reporting)
611{
612 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
613 int (*ioctl)(struct net_device *, struct ifreq *, int);
614 struct ifreq ifr;
615 struct mii_ioctl_data *mii;
616
617 if (!reporting && !netif_running(slave_dev))
618 return 0;
619
620 if (bond->params.use_carrier)
621 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
622
623 /* Try to get link status using Ethtool first. */
624 if (slave_dev->ethtool_ops) {
625 if (slave_dev->ethtool_ops->get_link) {
626 u32 link;
627
628 link = slave_dev->ethtool_ops->get_link(slave_dev);
629
630 return link ? BMSR_LSTATUS : 0;
631 }
632 }
633
634 /* Ethtool can't be used, fallback to MII ioctls. */
635 ioctl = slave_ops->ndo_do_ioctl;
636 if (ioctl) {
637 /* TODO: set pointer to correct ioctl on a per team member */
638 /* bases to make this more efficient. that is, once */
639 /* we determine the correct ioctl, we will always */
640 /* call it and not the others for that team */
641 /* member. */
642
643 /*
644 * We cannot assume that SIOCGMIIPHY will also read a
645 * register; not all network drivers (e.g., e100)
646 * support that.
647 */
648
649 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
650 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
651 mii = if_mii(&ifr);
652 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
653 mii->reg_num = MII_BMSR;
654 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
655 return mii->val_out & BMSR_LSTATUS;
656 }
657 }
658
659 /*
660 * If reporting, report that either there's no dev->do_ioctl,
661 * or both SIOCGMIIREG and get_link failed (meaning that we
662 * cannot report link status). If not reporting, pretend
663 * we're ok.
664 */
665 return reporting ? -1 : BMSR_LSTATUS;
666}
667
668/*----------------------------- Multicast list ------------------------------*/
669
670/*
671 * Push the promiscuity flag down to appropriate slaves
672 */
673static int bond_set_promiscuity(struct bonding *bond, int inc)
674{
675 int err = 0;
676 if (USES_PRIMARY(bond->params.mode)) {
677 /* write lock already acquired */
678 if (bond->curr_active_slave) {
679 err = dev_set_promiscuity(bond->curr_active_slave->dev,
680 inc);
681 }
682 } else {
683 struct slave *slave;
684 int i;
685 bond_for_each_slave(bond, slave, i) {
686 err = dev_set_promiscuity(slave->dev, inc);
687 if (err)
688 return err;
689 }
690 }
691 return err;
692}
693
694/*
695 * Push the allmulti flag down to all slaves
696 */
697static int bond_set_allmulti(struct bonding *bond, int inc)
698{
699 int err = 0;
700 if (USES_PRIMARY(bond->params.mode)) {
701 /* write lock already acquired */
702 if (bond->curr_active_slave) {
703 err = dev_set_allmulti(bond->curr_active_slave->dev,
704 inc);
705 }
706 } else {
707 struct slave *slave;
708 int i;
709 bond_for_each_slave(bond, slave, i) {
710 err = dev_set_allmulti(slave->dev, inc);
711 if (err)
712 return err;
713 }
714 }
715 return err;
716}
717
718/*
719 * Add a Multicast address to slaves
720 * according to mode
721 */
722static void bond_mc_add(struct bonding *bond, void *addr)
723{
724 if (USES_PRIMARY(bond->params.mode)) {
725 /* write lock already acquired */
726 if (bond->curr_active_slave)
727 dev_mc_add(bond->curr_active_slave->dev, addr);
728 } else {
729 struct slave *slave;
730 int i;
731
732 bond_for_each_slave(bond, slave, i)
733 dev_mc_add(slave->dev, addr);
734 }
735}
736
737/*
738 * Remove a multicast address from slave
739 * according to mode
740 */
741static void bond_mc_del(struct bonding *bond, void *addr)
742{
743 if (USES_PRIMARY(bond->params.mode)) {
744 /* write lock already acquired */
745 if (bond->curr_active_slave)
746 dev_mc_del(bond->curr_active_slave->dev, addr);
747 } else {
748 struct slave *slave;
749 int i;
750 bond_for_each_slave(bond, slave, i) {
751 dev_mc_del(slave->dev, addr);
752 }
753 }
754}
755
756
757static void __bond_resend_igmp_join_requests(struct net_device *dev)
758{
759 struct in_device *in_dev;
760
761 rcu_read_lock();
762 in_dev = __in_dev_get_rcu(dev);
763 if (in_dev)
764 ip_mc_rejoin_groups(in_dev);
765 rcu_read_unlock();
766}
767
768/*
769 * Retrieve the list of registered multicast addresses for the bonding
770 * device and retransmit an IGMP JOIN request to the current active
771 * slave.
772 */
773static void bond_resend_igmp_join_requests(struct bonding *bond)
774{
775 struct net_device *vlan_dev;
776 struct vlan_entry *vlan;
777
778 read_lock(&bond->lock);
779
780 if (bond->kill_timers)
781 goto out;
782
783 /* rejoin all groups on bond device */
784 __bond_resend_igmp_join_requests(bond->dev);
785
786 /* rejoin all groups on vlan devices */
787 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
788 rcu_read_lock();
789 vlan_dev = __vlan_find_dev_deep(bond->dev,
790 vlan->vlan_id);
791 rcu_read_unlock();
792 if (vlan_dev)
793 __bond_resend_igmp_join_requests(vlan_dev);
794 }
795
796 if ((--bond->igmp_retrans > 0) && !bond->kill_timers)
797 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
798out:
799 read_unlock(&bond->lock);
800}
801
802static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
803{
804 struct bonding *bond = container_of(work, struct bonding,
805 mcast_work.work);
806 bond_resend_igmp_join_requests(bond);
807}
808
809/*
810 * flush all members of flush->mc_list from device dev->mc_list
811 */
812static void bond_mc_list_flush(struct net_device *bond_dev,
813 struct net_device *slave_dev)
814{
815 struct bonding *bond = netdev_priv(bond_dev);
816 struct netdev_hw_addr *ha;
817
818 netdev_for_each_mc_addr(ha, bond_dev)
819 dev_mc_del(slave_dev, ha->addr);
820
821 if (bond->params.mode == BOND_MODE_8023AD) {
822 /* del lacpdu mc addr from mc list */
823 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
824
825 dev_mc_del(slave_dev, lacpdu_multicast);
826 }
827}
828
829/*--------------------------- Active slave change ---------------------------*/
830
831/*
832 * Update the mc list and multicast-related flags for the new and
833 * old active slaves (if any) according to the multicast mode, and
834 * promiscuous flags unconditionally.
835 */
836static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
837 struct slave *old_active)
838{
839 struct netdev_hw_addr *ha;
840
841 if (!USES_PRIMARY(bond->params.mode))
842 /* nothing to do - mc list is already up-to-date on
843 * all slaves
844 */
845 return;
846
847 if (old_active) {
848 if (bond->dev->flags & IFF_PROMISC)
849 dev_set_promiscuity(old_active->dev, -1);
850
851 if (bond->dev->flags & IFF_ALLMULTI)
852 dev_set_allmulti(old_active->dev, -1);
853
854 netdev_for_each_mc_addr(ha, bond->dev)
855 dev_mc_del(old_active->dev, ha->addr);
856 }
857
858 if (new_active) {
859 /* FIXME: Signal errors upstream. */
860 if (bond->dev->flags & IFF_PROMISC)
861 dev_set_promiscuity(new_active->dev, 1);
862
863 if (bond->dev->flags & IFF_ALLMULTI)
864 dev_set_allmulti(new_active->dev, 1);
865
866 netdev_for_each_mc_addr(ha, bond->dev)
867 dev_mc_add(new_active->dev, ha->addr);
868 }
869}
870
871/*
872 * bond_do_fail_over_mac
873 *
874 * Perform special MAC address swapping for fail_over_mac settings
875 *
876 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
877 */
878static void bond_do_fail_over_mac(struct bonding *bond,
879 struct slave *new_active,
880 struct slave *old_active)
881 __releases(&bond->curr_slave_lock)
882 __releases(&bond->lock)
883 __acquires(&bond->lock)
884 __acquires(&bond->curr_slave_lock)
885{
886 u8 tmp_mac[ETH_ALEN];
887 struct sockaddr saddr;
888 int rv;
889
890 switch (bond->params.fail_over_mac) {
891 case BOND_FOM_ACTIVE:
892 if (new_active)
893 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
894 new_active->dev->addr_len);
895 break;
896 case BOND_FOM_FOLLOW:
897 /*
898 * if new_active && old_active, swap them
899 * if just old_active, do nothing (going to no active slave)
900 * if just new_active, set new_active to bond's MAC
901 */
902 if (!new_active)
903 return;
904
905 write_unlock_bh(&bond->curr_slave_lock);
906 read_unlock(&bond->lock);
907
908 if (old_active) {
909 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
910 memcpy(saddr.sa_data, old_active->dev->dev_addr,
911 ETH_ALEN);
912 saddr.sa_family = new_active->dev->type;
913 } else {
914 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
915 saddr.sa_family = bond->dev->type;
916 }
917
918 rv = dev_set_mac_address(new_active->dev, &saddr);
919 if (rv) {
920 pr_err("%s: Error %d setting MAC of slave %s\n",
921 bond->dev->name, -rv, new_active->dev->name);
922 goto out;
923 }
924
925 if (!old_active)
926 goto out;
927
928 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
929 saddr.sa_family = old_active->dev->type;
930
931 rv = dev_set_mac_address(old_active->dev, &saddr);
932 if (rv)
933 pr_err("%s: Error %d setting MAC of slave %s\n",
934 bond->dev->name, -rv, new_active->dev->name);
935out:
936 read_lock(&bond->lock);
937 write_lock_bh(&bond->curr_slave_lock);
938 break;
939 default:
940 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
941 bond->dev->name, bond->params.fail_over_mac);
942 break;
943 }
944
945}
946
947static bool bond_should_change_active(struct bonding *bond)
948{
949 struct slave *prim = bond->primary_slave;
950 struct slave *curr = bond->curr_active_slave;
951
952 if (!prim || !curr || curr->link != BOND_LINK_UP)
953 return true;
954 if (bond->force_primary) {
955 bond->force_primary = false;
956 return true;
957 }
958 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
959 (prim->speed < curr->speed ||
960 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
961 return false;
962 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
963 return false;
964 return true;
965}
966
967/**
968 * find_best_interface - select the best available slave to be the active one
969 * @bond: our bonding struct
970 *
971 * Warning: Caller must hold curr_slave_lock for writing.
972 */
973static struct slave *bond_find_best_slave(struct bonding *bond)
974{
975 struct slave *new_active, *old_active;
976 struct slave *bestslave = NULL;
977 int mintime = bond->params.updelay;
978 int i;
979
980 new_active = bond->curr_active_slave;
981
982 if (!new_active) { /* there were no active slaves left */
983 if (bond->slave_cnt > 0) /* found one slave */
984 new_active = bond->first_slave;
985 else
986 return NULL; /* still no slave, return NULL */
987 }
988
989 if ((bond->primary_slave) &&
990 bond->primary_slave->link == BOND_LINK_UP &&
991 bond_should_change_active(bond)) {
992 new_active = bond->primary_slave;
993 }
994
995 /* remember where to stop iterating over the slaves */
996 old_active = new_active;
997
998 bond_for_each_slave_from(bond, new_active, i, old_active) {
999 if (new_active->link == BOND_LINK_UP) {
1000 return new_active;
1001 } else if (new_active->link == BOND_LINK_BACK &&
1002 IS_UP(new_active->dev)) {
1003 /* link up, but waiting for stabilization */
1004 if (new_active->delay < mintime) {
1005 mintime = new_active->delay;
1006 bestslave = new_active;
1007 }
1008 }
1009 }
1010
1011 return bestslave;
1012}
1013
1014static bool bond_should_notify_peers(struct bonding *bond)
1015{
1016 struct slave *slave = bond->curr_active_slave;
1017
1018 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1019 bond->dev->name, slave ? slave->dev->name : "NULL");
1020
1021 if (!slave || !bond->send_peer_notif ||
1022 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1023 return false;
1024
1025 bond->send_peer_notif--;
1026 return true;
1027}
1028
1029/**
1030 * change_active_interface - change the active slave into the specified one
1031 * @bond: our bonding struct
1032 * @new: the new slave to make the active one
1033 *
1034 * Set the new slave to the bond's settings and unset them on the old
1035 * curr_active_slave.
1036 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1037 *
1038 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1039 * because it is apparently the best available slave we have, even though its
1040 * updelay hasn't timed out yet.
1041 *
1042 * If new_active is not NULL, caller must hold bond->lock for read and
1043 * curr_slave_lock for write_bh.
1044 */
1045void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1046{
1047 struct slave *old_active = bond->curr_active_slave;
1048
1049 if (old_active == new_active)
1050 return;
1051
1052 if (new_active) {
1053 new_active->jiffies = jiffies;
1054
1055 if (new_active->link == BOND_LINK_BACK) {
1056 if (USES_PRIMARY(bond->params.mode)) {
1057 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1058 bond->dev->name, new_active->dev->name,
1059 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1060 }
1061
1062 new_active->delay = 0;
1063 new_active->link = BOND_LINK_UP;
1064
1065 if (bond->params.mode == BOND_MODE_8023AD)
1066 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1067
1068 if (bond_is_lb(bond))
1069 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1070 } else {
1071 if (USES_PRIMARY(bond->params.mode)) {
1072 pr_info("%s: making interface %s the new active one.\n",
1073 bond->dev->name, new_active->dev->name);
1074 }
1075 }
1076 }
1077
1078 if (USES_PRIMARY(bond->params.mode))
1079 bond_mc_swap(bond, new_active, old_active);
1080
1081 if (bond_is_lb(bond)) {
1082 bond_alb_handle_active_change(bond, new_active);
1083 if (old_active)
1084 bond_set_slave_inactive_flags(old_active);
1085 if (new_active)
1086 bond_set_slave_active_flags(new_active);
1087 } else {
1088 bond->curr_active_slave = new_active;
1089 }
1090
1091 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1092 if (old_active)
1093 bond_set_slave_inactive_flags(old_active);
1094
1095 if (new_active) {
1096 bool should_notify_peers = false;
1097
1098 bond_set_slave_active_flags(new_active);
1099
1100 if (bond->params.fail_over_mac)
1101 bond_do_fail_over_mac(bond, new_active,
1102 old_active);
1103
1104 if (netif_running(bond->dev)) {
1105 bond->send_peer_notif =
1106 bond->params.num_peer_notif;
1107 should_notify_peers =
1108 bond_should_notify_peers(bond);
1109 }
1110
1111 write_unlock_bh(&bond->curr_slave_lock);
1112 read_unlock(&bond->lock);
1113
1114 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1115 if (should_notify_peers)
1116 netdev_bonding_change(bond->dev,
1117 NETDEV_NOTIFY_PEERS);
1118
1119 read_lock(&bond->lock);
1120 write_lock_bh(&bond->curr_slave_lock);
1121 }
1122 }
1123
1124 /* resend IGMP joins since active slave has changed or
1125 * all were sent on curr_active_slave.
1126 * resend only if bond is brought up with the affected
1127 * bonding modes and the retransmission is enabled */
1128 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1129 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1130 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1131 bond->igmp_retrans = bond->params.resend_igmp;
1132 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1133 }
1134}
1135
1136/**
1137 * bond_select_active_slave - select a new active slave, if needed
1138 * @bond: our bonding struct
1139 *
1140 * This functions should be called when one of the following occurs:
1141 * - The old curr_active_slave has been released or lost its link.
1142 * - The primary_slave has got its link back.
1143 * - A slave has got its link back and there's no old curr_active_slave.
1144 *
1145 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1146 */
1147void bond_select_active_slave(struct bonding *bond)
1148{
1149 struct slave *best_slave;
1150 int rv;
1151
1152 best_slave = bond_find_best_slave(bond);
1153 if (best_slave != bond->curr_active_slave) {
1154 bond_change_active_slave(bond, best_slave);
1155 rv = bond_set_carrier(bond);
1156 if (!rv)
1157 return;
1158
1159 if (netif_carrier_ok(bond->dev)) {
1160 pr_info("%s: first active interface up!\n",
1161 bond->dev->name);
1162 } else {
1163 pr_info("%s: now running without any active interface !\n",
1164 bond->dev->name);
1165 }
1166 }
1167}
1168
1169/*--------------------------- slave list handling ---------------------------*/
1170
1171/*
1172 * This function attaches the slave to the end of list.
1173 *
1174 * bond->lock held for writing by caller.
1175 */
1176static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1177{
1178 if (bond->first_slave == NULL) { /* attaching the first slave */
1179 new_slave->next = new_slave;
1180 new_slave->prev = new_slave;
1181 bond->first_slave = new_slave;
1182 } else {
1183 new_slave->next = bond->first_slave;
1184 new_slave->prev = bond->first_slave->prev;
1185 new_slave->next->prev = new_slave;
1186 new_slave->prev->next = new_slave;
1187 }
1188
1189 bond->slave_cnt++;
1190}
1191
1192/*
1193 * This function detaches the slave from the list.
1194 * WARNING: no check is made to verify if the slave effectively
1195 * belongs to <bond>.
1196 * Nothing is freed on return, structures are just unchained.
1197 * If any slave pointer in bond was pointing to <slave>,
1198 * it should be changed by the calling function.
1199 *
1200 * bond->lock held for writing by caller.
1201 */
1202static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1203{
1204 if (slave->next)
1205 slave->next->prev = slave->prev;
1206
1207 if (slave->prev)
1208 slave->prev->next = slave->next;
1209
1210 if (bond->first_slave == slave) { /* slave is the first slave */
1211 if (bond->slave_cnt > 1) { /* there are more slave */
1212 bond->first_slave = slave->next;
1213 } else {
1214 bond->first_slave = NULL; /* slave was the last one */
1215 }
1216 }
1217
1218 slave->next = NULL;
1219 slave->prev = NULL;
1220 bond->slave_cnt--;
1221}
1222
1223#ifdef CONFIG_NET_POLL_CONTROLLER
1224static inline int slave_enable_netpoll(struct slave *slave)
1225{
1226 struct netpoll *np;
1227 int err = 0;
1228
1229 np = kzalloc(sizeof(*np), GFP_KERNEL);
1230 err = -ENOMEM;
1231 if (!np)
1232 goto out;
1233
1234 np->dev = slave->dev;
1235 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1236 err = __netpoll_setup(np);
1237 if (err) {
1238 kfree(np);
1239 goto out;
1240 }
1241 slave->np = np;
1242out:
1243 return err;
1244}
1245static inline void slave_disable_netpoll(struct slave *slave)
1246{
1247 struct netpoll *np = slave->np;
1248
1249 if (!np)
1250 return;
1251
1252 slave->np = NULL;
1253 synchronize_rcu_bh();
1254 __netpoll_cleanup(np);
1255 kfree(np);
1256}
1257static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1258{
1259 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1260 return false;
1261 if (!slave_dev->netdev_ops->ndo_poll_controller)
1262 return false;
1263 return true;
1264}
1265
1266static void bond_poll_controller(struct net_device *bond_dev)
1267{
1268}
1269
1270static void __bond_netpoll_cleanup(struct bonding *bond)
1271{
1272 struct slave *slave;
1273 int i;
1274
1275 bond_for_each_slave(bond, slave, i)
1276 if (IS_UP(slave->dev))
1277 slave_disable_netpoll(slave);
1278}
1279static void bond_netpoll_cleanup(struct net_device *bond_dev)
1280{
1281 struct bonding *bond = netdev_priv(bond_dev);
1282
1283 read_lock(&bond->lock);
1284 __bond_netpoll_cleanup(bond);
1285 read_unlock(&bond->lock);
1286}
1287
1288static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1289{
1290 struct bonding *bond = netdev_priv(dev);
1291 struct slave *slave;
1292 int i, err = 0;
1293
1294 read_lock(&bond->lock);
1295 bond_for_each_slave(bond, slave, i) {
1296 err = slave_enable_netpoll(slave);
1297 if (err) {
1298 __bond_netpoll_cleanup(bond);
1299 break;
1300 }
1301 }
1302 read_unlock(&bond->lock);
1303 return err;
1304}
1305
1306static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1307{
1308 return bond->dev->npinfo;
1309}
1310
1311#else
1312static inline int slave_enable_netpoll(struct slave *slave)
1313{
1314 return 0;
1315}
1316static inline void slave_disable_netpoll(struct slave *slave)
1317{
1318}
1319static void bond_netpoll_cleanup(struct net_device *bond_dev)
1320{
1321}
1322#endif
1323
1324/*---------------------------------- IOCTL ----------------------------------*/
1325
1326static int bond_sethwaddr(struct net_device *bond_dev,
1327 struct net_device *slave_dev)
1328{
1329 pr_debug("bond_dev=%p\n", bond_dev);
1330 pr_debug("slave_dev=%p\n", slave_dev);
1331 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1332 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1333 return 0;
1334}
1335
1336static u32 bond_fix_features(struct net_device *dev, u32 features)
1337{
1338 struct slave *slave;
1339 struct bonding *bond = netdev_priv(dev);
1340 u32 mask;
1341 int i;
1342
1343 read_lock(&bond->lock);
1344
1345 if (!bond->first_slave) {
1346 /* Disable adding VLANs to empty bond. But why? --mq */
1347 features |= NETIF_F_VLAN_CHALLENGED;
1348 goto out;
1349 }
1350
1351 mask = features;
1352 features &= ~NETIF_F_ONE_FOR_ALL;
1353 features |= NETIF_F_ALL_FOR_ALL;
1354
1355 bond_for_each_slave(bond, slave, i) {
1356 features = netdev_increment_features(features,
1357 slave->dev->features,
1358 mask);
1359 }
1360
1361out:
1362 read_unlock(&bond->lock);
1363 return features;
1364}
1365
1366#define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1367 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1368 NETIF_F_HIGHDMA | NETIF_F_LRO)
1369
1370static void bond_compute_features(struct bonding *bond)
1371{
1372 struct slave *slave;
1373 struct net_device *bond_dev = bond->dev;
1374 u32 vlan_features = BOND_VLAN_FEATURES;
1375 unsigned short max_hard_header_len = ETH_HLEN;
1376 int i;
1377
1378 read_lock(&bond->lock);
1379
1380 if (!bond->first_slave)
1381 goto done;
1382
1383 bond_for_each_slave(bond, slave, i) {
1384 vlan_features = netdev_increment_features(vlan_features,
1385 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1386
1387 if (slave->dev->hard_header_len > max_hard_header_len)
1388 max_hard_header_len = slave->dev->hard_header_len;
1389 }
1390
1391done:
1392 bond_dev->vlan_features = vlan_features;
1393 bond_dev->hard_header_len = max_hard_header_len;
1394
1395 read_unlock(&bond->lock);
1396
1397 netdev_change_features(bond_dev);
1398}
1399
1400static void bond_setup_by_slave(struct net_device *bond_dev,
1401 struct net_device *slave_dev)
1402{
1403 struct bonding *bond = netdev_priv(bond_dev);
1404
1405 bond_dev->header_ops = slave_dev->header_ops;
1406
1407 bond_dev->type = slave_dev->type;
1408 bond_dev->hard_header_len = slave_dev->hard_header_len;
1409 bond_dev->addr_len = slave_dev->addr_len;
1410
1411 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1412 slave_dev->addr_len);
1413 bond->setup_by_slave = 1;
1414}
1415
1416/* On bonding slaves other than the currently active slave, suppress
1417 * duplicates except for alb non-mcast/bcast.
1418 */
1419static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1420 struct slave *slave,
1421 struct bonding *bond)
1422{
1423 if (bond_is_slave_inactive(slave)) {
1424 if (bond->params.mode == BOND_MODE_ALB &&
1425 skb->pkt_type != PACKET_BROADCAST &&
1426 skb->pkt_type != PACKET_MULTICAST)
1427 return false;
1428 return true;
1429 }
1430 return false;
1431}
1432
1433static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1434{
1435 struct sk_buff *skb = *pskb;
1436 struct slave *slave;
1437 struct bonding *bond;
1438 void (*recv_probe)(struct sk_buff *, struct bonding *,
1439 struct slave *);
1440
1441 skb = skb_share_check(skb, GFP_ATOMIC);
1442 if (unlikely(!skb))
1443 return RX_HANDLER_CONSUMED;
1444
1445 *pskb = skb;
1446
1447 slave = bond_slave_get_rcu(skb->dev);
1448 bond = slave->bond;
1449
1450 if (bond->params.arp_interval)
1451 slave->dev->last_rx = jiffies;
1452
1453 recv_probe = ACCESS_ONCE(bond->recv_probe);
1454 if (recv_probe) {
1455 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1456
1457 if (likely(nskb)) {
1458 recv_probe(nskb, bond, slave);
1459 dev_kfree_skb(nskb);
1460 }
1461 }
1462
1463 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1464 return RX_HANDLER_EXACT;
1465 }
1466
1467 skb->dev = bond->dev;
1468
1469 if (bond->params.mode == BOND_MODE_ALB &&
1470 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1471 skb->pkt_type == PACKET_HOST) {
1472
1473 if (unlikely(skb_cow_head(skb,
1474 skb->data - skb_mac_header(skb)))) {
1475 kfree_skb(skb);
1476 return RX_HANDLER_CONSUMED;
1477 }
1478 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1479 }
1480
1481 return RX_HANDLER_ANOTHER;
1482}
1483
1484/* enslave device <slave> to bond device <master> */
1485int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1486{
1487 struct bonding *bond = netdev_priv(bond_dev);
1488 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1489 struct slave *new_slave = NULL;
1490 struct netdev_hw_addr *ha;
1491 struct sockaddr addr;
1492 int link_reporting;
1493 int res = 0;
1494
1495 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1496 slave_ops->ndo_do_ioctl == NULL) {
1497 pr_warning("%s: Warning: no link monitoring support for %s\n",
1498 bond_dev->name, slave_dev->name);
1499 }
1500
1501 /* already enslaved */
1502 if (slave_dev->flags & IFF_SLAVE) {
1503 pr_debug("Error, Device was already enslaved\n");
1504 return -EBUSY;
1505 }
1506
1507 /* vlan challenged mutual exclusion */
1508 /* no need to lock since we're protected by rtnl_lock */
1509 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1510 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1511 if (bond_vlan_used(bond)) {
1512 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1513 bond_dev->name, slave_dev->name, bond_dev->name);
1514 return -EPERM;
1515 } else {
1516 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1517 bond_dev->name, slave_dev->name,
1518 slave_dev->name, bond_dev->name);
1519 }
1520 } else {
1521 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1522 }
1523
1524 /*
1525 * Old ifenslave binaries are no longer supported. These can
1526 * be identified with moderate accuracy by the state of the slave:
1527 * the current ifenslave will set the interface down prior to
1528 * enslaving it; the old ifenslave will not.
1529 */
1530 if ((slave_dev->flags & IFF_UP)) {
1531 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1532 slave_dev->name);
1533 res = -EPERM;
1534 goto err_undo_flags;
1535 }
1536
1537 /* set bonding device ether type by slave - bonding netdevices are
1538 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1539 * there is a need to override some of the type dependent attribs/funcs.
1540 *
1541 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1542 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1543 */
1544 if (bond->slave_cnt == 0) {
1545 if (bond_dev->type != slave_dev->type) {
1546 pr_debug("%s: change device type from %d to %d\n",
1547 bond_dev->name,
1548 bond_dev->type, slave_dev->type);
1549
1550 res = netdev_bonding_change(bond_dev,
1551 NETDEV_PRE_TYPE_CHANGE);
1552 res = notifier_to_errno(res);
1553 if (res) {
1554 pr_err("%s: refused to change device type\n",
1555 bond_dev->name);
1556 res = -EBUSY;
1557 goto err_undo_flags;
1558 }
1559
1560 /* Flush unicast and multicast addresses */
1561 dev_uc_flush(bond_dev);
1562 dev_mc_flush(bond_dev);
1563
1564 if (slave_dev->type != ARPHRD_ETHER)
1565 bond_setup_by_slave(bond_dev, slave_dev);
1566 else {
1567 ether_setup(bond_dev);
1568 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1569 }
1570
1571 netdev_bonding_change(bond_dev,
1572 NETDEV_POST_TYPE_CHANGE);
1573 }
1574 } else if (bond_dev->type != slave_dev->type) {
1575 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1576 slave_dev->name,
1577 slave_dev->type, bond_dev->type);
1578 res = -EINVAL;
1579 goto err_undo_flags;
1580 }
1581
1582 if (slave_ops->ndo_set_mac_address == NULL) {
1583 if (bond->slave_cnt == 0) {
1584 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1585 bond_dev->name);
1586 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1587 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1588 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",
1589 bond_dev->name);
1590 res = -EOPNOTSUPP;
1591 goto err_undo_flags;
1592 }
1593 }
1594
1595 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1596
1597 /* If this is the first slave, then we need to set the master's hardware
1598 * address to be the same as the slave's. */
1599 if (is_zero_ether_addr(bond->dev->dev_addr))
1600 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1601 slave_dev->addr_len);
1602
1603
1604 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1605 if (!new_slave) {
1606 res = -ENOMEM;
1607 goto err_undo_flags;
1608 }
1609
1610 /*
1611 * Set the new_slave's queue_id to be zero. Queue ID mapping
1612 * is set via sysfs or module option if desired.
1613 */
1614 new_slave->queue_id = 0;
1615
1616 /* Save slave's original mtu and then set it to match the bond */
1617 new_slave->original_mtu = slave_dev->mtu;
1618 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1619 if (res) {
1620 pr_debug("Error %d calling dev_set_mtu\n", res);
1621 goto err_free;
1622 }
1623
1624 /*
1625 * Save slave's original ("permanent") mac address for modes
1626 * that need it, and for restoring it upon release, and then
1627 * set it to the master's address
1628 */
1629 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1630
1631 if (!bond->params.fail_over_mac) {
1632 /*
1633 * Set slave to master's mac address. The application already
1634 * set the master's mac address to that of the first slave
1635 */
1636 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1637 addr.sa_family = slave_dev->type;
1638 res = dev_set_mac_address(slave_dev, &addr);
1639 if (res) {
1640 pr_debug("Error %d calling set_mac_address\n", res);
1641 goto err_restore_mtu;
1642 }
1643 }
1644
1645 res = netdev_set_bond_master(slave_dev, bond_dev);
1646 if (res) {
1647 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1648 goto err_restore_mac;
1649 }
1650
1651 /* open the slave since the application closed it */
1652 res = dev_open(slave_dev);
1653 if (res) {
1654 pr_debug("Opening slave %s failed\n", slave_dev->name);
1655 goto err_unset_master;
1656 }
1657
1658 new_slave->bond = bond;
1659 new_slave->dev = slave_dev;
1660 slave_dev->priv_flags |= IFF_BONDING;
1661
1662 if (bond_is_lb(bond)) {
1663 /* bond_alb_init_slave() must be called before all other stages since
1664 * it might fail and we do not want to have to undo everything
1665 */
1666 res = bond_alb_init_slave(bond, new_slave);
1667 if (res)
1668 goto err_close;
1669 }
1670
1671 /* If the mode USES_PRIMARY, then the new slave gets the
1672 * master's promisc (and mc) settings only if it becomes the
1673 * curr_active_slave, and that is taken care of later when calling
1674 * bond_change_active()
1675 */
1676 if (!USES_PRIMARY(bond->params.mode)) {
1677 /* set promiscuity level to new slave */
1678 if (bond_dev->flags & IFF_PROMISC) {
1679 res = dev_set_promiscuity(slave_dev, 1);
1680 if (res)
1681 goto err_close;
1682 }
1683
1684 /* set allmulti level to new slave */
1685 if (bond_dev->flags & IFF_ALLMULTI) {
1686 res = dev_set_allmulti(slave_dev, 1);
1687 if (res)
1688 goto err_close;
1689 }
1690
1691 netif_addr_lock_bh(bond_dev);
1692 /* upload master's mc_list to new slave */
1693 netdev_for_each_mc_addr(ha, bond_dev)
1694 dev_mc_add(slave_dev, ha->addr);
1695 netif_addr_unlock_bh(bond_dev);
1696 }
1697
1698 if (bond->params.mode == BOND_MODE_8023AD) {
1699 /* add lacpdu mc addr to mc list */
1700 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1701
1702 dev_mc_add(slave_dev, lacpdu_multicast);
1703 }
1704
1705 bond_add_vlans_on_slave(bond, slave_dev);
1706
1707 write_lock_bh(&bond->lock);
1708
1709 bond_attach_slave(bond, new_slave);
1710
1711 new_slave->delay = 0;
1712 new_slave->link_failure_count = 0;
1713
1714 write_unlock_bh(&bond->lock);
1715
1716 bond_compute_features(bond);
1717
1718 read_lock(&bond->lock);
1719
1720 new_slave->last_arp_rx = jiffies;
1721
1722 if (bond->params.miimon && !bond->params.use_carrier) {
1723 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1724
1725 if ((link_reporting == -1) && !bond->params.arp_interval) {
1726 /*
1727 * miimon is set but a bonded network driver
1728 * does not support ETHTOOL/MII and
1729 * arp_interval is not set. Note: if
1730 * use_carrier is enabled, we will never go
1731 * here (because netif_carrier is always
1732 * supported); thus, we don't need to change
1733 * the messages for netif_carrier.
1734 */
1735 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1736 bond_dev->name, slave_dev->name);
1737 } else if (link_reporting == -1) {
1738 /* unable get link status using mii/ethtool */
1739 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1740 bond_dev->name, slave_dev->name);
1741 }
1742 }
1743
1744 /* check for initial state */
1745 if (!bond->params.miimon ||
1746 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1747 if (bond->params.updelay) {
1748 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1749 new_slave->link = BOND_LINK_BACK;
1750 new_slave->delay = bond->params.updelay;
1751 } else {
1752 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1753 new_slave->link = BOND_LINK_UP;
1754 }
1755 new_slave->jiffies = jiffies;
1756 } else {
1757 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1758 new_slave->link = BOND_LINK_DOWN;
1759 }
1760
1761 if (bond_update_speed_duplex(new_slave) &&
1762 (new_slave->link != BOND_LINK_DOWN)) {
1763 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1764 bond_dev->name, new_slave->dev->name);
1765
1766 if (bond->params.mode == BOND_MODE_8023AD) {
1767 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1768 bond_dev->name);
1769 }
1770 }
1771
1772 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1773 /* if there is a primary slave, remember it */
1774 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1775 bond->primary_slave = new_slave;
1776 bond->force_primary = true;
1777 }
1778 }
1779
1780 write_lock_bh(&bond->curr_slave_lock);
1781
1782 switch (bond->params.mode) {
1783 case BOND_MODE_ACTIVEBACKUP:
1784 bond_set_slave_inactive_flags(new_slave);
1785 bond_select_active_slave(bond);
1786 break;
1787 case BOND_MODE_8023AD:
1788 /* in 802.3ad mode, the internal mechanism
1789 * will activate the slaves in the selected
1790 * aggregator
1791 */
1792 bond_set_slave_inactive_flags(new_slave);
1793 /* if this is the first slave */
1794 if (bond->slave_cnt == 1) {
1795 SLAVE_AD_INFO(new_slave).id = 1;
1796 /* Initialize AD with the number of times that the AD timer is called in 1 second
1797 * can be called only after the mac address of the bond is set
1798 */
1799 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1800 } else {
1801 SLAVE_AD_INFO(new_slave).id =
1802 SLAVE_AD_INFO(new_slave->prev).id + 1;
1803 }
1804
1805 bond_3ad_bind_slave(new_slave);
1806 break;
1807 case BOND_MODE_TLB:
1808 case BOND_MODE_ALB:
1809 bond_set_active_slave(new_slave);
1810 bond_set_slave_inactive_flags(new_slave);
1811 bond_select_active_slave(bond);
1812 break;
1813 default:
1814 pr_debug("This slave is always active in trunk mode\n");
1815
1816 /* always active in trunk mode */
1817 bond_set_active_slave(new_slave);
1818
1819 /* In trunking mode there is little meaning to curr_active_slave
1820 * anyway (it holds no special properties of the bond device),
1821 * so we can change it without calling change_active_interface()
1822 */
1823 if (!bond->curr_active_slave)
1824 bond->curr_active_slave = new_slave;
1825
1826 break;
1827 } /* switch(bond_mode) */
1828
1829 write_unlock_bh(&bond->curr_slave_lock);
1830
1831 bond_set_carrier(bond);
1832
1833#ifdef CONFIG_NET_POLL_CONTROLLER
1834 slave_dev->npinfo = bond_netpoll_info(bond);
1835 if (slave_dev->npinfo) {
1836 if (slave_enable_netpoll(new_slave)) {
1837 read_unlock(&bond->lock);
1838 pr_info("Error, %s: master_dev is using netpoll, "
1839 "but new slave device does not support netpoll.\n",
1840 bond_dev->name);
1841 res = -EBUSY;
1842 goto err_close;
1843 }
1844 }
1845#endif
1846
1847 read_unlock(&bond->lock);
1848
1849 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1850 if (res)
1851 goto err_close;
1852
1853 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1854 new_slave);
1855 if (res) {
1856 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1857 goto err_dest_symlinks;
1858 }
1859
1860 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1861 bond_dev->name, slave_dev->name,
1862 bond_is_active_slave(new_slave) ? "n active" : " backup",
1863 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1864
1865 /* enslave is successful */
1866 return 0;
1867
1868/* Undo stages on error */
1869err_dest_symlinks:
1870 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1871
1872err_close:
1873 dev_close(slave_dev);
1874
1875err_unset_master:
1876 netdev_set_bond_master(slave_dev, NULL);
1877
1878err_restore_mac:
1879 if (!bond->params.fail_over_mac) {
1880 /* XXX TODO - fom follow mode needs to change master's
1881 * MAC if this slave's MAC is in use by the bond, or at
1882 * least print a warning.
1883 */
1884 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1885 addr.sa_family = slave_dev->type;
1886 dev_set_mac_address(slave_dev, &addr);
1887 }
1888
1889err_restore_mtu:
1890 dev_set_mtu(slave_dev, new_slave->original_mtu);
1891
1892err_free:
1893 kfree(new_slave);
1894
1895err_undo_flags:
1896 bond_compute_features(bond);
1897
1898 return res;
1899}
1900
1901/*
1902 * Try to release the slave device <slave> from the bond device <master>
1903 * It is legal to access curr_active_slave without a lock because all the function
1904 * is write-locked.
1905 *
1906 * The rules for slave state should be:
1907 * for Active/Backup:
1908 * Active stays on all backups go down
1909 * for Bonded connections:
1910 * The first up interface should be left on and all others downed.
1911 */
1912int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1913{
1914 struct bonding *bond = netdev_priv(bond_dev);
1915 struct slave *slave, *oldcurrent;
1916 struct sockaddr addr;
1917 u32 old_features = bond_dev->features;
1918
1919 /* slave is not a slave or master is not master of this slave */
1920 if (!(slave_dev->flags & IFF_SLAVE) ||
1921 (slave_dev->master != bond_dev)) {
1922 pr_err("%s: Error: cannot release %s.\n",
1923 bond_dev->name, slave_dev->name);
1924 return -EINVAL;
1925 }
1926
1927 block_netpoll_tx();
1928 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1929 write_lock_bh(&bond->lock);
1930
1931 slave = bond_get_slave_by_dev(bond, slave_dev);
1932 if (!slave) {
1933 /* not a slave of this bond */
1934 pr_info("%s: %s not enslaved\n",
1935 bond_dev->name, slave_dev->name);
1936 write_unlock_bh(&bond->lock);
1937 unblock_netpoll_tx();
1938 return -EINVAL;
1939 }
1940
1941 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1942 * for this slave anymore.
1943 */
1944 netdev_rx_handler_unregister(slave_dev);
1945 write_unlock_bh(&bond->lock);
1946 synchronize_net();
1947 write_lock_bh(&bond->lock);
1948
1949 if (!bond->params.fail_over_mac) {
1950 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1951 bond->slave_cnt > 1)
1952 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1953 bond_dev->name, slave_dev->name,
1954 slave->perm_hwaddr,
1955 bond_dev->name, slave_dev->name);
1956 }
1957
1958 /* Inform AD package of unbinding of slave. */
1959 if (bond->params.mode == BOND_MODE_8023AD) {
1960 /* must be called before the slave is
1961 * detached from the list
1962 */
1963 bond_3ad_unbind_slave(slave);
1964 }
1965
1966 pr_info("%s: releasing %s interface %s\n",
1967 bond_dev->name,
1968 bond_is_active_slave(slave) ? "active" : "backup",
1969 slave_dev->name);
1970
1971 oldcurrent = bond->curr_active_slave;
1972
1973 bond->current_arp_slave = NULL;
1974
1975 /* release the slave from its bond */
1976 bond_detach_slave(bond, slave);
1977
1978 if (bond->primary_slave == slave)
1979 bond->primary_slave = NULL;
1980
1981 if (oldcurrent == slave)
1982 bond_change_active_slave(bond, NULL);
1983
1984 if (bond_is_lb(bond)) {
1985 /* Must be called only after the slave has been
1986 * detached from the list and the curr_active_slave
1987 * has been cleared (if our_slave == old_current),
1988 * but before a new active slave is selected.
1989 */
1990 write_unlock_bh(&bond->lock);
1991 bond_alb_deinit_slave(bond, slave);
1992 write_lock_bh(&bond->lock);
1993 }
1994
1995 if (oldcurrent == slave) {
1996 /*
1997 * Note that we hold RTNL over this sequence, so there
1998 * is no concern that another slave add/remove event
1999 * will interfere.
2000 */
2001 write_unlock_bh(&bond->lock);
2002 read_lock(&bond->lock);
2003 write_lock_bh(&bond->curr_slave_lock);
2004
2005 bond_select_active_slave(bond);
2006
2007 write_unlock_bh(&bond->curr_slave_lock);
2008 read_unlock(&bond->lock);
2009 write_lock_bh(&bond->lock);
2010 }
2011
2012 if (bond->slave_cnt == 0) {
2013 bond_set_carrier(bond);
2014
2015 /* if the last slave was removed, zero the mac address
2016 * of the master so it will be set by the application
2017 * to the mac address of the first slave
2018 */
2019 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2020
2021 if (bond_vlan_used(bond)) {
2022 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2023 bond_dev->name, bond_dev->name);
2024 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2025 bond_dev->name);
2026 }
2027 }
2028
2029 write_unlock_bh(&bond->lock);
2030 unblock_netpoll_tx();
2031
2032 bond_compute_features(bond);
2033 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2034 (old_features & NETIF_F_VLAN_CHALLENGED))
2035 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2036 bond_dev->name, slave_dev->name, bond_dev->name);
2037
2038 /* must do this from outside any spinlocks */
2039 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2040
2041 bond_del_vlans_from_slave(bond, slave_dev);
2042
2043 /* If the mode USES_PRIMARY, then we should only remove its
2044 * promisc and mc settings if it was the curr_active_slave, but that was
2045 * already taken care of above when we detached the slave
2046 */
2047 if (!USES_PRIMARY(bond->params.mode)) {
2048 /* unset promiscuity level from slave */
2049 if (bond_dev->flags & IFF_PROMISC)
2050 dev_set_promiscuity(slave_dev, -1);
2051
2052 /* unset allmulti level from slave */
2053 if (bond_dev->flags & IFF_ALLMULTI)
2054 dev_set_allmulti(slave_dev, -1);
2055
2056 /* flush master's mc_list from slave */
2057 netif_addr_lock_bh(bond_dev);
2058 bond_mc_list_flush(bond_dev, slave_dev);
2059 netif_addr_unlock_bh(bond_dev);
2060 }
2061
2062 netdev_set_bond_master(slave_dev, NULL);
2063
2064 slave_disable_netpoll(slave);
2065
2066 /* close slave before restoring its mac address */
2067 dev_close(slave_dev);
2068
2069 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2070 /* restore original ("permanent") mac address */
2071 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2072 addr.sa_family = slave_dev->type;
2073 dev_set_mac_address(slave_dev, &addr);
2074 }
2075
2076 dev_set_mtu(slave_dev, slave->original_mtu);
2077
2078 slave_dev->priv_flags &= ~IFF_BONDING;
2079
2080 kfree(slave);
2081
2082 return 0; /* deletion OK */
2083}
2084
2085/*
2086* First release a slave and then destroy the bond if no more slaves are left.
2087* Must be under rtnl_lock when this function is called.
2088*/
2089static int bond_release_and_destroy(struct net_device *bond_dev,
2090 struct net_device *slave_dev)
2091{
2092 struct bonding *bond = netdev_priv(bond_dev);
2093 int ret;
2094
2095 ret = bond_release(bond_dev, slave_dev);
2096 if ((ret == 0) && (bond->slave_cnt == 0)) {
2097 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2098 pr_info("%s: destroying bond %s.\n",
2099 bond_dev->name, bond_dev->name);
2100 unregister_netdevice(bond_dev);
2101 }
2102 return ret;
2103}
2104
2105/*
2106 * This function releases all slaves.
2107 */
2108static int bond_release_all(struct net_device *bond_dev)
2109{
2110 struct bonding *bond = netdev_priv(bond_dev);
2111 struct slave *slave;
2112 struct net_device *slave_dev;
2113 struct sockaddr addr;
2114
2115 write_lock_bh(&bond->lock);
2116
2117 netif_carrier_off(bond_dev);
2118
2119 if (bond->slave_cnt == 0)
2120 goto out;
2121
2122 bond->current_arp_slave = NULL;
2123 bond->primary_slave = NULL;
2124 bond_change_active_slave(bond, NULL);
2125
2126 while ((slave = bond->first_slave) != NULL) {
2127 /* Inform AD package of unbinding of slave
2128 * before slave is detached from the list.
2129 */
2130 if (bond->params.mode == BOND_MODE_8023AD)
2131 bond_3ad_unbind_slave(slave);
2132
2133 slave_dev = slave->dev;
2134 bond_detach_slave(bond, slave);
2135
2136 /* now that the slave is detached, unlock and perform
2137 * all the undo steps that should not be called from
2138 * within a lock.
2139 */
2140 write_unlock_bh(&bond->lock);
2141
2142 /* unregister rx_handler early so bond_handle_frame wouldn't
2143 * be called for this slave anymore.
2144 */
2145 netdev_rx_handler_unregister(slave_dev);
2146 synchronize_net();
2147
2148 if (bond_is_lb(bond)) {
2149 /* must be called only after the slave
2150 * has been detached from the list
2151 */
2152 bond_alb_deinit_slave(bond, slave);
2153 }
2154
2155 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2156 bond_del_vlans_from_slave(bond, slave_dev);
2157
2158 /* If the mode USES_PRIMARY, then we should only remove its
2159 * promisc and mc settings if it was the curr_active_slave, but that was
2160 * already taken care of above when we detached the slave
2161 */
2162 if (!USES_PRIMARY(bond->params.mode)) {
2163 /* unset promiscuity level from slave */
2164 if (bond_dev->flags & IFF_PROMISC)
2165 dev_set_promiscuity(slave_dev, -1);
2166
2167 /* unset allmulti level from slave */
2168 if (bond_dev->flags & IFF_ALLMULTI)
2169 dev_set_allmulti(slave_dev, -1);
2170
2171 /* flush master's mc_list from slave */
2172 netif_addr_lock_bh(bond_dev);
2173 bond_mc_list_flush(bond_dev, slave_dev);
2174 netif_addr_unlock_bh(bond_dev);
2175 }
2176
2177 netdev_set_bond_master(slave_dev, NULL);
2178
2179 slave_disable_netpoll(slave);
2180
2181 /* close slave before restoring its mac address */
2182 dev_close(slave_dev);
2183
2184 if (!bond->params.fail_over_mac) {
2185 /* restore original ("permanent") mac address*/
2186 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2187 addr.sa_family = slave_dev->type;
2188 dev_set_mac_address(slave_dev, &addr);
2189 }
2190
2191 kfree(slave);
2192
2193 /* re-acquire the lock before getting the next slave */
2194 write_lock_bh(&bond->lock);
2195 }
2196
2197 /* zero the mac address of the master so it will be
2198 * set by the application to the mac address of the
2199 * first slave
2200 */
2201 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2202
2203 if (bond_vlan_used(bond)) {
2204 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2205 bond_dev->name, bond_dev->name);
2206 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2207 bond_dev->name);
2208 }
2209
2210 pr_info("%s: released all slaves\n", bond_dev->name);
2211
2212out:
2213 write_unlock_bh(&bond->lock);
2214
2215 bond_compute_features(bond);
2216
2217 return 0;
2218}
2219
2220/*
2221 * This function changes the active slave to slave <slave_dev>.
2222 * It returns -EINVAL in the following cases.
2223 * - <slave_dev> is not found in the list.
2224 * - There is not active slave now.
2225 * - <slave_dev> is already active.
2226 * - The link state of <slave_dev> is not BOND_LINK_UP.
2227 * - <slave_dev> is not running.
2228 * In these cases, this function does nothing.
2229 * In the other cases, current_slave pointer is changed and 0 is returned.
2230 */
2231static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2232{
2233 struct bonding *bond = netdev_priv(bond_dev);
2234 struct slave *old_active = NULL;
2235 struct slave *new_active = NULL;
2236 int res = 0;
2237
2238 if (!USES_PRIMARY(bond->params.mode))
2239 return -EINVAL;
2240
2241 /* Verify that master_dev is indeed the master of slave_dev */
2242 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2243 return -EINVAL;
2244
2245 read_lock(&bond->lock);
2246
2247 read_lock(&bond->curr_slave_lock);
2248 old_active = bond->curr_active_slave;
2249 read_unlock(&bond->curr_slave_lock);
2250
2251 new_active = bond_get_slave_by_dev(bond, slave_dev);
2252
2253 /*
2254 * Changing to the current active: do nothing; return success.
2255 */
2256 if (new_active && (new_active == old_active)) {
2257 read_unlock(&bond->lock);
2258 return 0;
2259 }
2260
2261 if ((new_active) &&
2262 (old_active) &&
2263 (new_active->link == BOND_LINK_UP) &&
2264 IS_UP(new_active->dev)) {
2265 block_netpoll_tx();
2266 write_lock_bh(&bond->curr_slave_lock);
2267 bond_change_active_slave(bond, new_active);
2268 write_unlock_bh(&bond->curr_slave_lock);
2269 unblock_netpoll_tx();
2270 } else
2271 res = -EINVAL;
2272
2273 read_unlock(&bond->lock);
2274
2275 return res;
2276}
2277
2278static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2279{
2280 struct bonding *bond = netdev_priv(bond_dev);
2281
2282 info->bond_mode = bond->params.mode;
2283 info->miimon = bond->params.miimon;
2284
2285 read_lock(&bond->lock);
2286 info->num_slaves = bond->slave_cnt;
2287 read_unlock(&bond->lock);
2288
2289 return 0;
2290}
2291
2292static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2293{
2294 struct bonding *bond = netdev_priv(bond_dev);
2295 struct slave *slave;
2296 int i, res = -ENODEV;
2297
2298 read_lock(&bond->lock);
2299
2300 bond_for_each_slave(bond, slave, i) {
2301 if (i == (int)info->slave_id) {
2302 res = 0;
2303 strcpy(info->slave_name, slave->dev->name);
2304 info->link = slave->link;
2305 info->state = bond_slave_state(slave);
2306 info->link_failure_count = slave->link_failure_count;
2307 break;
2308 }
2309 }
2310
2311 read_unlock(&bond->lock);
2312
2313 return res;
2314}
2315
2316/*-------------------------------- Monitoring -------------------------------*/
2317
2318
2319static int bond_miimon_inspect(struct bonding *bond)
2320{
2321 struct slave *slave;
2322 int i, link_state, commit = 0;
2323 bool ignore_updelay;
2324
2325 ignore_updelay = !bond->curr_active_slave ? true : false;
2326
2327 bond_for_each_slave(bond, slave, i) {
2328 slave->new_link = BOND_LINK_NOCHANGE;
2329
2330 link_state = bond_check_dev_link(bond, slave->dev, 0);
2331
2332 switch (slave->link) {
2333 case BOND_LINK_UP:
2334 if (link_state)
2335 continue;
2336
2337 slave->link = BOND_LINK_FAIL;
2338 slave->delay = bond->params.downdelay;
2339 if (slave->delay) {
2340 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2341 bond->dev->name,
2342 (bond->params.mode ==
2343 BOND_MODE_ACTIVEBACKUP) ?
2344 (bond_is_active_slave(slave) ?
2345 "active " : "backup ") : "",
2346 slave->dev->name,
2347 bond->params.downdelay * bond->params.miimon);
2348 }
2349 /*FALLTHRU*/
2350 case BOND_LINK_FAIL:
2351 if (link_state) {
2352 /*
2353 * recovered before downdelay expired
2354 */
2355 slave->link = BOND_LINK_UP;
2356 slave->jiffies = jiffies;
2357 pr_info("%s: link status up again after %d ms for interface %s.\n",
2358 bond->dev->name,
2359 (bond->params.downdelay - slave->delay) *
2360 bond->params.miimon,
2361 slave->dev->name);
2362 continue;
2363 }
2364
2365 if (slave->delay <= 0) {
2366 slave->new_link = BOND_LINK_DOWN;
2367 commit++;
2368 continue;
2369 }
2370
2371 slave->delay--;
2372 break;
2373
2374 case BOND_LINK_DOWN:
2375 if (!link_state)
2376 continue;
2377
2378 slave->link = BOND_LINK_BACK;
2379 slave->delay = bond->params.updelay;
2380
2381 if (slave->delay) {
2382 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2383 bond->dev->name, slave->dev->name,
2384 ignore_updelay ? 0 :
2385 bond->params.updelay *
2386 bond->params.miimon);
2387 }
2388 /*FALLTHRU*/
2389 case BOND_LINK_BACK:
2390 if (!link_state) {
2391 slave->link = BOND_LINK_DOWN;
2392 pr_info("%s: link status down again after %d ms for interface %s.\n",
2393 bond->dev->name,
2394 (bond->params.updelay - slave->delay) *
2395 bond->params.miimon,
2396 slave->dev->name);
2397
2398 continue;
2399 }
2400
2401 if (ignore_updelay)
2402 slave->delay = 0;
2403
2404 if (slave->delay <= 0) {
2405 slave->new_link = BOND_LINK_UP;
2406 commit++;
2407 ignore_updelay = false;
2408 continue;
2409 }
2410
2411 slave->delay--;
2412 break;
2413 }
2414 }
2415
2416 return commit;
2417}
2418
2419static void bond_miimon_commit(struct bonding *bond)
2420{
2421 struct slave *slave;
2422 int i;
2423
2424 bond_for_each_slave(bond, slave, i) {
2425 switch (slave->new_link) {
2426 case BOND_LINK_NOCHANGE:
2427 continue;
2428
2429 case BOND_LINK_UP:
2430 slave->link = BOND_LINK_UP;
2431 slave->jiffies = jiffies;
2432
2433 if (bond->params.mode == BOND_MODE_8023AD) {
2434 /* prevent it from being the active one */
2435 bond_set_backup_slave(slave);
2436 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2437 /* make it immediately active */
2438 bond_set_active_slave(slave);
2439 } else if (slave != bond->primary_slave) {
2440 /* prevent it from being the active one */
2441 bond_set_backup_slave(slave);
2442 }
2443
2444 bond_update_speed_duplex(slave);
2445
2446 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2447 bond->dev->name, slave->dev->name,
2448 slave->speed, slave->duplex ? "full" : "half");
2449
2450 /* notify ad that the link status has changed */
2451 if (bond->params.mode == BOND_MODE_8023AD)
2452 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2453
2454 if (bond_is_lb(bond))
2455 bond_alb_handle_link_change(bond, slave,
2456 BOND_LINK_UP);
2457
2458 if (!bond->curr_active_slave ||
2459 (slave == bond->primary_slave))
2460 goto do_failover;
2461
2462 continue;
2463
2464 case BOND_LINK_DOWN:
2465 if (slave->link_failure_count < UINT_MAX)
2466 slave->link_failure_count++;
2467
2468 slave->link = BOND_LINK_DOWN;
2469
2470 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2471 bond->params.mode == BOND_MODE_8023AD)
2472 bond_set_slave_inactive_flags(slave);
2473
2474 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2475 bond->dev->name, slave->dev->name);
2476
2477 if (bond->params.mode == BOND_MODE_8023AD)
2478 bond_3ad_handle_link_change(slave,
2479 BOND_LINK_DOWN);
2480
2481 if (bond_is_lb(bond))
2482 bond_alb_handle_link_change(bond, slave,
2483 BOND_LINK_DOWN);
2484
2485 if (slave == bond->curr_active_slave)
2486 goto do_failover;
2487
2488 continue;
2489
2490 default:
2491 pr_err("%s: invalid new link %d on slave %s\n",
2492 bond->dev->name, slave->new_link,
2493 slave->dev->name);
2494 slave->new_link = BOND_LINK_NOCHANGE;
2495
2496 continue;
2497 }
2498
2499do_failover:
2500 ASSERT_RTNL();
2501 block_netpoll_tx();
2502 write_lock_bh(&bond->curr_slave_lock);
2503 bond_select_active_slave(bond);
2504 write_unlock_bh(&bond->curr_slave_lock);
2505 unblock_netpoll_tx();
2506 }
2507
2508 bond_set_carrier(bond);
2509}
2510
2511/*
2512 * bond_mii_monitor
2513 *
2514 * Really a wrapper that splits the mii monitor into two phases: an
2515 * inspection, then (if inspection indicates something needs to be done)
2516 * an acquisition of appropriate locks followed by a commit phase to
2517 * implement whatever link state changes are indicated.
2518 */
2519void bond_mii_monitor(struct work_struct *work)
2520{
2521 struct bonding *bond = container_of(work, struct bonding,
2522 mii_work.work);
2523 bool should_notify_peers = false;
2524
2525 read_lock(&bond->lock);
2526 if (bond->kill_timers)
2527 goto out;
2528
2529 if (bond->slave_cnt == 0)
2530 goto re_arm;
2531
2532 should_notify_peers = bond_should_notify_peers(bond);
2533
2534 if (bond_miimon_inspect(bond)) {
2535 read_unlock(&bond->lock);
2536 rtnl_lock();
2537 read_lock(&bond->lock);
2538
2539 bond_miimon_commit(bond);
2540
2541 read_unlock(&bond->lock);
2542 rtnl_unlock(); /* might sleep, hold no other locks */
2543 read_lock(&bond->lock);
2544 }
2545
2546re_arm:
2547 if (bond->params.miimon && !bond->kill_timers)
2548 queue_delayed_work(bond->wq, &bond->mii_work,
2549 msecs_to_jiffies(bond->params.miimon));
2550out:
2551 read_unlock(&bond->lock);
2552
2553 if (should_notify_peers) {
2554 rtnl_lock();
2555 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2556 rtnl_unlock();
2557 }
2558}
2559
2560static __be32 bond_glean_dev_ip(struct net_device *dev)
2561{
2562 struct in_device *idev;
2563 struct in_ifaddr *ifa;
2564 __be32 addr = 0;
2565
2566 if (!dev)
2567 return 0;
2568
2569 rcu_read_lock();
2570 idev = __in_dev_get_rcu(dev);
2571 if (!idev)
2572 goto out;
2573
2574 ifa = idev->ifa_list;
2575 if (!ifa)
2576 goto out;
2577
2578 addr = ifa->ifa_local;
2579out:
2580 rcu_read_unlock();
2581 return addr;
2582}
2583
2584static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2585{
2586 struct vlan_entry *vlan;
2587
2588 if (ip == bond->master_ip)
2589 return 1;
2590
2591 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2592 if (ip == vlan->vlan_ip)
2593 return 1;
2594 }
2595
2596 return 0;
2597}
2598
2599/*
2600 * We go to the (large) trouble of VLAN tagging ARP frames because
2601 * switches in VLAN mode (especially if ports are configured as
2602 * "native" to a VLAN) might not pass non-tagged frames.
2603 */
2604static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2605{
2606 struct sk_buff *skb;
2607
2608 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2609 slave_dev->name, dest_ip, src_ip, vlan_id);
2610
2611 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2612 NULL, slave_dev->dev_addr, NULL);
2613
2614 if (!skb) {
2615 pr_err("ARP packet allocation failed\n");
2616 return;
2617 }
2618 if (vlan_id) {
2619 skb = vlan_put_tag(skb, vlan_id);
2620 if (!skb) {
2621 pr_err("failed to insert VLAN tag\n");
2622 return;
2623 }
2624 }
2625 arp_xmit(skb);
2626}
2627
2628
2629static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2630{
2631 int i, vlan_id;
2632 __be32 *targets = bond->params.arp_targets;
2633 struct vlan_entry *vlan;
2634 struct net_device *vlan_dev;
2635 struct rtable *rt;
2636
2637 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2638 if (!targets[i])
2639 break;
2640 pr_debug("basa: target %x\n", targets[i]);
2641 if (!bond_vlan_used(bond)) {
2642 pr_debug("basa: empty vlan: arp_send\n");
2643 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2644 bond->master_ip, 0);
2645 continue;
2646 }
2647
2648 /*
2649 * If VLANs are configured, we do a route lookup to
2650 * determine which VLAN interface would be used, so we
2651 * can tag the ARP with the proper VLAN tag.
2652 */
2653 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2654 RTO_ONLINK, 0);
2655 if (IS_ERR(rt)) {
2656 if (net_ratelimit()) {
2657 pr_warning("%s: no route to arp_ip_target %pI4\n",
2658 bond->dev->name, &targets[i]);
2659 }
2660 continue;
2661 }
2662
2663 /*
2664 * This target is not on a VLAN
2665 */
2666 if (rt->dst.dev == bond->dev) {
2667 ip_rt_put(rt);
2668 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2669 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2670 bond->master_ip, 0);
2671 continue;
2672 }
2673
2674 vlan_id = 0;
2675 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2676 rcu_read_lock();
2677 vlan_dev = __vlan_find_dev_deep(bond->dev,
2678 vlan->vlan_id);
2679 rcu_read_unlock();
2680 if (vlan_dev == rt->dst.dev) {
2681 vlan_id = vlan->vlan_id;
2682 pr_debug("basa: vlan match on %s %d\n",
2683 vlan_dev->name, vlan_id);
2684 break;
2685 }
2686 }
2687
2688 if (vlan_id) {
2689 ip_rt_put(rt);
2690 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2691 vlan->vlan_ip, vlan_id);
2692 continue;
2693 }
2694
2695 if (net_ratelimit()) {
2696 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2697 bond->dev->name, &targets[i],
2698 rt->dst.dev ? rt->dst.dev->name : "NULL");
2699 }
2700 ip_rt_put(rt);
2701 }
2702}
2703
2704static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2705{
2706 int i;
2707 __be32 *targets = bond->params.arp_targets;
2708
2709 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2710 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2711 &sip, &tip, i, &targets[i],
2712 bond_has_this_ip(bond, tip));
2713 if (sip == targets[i]) {
2714 if (bond_has_this_ip(bond, tip))
2715 slave->last_arp_rx = jiffies;
2716 return;
2717 }
2718 }
2719}
2720
2721static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2722 struct slave *slave)
2723{
2724 struct arphdr *arp;
2725 unsigned char *arp_ptr;
2726 __be32 sip, tip;
2727
2728 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2729 return;
2730
2731 read_lock(&bond->lock);
2732
2733 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2734 bond->dev->name, skb->dev->name);
2735
2736 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2737 goto out_unlock;
2738
2739 arp = arp_hdr(skb);
2740 if (arp->ar_hln != bond->dev->addr_len ||
2741 skb->pkt_type == PACKET_OTHERHOST ||
2742 skb->pkt_type == PACKET_LOOPBACK ||
2743 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2744 arp->ar_pro != htons(ETH_P_IP) ||
2745 arp->ar_pln != 4)
2746 goto out_unlock;
2747
2748 arp_ptr = (unsigned char *)(arp + 1);
2749 arp_ptr += bond->dev->addr_len;
2750 memcpy(&sip, arp_ptr, 4);
2751 arp_ptr += 4 + bond->dev->addr_len;
2752 memcpy(&tip, arp_ptr, 4);
2753
2754 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2755 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2756 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2757 &sip, &tip);
2758
2759 /*
2760 * Backup slaves won't see the ARP reply, but do come through
2761 * here for each ARP probe (so we swap the sip/tip to validate
2762 * the probe). In a "redundant switch, common router" type of
2763 * configuration, the ARP probe will (hopefully) travel from
2764 * the active, through one switch, the router, then the other
2765 * switch before reaching the backup.
2766 */
2767 if (bond_is_active_slave(slave))
2768 bond_validate_arp(bond, slave, sip, tip);
2769 else
2770 bond_validate_arp(bond, slave, tip, sip);
2771
2772out_unlock:
2773 read_unlock(&bond->lock);
2774}
2775
2776/*
2777 * this function is called regularly to monitor each slave's link
2778 * ensuring that traffic is being sent and received when arp monitoring
2779 * is used in load-balancing mode. if the adapter has been dormant, then an
2780 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2781 * arp monitoring in active backup mode.
2782 */
2783void bond_loadbalance_arp_mon(struct work_struct *work)
2784{
2785 struct bonding *bond = container_of(work, struct bonding,
2786 arp_work.work);
2787 struct slave *slave, *oldcurrent;
2788 int do_failover = 0;
2789 int delta_in_ticks;
2790 int i;
2791
2792 read_lock(&bond->lock);
2793
2794 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2795
2796 if (bond->kill_timers)
2797 goto out;
2798
2799 if (bond->slave_cnt == 0)
2800 goto re_arm;
2801
2802 read_lock(&bond->curr_slave_lock);
2803 oldcurrent = bond->curr_active_slave;
2804 read_unlock(&bond->curr_slave_lock);
2805
2806 /* see if any of the previous devices are up now (i.e. they have
2807 * xmt and rcv traffic). the curr_active_slave does not come into
2808 * the picture unless it is null. also, slave->jiffies is not needed
2809 * here because we send an arp on each slave and give a slave as
2810 * long as it needs to get the tx/rx within the delta.
2811 * TODO: what about up/down delay in arp mode? it wasn't here before
2812 * so it can wait
2813 */
2814 bond_for_each_slave(bond, slave, i) {
2815 unsigned long trans_start = dev_trans_start(slave->dev);
2816
2817 if (slave->link != BOND_LINK_UP) {
2818 if (time_in_range(jiffies,
2819 trans_start - delta_in_ticks,
2820 trans_start + delta_in_ticks) &&
2821 time_in_range(jiffies,
2822 slave->dev->last_rx - delta_in_ticks,
2823 slave->dev->last_rx + delta_in_ticks)) {
2824
2825 slave->link = BOND_LINK_UP;
2826 bond_set_active_slave(slave);
2827
2828 /* primary_slave has no meaning in round-robin
2829 * mode. the window of a slave being up and
2830 * curr_active_slave being null after enslaving
2831 * is closed.
2832 */
2833 if (!oldcurrent) {
2834 pr_info("%s: link status definitely up for interface %s, ",
2835 bond->dev->name,
2836 slave->dev->name);
2837 do_failover = 1;
2838 } else {
2839 pr_info("%s: interface %s is now up\n",
2840 bond->dev->name,
2841 slave->dev->name);
2842 }
2843 }
2844 } else {
2845 /* slave->link == BOND_LINK_UP */
2846
2847 /* not all switches will respond to an arp request
2848 * when the source ip is 0, so don't take the link down
2849 * if we don't know our ip yet
2850 */
2851 if (!time_in_range(jiffies,
2852 trans_start - delta_in_ticks,
2853 trans_start + 2 * delta_in_ticks) ||
2854 !time_in_range(jiffies,
2855 slave->dev->last_rx - delta_in_ticks,
2856 slave->dev->last_rx + 2 * delta_in_ticks)) {
2857
2858 slave->link = BOND_LINK_DOWN;
2859 bond_set_backup_slave(slave);
2860
2861 if (slave->link_failure_count < UINT_MAX)
2862 slave->link_failure_count++;
2863
2864 pr_info("%s: interface %s is now down.\n",
2865 bond->dev->name,
2866 slave->dev->name);
2867
2868 if (slave == oldcurrent)
2869 do_failover = 1;
2870 }
2871 }
2872
2873 /* note: if switch is in round-robin mode, all links
2874 * must tx arp to ensure all links rx an arp - otherwise
2875 * links may oscillate or not come up at all; if switch is
2876 * in something like xor mode, there is nothing we can
2877 * do - all replies will be rx'ed on same link causing slaves
2878 * to be unstable during low/no traffic periods
2879 */
2880 if (IS_UP(slave->dev))
2881 bond_arp_send_all(bond, slave);
2882 }
2883
2884 if (do_failover) {
2885 block_netpoll_tx();
2886 write_lock_bh(&bond->curr_slave_lock);
2887
2888 bond_select_active_slave(bond);
2889
2890 write_unlock_bh(&bond->curr_slave_lock);
2891 unblock_netpoll_tx();
2892 }
2893
2894re_arm:
2895 if (bond->params.arp_interval && !bond->kill_timers)
2896 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2897out:
2898 read_unlock(&bond->lock);
2899}
2900
2901/*
2902 * Called to inspect slaves for active-backup mode ARP monitor link state
2903 * changes. Sets new_link in slaves to specify what action should take
2904 * place for the slave. Returns 0 if no changes are found, >0 if changes
2905 * to link states must be committed.
2906 *
2907 * Called with bond->lock held for read.
2908 */
2909static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2910{
2911 struct slave *slave;
2912 int i, commit = 0;
2913 unsigned long trans_start;
2914
2915 bond_for_each_slave(bond, slave, i) {
2916 slave->new_link = BOND_LINK_NOCHANGE;
2917
2918 if (slave->link != BOND_LINK_UP) {
2919 if (time_in_range(jiffies,
2920 slave_last_rx(bond, slave) - delta_in_ticks,
2921 slave_last_rx(bond, slave) + delta_in_ticks)) {
2922
2923 slave->new_link = BOND_LINK_UP;
2924 commit++;
2925 }
2926
2927 continue;
2928 }
2929
2930 /*
2931 * Give slaves 2*delta after being enslaved or made
2932 * active. This avoids bouncing, as the last receive
2933 * times need a full ARP monitor cycle to be updated.
2934 */
2935 if (time_in_range(jiffies,
2936 slave->jiffies - delta_in_ticks,
2937 slave->jiffies + 2 * delta_in_ticks))
2938 continue;
2939
2940 /*
2941 * Backup slave is down if:
2942 * - No current_arp_slave AND
2943 * - more than 3*delta since last receive AND
2944 * - the bond has an IP address
2945 *
2946 * Note: a non-null current_arp_slave indicates
2947 * the curr_active_slave went down and we are
2948 * searching for a new one; under this condition
2949 * we only take the curr_active_slave down - this
2950 * gives each slave a chance to tx/rx traffic
2951 * before being taken out
2952 */
2953 if (!bond_is_active_slave(slave) &&
2954 !bond->current_arp_slave &&
2955 !time_in_range(jiffies,
2956 slave_last_rx(bond, slave) - delta_in_ticks,
2957 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2958
2959 slave->new_link = BOND_LINK_DOWN;
2960 commit++;
2961 }
2962
2963 /*
2964 * Active slave is down if:
2965 * - more than 2*delta since transmitting OR
2966 * - (more than 2*delta since receive AND
2967 * the bond has an IP address)
2968 */
2969 trans_start = dev_trans_start(slave->dev);
2970 if (bond_is_active_slave(slave) &&
2971 (!time_in_range(jiffies,
2972 trans_start - delta_in_ticks,
2973 trans_start + 2 * delta_in_ticks) ||
2974 !time_in_range(jiffies,
2975 slave_last_rx(bond, slave) - delta_in_ticks,
2976 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2977
2978 slave->new_link = BOND_LINK_DOWN;
2979 commit++;
2980 }
2981 }
2982
2983 return commit;
2984}
2985
2986/*
2987 * Called to commit link state changes noted by inspection step of
2988 * active-backup mode ARP monitor.
2989 *
2990 * Called with RTNL and bond->lock for read.
2991 */
2992static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2993{
2994 struct slave *slave;
2995 int i;
2996 unsigned long trans_start;
2997
2998 bond_for_each_slave(bond, slave, i) {
2999 switch (slave->new_link) {
3000 case BOND_LINK_NOCHANGE:
3001 continue;
3002
3003 case BOND_LINK_UP:
3004 trans_start = dev_trans_start(slave->dev);
3005 if ((!bond->curr_active_slave &&
3006 time_in_range(jiffies,
3007 trans_start - delta_in_ticks,
3008 trans_start + delta_in_ticks)) ||
3009 bond->curr_active_slave != slave) {
3010 slave->link = BOND_LINK_UP;
3011 bond->current_arp_slave = NULL;
3012
3013 pr_info("%s: link status definitely up for interface %s.\n",
3014 bond->dev->name, slave->dev->name);
3015
3016 if (!bond->curr_active_slave ||
3017 (slave == bond->primary_slave))
3018 goto do_failover;
3019
3020 }
3021
3022 continue;
3023
3024 case BOND_LINK_DOWN:
3025 if (slave->link_failure_count < UINT_MAX)
3026 slave->link_failure_count++;
3027
3028 slave->link = BOND_LINK_DOWN;
3029 bond_set_slave_inactive_flags(slave);
3030
3031 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3032 bond->dev->name, slave->dev->name);
3033
3034 if (slave == bond->curr_active_slave) {
3035 bond->current_arp_slave = NULL;
3036 goto do_failover;
3037 }
3038
3039 continue;
3040
3041 default:
3042 pr_err("%s: impossible: new_link %d on slave %s\n",
3043 bond->dev->name, slave->new_link,
3044 slave->dev->name);
3045 continue;
3046 }
3047
3048do_failover:
3049 ASSERT_RTNL();
3050 block_netpoll_tx();
3051 write_lock_bh(&bond->curr_slave_lock);
3052 bond_select_active_slave(bond);
3053 write_unlock_bh(&bond->curr_slave_lock);
3054 unblock_netpoll_tx();
3055 }
3056
3057 bond_set_carrier(bond);
3058}
3059
3060/*
3061 * Send ARP probes for active-backup mode ARP monitor.
3062 *
3063 * Called with bond->lock held for read.
3064 */
3065static void bond_ab_arp_probe(struct bonding *bond)
3066{
3067 struct slave *slave;
3068 int i;
3069
3070 read_lock(&bond->curr_slave_lock);
3071
3072 if (bond->current_arp_slave && bond->curr_active_slave)
3073 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3074 bond->current_arp_slave->dev->name,
3075 bond->curr_active_slave->dev->name);
3076
3077 if (bond->curr_active_slave) {
3078 bond_arp_send_all(bond, bond->curr_active_slave);
3079 read_unlock(&bond->curr_slave_lock);
3080 return;
3081 }
3082
3083 read_unlock(&bond->curr_slave_lock);
3084
3085 /* if we don't have a curr_active_slave, search for the next available
3086 * backup slave from the current_arp_slave and make it the candidate
3087 * for becoming the curr_active_slave
3088 */
3089
3090 if (!bond->current_arp_slave) {
3091 bond->current_arp_slave = bond->first_slave;
3092 if (!bond->current_arp_slave)
3093 return;
3094 }
3095
3096 bond_set_slave_inactive_flags(bond->current_arp_slave);
3097
3098 /* search for next candidate */
3099 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3100 if (IS_UP(slave->dev)) {
3101 slave->link = BOND_LINK_BACK;
3102 bond_set_slave_active_flags(slave);
3103 bond_arp_send_all(bond, slave);
3104 slave->jiffies = jiffies;
3105 bond->current_arp_slave = slave;
3106 break;
3107 }
3108
3109 /* if the link state is up at this point, we
3110 * mark it down - this can happen if we have
3111 * simultaneous link failures and
3112 * reselect_active_interface doesn't make this
3113 * one the current slave so it is still marked
3114 * up when it is actually down
3115 */
3116 if (slave->link == BOND_LINK_UP) {
3117 slave->link = BOND_LINK_DOWN;
3118 if (slave->link_failure_count < UINT_MAX)
3119 slave->link_failure_count++;
3120
3121 bond_set_slave_inactive_flags(slave);
3122
3123 pr_info("%s: backup interface %s is now down.\n",
3124 bond->dev->name, slave->dev->name);
3125 }
3126 }
3127}
3128
3129void bond_activebackup_arp_mon(struct work_struct *work)
3130{
3131 struct bonding *bond = container_of(work, struct bonding,
3132 arp_work.work);
3133 bool should_notify_peers = false;
3134 int delta_in_ticks;
3135
3136 read_lock(&bond->lock);
3137
3138 if (bond->kill_timers)
3139 goto out;
3140
3141 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3142
3143 if (bond->slave_cnt == 0)
3144 goto re_arm;
3145
3146 should_notify_peers = bond_should_notify_peers(bond);
3147
3148 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3149 read_unlock(&bond->lock);
3150 rtnl_lock();
3151 read_lock(&bond->lock);
3152
3153 bond_ab_arp_commit(bond, delta_in_ticks);
3154
3155 read_unlock(&bond->lock);
3156 rtnl_unlock();
3157 read_lock(&bond->lock);
3158 }
3159
3160 bond_ab_arp_probe(bond);
3161
3162re_arm:
3163 if (bond->params.arp_interval && !bond->kill_timers)
3164 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3165out:
3166 read_unlock(&bond->lock);
3167
3168 if (should_notify_peers) {
3169 rtnl_lock();
3170 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3171 rtnl_unlock();
3172 }
3173}
3174
3175/*-------------------------- netdev event handling --------------------------*/
3176
3177/*
3178 * Change device name
3179 */
3180static int bond_event_changename(struct bonding *bond)
3181{
3182 bond_remove_proc_entry(bond);
3183 bond_create_proc_entry(bond);
3184
3185 bond_debug_reregister(bond);
3186
3187 return NOTIFY_DONE;
3188}
3189
3190static int bond_master_netdev_event(unsigned long event,
3191 struct net_device *bond_dev)
3192{
3193 struct bonding *event_bond = netdev_priv(bond_dev);
3194
3195 switch (event) {
3196 case NETDEV_CHANGENAME:
3197 return bond_event_changename(event_bond);
3198 default:
3199 break;
3200 }
3201
3202 return NOTIFY_DONE;
3203}
3204
3205static int bond_slave_netdev_event(unsigned long event,
3206 struct net_device *slave_dev)
3207{
3208 struct net_device *bond_dev = slave_dev->master;
3209 struct bonding *bond = netdev_priv(bond_dev);
3210
3211 switch (event) {
3212 case NETDEV_UNREGISTER:
3213 if (bond_dev) {
3214 if (bond->setup_by_slave)
3215 bond_release_and_destroy(bond_dev, slave_dev);
3216 else
3217 bond_release(bond_dev, slave_dev);
3218 }
3219 break;
3220 case NETDEV_CHANGE:
3221 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3222 struct slave *slave;
3223
3224 slave = bond_get_slave_by_dev(bond, slave_dev);
3225 if (slave) {
3226 u32 old_speed = slave->speed;
3227 u8 old_duplex = slave->duplex;
3228
3229 bond_update_speed_duplex(slave);
3230
3231 if (bond_is_lb(bond))
3232 break;
3233
3234 if (old_speed != slave->speed)
3235 bond_3ad_adapter_speed_changed(slave);
3236 if (old_duplex != slave->duplex)
3237 bond_3ad_adapter_duplex_changed(slave);
3238 }
3239 }
3240
3241 break;
3242 case NETDEV_DOWN:
3243 /*
3244 * ... Or is it this?
3245 */
3246 break;
3247 case NETDEV_CHANGEMTU:
3248 /*
3249 * TODO: Should slaves be allowed to
3250 * independently alter their MTU? For
3251 * an active-backup bond, slaves need
3252 * not be the same type of device, so
3253 * MTUs may vary. For other modes,
3254 * slaves arguably should have the
3255 * same MTUs. To do this, we'd need to
3256 * take over the slave's change_mtu
3257 * function for the duration of their
3258 * servitude.
3259 */
3260 break;
3261 case NETDEV_CHANGENAME:
3262 /*
3263 * TODO: handle changing the primary's name
3264 */
3265 break;
3266 case NETDEV_FEAT_CHANGE:
3267 bond_compute_features(bond);
3268 break;
3269 default:
3270 break;
3271 }
3272
3273 return NOTIFY_DONE;
3274}
3275
3276/*
3277 * bond_netdev_event: handle netdev notifier chain events.
3278 *
3279 * This function receives events for the netdev chain. The caller (an
3280 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3281 * locks for us to safely manipulate the slave devices (RTNL lock,
3282 * dev_probe_lock).
3283 */
3284static int bond_netdev_event(struct notifier_block *this,
3285 unsigned long event, void *ptr)
3286{
3287 struct net_device *event_dev = (struct net_device *)ptr;
3288
3289 pr_debug("event_dev: %s, event: %lx\n",
3290 event_dev ? event_dev->name : "None",
3291 event);
3292
3293 if (!(event_dev->priv_flags & IFF_BONDING))
3294 return NOTIFY_DONE;
3295
3296 if (event_dev->flags & IFF_MASTER) {
3297 pr_debug("IFF_MASTER\n");
3298 return bond_master_netdev_event(event, event_dev);
3299 }
3300
3301 if (event_dev->flags & IFF_SLAVE) {
3302 pr_debug("IFF_SLAVE\n");
3303 return bond_slave_netdev_event(event, event_dev);
3304 }
3305
3306 return NOTIFY_DONE;
3307}
3308
3309/*
3310 * bond_inetaddr_event: handle inetaddr notifier chain events.
3311 *
3312 * We keep track of device IPs primarily to use as source addresses in
3313 * ARP monitor probes (rather than spewing out broadcasts all the time).
3314 *
3315 * We track one IP for the main device (if it has one), plus one per VLAN.
3316 */
3317static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3318{
3319 struct in_ifaddr *ifa = ptr;
3320 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3321 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3322 struct bonding *bond;
3323 struct vlan_entry *vlan;
3324
3325 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3326 if (bond->dev == event_dev) {
3327 switch (event) {
3328 case NETDEV_UP:
3329 bond->master_ip = ifa->ifa_local;
3330 return NOTIFY_OK;
3331 case NETDEV_DOWN:
3332 bond->master_ip = bond_glean_dev_ip(bond->dev);
3333 return NOTIFY_OK;
3334 default:
3335 return NOTIFY_DONE;
3336 }
3337 }
3338
3339 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3340 vlan_dev = __vlan_find_dev_deep(bond->dev,
3341 vlan->vlan_id);
3342 if (vlan_dev == event_dev) {
3343 switch (event) {
3344 case NETDEV_UP:
3345 vlan->vlan_ip = ifa->ifa_local;
3346 return NOTIFY_OK;
3347 case NETDEV_DOWN:
3348 vlan->vlan_ip =
3349 bond_glean_dev_ip(vlan_dev);
3350 return NOTIFY_OK;
3351 default:
3352 return NOTIFY_DONE;
3353 }
3354 }
3355 }
3356 }
3357 return NOTIFY_DONE;
3358}
3359
3360static struct notifier_block bond_netdev_notifier = {
3361 .notifier_call = bond_netdev_event,
3362};
3363
3364static struct notifier_block bond_inetaddr_notifier = {
3365 .notifier_call = bond_inetaddr_event,
3366};
3367
3368/*---------------------------- Hashing Policies -----------------------------*/
3369
3370/*
3371 * Hash for the output device based upon layer 2 and layer 3 data. If
3372 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3373 */
3374static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3375{
3376 struct ethhdr *data = (struct ethhdr *)skb->data;
3377 struct iphdr *iph = ip_hdr(skb);
3378
3379 if (skb->protocol == htons(ETH_P_IP)) {
3380 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3381 (data->h_dest[5] ^ data->h_source[5])) % count;
3382 }
3383
3384 return (data->h_dest[5] ^ data->h_source[5]) % count;
3385}
3386
3387/*
3388 * Hash for the output device based upon layer 3 and layer 4 data. If
3389 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3390 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3391 */
3392static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3393{
3394 struct ethhdr *data = (struct ethhdr *)skb->data;
3395 struct iphdr *iph = ip_hdr(skb);
3396 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3397 int layer4_xor = 0;
3398
3399 if (skb->protocol == htons(ETH_P_IP)) {
3400 if (!ip_is_fragment(iph) &&
3401 (iph->protocol == IPPROTO_TCP ||
3402 iph->protocol == IPPROTO_UDP)) {
3403 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3404 }
3405 return (layer4_xor ^
3406 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3407
3408 }
3409
3410 return (data->h_dest[5] ^ data->h_source[5]) % count;
3411}
3412
3413/*
3414 * Hash for the output device based upon layer 2 data
3415 */
3416static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3417{
3418 struct ethhdr *data = (struct ethhdr *)skb->data;
3419
3420 return (data->h_dest[5] ^ data->h_source[5]) % count;
3421}
3422
3423/*-------------------------- Device entry points ----------------------------*/
3424
3425static int bond_open(struct net_device *bond_dev)
3426{
3427 struct bonding *bond = netdev_priv(bond_dev);
3428 struct slave *slave;
3429 int i;
3430
3431 bond->kill_timers = 0;
3432
3433 /* reset slave->backup and slave->inactive */
3434 read_lock(&bond->lock);
3435 if (bond->slave_cnt > 0) {
3436 read_lock(&bond->curr_slave_lock);
3437 bond_for_each_slave(bond, slave, i) {
3438 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3439 && (slave != bond->curr_active_slave)) {
3440 bond_set_slave_inactive_flags(slave);
3441 } else {
3442 bond_set_slave_active_flags(slave);
3443 }
3444 }
3445 read_unlock(&bond->curr_slave_lock);
3446 }
3447 read_unlock(&bond->lock);
3448
3449 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3450
3451 if (bond_is_lb(bond)) {
3452 /* bond_alb_initialize must be called before the timer
3453 * is started.
3454 */
3455 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3456 /* something went wrong - fail the open operation */
3457 return -ENOMEM;
3458 }
3459
3460 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3461 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3462 }
3463
3464 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3465 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3466 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3467 }
3468
3469 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3470 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3471 INIT_DELAYED_WORK(&bond->arp_work,
3472 bond_activebackup_arp_mon);
3473 else
3474 INIT_DELAYED_WORK(&bond->arp_work,
3475 bond_loadbalance_arp_mon);
3476
3477 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3478 if (bond->params.arp_validate)
3479 bond->recv_probe = bond_arp_rcv;
3480 }
3481
3482 if (bond->params.mode == BOND_MODE_8023AD) {
3483 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3484 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3485 /* register to receive LACPDUs */
3486 bond->recv_probe = bond_3ad_lacpdu_recv;
3487 bond_3ad_initiate_agg_selection(bond, 1);
3488 }
3489
3490 return 0;
3491}
3492
3493static int bond_close(struct net_device *bond_dev)
3494{
3495 struct bonding *bond = netdev_priv(bond_dev);
3496
3497 write_lock_bh(&bond->lock);
3498
3499 bond->send_peer_notif = 0;
3500
3501 /* signal timers not to re-arm */
3502 bond->kill_timers = 1;
3503
3504 write_unlock_bh(&bond->lock);
3505
3506 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3507 cancel_delayed_work(&bond->mii_work);
3508 }
3509
3510 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3511 cancel_delayed_work(&bond->arp_work);
3512 }
3513
3514 switch (bond->params.mode) {
3515 case BOND_MODE_8023AD:
3516 cancel_delayed_work(&bond->ad_work);
3517 break;
3518 case BOND_MODE_TLB:
3519 case BOND_MODE_ALB:
3520 cancel_delayed_work(&bond->alb_work);
3521 break;
3522 default:
3523 break;
3524 }
3525
3526 if (delayed_work_pending(&bond->mcast_work))
3527 cancel_delayed_work(&bond->mcast_work);
3528
3529 if (bond_is_lb(bond)) {
3530 /* Must be called only after all
3531 * slaves have been released
3532 */
3533 bond_alb_deinitialize(bond);
3534 }
3535 bond->recv_probe = NULL;
3536
3537 return 0;
3538}
3539
3540static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3541 struct rtnl_link_stats64 *stats)
3542{
3543 struct bonding *bond = netdev_priv(bond_dev);
3544 struct rtnl_link_stats64 temp;
3545 struct slave *slave;
3546 int i;
3547
3548 memset(stats, 0, sizeof(*stats));
3549
3550 read_lock_bh(&bond->lock);
3551
3552 bond_for_each_slave(bond, slave, i) {
3553 const struct rtnl_link_stats64 *sstats =
3554 dev_get_stats(slave->dev, &temp);
3555
3556 stats->rx_packets += sstats->rx_packets;
3557 stats->rx_bytes += sstats->rx_bytes;
3558 stats->rx_errors += sstats->rx_errors;
3559 stats->rx_dropped += sstats->rx_dropped;
3560
3561 stats->tx_packets += sstats->tx_packets;
3562 stats->tx_bytes += sstats->tx_bytes;
3563 stats->tx_errors += sstats->tx_errors;
3564 stats->tx_dropped += sstats->tx_dropped;
3565
3566 stats->multicast += sstats->multicast;
3567 stats->collisions += sstats->collisions;
3568
3569 stats->rx_length_errors += sstats->rx_length_errors;
3570 stats->rx_over_errors += sstats->rx_over_errors;
3571 stats->rx_crc_errors += sstats->rx_crc_errors;
3572 stats->rx_frame_errors += sstats->rx_frame_errors;
3573 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3574 stats->rx_missed_errors += sstats->rx_missed_errors;
3575
3576 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3577 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3578 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3579 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3580 stats->tx_window_errors += sstats->tx_window_errors;
3581 }
3582
3583 read_unlock_bh(&bond->lock);
3584
3585 return stats;
3586}
3587
3588static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3589{
3590 struct net_device *slave_dev = NULL;
3591 struct ifbond k_binfo;
3592 struct ifbond __user *u_binfo = NULL;
3593 struct ifslave k_sinfo;
3594 struct ifslave __user *u_sinfo = NULL;
3595 struct mii_ioctl_data *mii = NULL;
3596 int res = 0;
3597
3598 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3599
3600 switch (cmd) {
3601 case SIOCGMIIPHY:
3602 mii = if_mii(ifr);
3603 if (!mii)
3604 return -EINVAL;
3605
3606 mii->phy_id = 0;
3607 /* Fall Through */
3608 case SIOCGMIIREG:
3609 /*
3610 * We do this again just in case we were called by SIOCGMIIREG
3611 * instead of SIOCGMIIPHY.
3612 */
3613 mii = if_mii(ifr);
3614 if (!mii)
3615 return -EINVAL;
3616
3617
3618 if (mii->reg_num == 1) {
3619 struct bonding *bond = netdev_priv(bond_dev);
3620 mii->val_out = 0;
3621 read_lock(&bond->lock);
3622 read_lock(&bond->curr_slave_lock);
3623 if (netif_carrier_ok(bond->dev))
3624 mii->val_out = BMSR_LSTATUS;
3625
3626 read_unlock(&bond->curr_slave_lock);
3627 read_unlock(&bond->lock);
3628 }
3629
3630 return 0;
3631 case BOND_INFO_QUERY_OLD:
3632 case SIOCBONDINFOQUERY:
3633 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3634
3635 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3636 return -EFAULT;
3637
3638 res = bond_info_query(bond_dev, &k_binfo);
3639 if (res == 0 &&
3640 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3641 return -EFAULT;
3642
3643 return res;
3644 case BOND_SLAVE_INFO_QUERY_OLD:
3645 case SIOCBONDSLAVEINFOQUERY:
3646 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3647
3648 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3649 return -EFAULT;
3650
3651 res = bond_slave_info_query(bond_dev, &k_sinfo);
3652 if (res == 0 &&
3653 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3654 return -EFAULT;
3655
3656 return res;
3657 default:
3658 /* Go on */
3659 break;
3660 }
3661
3662 if (!capable(CAP_NET_ADMIN))
3663 return -EPERM;
3664
3665 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3666
3667 pr_debug("slave_dev=%p:\n", slave_dev);
3668
3669 if (!slave_dev)
3670 res = -ENODEV;
3671 else {
3672 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3673 switch (cmd) {
3674 case BOND_ENSLAVE_OLD:
3675 case SIOCBONDENSLAVE:
3676 res = bond_enslave(bond_dev, slave_dev);
3677 break;
3678 case BOND_RELEASE_OLD:
3679 case SIOCBONDRELEASE:
3680 res = bond_release(bond_dev, slave_dev);
3681 break;
3682 case BOND_SETHWADDR_OLD:
3683 case SIOCBONDSETHWADDR:
3684 res = bond_sethwaddr(bond_dev, slave_dev);
3685 break;
3686 case BOND_CHANGE_ACTIVE_OLD:
3687 case SIOCBONDCHANGEACTIVE:
3688 res = bond_ioctl_change_active(bond_dev, slave_dev);
3689 break;
3690 default:
3691 res = -EOPNOTSUPP;
3692 }
3693
3694 dev_put(slave_dev);
3695 }
3696
3697 return res;
3698}
3699
3700static bool bond_addr_in_mc_list(unsigned char *addr,
3701 struct netdev_hw_addr_list *list,
3702 int addrlen)
3703{
3704 struct netdev_hw_addr *ha;
3705
3706 netdev_hw_addr_list_for_each(ha, list)
3707 if (!memcmp(ha->addr, addr, addrlen))
3708 return true;
3709
3710 return false;
3711}
3712
3713static void bond_set_multicast_list(struct net_device *bond_dev)
3714{
3715 struct bonding *bond = netdev_priv(bond_dev);
3716 struct netdev_hw_addr *ha;
3717 bool found;
3718
3719 /*
3720 * Do promisc before checking multicast_mode
3721 */
3722 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3723 /*
3724 * FIXME: Need to handle the error when one of the multi-slaves
3725 * encounters error.
3726 */
3727 bond_set_promiscuity(bond, 1);
3728
3729
3730 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3731 bond_set_promiscuity(bond, -1);
3732
3733
3734 /* set allmulti flag to slaves */
3735 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3736 /*
3737 * FIXME: Need to handle the error when one of the multi-slaves
3738 * encounters error.
3739 */
3740 bond_set_allmulti(bond, 1);
3741
3742
3743 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3744 bond_set_allmulti(bond, -1);
3745
3746
3747 read_lock(&bond->lock);
3748
3749 bond->flags = bond_dev->flags;
3750
3751 /* looking for addresses to add to slaves' mc list */
3752 netdev_for_each_mc_addr(ha, bond_dev) {
3753 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3754 bond_dev->addr_len);
3755 if (!found)
3756 bond_mc_add(bond, ha->addr);
3757 }
3758
3759 /* looking for addresses to delete from slaves' list */
3760 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3761 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3762 bond_dev->addr_len);
3763 if (!found)
3764 bond_mc_del(bond, ha->addr);
3765 }
3766
3767 /* save master's multicast list */
3768 __hw_addr_flush(&bond->mc_list);
3769 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3770 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3771
3772 read_unlock(&bond->lock);
3773}
3774
3775static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3776{
3777 struct bonding *bond = netdev_priv(dev);
3778 struct slave *slave = bond->first_slave;
3779
3780 if (slave) {
3781 const struct net_device_ops *slave_ops
3782 = slave->dev->netdev_ops;
3783 if (slave_ops->ndo_neigh_setup)
3784 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3785 }
3786 return 0;
3787}
3788
3789/*
3790 * Change the MTU of all of a master's slaves to match the master
3791 */
3792static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3793{
3794 struct bonding *bond = netdev_priv(bond_dev);
3795 struct slave *slave, *stop_at;
3796 int res = 0;
3797 int i;
3798
3799 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3800 (bond_dev ? bond_dev->name : "None"), new_mtu);
3801
3802 /* Can't hold bond->lock with bh disabled here since
3803 * some base drivers panic. On the other hand we can't
3804 * hold bond->lock without bh disabled because we'll
3805 * deadlock. The only solution is to rely on the fact
3806 * that we're under rtnl_lock here, and the slaves
3807 * list won't change. This doesn't solve the problem
3808 * of setting the slave's MTU while it is
3809 * transmitting, but the assumption is that the base
3810 * driver can handle that.
3811 *
3812 * TODO: figure out a way to safely iterate the slaves
3813 * list, but without holding a lock around the actual
3814 * call to the base driver.
3815 */
3816
3817 bond_for_each_slave(bond, slave, i) {
3818 pr_debug("s %p s->p %p c_m %p\n",
3819 slave,
3820 slave->prev,
3821 slave->dev->netdev_ops->ndo_change_mtu);
3822
3823 res = dev_set_mtu(slave->dev, new_mtu);
3824
3825 if (res) {
3826 /* If we failed to set the slave's mtu to the new value
3827 * we must abort the operation even in ACTIVE_BACKUP
3828 * mode, because if we allow the backup slaves to have
3829 * different mtu values than the active slave we'll
3830 * need to change their mtu when doing a failover. That
3831 * means changing their mtu from timer context, which
3832 * is probably not a good idea.
3833 */
3834 pr_debug("err %d %s\n", res, slave->dev->name);
3835 goto unwind;
3836 }
3837 }
3838
3839 bond_dev->mtu = new_mtu;
3840
3841 return 0;
3842
3843unwind:
3844 /* unwind from head to the slave that failed */
3845 stop_at = slave;
3846 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3847 int tmp_res;
3848
3849 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3850 if (tmp_res) {
3851 pr_debug("unwind err %d dev %s\n",
3852 tmp_res, slave->dev->name);
3853 }
3854 }
3855
3856 return res;
3857}
3858
3859/*
3860 * Change HW address
3861 *
3862 * Note that many devices must be down to change the HW address, and
3863 * downing the master releases all slaves. We can make bonds full of
3864 * bonding devices to test this, however.
3865 */
3866static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3867{
3868 struct bonding *bond = netdev_priv(bond_dev);
3869 struct sockaddr *sa = addr, tmp_sa;
3870 struct slave *slave, *stop_at;
3871 int res = 0;
3872 int i;
3873
3874 if (bond->params.mode == BOND_MODE_ALB)
3875 return bond_alb_set_mac_address(bond_dev, addr);
3876
3877
3878 pr_debug("bond=%p, name=%s\n",
3879 bond, bond_dev ? bond_dev->name : "None");
3880
3881 /*
3882 * If fail_over_mac is set to active, do nothing and return
3883 * success. Returning an error causes ifenslave to fail.
3884 */
3885 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3886 return 0;
3887
3888 if (!is_valid_ether_addr(sa->sa_data))
3889 return -EADDRNOTAVAIL;
3890
3891 /* Can't hold bond->lock with bh disabled here since
3892 * some base drivers panic. On the other hand we can't
3893 * hold bond->lock without bh disabled because we'll
3894 * deadlock. The only solution is to rely on the fact
3895 * that we're under rtnl_lock here, and the slaves
3896 * list won't change. This doesn't solve the problem
3897 * of setting the slave's hw address while it is
3898 * transmitting, but the assumption is that the base
3899 * driver can handle that.
3900 *
3901 * TODO: figure out a way to safely iterate the slaves
3902 * list, but without holding a lock around the actual
3903 * call to the base driver.
3904 */
3905
3906 bond_for_each_slave(bond, slave, i) {
3907 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3908 pr_debug("slave %p %s\n", slave, slave->dev->name);
3909
3910 if (slave_ops->ndo_set_mac_address == NULL) {
3911 res = -EOPNOTSUPP;
3912 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3913 goto unwind;
3914 }
3915
3916 res = dev_set_mac_address(slave->dev, addr);
3917 if (res) {
3918 /* TODO: consider downing the slave
3919 * and retry ?
3920 * User should expect communications
3921 * breakage anyway until ARP finish
3922 * updating, so...
3923 */
3924 pr_debug("err %d %s\n", res, slave->dev->name);
3925 goto unwind;
3926 }
3927 }
3928
3929 /* success */
3930 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3931 return 0;
3932
3933unwind:
3934 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3935 tmp_sa.sa_family = bond_dev->type;
3936
3937 /* unwind from head to the slave that failed */
3938 stop_at = slave;
3939 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3940 int tmp_res;
3941
3942 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3943 if (tmp_res) {
3944 pr_debug("unwind err %d dev %s\n",
3945 tmp_res, slave->dev->name);
3946 }
3947 }
3948
3949 return res;
3950}
3951
3952static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3953{
3954 struct bonding *bond = netdev_priv(bond_dev);
3955 struct slave *slave, *start_at;
3956 int i, slave_no, res = 1;
3957 struct iphdr *iph = ip_hdr(skb);
3958
3959 /*
3960 * Start with the curr_active_slave that joined the bond as the
3961 * default for sending IGMP traffic. For failover purposes one
3962 * needs to maintain some consistency for the interface that will
3963 * send the join/membership reports. The curr_active_slave found
3964 * will send all of this type of traffic.
3965 */
3966 if ((iph->protocol == IPPROTO_IGMP) &&
3967 (skb->protocol == htons(ETH_P_IP))) {
3968
3969 read_lock(&bond->curr_slave_lock);
3970 slave = bond->curr_active_slave;
3971 read_unlock(&bond->curr_slave_lock);
3972
3973 if (!slave)
3974 goto out;
3975 } else {
3976 /*
3977 * Concurrent TX may collide on rr_tx_counter; we accept
3978 * that as being rare enough not to justify using an
3979 * atomic op here.
3980 */
3981 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3982
3983 bond_for_each_slave(bond, slave, i) {
3984 slave_no--;
3985 if (slave_no < 0)
3986 break;
3987 }
3988 }
3989
3990 start_at = slave;
3991 bond_for_each_slave_from(bond, slave, i, start_at) {
3992 if (IS_UP(slave->dev) &&
3993 (slave->link == BOND_LINK_UP) &&
3994 bond_is_active_slave(slave)) {
3995 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3996 break;
3997 }
3998 }
3999
4000out:
4001 if (res) {
4002 /* no suitable interface, frame not sent */
4003 dev_kfree_skb(skb);
4004 }
4005
4006 return NETDEV_TX_OK;
4007}
4008
4009
4010/*
4011 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4012 * the bond has a usable interface.
4013 */
4014static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4015{
4016 struct bonding *bond = netdev_priv(bond_dev);
4017 int res = 1;
4018
4019 read_lock(&bond->curr_slave_lock);
4020
4021 if (bond->curr_active_slave)
4022 res = bond_dev_queue_xmit(bond, skb,
4023 bond->curr_active_slave->dev);
4024
4025 if (res)
4026 /* no suitable interface, frame not sent */
4027 dev_kfree_skb(skb);
4028
4029 read_unlock(&bond->curr_slave_lock);
4030
4031 return NETDEV_TX_OK;
4032}
4033
4034/*
4035 * In bond_xmit_xor() , we determine the output device by using a pre-
4036 * determined xmit_hash_policy(), If the selected device is not enabled,
4037 * find the next active slave.
4038 */
4039static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4040{
4041 struct bonding *bond = netdev_priv(bond_dev);
4042 struct slave *slave, *start_at;
4043 int slave_no;
4044 int i;
4045 int res = 1;
4046
4047 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4048
4049 bond_for_each_slave(bond, slave, i) {
4050 slave_no--;
4051 if (slave_no < 0)
4052 break;
4053 }
4054
4055 start_at = slave;
4056
4057 bond_for_each_slave_from(bond, slave, i, start_at) {
4058 if (IS_UP(slave->dev) &&
4059 (slave->link == BOND_LINK_UP) &&
4060 bond_is_active_slave(slave)) {
4061 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4062 break;
4063 }
4064 }
4065
4066 if (res) {
4067 /* no suitable interface, frame not sent */
4068 dev_kfree_skb(skb);
4069 }
4070
4071 return NETDEV_TX_OK;
4072}
4073
4074/*
4075 * in broadcast mode, we send everything to all usable interfaces.
4076 */
4077static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4078{
4079 struct bonding *bond = netdev_priv(bond_dev);
4080 struct slave *slave, *start_at;
4081 struct net_device *tx_dev = NULL;
4082 int i;
4083 int res = 1;
4084
4085 read_lock(&bond->curr_slave_lock);
4086 start_at = bond->curr_active_slave;
4087 read_unlock(&bond->curr_slave_lock);
4088
4089 if (!start_at)
4090 goto out;
4091
4092 bond_for_each_slave_from(bond, slave, i, start_at) {
4093 if (IS_UP(slave->dev) &&
4094 (slave->link == BOND_LINK_UP) &&
4095 bond_is_active_slave(slave)) {
4096 if (tx_dev) {
4097 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4098 if (!skb2) {
4099 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4100 bond_dev->name);
4101 continue;
4102 }
4103
4104 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4105 if (res) {
4106 dev_kfree_skb(skb2);
4107 continue;
4108 }
4109 }
4110 tx_dev = slave->dev;
4111 }
4112 }
4113
4114 if (tx_dev)
4115 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4116
4117out:
4118 if (res)
4119 /* no suitable interface, frame not sent */
4120 dev_kfree_skb(skb);
4121
4122 /* frame sent to all suitable interfaces */
4123 return NETDEV_TX_OK;
4124}
4125
4126/*------------------------- Device initialization ---------------------------*/
4127
4128static void bond_set_xmit_hash_policy(struct bonding *bond)
4129{
4130 switch (bond->params.xmit_policy) {
4131 case BOND_XMIT_POLICY_LAYER23:
4132 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4133 break;
4134 case BOND_XMIT_POLICY_LAYER34:
4135 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4136 break;
4137 case BOND_XMIT_POLICY_LAYER2:
4138 default:
4139 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4140 break;
4141 }
4142}
4143
4144/*
4145 * Lookup the slave that corresponds to a qid
4146 */
4147static inline int bond_slave_override(struct bonding *bond,
4148 struct sk_buff *skb)
4149{
4150 int i, res = 1;
4151 struct slave *slave = NULL;
4152 struct slave *check_slave;
4153
4154 if (!skb->queue_mapping)
4155 return 1;
4156
4157 /* Find out if any slaves have the same mapping as this skb. */
4158 bond_for_each_slave(bond, check_slave, i) {
4159 if (check_slave->queue_id == skb->queue_mapping) {
4160 slave = check_slave;
4161 break;
4162 }
4163 }
4164
4165 /* If the slave isn't UP, use default transmit policy. */
4166 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4167 (slave->link == BOND_LINK_UP)) {
4168 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4169 }
4170
4171 return res;
4172}
4173
4174
4175static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4176{
4177 /*
4178 * This helper function exists to help dev_pick_tx get the correct
4179 * destination queue. Using a helper function skips a call to
4180 * skb_tx_hash and will put the skbs in the queue we expect on their
4181 * way down to the bonding driver.
4182 */
4183 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4184
4185 /*
4186 * Save the original txq to restore before passing to the driver
4187 */
4188 bond_queue_mapping(skb) = skb->queue_mapping;
4189
4190 if (unlikely(txq >= dev->real_num_tx_queues)) {
4191 do {
4192 txq -= dev->real_num_tx_queues;
4193 } while (txq >= dev->real_num_tx_queues);
4194 }
4195 return txq;
4196}
4197
4198static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4199{
4200 struct bonding *bond = netdev_priv(dev);
4201
4202 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4203 if (!bond_slave_override(bond, skb))
4204 return NETDEV_TX_OK;
4205 }
4206
4207 switch (bond->params.mode) {
4208 case BOND_MODE_ROUNDROBIN:
4209 return bond_xmit_roundrobin(skb, dev);
4210 case BOND_MODE_ACTIVEBACKUP:
4211 return bond_xmit_activebackup(skb, dev);
4212 case BOND_MODE_XOR:
4213 return bond_xmit_xor(skb, dev);
4214 case BOND_MODE_BROADCAST:
4215 return bond_xmit_broadcast(skb, dev);
4216 case BOND_MODE_8023AD:
4217 return bond_3ad_xmit_xor(skb, dev);
4218 case BOND_MODE_ALB:
4219 case BOND_MODE_TLB:
4220 return bond_alb_xmit(skb, dev);
4221 default:
4222 /* Should never happen, mode already checked */
4223 pr_err("%s: Error: Unknown bonding mode %d\n",
4224 dev->name, bond->params.mode);
4225 WARN_ON_ONCE(1);
4226 dev_kfree_skb(skb);
4227 return NETDEV_TX_OK;
4228 }
4229}
4230
4231static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4232{
4233 struct bonding *bond = netdev_priv(dev);
4234 netdev_tx_t ret = NETDEV_TX_OK;
4235
4236 /*
4237 * If we risk deadlock from transmitting this in the
4238 * netpoll path, tell netpoll to queue the frame for later tx
4239 */
4240 if (is_netpoll_tx_blocked(dev))
4241 return NETDEV_TX_BUSY;
4242
4243 read_lock(&bond->lock);
4244
4245 if (bond->slave_cnt)
4246 ret = __bond_start_xmit(skb, dev);
4247 else
4248 dev_kfree_skb(skb);
4249
4250 read_unlock(&bond->lock);
4251
4252 return ret;
4253}
4254
4255/*
4256 * set bond mode specific net device operations
4257 */
4258void bond_set_mode_ops(struct bonding *bond, int mode)
4259{
4260 struct net_device *bond_dev = bond->dev;
4261
4262 switch (mode) {
4263 case BOND_MODE_ROUNDROBIN:
4264 break;
4265 case BOND_MODE_ACTIVEBACKUP:
4266 break;
4267 case BOND_MODE_XOR:
4268 bond_set_xmit_hash_policy(bond);
4269 break;
4270 case BOND_MODE_BROADCAST:
4271 break;
4272 case BOND_MODE_8023AD:
4273 bond_set_xmit_hash_policy(bond);
4274 break;
4275 case BOND_MODE_ALB:
4276 /* FALLTHRU */
4277 case BOND_MODE_TLB:
4278 break;
4279 default:
4280 /* Should never happen, mode already checked */
4281 pr_err("%s: Error: Unknown bonding mode %d\n",
4282 bond_dev->name, mode);
4283 break;
4284 }
4285}
4286
4287static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4288 struct ethtool_drvinfo *drvinfo)
4289{
4290 strncpy(drvinfo->driver, DRV_NAME, 32);
4291 strncpy(drvinfo->version, DRV_VERSION, 32);
4292 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4293}
4294
4295static const struct ethtool_ops bond_ethtool_ops = {
4296 .get_drvinfo = bond_ethtool_get_drvinfo,
4297 .get_link = ethtool_op_get_link,
4298};
4299
4300static const struct net_device_ops bond_netdev_ops = {
4301 .ndo_init = bond_init,
4302 .ndo_uninit = bond_uninit,
4303 .ndo_open = bond_open,
4304 .ndo_stop = bond_close,
4305 .ndo_start_xmit = bond_start_xmit,
4306 .ndo_select_queue = bond_select_queue,
4307 .ndo_get_stats64 = bond_get_stats,
4308 .ndo_do_ioctl = bond_do_ioctl,
4309 .ndo_set_multicast_list = bond_set_multicast_list,
4310 .ndo_change_mtu = bond_change_mtu,
4311 .ndo_set_mac_address = bond_set_mac_address,
4312 .ndo_neigh_setup = bond_neigh_setup,
4313 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4314 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4315#ifdef CONFIG_NET_POLL_CONTROLLER
4316 .ndo_netpoll_setup = bond_netpoll_setup,
4317 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4318 .ndo_poll_controller = bond_poll_controller,
4319#endif
4320 .ndo_add_slave = bond_enslave,
4321 .ndo_del_slave = bond_release,
4322 .ndo_fix_features = bond_fix_features,
4323};
4324
4325static void bond_destructor(struct net_device *bond_dev)
4326{
4327 struct bonding *bond = netdev_priv(bond_dev);
4328 if (bond->wq)
4329 destroy_workqueue(bond->wq);
4330 free_netdev(bond_dev);
4331}
4332
4333static void bond_setup(struct net_device *bond_dev)
4334{
4335 struct bonding *bond = netdev_priv(bond_dev);
4336
4337 /* initialize rwlocks */
4338 rwlock_init(&bond->lock);
4339 rwlock_init(&bond->curr_slave_lock);
4340
4341 bond->params = bonding_defaults;
4342
4343 /* Initialize pointers */
4344 bond->dev = bond_dev;
4345 INIT_LIST_HEAD(&bond->vlan_list);
4346
4347 /* Initialize the device entry points */
4348 ether_setup(bond_dev);
4349 bond_dev->netdev_ops = &bond_netdev_ops;
4350 bond_dev->ethtool_ops = &bond_ethtool_ops;
4351 bond_set_mode_ops(bond, bond->params.mode);
4352
4353 bond_dev->destructor = bond_destructor;
4354
4355 /* Initialize the device options */
4356 bond_dev->tx_queue_len = 0;
4357 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4358 bond_dev->priv_flags |= IFF_BONDING;
4359 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4360
4361 /* At first, we block adding VLANs. That's the only way to
4362 * prevent problems that occur when adding VLANs over an
4363 * empty bond. The block will be removed once non-challenged
4364 * slaves are enslaved.
4365 */
4366 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4367
4368 /* don't acquire bond device's netif_tx_lock when
4369 * transmitting */
4370 bond_dev->features |= NETIF_F_LLTX;
4371
4372 /* By default, we declare the bond to be fully
4373 * VLAN hardware accelerated capable. Special
4374 * care is taken in the various xmit functions
4375 * when there are slaves that are not hw accel
4376 * capable
4377 */
4378
4379 bond_dev->hw_features = BOND_VLAN_FEATURES |
4380 NETIF_F_HW_VLAN_TX |
4381 NETIF_F_HW_VLAN_RX |
4382 NETIF_F_HW_VLAN_FILTER;
4383
4384 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4385 bond_dev->features |= bond_dev->hw_features;
4386}
4387
4388static void bond_work_cancel_all(struct bonding *bond)
4389{
4390 write_lock_bh(&bond->lock);
4391 bond->kill_timers = 1;
4392 write_unlock_bh(&bond->lock);
4393
4394 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4395 cancel_delayed_work(&bond->mii_work);
4396
4397 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4398 cancel_delayed_work(&bond->arp_work);
4399
4400 if (bond->params.mode == BOND_MODE_ALB &&
4401 delayed_work_pending(&bond->alb_work))
4402 cancel_delayed_work(&bond->alb_work);
4403
4404 if (bond->params.mode == BOND_MODE_8023AD &&
4405 delayed_work_pending(&bond->ad_work))
4406 cancel_delayed_work(&bond->ad_work);
4407
4408 if (delayed_work_pending(&bond->mcast_work))
4409 cancel_delayed_work(&bond->mcast_work);
4410}
4411
4412/*
4413* Destroy a bonding device.
4414* Must be under rtnl_lock when this function is called.
4415*/
4416static void bond_uninit(struct net_device *bond_dev)
4417{
4418 struct bonding *bond = netdev_priv(bond_dev);
4419 struct vlan_entry *vlan, *tmp;
4420
4421 bond_netpoll_cleanup(bond_dev);
4422
4423 /* Release the bonded slaves */
4424 bond_release_all(bond_dev);
4425
4426 list_del(&bond->bond_list);
4427
4428 bond_work_cancel_all(bond);
4429
4430 bond_remove_proc_entry(bond);
4431
4432 bond_debug_unregister(bond);
4433
4434 __hw_addr_flush(&bond->mc_list);
4435
4436 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4437 list_del(&vlan->vlan_list);
4438 kfree(vlan);
4439 }
4440}
4441
4442/*------------------------- Module initialization ---------------------------*/
4443
4444/*
4445 * Convert string input module parms. Accept either the
4446 * number of the mode or its string name. A bit complicated because
4447 * some mode names are substrings of other names, and calls from sysfs
4448 * may have whitespace in the name (trailing newlines, for example).
4449 */
4450int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4451{
4452 int modeint = -1, i, rv;
4453 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4454
4455 for (p = (char *)buf; *p; p++)
4456 if (!(isdigit(*p) || isspace(*p)))
4457 break;
4458
4459 if (*p)
4460 rv = sscanf(buf, "%20s", modestr);
4461 else
4462 rv = sscanf(buf, "%d", &modeint);
4463
4464 if (!rv)
4465 return -1;
4466
4467 for (i = 0; tbl[i].modename; i++) {
4468 if (modeint == tbl[i].mode)
4469 return tbl[i].mode;
4470 if (strcmp(modestr, tbl[i].modename) == 0)
4471 return tbl[i].mode;
4472 }
4473
4474 return -1;
4475}
4476
4477static int bond_check_params(struct bond_params *params)
4478{
4479 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4480
4481 /*
4482 * Convert string parameters.
4483 */
4484 if (mode) {
4485 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4486 if (bond_mode == -1) {
4487 pr_err("Error: Invalid bonding mode \"%s\"\n",
4488 mode == NULL ? "NULL" : mode);
4489 return -EINVAL;
4490 }
4491 }
4492
4493 if (xmit_hash_policy) {
4494 if ((bond_mode != BOND_MODE_XOR) &&
4495 (bond_mode != BOND_MODE_8023AD)) {
4496 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4497 bond_mode_name(bond_mode));
4498 } else {
4499 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4500 xmit_hashtype_tbl);
4501 if (xmit_hashtype == -1) {
4502 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4503 xmit_hash_policy == NULL ? "NULL" :
4504 xmit_hash_policy);
4505 return -EINVAL;
4506 }
4507 }
4508 }
4509
4510 if (lacp_rate) {
4511 if (bond_mode != BOND_MODE_8023AD) {
4512 pr_info("lacp_rate param is irrelevant in mode %s\n",
4513 bond_mode_name(bond_mode));
4514 } else {
4515 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4516 if (lacp_fast == -1) {
4517 pr_err("Error: Invalid lacp rate \"%s\"\n",
4518 lacp_rate == NULL ? "NULL" : lacp_rate);
4519 return -EINVAL;
4520 }
4521 }
4522 }
4523
4524 if (ad_select) {
4525 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4526 if (params->ad_select == -1) {
4527 pr_err("Error: Invalid ad_select \"%s\"\n",
4528 ad_select == NULL ? "NULL" : ad_select);
4529 return -EINVAL;
4530 }
4531
4532 if (bond_mode != BOND_MODE_8023AD) {
4533 pr_warning("ad_select param only affects 802.3ad mode\n");
4534 }
4535 } else {
4536 params->ad_select = BOND_AD_STABLE;
4537 }
4538
4539 if (max_bonds < 0) {
4540 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4541 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4542 max_bonds = BOND_DEFAULT_MAX_BONDS;
4543 }
4544
4545 if (miimon < 0) {
4546 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4547 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4548 miimon = BOND_LINK_MON_INTERV;
4549 }
4550
4551 if (updelay < 0) {
4552 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4553 updelay, INT_MAX);
4554 updelay = 0;
4555 }
4556
4557 if (downdelay < 0) {
4558 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4559 downdelay, INT_MAX);
4560 downdelay = 0;
4561 }
4562
4563 if ((use_carrier != 0) && (use_carrier != 1)) {
4564 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4565 use_carrier);
4566 use_carrier = 1;
4567 }
4568
4569 if (num_peer_notif < 0 || num_peer_notif > 255) {
4570 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4571 num_peer_notif);
4572 num_peer_notif = 1;
4573 }
4574
4575 /* reset values for 802.3ad */
4576 if (bond_mode == BOND_MODE_8023AD) {
4577 if (!miimon) {
4578 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4579 pr_warning("Forcing miimon to 100msec\n");
4580 miimon = 100;
4581 }
4582 }
4583
4584 if (tx_queues < 1 || tx_queues > 255) {
4585 pr_warning("Warning: tx_queues (%d) should be between "
4586 "1 and 255, resetting to %d\n",
4587 tx_queues, BOND_DEFAULT_TX_QUEUES);
4588 tx_queues = BOND_DEFAULT_TX_QUEUES;
4589 }
4590
4591 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4592 pr_warning("Warning: all_slaves_active module parameter (%d), "
4593 "not of valid value (0/1), so it was set to "
4594 "0\n", all_slaves_active);
4595 all_slaves_active = 0;
4596 }
4597
4598 if (resend_igmp < 0 || resend_igmp > 255) {
4599 pr_warning("Warning: resend_igmp (%d) should be between "
4600 "0 and 255, resetting to %d\n",
4601 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4602 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4603 }
4604
4605 /* reset values for TLB/ALB */
4606 if ((bond_mode == BOND_MODE_TLB) ||
4607 (bond_mode == BOND_MODE_ALB)) {
4608 if (!miimon) {
4609 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4610 pr_warning("Forcing miimon to 100msec\n");
4611 miimon = 100;
4612 }
4613 }
4614
4615 if (bond_mode == BOND_MODE_ALB) {
4616 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",
4617 updelay);
4618 }
4619
4620 if (!miimon) {
4621 if (updelay || downdelay) {
4622 /* just warn the user the up/down delay will have
4623 * no effect since miimon is zero...
4624 */
4625 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4626 updelay, downdelay);
4627 }
4628 } else {
4629 /* don't allow arp monitoring */
4630 if (arp_interval) {
4631 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4632 miimon, arp_interval);
4633 arp_interval = 0;
4634 }
4635
4636 if ((updelay % miimon) != 0) {
4637 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4638 updelay, miimon,
4639 (updelay / miimon) * miimon);
4640 }
4641
4642 updelay /= miimon;
4643
4644 if ((downdelay % miimon) != 0) {
4645 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4646 downdelay, miimon,
4647 (downdelay / miimon) * miimon);
4648 }
4649
4650 downdelay /= miimon;
4651 }
4652
4653 if (arp_interval < 0) {
4654 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4655 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4656 arp_interval = BOND_LINK_ARP_INTERV;
4657 }
4658
4659 for (arp_ip_count = 0;
4660 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4661 arp_ip_count++) {
4662 /* not complete check, but should be good enough to
4663 catch mistakes */
4664 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4665 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4666 arp_ip_target[arp_ip_count]);
4667 arp_interval = 0;
4668 } else {
4669 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4670 arp_target[arp_ip_count] = ip;
4671 }
4672 }
4673
4674 if (arp_interval && !arp_ip_count) {
4675 /* don't allow arping if no arp_ip_target given... */
4676 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4677 arp_interval);
4678 arp_interval = 0;
4679 }
4680
4681 if (arp_validate) {
4682 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4683 pr_err("arp_validate only supported in active-backup mode\n");
4684 return -EINVAL;
4685 }
4686 if (!arp_interval) {
4687 pr_err("arp_validate requires arp_interval\n");
4688 return -EINVAL;
4689 }
4690
4691 arp_validate_value = bond_parse_parm(arp_validate,
4692 arp_validate_tbl);
4693 if (arp_validate_value == -1) {
4694 pr_err("Error: invalid arp_validate \"%s\"\n",
4695 arp_validate == NULL ? "NULL" : arp_validate);
4696 return -EINVAL;
4697 }
4698 } else
4699 arp_validate_value = 0;
4700
4701 if (miimon) {
4702 pr_info("MII link monitoring set to %d ms\n", miimon);
4703 } else if (arp_interval) {
4704 int i;
4705
4706 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4707 arp_interval,
4708 arp_validate_tbl[arp_validate_value].modename,
4709 arp_ip_count);
4710
4711 for (i = 0; i < arp_ip_count; i++)
4712 pr_info(" %s", arp_ip_target[i]);
4713
4714 pr_info("\n");
4715
4716 } else if (max_bonds) {
4717 /* miimon and arp_interval not set, we need one so things
4718 * work as expected, see bonding.txt for details
4719 */
4720 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");
4721 }
4722
4723 if (primary && !USES_PRIMARY(bond_mode)) {
4724 /* currently, using a primary only makes sense
4725 * in active backup, TLB or ALB modes
4726 */
4727 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4728 primary, bond_mode_name(bond_mode));
4729 primary = NULL;
4730 }
4731
4732 if (primary && primary_reselect) {
4733 primary_reselect_value = bond_parse_parm(primary_reselect,
4734 pri_reselect_tbl);
4735 if (primary_reselect_value == -1) {
4736 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4737 primary_reselect ==
4738 NULL ? "NULL" : primary_reselect);
4739 return -EINVAL;
4740 }
4741 } else {
4742 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4743 }
4744
4745 if (fail_over_mac) {
4746 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4747 fail_over_mac_tbl);
4748 if (fail_over_mac_value == -1) {
4749 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4750 arp_validate == NULL ? "NULL" : arp_validate);
4751 return -EINVAL;
4752 }
4753
4754 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4755 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4756 } else {
4757 fail_over_mac_value = BOND_FOM_NONE;
4758 }
4759
4760 /* fill params struct with the proper values */
4761 params->mode = bond_mode;
4762 params->xmit_policy = xmit_hashtype;
4763 params->miimon = miimon;
4764 params->num_peer_notif = num_peer_notif;
4765 params->arp_interval = arp_interval;
4766 params->arp_validate = arp_validate_value;
4767 params->updelay = updelay;
4768 params->downdelay = downdelay;
4769 params->use_carrier = use_carrier;
4770 params->lacp_fast = lacp_fast;
4771 params->primary[0] = 0;
4772 params->primary_reselect = primary_reselect_value;
4773 params->fail_over_mac = fail_over_mac_value;
4774 params->tx_queues = tx_queues;
4775 params->all_slaves_active = all_slaves_active;
4776 params->resend_igmp = resend_igmp;
4777 params->min_links = min_links;
4778
4779 if (primary) {
4780 strncpy(params->primary, primary, IFNAMSIZ);
4781 params->primary[IFNAMSIZ - 1] = 0;
4782 }
4783
4784 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4785
4786 return 0;
4787}
4788
4789static struct lock_class_key bonding_netdev_xmit_lock_key;
4790static struct lock_class_key bonding_netdev_addr_lock_key;
4791
4792static void bond_set_lockdep_class_one(struct net_device *dev,
4793 struct netdev_queue *txq,
4794 void *_unused)
4795{
4796 lockdep_set_class(&txq->_xmit_lock,
4797 &bonding_netdev_xmit_lock_key);
4798}
4799
4800static void bond_set_lockdep_class(struct net_device *dev)
4801{
4802 lockdep_set_class(&dev->addr_list_lock,
4803 &bonding_netdev_addr_lock_key);
4804 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4805}
4806
4807/*
4808 * Called from registration process
4809 */
4810static int bond_init(struct net_device *bond_dev)
4811{
4812 struct bonding *bond = netdev_priv(bond_dev);
4813 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4814 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4815
4816 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4817
4818 /*
4819 * Initialize locks that may be required during
4820 * en/deslave operations. All of the bond_open work
4821 * (of which this is part) should really be moved to
4822 * a phase prior to dev_open
4823 */
4824 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4825 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4826
4827 bond->wq = create_singlethread_workqueue(bond_dev->name);
4828 if (!bond->wq)
4829 return -ENOMEM;
4830
4831 bond_set_lockdep_class(bond_dev);
4832
4833 bond_create_proc_entry(bond);
4834 list_add_tail(&bond->bond_list, &bn->dev_list);
4835
4836 bond_prepare_sysfs_group(bond);
4837
4838 bond_debug_register(bond);
4839
4840 __hw_addr_init(&bond->mc_list);
4841 return 0;
4842}
4843
4844static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4845{
4846 if (tb[IFLA_ADDRESS]) {
4847 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4848 return -EINVAL;
4849 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4850 return -EADDRNOTAVAIL;
4851 }
4852 return 0;
4853}
4854
4855static struct rtnl_link_ops bond_link_ops __read_mostly = {
4856 .kind = "bond",
4857 .priv_size = sizeof(struct bonding),
4858 .setup = bond_setup,
4859 .validate = bond_validate,
4860};
4861
4862/* Create a new bond based on the specified name and bonding parameters.
4863 * If name is NULL, obtain a suitable "bond%d" name for us.
4864 * Caller must NOT hold rtnl_lock; we need to release it here before we
4865 * set up our sysfs entries.
4866 */
4867int bond_create(struct net *net, const char *name)
4868{
4869 struct net_device *bond_dev;
4870 int res;
4871
4872 rtnl_lock();
4873
4874 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4875 name ? name : "bond%d",
4876 bond_setup, tx_queues);
4877 if (!bond_dev) {
4878 pr_err("%s: eek! can't alloc netdev!\n", name);
4879 rtnl_unlock();
4880 return -ENOMEM;
4881 }
4882
4883 dev_net_set(bond_dev, net);
4884 bond_dev->rtnl_link_ops = &bond_link_ops;
4885
4886 res = register_netdevice(bond_dev);
4887
4888 netif_carrier_off(bond_dev);
4889
4890 rtnl_unlock();
4891 if (res < 0)
4892 bond_destructor(bond_dev);
4893 return res;
4894}
4895
4896static int __net_init bond_net_init(struct net *net)
4897{
4898 struct bond_net *bn = net_generic(net, bond_net_id);
4899
4900 bn->net = net;
4901 INIT_LIST_HEAD(&bn->dev_list);
4902
4903 bond_create_proc_dir(bn);
4904
4905 return 0;
4906}
4907
4908static void __net_exit bond_net_exit(struct net *net)
4909{
4910 struct bond_net *bn = net_generic(net, bond_net_id);
4911
4912 bond_destroy_proc_dir(bn);
4913}
4914
4915static struct pernet_operations bond_net_ops = {
4916 .init = bond_net_init,
4917 .exit = bond_net_exit,
4918 .id = &bond_net_id,
4919 .size = sizeof(struct bond_net),
4920};
4921
4922static int __init bonding_init(void)
4923{
4924 int i;
4925 int res;
4926
4927 pr_info("%s", bond_version);
4928
4929 res = bond_check_params(&bonding_defaults);
4930 if (res)
4931 goto out;
4932
4933 res = register_pernet_subsys(&bond_net_ops);
4934 if (res)
4935 goto out;
4936
4937 res = rtnl_link_register(&bond_link_ops);
4938 if (res)
4939 goto err_link;
4940
4941 bond_create_debugfs();
4942
4943 for (i = 0; i < max_bonds; i++) {
4944 res = bond_create(&init_net, NULL);
4945 if (res)
4946 goto err;
4947 }
4948
4949 res = bond_create_sysfs();
4950 if (res)
4951 goto err;
4952
4953 register_netdevice_notifier(&bond_netdev_notifier);
4954 register_inetaddr_notifier(&bond_inetaddr_notifier);
4955out:
4956 return res;
4957err:
4958 rtnl_link_unregister(&bond_link_ops);
4959err_link:
4960 unregister_pernet_subsys(&bond_net_ops);
4961 goto out;
4962
4963}
4964
4965static void __exit bonding_exit(void)
4966{
4967 unregister_netdevice_notifier(&bond_netdev_notifier);
4968 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4969
4970 bond_destroy_sysfs();
4971 bond_destroy_debugfs();
4972
4973 rtnl_link_unregister(&bond_link_ops);
4974 unregister_pernet_subsys(&bond_net_ops);
4975
4976#ifdef CONFIG_NET_POLL_CONTROLLER
4977 /*
4978 * Make sure we don't have an imbalance on our netpoll blocking
4979 */
4980 WARN_ON(atomic_read(&netpoll_block_tx));
4981#endif
4982}
4983
4984module_init(bonding_init);
4985module_exit(bonding_exit);
4986MODULE_LICENSE("GPL");
4987MODULE_VERSION(DRV_VERSION);
4988MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4989MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4990MODULE_ALIAS_RTNL_LINK("bond");