2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
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.
24 * will release all slaves, marking them as down.
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.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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>
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>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.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>
84 /*---------------------------- Module parameters ----------------------------*/
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
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon = BOND_LINK_MON_INTERV;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_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");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_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");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_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");
173 module_param(all_slaves_active, int, 0);
174 MODULE_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.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
187 int bond_net_id __read_mostly;
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode = BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 { "slow", AD_LACP_SLOW},
197 { "fast", AD_LACP_FAST},
201 const 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},
212 const 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},
219 const 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},
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 { "none", BOND_FOM_NONE},
229 { "active", BOND_FOM_ACTIVE},
230 { "follow", BOND_FOM_FOLLOW},
234 const 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},
241 struct bond_parm_tbl ad_select_tbl[] = {
242 { "stable", BOND_AD_STABLE},
243 { "bandwidth", BOND_AD_BANDWIDTH},
244 { "count", BOND_AD_COUNT},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode)
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",
267 if (mode < 0 || mode > BOND_MODE_ALB)
273 /*---------------------------------- VLAN -----------------------------------*/
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
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
284 struct vlan_entry *vlan;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond ? bond->dev->name : "None"), vlan_id);
289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
296 write_lock_bh(&bond->lock);
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
300 write_unlock_bh(&bond->lock);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
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
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
316 struct vlan_entry *vlan;
319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
322 write_lock_bh(&bond->lock);
324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vlan_id);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
345 write_unlock_bh(&bond->lock);
346 unblock_netpoll_tx();
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
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).
358 * Caller must hold bond->lock
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
362 struct vlan_entry *next, *last;
364 if (list_empty(&bond->vlan_list))
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
386 * bond_dev_queue_xmit - Prepare skb for xmit.
388 * @bond: bond device that got this skb for tx.
389 * @skb: hw accel VLAN tagged skb to transmit
390 * @slave_dev: slave that is supposed to xmit this skbuff
392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
393 struct net_device *slave_dev)
395 skb->dev = slave_dev;
397 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
398 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
399 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
401 if (unlikely(netpoll_tx_running(bond->dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433 struct bonding *bond = netdev_priv(bond_dev);
434 struct slave *slave, *stop_at;
437 bond_for_each_slave(bond, slave, i) {
438 res = vlan_vid_add(slave->dev, vid);
443 res = bond_add_vlan(bond, vid);
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev->name, vid);
453 /* unwind from head to the slave that failed */
455 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
456 vlan_vid_del(slave->dev, vid);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468 struct bonding *bond = netdev_priv(bond_dev);
472 bond_for_each_slave(bond, slave, i)
473 vlan_vid_del(slave->dev, vid);
475 res = bond_del_vlan(bond, vid);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev->name, vid);
485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
487 struct vlan_entry *vlan;
490 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
491 res = vlan_vid_add(slave_dev, vlan->vlan_id);
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond->dev->name, vlan->vlan_id,
499 static void bond_del_vlans_from_slave(struct bonding *bond,
500 struct net_device *slave_dev)
502 struct vlan_entry *vlan;
504 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
507 vlan_vid_del(slave_dev, vlan->vlan_id);
511 /*------------------------------- Link status -------------------------------*/
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
518 * Returns zero if carrier state does not change, nonzero if it does.
520 static int bond_set_carrier(struct bonding *bond)
525 if (bond->slave_cnt == 0)
528 if (bond->params.mode == BOND_MODE_8023AD)
529 return bond_3ad_set_carrier(bond);
531 bond_for_each_slave(bond, slave, i) {
532 if (slave->link == BOND_LINK_UP) {
533 if (!netif_carrier_ok(bond->dev)) {
534 netif_carrier_on(bond->dev);
542 if (netif_carrier_ok(bond->dev)) {
543 netif_carrier_off(bond->dev);
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
555 static void bond_update_speed_duplex(struct slave *slave)
557 struct net_device *slave_dev = slave->dev;
558 struct ethtool_cmd ecmd;
562 slave->speed = SPEED_UNKNOWN;
563 slave->duplex = DUPLEX_UNKNOWN;
565 res = __ethtool_get_settings(slave_dev, &ecmd);
569 slave_speed = ethtool_cmd_speed(&ecmd);
570 if (slave_speed == 0 || slave_speed == ((__u32) -1))
573 switch (ecmd.duplex) {
581 slave->speed = slave_speed;
582 slave->duplex = ecmd.duplex;
588 * if <dev> supports MII link status reporting, check its link status.
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
603 static int bond_check_dev_link(struct bonding *bond,
604 struct net_device *slave_dev, int reporting)
606 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
607 int (*ioctl)(struct net_device *, struct ifreq *, int);
609 struct mii_ioctl_data *mii;
611 if (!reporting && !netif_running(slave_dev))
614 if (bond->params.use_carrier)
615 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev->ethtool_ops->get_link)
619 return slave_dev->ethtool_ops->get_link(slave_dev) ?
622 /* Ethtool can't be used, fallback to MII ioctls. */
623 ioctl = slave_ops->ndo_do_ioctl;
625 /* TODO: set pointer to correct ioctl on a per team member */
626 /* bases to make this more efficient. that is, once */
627 /* we determine the correct ioctl, we will always */
628 /* call it and not the others for that team */
632 * We cannot assume that SIOCGMIIPHY will also read a
633 * register; not all network drivers (e.g., e100)
637 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
638 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
640 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
641 mii->reg_num = MII_BMSR;
642 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
643 return mii->val_out & BMSR_LSTATUS;
648 * If reporting, report that either there's no dev->do_ioctl,
649 * or both SIOCGMIIREG and get_link failed (meaning that we
650 * cannot report link status). If not reporting, pretend
653 return reporting ? -1 : BMSR_LSTATUS;
656 /*----------------------------- Multicast list ------------------------------*/
659 * Push the promiscuity flag down to appropriate slaves
661 static int bond_set_promiscuity(struct bonding *bond, int inc)
664 if (USES_PRIMARY(bond->params.mode)) {
665 /* write lock already acquired */
666 if (bond->curr_active_slave) {
667 err = dev_set_promiscuity(bond->curr_active_slave->dev,
673 bond_for_each_slave(bond, slave, i) {
674 err = dev_set_promiscuity(slave->dev, inc);
683 * Push the allmulti flag down to all slaves
685 static int bond_set_allmulti(struct bonding *bond, int inc)
688 if (USES_PRIMARY(bond->params.mode)) {
689 /* write lock already acquired */
690 if (bond->curr_active_slave) {
691 err = dev_set_allmulti(bond->curr_active_slave->dev,
697 bond_for_each_slave(bond, slave, i) {
698 err = dev_set_allmulti(slave->dev, inc);
707 * Add a Multicast address to slaves
710 static void bond_mc_add(struct bonding *bond, void *addr)
712 if (USES_PRIMARY(bond->params.mode)) {
713 /* write lock already acquired */
714 if (bond->curr_active_slave)
715 dev_mc_add(bond->curr_active_slave->dev, addr);
720 bond_for_each_slave(bond, slave, i)
721 dev_mc_add(slave->dev, addr);
726 * Remove a multicast address from slave
729 static void bond_mc_del(struct bonding *bond, void *addr)
731 if (USES_PRIMARY(bond->params.mode)) {
732 /* write lock already acquired */
733 if (bond->curr_active_slave)
734 dev_mc_del(bond->curr_active_slave->dev, addr);
738 bond_for_each_slave(bond, slave, i) {
739 dev_mc_del(slave->dev, addr);
745 static void __bond_resend_igmp_join_requests(struct net_device *dev)
747 struct in_device *in_dev;
749 in_dev = __in_dev_get_rcu(dev);
751 ip_mc_rejoin_groups(in_dev);
755 * Retrieve the list of registered multicast addresses for the bonding
756 * device and retransmit an IGMP JOIN request to the current active
759 static void bond_resend_igmp_join_requests(struct bonding *bond)
761 struct net_device *bond_dev, *vlan_dev, *upper_dev;
762 struct vlan_entry *vlan;
765 read_lock(&bond->lock);
767 bond_dev = bond->dev;
769 /* rejoin all groups on bond device */
770 __bond_resend_igmp_join_requests(bond_dev);
773 * if bond is enslaved to a bridge,
774 * then rejoin all groups on its master
776 upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
777 if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
778 __bond_resend_igmp_join_requests(upper_dev);
780 /* rejoin all groups on vlan devices */
781 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
782 vlan_dev = __vlan_find_dev_deep(bond_dev,
785 __bond_resend_igmp_join_requests(vlan_dev);
788 if (--bond->igmp_retrans > 0)
789 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
791 read_unlock(&bond->lock);
795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
797 struct bonding *bond = container_of(work, struct bonding,
800 bond_resend_igmp_join_requests(bond);
804 * flush all members of flush->mc_list from device dev->mc_list
806 static void bond_mc_list_flush(struct net_device *bond_dev,
807 struct net_device *slave_dev)
809 struct bonding *bond = netdev_priv(bond_dev);
810 struct netdev_hw_addr *ha;
812 netdev_for_each_mc_addr(ha, bond_dev)
813 dev_mc_del(slave_dev, ha->addr);
815 if (bond->params.mode == BOND_MODE_8023AD) {
816 /* del lacpdu mc addr from mc list */
817 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
819 dev_mc_del(slave_dev, lacpdu_multicast);
823 /*--------------------------- Active slave change ---------------------------*/
826 * Update the mc list and multicast-related flags for the new and
827 * old active slaves (if any) according to the multicast mode, and
828 * promiscuous flags unconditionally.
830 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
831 struct slave *old_active)
833 struct netdev_hw_addr *ha;
835 if (!USES_PRIMARY(bond->params.mode))
836 /* nothing to do - mc list is already up-to-date on
842 if (bond->dev->flags & IFF_PROMISC)
843 dev_set_promiscuity(old_active->dev, -1);
845 if (bond->dev->flags & IFF_ALLMULTI)
846 dev_set_allmulti(old_active->dev, -1);
848 netdev_for_each_mc_addr(ha, bond->dev)
849 dev_mc_del(old_active->dev, ha->addr);
853 /* FIXME: Signal errors upstream. */
854 if (bond->dev->flags & IFF_PROMISC)
855 dev_set_promiscuity(new_active->dev, 1);
857 if (bond->dev->flags & IFF_ALLMULTI)
858 dev_set_allmulti(new_active->dev, 1);
860 netdev_for_each_mc_addr(ha, bond->dev)
861 dev_mc_add(new_active->dev, ha->addr);
866 * bond_do_fail_over_mac
868 * Perform special MAC address swapping for fail_over_mac settings
870 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
872 static void bond_do_fail_over_mac(struct bonding *bond,
873 struct slave *new_active,
874 struct slave *old_active)
875 __releases(&bond->curr_slave_lock)
876 __releases(&bond->lock)
877 __acquires(&bond->lock)
878 __acquires(&bond->curr_slave_lock)
880 u8 tmp_mac[ETH_ALEN];
881 struct sockaddr saddr;
884 switch (bond->params.fail_over_mac) {
885 case BOND_FOM_ACTIVE:
887 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
888 new_active->dev->addr_len);
889 write_unlock_bh(&bond->curr_slave_lock);
890 read_unlock(&bond->lock);
891 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
892 read_lock(&bond->lock);
893 write_lock_bh(&bond->curr_slave_lock);
896 case BOND_FOM_FOLLOW:
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
905 write_unlock_bh(&bond->curr_slave_lock);
906 read_unlock(&bond->lock);
909 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
910 memcpy(saddr.sa_data, old_active->dev->dev_addr,
912 saddr.sa_family = new_active->dev->type;
914 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
915 saddr.sa_family = bond->dev->type;
918 rv = dev_set_mac_address(new_active->dev, &saddr);
920 pr_err("%s: Error %d setting MAC of slave %s\n",
921 bond->dev->name, -rv, new_active->dev->name);
928 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
929 saddr.sa_family = old_active->dev->type;
931 rv = dev_set_mac_address(old_active->dev, &saddr);
933 pr_err("%s: Error %d setting MAC of slave %s\n",
934 bond->dev->name, -rv, new_active->dev->name);
936 read_lock(&bond->lock);
937 write_lock_bh(&bond->curr_slave_lock);
940 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
941 bond->dev->name, bond->params.fail_over_mac);
947 static bool bond_should_change_active(struct bonding *bond)
949 struct slave *prim = bond->primary_slave;
950 struct slave *curr = bond->curr_active_slave;
952 if (!prim || !curr || curr->link != BOND_LINK_UP)
954 if (bond->force_primary) {
955 bond->force_primary = false;
958 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
959 (prim->speed < curr->speed ||
960 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
962 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
968 * find_best_interface - select the best available slave to be the active one
969 * @bond: our bonding struct
971 * Warning: Caller must hold curr_slave_lock for writing.
973 static struct slave *bond_find_best_slave(struct bonding *bond)
975 struct slave *new_active, *old_active;
976 struct slave *bestslave = NULL;
977 int mintime = bond->params.updelay;
980 new_active = bond->curr_active_slave;
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;
986 return NULL; /* still no slave, return NULL */
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;
995 /* remember where to stop iterating over the slaves */
996 old_active = new_active;
998 bond_for_each_slave_from(bond, new_active, i, old_active) {
999 if (new_active->link == BOND_LINK_UP) {
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;
1014 static bool bond_should_notify_peers(struct bonding *bond)
1016 struct slave *slave = bond->curr_active_slave;
1018 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1019 bond->dev->name, slave ? slave->dev->name : "NULL");
1021 if (!slave || !bond->send_peer_notif ||
1022 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1025 bond->send_peer_notif--;
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
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.
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.
1042 * If new_active is not NULL, caller must hold bond->lock for read and
1043 * curr_slave_lock for write_bh.
1045 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1047 struct slave *old_active = bond->curr_active_slave;
1049 if (old_active == new_active)
1053 new_active->jiffies = jiffies;
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);
1062 new_active->delay = 0;
1063 new_active->link = BOND_LINK_UP;
1065 if (bond->params.mode == BOND_MODE_8023AD)
1066 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1068 if (bond_is_lb(bond))
1069 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
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);
1078 if (USES_PRIMARY(bond->params.mode))
1079 bond_mc_swap(bond, new_active, old_active);
1081 if (bond_is_lb(bond)) {
1082 bond_alb_handle_active_change(bond, new_active);
1084 bond_set_slave_inactive_flags(old_active);
1086 bond_set_slave_active_flags(new_active);
1088 bond->curr_active_slave = new_active;
1091 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1093 bond_set_slave_inactive_flags(old_active);
1096 bool should_notify_peers = false;
1098 bond_set_slave_active_flags(new_active);
1100 if (bond->params.fail_over_mac)
1101 bond_do_fail_over_mac(bond, new_active,
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);
1111 write_unlock_bh(&bond->curr_slave_lock);
1112 read_unlock(&bond->lock);
1114 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1115 if (should_notify_peers)
1116 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1119 read_lock(&bond->lock);
1120 write_lock_bh(&bond->curr_slave_lock);
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);
1137 * bond_select_active_slave - select a new active slave, if needed
1138 * @bond: our bonding struct
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.
1145 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1147 void bond_select_active_slave(struct bonding *bond)
1149 struct slave *best_slave;
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);
1159 if (netif_carrier_ok(bond->dev)) {
1160 pr_info("%s: first active interface up!\n",
1163 pr_info("%s: now running without any active interface !\n",
1169 /*--------------------------- slave list handling ---------------------------*/
1172 * This function attaches the slave to the end of list.
1174 * bond->lock held for writing by caller.
1176 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
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;
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;
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.
1200 * bond->lock held for writing by caller.
1202 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1205 slave->next->prev = slave->prev;
1208 slave->prev->next = slave->next;
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;
1214 bond->first_slave = NULL; /* slave was the last one */
1223 #ifdef CONFIG_NET_POLL_CONTROLLER
1224 static inline int slave_enable_netpoll(struct slave *slave)
1229 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1234 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1243 static inline void slave_disable_netpoll(struct slave *slave)
1245 struct netpoll *np = slave->np;
1251 __netpoll_free_async(np);
1253 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257 if (!slave_dev->netdev_ops->ndo_poll_controller)
1262 static void bond_poll_controller(struct net_device *bond_dev)
1266 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 struct slave *slave;
1271 bond_for_each_slave(bond, slave, i)
1272 if (IS_UP(slave->dev))
1273 slave_disable_netpoll(slave);
1275 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 struct bonding *bond = netdev_priv(bond_dev);
1279 read_lock(&bond->lock);
1280 __bond_netpoll_cleanup(bond);
1281 read_unlock(&bond->lock);
1284 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1286 struct bonding *bond = netdev_priv(dev);
1287 struct slave *slave;
1290 read_lock(&bond->lock);
1291 bond_for_each_slave(bond, slave, i) {
1292 err = slave_enable_netpoll(slave);
1294 __bond_netpoll_cleanup(bond);
1298 read_unlock(&bond->lock);
1302 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 return bond->dev->npinfo;
1308 static inline int slave_enable_netpoll(struct slave *slave)
1312 static inline void slave_disable_netpoll(struct slave *slave)
1315 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1320 /*---------------------------------- IOCTL ----------------------------------*/
1322 static void bond_set_dev_addr(struct net_device *bond_dev,
1323 struct net_device *slave_dev)
1325 pr_debug("bond_dev=%p\n", bond_dev);
1326 pr_debug("slave_dev=%p\n", slave_dev);
1327 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1328 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1329 bond_dev->addr_assign_type = NET_ADDR_SET;
1330 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1333 static netdev_features_t bond_fix_features(struct net_device *dev,
1334 netdev_features_t features)
1336 struct slave *slave;
1337 struct bonding *bond = netdev_priv(dev);
1338 netdev_features_t mask;
1341 read_lock(&bond->lock);
1343 if (!bond->first_slave) {
1344 /* Disable adding VLANs to empty bond. But why? --mq */
1345 features |= NETIF_F_VLAN_CHALLENGED;
1350 features &= ~NETIF_F_ONE_FOR_ALL;
1351 features |= NETIF_F_ALL_FOR_ALL;
1353 bond_for_each_slave(bond, slave, i) {
1354 features = netdev_increment_features(features,
1355 slave->dev->features,
1360 read_unlock(&bond->lock);
1364 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1365 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1366 NETIF_F_HIGHDMA | NETIF_F_LRO)
1368 static void bond_compute_features(struct bonding *bond)
1370 struct slave *slave;
1371 struct net_device *bond_dev = bond->dev;
1372 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1373 unsigned short max_hard_header_len = ETH_HLEN;
1374 unsigned int gso_max_size = GSO_MAX_SIZE;
1375 u16 gso_max_segs = GSO_MAX_SEGS;
1377 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1379 read_lock(&bond->lock);
1381 if (!bond->first_slave)
1384 bond_for_each_slave(bond, slave, i) {
1385 vlan_features = netdev_increment_features(vlan_features,
1386 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1388 dst_release_flag &= slave->dev->priv_flags;
1389 if (slave->dev->hard_header_len > max_hard_header_len)
1390 max_hard_header_len = slave->dev->hard_header_len;
1392 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1393 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1397 bond_dev->vlan_features = vlan_features;
1398 bond_dev->hard_header_len = max_hard_header_len;
1399 bond_dev->gso_max_segs = gso_max_segs;
1400 netif_set_gso_max_size(bond_dev, gso_max_size);
1402 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1403 bond_dev->priv_flags = flags | dst_release_flag;
1405 read_unlock(&bond->lock);
1407 netdev_change_features(bond_dev);
1410 static void bond_setup_by_slave(struct net_device *bond_dev,
1411 struct net_device *slave_dev)
1413 struct bonding *bond = netdev_priv(bond_dev);
1415 bond_dev->header_ops = slave_dev->header_ops;
1417 bond_dev->type = slave_dev->type;
1418 bond_dev->hard_header_len = slave_dev->hard_header_len;
1419 bond_dev->addr_len = slave_dev->addr_len;
1421 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422 slave_dev->addr_len);
1423 bond->setup_by_slave = 1;
1426 /* On bonding slaves other than the currently active slave, suppress
1427 * duplicates except for alb non-mcast/bcast.
1429 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1430 struct slave *slave,
1431 struct bonding *bond)
1433 if (bond_is_slave_inactive(slave)) {
1434 if (bond->params.mode == BOND_MODE_ALB &&
1435 skb->pkt_type != PACKET_BROADCAST &&
1436 skb->pkt_type != PACKET_MULTICAST)
1443 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1445 struct sk_buff *skb = *pskb;
1446 struct slave *slave;
1447 struct bonding *bond;
1448 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1450 int ret = RX_HANDLER_ANOTHER;
1452 skb = skb_share_check(skb, GFP_ATOMIC);
1454 return RX_HANDLER_CONSUMED;
1458 slave = bond_slave_get_rcu(skb->dev);
1461 if (bond->params.arp_interval)
1462 slave->dev->last_rx = jiffies;
1464 recv_probe = ACCESS_ONCE(bond->recv_probe);
1466 ret = recv_probe(skb, bond, slave);
1467 if (ret == RX_HANDLER_CONSUMED) {
1473 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1474 return RX_HANDLER_EXACT;
1477 skb->dev = bond->dev;
1479 if (bond->params.mode == BOND_MODE_ALB &&
1480 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1481 skb->pkt_type == PACKET_HOST) {
1483 if (unlikely(skb_cow_head(skb,
1484 skb->data - skb_mac_header(skb)))) {
1486 return RX_HANDLER_CONSUMED;
1488 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1495 struct net_device *slave_dev)
1499 err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1502 slave_dev->flags |= IFF_SLAVE;
1503 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1507 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1508 struct net_device *slave_dev)
1510 netdev_upper_dev_unlink(slave_dev, bond_dev);
1511 slave_dev->flags &= ~IFF_SLAVE;
1512 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1515 /* enslave device <slave> to bond device <master> */
1516 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1518 struct bonding *bond = netdev_priv(bond_dev);
1519 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1520 struct slave *new_slave = NULL;
1521 struct netdev_hw_addr *ha;
1522 struct sockaddr addr;
1526 if (!bond->params.use_carrier &&
1527 slave_dev->ethtool_ops->get_link == NULL &&
1528 slave_ops->ndo_do_ioctl == NULL) {
1529 pr_warning("%s: Warning: no link monitoring support for %s\n",
1530 bond_dev->name, slave_dev->name);
1533 /* already enslaved */
1534 if (slave_dev->flags & IFF_SLAVE) {
1535 pr_debug("Error, Device was already enslaved\n");
1539 /* vlan challenged mutual exclusion */
1540 /* no need to lock since we're protected by rtnl_lock */
1541 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1542 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1543 if (vlan_uses_dev(bond_dev)) {
1544 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1545 bond_dev->name, slave_dev->name, bond_dev->name);
1548 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1549 bond_dev->name, slave_dev->name,
1550 slave_dev->name, bond_dev->name);
1553 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1557 * Old ifenslave binaries are no longer supported. These can
1558 * be identified with moderate accuracy by the state of the slave:
1559 * the current ifenslave will set the interface down prior to
1560 * enslaving it; the old ifenslave will not.
1562 if ((slave_dev->flags & IFF_UP)) {
1563 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1566 goto err_undo_flags;
1569 /* set bonding device ether type by slave - bonding netdevices are
1570 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1571 * there is a need to override some of the type dependent attribs/funcs.
1573 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1574 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1576 if (bond->slave_cnt == 0) {
1577 if (bond_dev->type != slave_dev->type) {
1578 pr_debug("%s: change device type from %d to %d\n",
1580 bond_dev->type, slave_dev->type);
1582 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1584 res = notifier_to_errno(res);
1586 pr_err("%s: refused to change device type\n",
1589 goto err_undo_flags;
1592 /* Flush unicast and multicast addresses */
1593 dev_uc_flush(bond_dev);
1594 dev_mc_flush(bond_dev);
1596 if (slave_dev->type != ARPHRD_ETHER)
1597 bond_setup_by_slave(bond_dev, slave_dev);
1599 ether_setup(bond_dev);
1600 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1603 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1606 } else if (bond_dev->type != slave_dev->type) {
1607 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1609 slave_dev->type, bond_dev->type);
1611 goto err_undo_flags;
1614 if (slave_ops->ndo_set_mac_address == NULL) {
1615 if (bond->slave_cnt == 0) {
1616 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1618 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1619 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1620 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",
1623 goto err_undo_flags;
1627 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1629 /* If this is the first slave, then we need to set the master's hardware
1630 * address to be the same as the slave's. */
1631 if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1632 bond_set_dev_addr(bond->dev, slave_dev);
1634 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1637 goto err_undo_flags;
1641 * Set the new_slave's queue_id to be zero. Queue ID mapping
1642 * is set via sysfs or module option if desired.
1644 new_slave->queue_id = 0;
1646 /* Save slave's original mtu and then set it to match the bond */
1647 new_slave->original_mtu = slave_dev->mtu;
1648 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1650 pr_debug("Error %d calling dev_set_mtu\n", res);
1655 * Save slave's original ("permanent") mac address for modes
1656 * that need it, and for restoring it upon release, and then
1657 * set it to the master's address
1659 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1661 if (!bond->params.fail_over_mac) {
1663 * Set slave to master's mac address. The application already
1664 * set the master's mac address to that of the first slave
1666 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1667 addr.sa_family = slave_dev->type;
1668 res = dev_set_mac_address(slave_dev, &addr);
1670 pr_debug("Error %d calling set_mac_address\n", res);
1671 goto err_restore_mtu;
1675 res = bond_master_upper_dev_link(bond_dev, slave_dev);
1677 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1678 goto err_restore_mac;
1681 /* open the slave since the application closed it */
1682 res = dev_open(slave_dev);
1684 pr_debug("Opening slave %s failed\n", slave_dev->name);
1685 goto err_unset_master;
1688 new_slave->bond = bond;
1689 new_slave->dev = slave_dev;
1690 slave_dev->priv_flags |= IFF_BONDING;
1692 if (bond_is_lb(bond)) {
1693 /* bond_alb_init_slave() must be called before all other stages since
1694 * it might fail and we do not want to have to undo everything
1696 res = bond_alb_init_slave(bond, new_slave);
1701 /* If the mode USES_PRIMARY, then the new slave gets the
1702 * master's promisc (and mc) settings only if it becomes the
1703 * curr_active_slave, and that is taken care of later when calling
1704 * bond_change_active()
1706 if (!USES_PRIMARY(bond->params.mode)) {
1707 /* set promiscuity level to new slave */
1708 if (bond_dev->flags & IFF_PROMISC) {
1709 res = dev_set_promiscuity(slave_dev, 1);
1714 /* set allmulti level to new slave */
1715 if (bond_dev->flags & IFF_ALLMULTI) {
1716 res = dev_set_allmulti(slave_dev, 1);
1721 netif_addr_lock_bh(bond_dev);
1722 /* upload master's mc_list to new slave */
1723 netdev_for_each_mc_addr(ha, bond_dev)
1724 dev_mc_add(slave_dev, ha->addr);
1725 netif_addr_unlock_bh(bond_dev);
1728 if (bond->params.mode == BOND_MODE_8023AD) {
1729 /* add lacpdu mc addr to mc list */
1730 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1732 dev_mc_add(slave_dev, lacpdu_multicast);
1735 bond_add_vlans_on_slave(bond, slave_dev);
1737 write_lock_bh(&bond->lock);
1739 bond_attach_slave(bond, new_slave);
1741 new_slave->delay = 0;
1742 new_slave->link_failure_count = 0;
1744 write_unlock_bh(&bond->lock);
1746 bond_compute_features(bond);
1748 bond_update_speed_duplex(new_slave);
1750 read_lock(&bond->lock);
1752 new_slave->last_arp_rx = jiffies -
1753 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1755 if (bond->params.miimon && !bond->params.use_carrier) {
1756 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1758 if ((link_reporting == -1) && !bond->params.arp_interval) {
1760 * miimon is set but a bonded network driver
1761 * does not support ETHTOOL/MII and
1762 * arp_interval is not set. Note: if
1763 * use_carrier is enabled, we will never go
1764 * here (because netif_carrier is always
1765 * supported); thus, we don't need to change
1766 * the messages for netif_carrier.
1768 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",
1769 bond_dev->name, slave_dev->name);
1770 } else if (link_reporting == -1) {
1771 /* unable get link status using mii/ethtool */
1772 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",
1773 bond_dev->name, slave_dev->name);
1777 /* check for initial state */
1778 if (bond->params.miimon) {
1779 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1780 if (bond->params.updelay) {
1781 new_slave->link = BOND_LINK_BACK;
1782 new_slave->delay = bond->params.updelay;
1784 new_slave->link = BOND_LINK_UP;
1787 new_slave->link = BOND_LINK_DOWN;
1789 } else if (bond->params.arp_interval) {
1790 new_slave->link = (netif_carrier_ok(slave_dev) ?
1791 BOND_LINK_UP : BOND_LINK_DOWN);
1793 new_slave->link = BOND_LINK_UP;
1796 if (new_slave->link != BOND_LINK_DOWN)
1797 new_slave->jiffies = jiffies;
1798 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1799 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1800 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1802 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1803 /* if there is a primary slave, remember it */
1804 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1805 bond->primary_slave = new_slave;
1806 bond->force_primary = true;
1810 write_lock_bh(&bond->curr_slave_lock);
1812 switch (bond->params.mode) {
1813 case BOND_MODE_ACTIVEBACKUP:
1814 bond_set_slave_inactive_flags(new_slave);
1815 bond_select_active_slave(bond);
1817 case BOND_MODE_8023AD:
1818 /* in 802.3ad mode, the internal mechanism
1819 * will activate the slaves in the selected
1822 bond_set_slave_inactive_flags(new_slave);
1823 /* if this is the first slave */
1824 if (bond->slave_cnt == 1) {
1825 SLAVE_AD_INFO(new_slave).id = 1;
1826 /* Initialize AD with the number of times that the AD timer is called in 1 second
1827 * can be called only after the mac address of the bond is set
1829 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1831 SLAVE_AD_INFO(new_slave).id =
1832 SLAVE_AD_INFO(new_slave->prev).id + 1;
1835 bond_3ad_bind_slave(new_slave);
1839 bond_set_active_slave(new_slave);
1840 bond_set_slave_inactive_flags(new_slave);
1841 bond_select_active_slave(bond);
1844 pr_debug("This slave is always active in trunk mode\n");
1846 /* always active in trunk mode */
1847 bond_set_active_slave(new_slave);
1849 /* In trunking mode there is little meaning to curr_active_slave
1850 * anyway (it holds no special properties of the bond device),
1851 * so we can change it without calling change_active_interface()
1853 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1854 bond->curr_active_slave = new_slave;
1857 } /* switch(bond_mode) */
1859 write_unlock_bh(&bond->curr_slave_lock);
1861 bond_set_carrier(bond);
1863 #ifdef CONFIG_NET_POLL_CONTROLLER
1864 slave_dev->npinfo = bond_netpoll_info(bond);
1865 if (slave_dev->npinfo) {
1866 if (slave_enable_netpoll(new_slave)) {
1867 read_unlock(&bond->lock);
1868 pr_info("Error, %s: master_dev is using netpoll, "
1869 "but new slave device does not support netpoll.\n",
1877 read_unlock(&bond->lock);
1879 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1883 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1886 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1887 goto err_dest_symlinks;
1890 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1891 bond_dev->name, slave_dev->name,
1892 bond_is_active_slave(new_slave) ? "n active" : " backup",
1893 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1895 /* enslave is successful */
1898 /* Undo stages on error */
1900 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1903 write_lock_bh(&bond->lock);
1904 bond_detach_slave(bond, new_slave);
1905 write_unlock_bh(&bond->lock);
1908 dev_close(slave_dev);
1911 bond_upper_dev_unlink(bond_dev, slave_dev);
1914 if (!bond->params.fail_over_mac) {
1915 /* XXX TODO - fom follow mode needs to change master's
1916 * MAC if this slave's MAC is in use by the bond, or at
1917 * least print a warning.
1919 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1920 addr.sa_family = slave_dev->type;
1921 dev_set_mac_address(slave_dev, &addr);
1925 dev_set_mtu(slave_dev, new_slave->original_mtu);
1931 bond_compute_features(bond);
1937 * Try to release the slave device <slave> from the bond device <master>
1938 * It is legal to access curr_active_slave without a lock because all the function
1939 * is write-locked. If "all" is true it means that the function is being called
1940 * while destroying a bond interface and all slaves are being released.
1942 * The rules for slave state should be:
1943 * for Active/Backup:
1944 * Active stays on all backups go down
1945 * for Bonded connections:
1946 * The first up interface should be left on and all others downed.
1948 static int __bond_release_one(struct net_device *bond_dev,
1949 struct net_device *slave_dev,
1952 struct bonding *bond = netdev_priv(bond_dev);
1953 struct slave *slave, *oldcurrent;
1954 struct sockaddr addr;
1955 netdev_features_t old_features = bond_dev->features;
1957 /* slave is not a slave or master is not master of this slave */
1958 if (!(slave_dev->flags & IFF_SLAVE) ||
1959 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1960 pr_err("%s: Error: cannot release %s.\n",
1961 bond_dev->name, slave_dev->name);
1966 write_lock_bh(&bond->lock);
1968 slave = bond_get_slave_by_dev(bond, slave_dev);
1970 /* not a slave of this bond */
1971 pr_info("%s: %s not enslaved\n",
1972 bond_dev->name, slave_dev->name);
1973 write_unlock_bh(&bond->lock);
1974 unblock_netpoll_tx();
1978 write_unlock_bh(&bond->lock);
1979 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1980 * for this slave anymore.
1982 netdev_rx_handler_unregister(slave_dev);
1983 write_lock_bh(&bond->lock);
1985 if (!all && !bond->params.fail_over_mac) {
1986 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1987 bond->slave_cnt > 1)
1988 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",
1989 bond_dev->name, slave_dev->name,
1991 bond_dev->name, slave_dev->name);
1994 /* Inform AD package of unbinding of slave. */
1995 if (bond->params.mode == BOND_MODE_8023AD) {
1996 /* must be called before the slave is
1997 * detached from the list
1999 bond_3ad_unbind_slave(slave);
2002 pr_info("%s: releasing %s interface %s\n",
2004 bond_is_active_slave(slave) ? "active" : "backup",
2007 oldcurrent = bond->curr_active_slave;
2009 bond->current_arp_slave = NULL;
2011 /* release the slave from its bond */
2012 bond_detach_slave(bond, slave);
2014 if (bond->primary_slave == slave)
2015 bond->primary_slave = NULL;
2017 if (oldcurrent == slave)
2018 bond_change_active_slave(bond, NULL);
2020 if (bond_is_lb(bond)) {
2021 /* Must be called only after the slave has been
2022 * detached from the list and the curr_active_slave
2023 * has been cleared (if our_slave == old_current),
2024 * but before a new active slave is selected.
2026 write_unlock_bh(&bond->lock);
2027 bond_alb_deinit_slave(bond, slave);
2028 write_lock_bh(&bond->lock);
2032 bond->curr_active_slave = NULL;
2033 } else if (oldcurrent == slave) {
2035 * Note that we hold RTNL over this sequence, so there
2036 * is no concern that another slave add/remove event
2039 write_unlock_bh(&bond->lock);
2040 read_lock(&bond->lock);
2041 write_lock_bh(&bond->curr_slave_lock);
2043 bond_select_active_slave(bond);
2045 write_unlock_bh(&bond->curr_slave_lock);
2046 read_unlock(&bond->lock);
2047 write_lock_bh(&bond->lock);
2050 if (bond->slave_cnt == 0) {
2051 bond_set_carrier(bond);
2052 eth_hw_addr_random(bond_dev);
2053 bond->dev_addr_from_first = true;
2055 if (bond_vlan_used(bond)) {
2056 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2057 bond_dev->name, bond_dev->name);
2058 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2063 write_unlock_bh(&bond->lock);
2064 unblock_netpoll_tx();
2066 if (bond->slave_cnt == 0) {
2067 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2068 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2071 bond_compute_features(bond);
2072 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2073 (old_features & NETIF_F_VLAN_CHALLENGED))
2074 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2075 bond_dev->name, slave_dev->name, bond_dev->name);
2077 /* must do this from outside any spinlocks */
2078 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2080 bond_del_vlans_from_slave(bond, slave_dev);
2082 /* If the mode USES_PRIMARY, then we should only remove its
2083 * promisc and mc settings if it was the curr_active_slave, but that was
2084 * already taken care of above when we detached the slave
2086 if (!USES_PRIMARY(bond->params.mode)) {
2087 /* unset promiscuity level from slave */
2088 if (bond_dev->flags & IFF_PROMISC)
2089 dev_set_promiscuity(slave_dev, -1);
2091 /* unset allmulti level from slave */
2092 if (bond_dev->flags & IFF_ALLMULTI)
2093 dev_set_allmulti(slave_dev, -1);
2095 /* flush master's mc_list from slave */
2096 netif_addr_lock_bh(bond_dev);
2097 bond_mc_list_flush(bond_dev, slave_dev);
2098 netif_addr_unlock_bh(bond_dev);
2101 bond_upper_dev_unlink(bond_dev, slave_dev);
2103 slave_disable_netpoll(slave);
2105 /* close slave before restoring its mac address */
2106 dev_close(slave_dev);
2108 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2109 /* restore original ("permanent") mac address */
2110 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2111 addr.sa_family = slave_dev->type;
2112 dev_set_mac_address(slave_dev, &addr);
2115 dev_set_mtu(slave_dev, slave->original_mtu);
2117 slave_dev->priv_flags &= ~IFF_BONDING;
2121 return 0; /* deletion OK */
2124 /* A wrapper used because of ndo_del_link */
2125 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2127 return __bond_release_one(bond_dev, slave_dev, false);
2131 * First release a slave and then destroy the bond if no more slaves are left.
2132 * Must be under rtnl_lock when this function is called.
2134 static int bond_release_and_destroy(struct net_device *bond_dev,
2135 struct net_device *slave_dev)
2137 struct bonding *bond = netdev_priv(bond_dev);
2140 ret = bond_release(bond_dev, slave_dev);
2141 if ((ret == 0) && (bond->slave_cnt == 0)) {
2142 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2143 pr_info("%s: destroying bond %s.\n",
2144 bond_dev->name, bond_dev->name);
2145 unregister_netdevice(bond_dev);
2151 * This function changes the active slave to slave <slave_dev>.
2152 * It returns -EINVAL in the following cases.
2153 * - <slave_dev> is not found in the list.
2154 * - There is not active slave now.
2155 * - <slave_dev> is already active.
2156 * - The link state of <slave_dev> is not BOND_LINK_UP.
2157 * - <slave_dev> is not running.
2158 * In these cases, this function does nothing.
2159 * In the other cases, current_slave pointer is changed and 0 is returned.
2161 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2163 struct bonding *bond = netdev_priv(bond_dev);
2164 struct slave *old_active = NULL;
2165 struct slave *new_active = NULL;
2168 if (!USES_PRIMARY(bond->params.mode))
2171 /* Verify that bond_dev is indeed the master of slave_dev */
2172 if (!(slave_dev->flags & IFF_SLAVE) ||
2173 !netdev_has_upper_dev(slave_dev, bond_dev))
2176 read_lock(&bond->lock);
2178 read_lock(&bond->curr_slave_lock);
2179 old_active = bond->curr_active_slave;
2180 read_unlock(&bond->curr_slave_lock);
2182 new_active = bond_get_slave_by_dev(bond, slave_dev);
2185 * Changing to the current active: do nothing; return success.
2187 if (new_active && (new_active == old_active)) {
2188 read_unlock(&bond->lock);
2194 (new_active->link == BOND_LINK_UP) &&
2195 IS_UP(new_active->dev)) {
2197 write_lock_bh(&bond->curr_slave_lock);
2198 bond_change_active_slave(bond, new_active);
2199 write_unlock_bh(&bond->curr_slave_lock);
2200 unblock_netpoll_tx();
2204 read_unlock(&bond->lock);
2209 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2211 struct bonding *bond = netdev_priv(bond_dev);
2213 info->bond_mode = bond->params.mode;
2214 info->miimon = bond->params.miimon;
2216 read_lock(&bond->lock);
2217 info->num_slaves = bond->slave_cnt;
2218 read_unlock(&bond->lock);
2223 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2225 struct bonding *bond = netdev_priv(bond_dev);
2226 struct slave *slave;
2227 int i, res = -ENODEV;
2229 read_lock(&bond->lock);
2231 bond_for_each_slave(bond, slave, i) {
2232 if (i == (int)info->slave_id) {
2234 strcpy(info->slave_name, slave->dev->name);
2235 info->link = slave->link;
2236 info->state = bond_slave_state(slave);
2237 info->link_failure_count = slave->link_failure_count;
2242 read_unlock(&bond->lock);
2247 /*-------------------------------- Monitoring -------------------------------*/
2250 static int bond_miimon_inspect(struct bonding *bond)
2252 struct slave *slave;
2253 int i, link_state, commit = 0;
2254 bool ignore_updelay;
2256 ignore_updelay = !bond->curr_active_slave ? true : false;
2258 bond_for_each_slave(bond, slave, i) {
2259 slave->new_link = BOND_LINK_NOCHANGE;
2261 link_state = bond_check_dev_link(bond, slave->dev, 0);
2263 switch (slave->link) {
2268 slave->link = BOND_LINK_FAIL;
2269 slave->delay = bond->params.downdelay;
2271 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2273 (bond->params.mode ==
2274 BOND_MODE_ACTIVEBACKUP) ?
2275 (bond_is_active_slave(slave) ?
2276 "active " : "backup ") : "",
2278 bond->params.downdelay * bond->params.miimon);
2281 case BOND_LINK_FAIL:
2284 * recovered before downdelay expired
2286 slave->link = BOND_LINK_UP;
2287 slave->jiffies = jiffies;
2288 pr_info("%s: link status up again after %d ms for interface %s.\n",
2290 (bond->params.downdelay - slave->delay) *
2291 bond->params.miimon,
2296 if (slave->delay <= 0) {
2297 slave->new_link = BOND_LINK_DOWN;
2305 case BOND_LINK_DOWN:
2309 slave->link = BOND_LINK_BACK;
2310 slave->delay = bond->params.updelay;
2313 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2314 bond->dev->name, slave->dev->name,
2315 ignore_updelay ? 0 :
2316 bond->params.updelay *
2317 bond->params.miimon);
2320 case BOND_LINK_BACK:
2322 slave->link = BOND_LINK_DOWN;
2323 pr_info("%s: link status down again after %d ms for interface %s.\n",
2325 (bond->params.updelay - slave->delay) *
2326 bond->params.miimon,
2335 if (slave->delay <= 0) {
2336 slave->new_link = BOND_LINK_UP;
2338 ignore_updelay = false;
2350 static void bond_miimon_commit(struct bonding *bond)
2352 struct slave *slave;
2355 bond_for_each_slave(bond, slave, i) {
2356 switch (slave->new_link) {
2357 case BOND_LINK_NOCHANGE:
2361 slave->link = BOND_LINK_UP;
2362 slave->jiffies = jiffies;
2364 if (bond->params.mode == BOND_MODE_8023AD) {
2365 /* prevent it from being the active one */
2366 bond_set_backup_slave(slave);
2367 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2368 /* make it immediately active */
2369 bond_set_active_slave(slave);
2370 } else if (slave != bond->primary_slave) {
2371 /* prevent it from being the active one */
2372 bond_set_backup_slave(slave);
2375 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2376 bond->dev->name, slave->dev->name,
2377 slave->speed, slave->duplex ? "full" : "half");
2379 /* notify ad that the link status has changed */
2380 if (bond->params.mode == BOND_MODE_8023AD)
2381 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2383 if (bond_is_lb(bond))
2384 bond_alb_handle_link_change(bond, slave,
2387 if (!bond->curr_active_slave ||
2388 (slave == bond->primary_slave))
2393 case BOND_LINK_DOWN:
2394 if (slave->link_failure_count < UINT_MAX)
2395 slave->link_failure_count++;
2397 slave->link = BOND_LINK_DOWN;
2399 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2400 bond->params.mode == BOND_MODE_8023AD)
2401 bond_set_slave_inactive_flags(slave);
2403 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2404 bond->dev->name, slave->dev->name);
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave,
2410 if (bond_is_lb(bond))
2411 bond_alb_handle_link_change(bond, slave,
2414 if (slave == bond->curr_active_slave)
2420 pr_err("%s: invalid new link %d on slave %s\n",
2421 bond->dev->name, slave->new_link,
2423 slave->new_link = BOND_LINK_NOCHANGE;
2431 write_lock_bh(&bond->curr_slave_lock);
2432 bond_select_active_slave(bond);
2433 write_unlock_bh(&bond->curr_slave_lock);
2434 unblock_netpoll_tx();
2437 bond_set_carrier(bond);
2443 * Really a wrapper that splits the mii monitor into two phases: an
2444 * inspection, then (if inspection indicates something needs to be done)
2445 * an acquisition of appropriate locks followed by a commit phase to
2446 * implement whatever link state changes are indicated.
2448 void bond_mii_monitor(struct work_struct *work)
2450 struct bonding *bond = container_of(work, struct bonding,
2452 bool should_notify_peers = false;
2453 unsigned long delay;
2455 read_lock(&bond->lock);
2457 delay = msecs_to_jiffies(bond->params.miimon);
2459 if (bond->slave_cnt == 0)
2462 should_notify_peers = bond_should_notify_peers(bond);
2464 if (bond_miimon_inspect(bond)) {
2465 read_unlock(&bond->lock);
2467 /* Race avoidance with bond_close cancel of workqueue */
2468 if (!rtnl_trylock()) {
2469 read_lock(&bond->lock);
2471 should_notify_peers = false;
2475 read_lock(&bond->lock);
2477 bond_miimon_commit(bond);
2479 read_unlock(&bond->lock);
2480 rtnl_unlock(); /* might sleep, hold no other locks */
2481 read_lock(&bond->lock);
2485 if (bond->params.miimon)
2486 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2488 read_unlock(&bond->lock);
2490 if (should_notify_peers) {
2491 if (!rtnl_trylock()) {
2492 read_lock(&bond->lock);
2493 bond->send_peer_notif++;
2494 read_unlock(&bond->lock);
2497 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2502 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2504 struct vlan_entry *vlan;
2505 struct net_device *vlan_dev;
2507 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2510 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2512 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2514 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2522 * We go to the (large) trouble of VLAN tagging ARP frames because
2523 * switches in VLAN mode (especially if ports are configured as
2524 * "native" to a VLAN) might not pass non-tagged frames.
2526 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2528 struct sk_buff *skb;
2530 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2531 slave_dev->name, dest_ip, src_ip, vlan_id);
2533 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2534 NULL, slave_dev->dev_addr, NULL);
2537 pr_err("ARP packet allocation failed\n");
2541 skb = vlan_put_tag(skb, vlan_id);
2543 pr_err("failed to insert VLAN tag\n");
2551 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2554 __be32 *targets = bond->params.arp_targets;
2555 struct vlan_entry *vlan;
2556 struct net_device *vlan_dev = NULL;
2559 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2563 pr_debug("basa: target %x\n", targets[i]);
2564 if (!bond_vlan_used(bond)) {
2565 pr_debug("basa: empty vlan: arp_send\n");
2566 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2567 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2573 * If VLANs are configured, we do a route lookup to
2574 * determine which VLAN interface would be used, so we
2575 * can tag the ARP with the proper VLAN tag.
2577 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2580 if (net_ratelimit()) {
2581 pr_warning("%s: no route to arp_ip_target %pI4\n",
2582 bond->dev->name, &targets[i]);
2588 * This target is not on a VLAN
2590 if (rt->dst.dev == bond->dev) {
2592 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2593 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2594 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2600 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2602 vlan_dev = __vlan_find_dev_deep(bond->dev,
2605 if (vlan_dev == rt->dst.dev) {
2606 vlan_id = vlan->vlan_id;
2607 pr_debug("basa: vlan match on %s %d\n",
2608 vlan_dev->name, vlan_id);
2613 if (vlan_id && vlan_dev) {
2615 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2616 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2621 if (net_ratelimit()) {
2622 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2623 bond->dev->name, &targets[i],
2624 rt->dst.dev ? rt->dst.dev->name : "NULL");
2630 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2633 __be32 *targets = bond->params.arp_targets;
2635 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2636 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2637 &sip, &tip, i, &targets[i],
2638 bond_has_this_ip(bond, tip));
2639 if (sip == targets[i]) {
2640 if (bond_has_this_ip(bond, tip))
2641 slave->last_arp_rx = jiffies;
2647 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2648 struct slave *slave)
2650 struct arphdr *arp = (struct arphdr *)skb->data;
2651 unsigned char *arp_ptr;
2655 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2656 return RX_HANDLER_ANOTHER;
2658 read_lock(&bond->lock);
2659 alen = arp_hdr_len(bond->dev);
2661 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2662 bond->dev->name, skb->dev->name);
2664 if (alen > skb_headlen(skb)) {
2665 arp = kmalloc(alen, GFP_ATOMIC);
2668 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2672 if (arp->ar_hln != bond->dev->addr_len ||
2673 skb->pkt_type == PACKET_OTHERHOST ||
2674 skb->pkt_type == PACKET_LOOPBACK ||
2675 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2676 arp->ar_pro != htons(ETH_P_IP) ||
2680 arp_ptr = (unsigned char *)(arp + 1);
2681 arp_ptr += bond->dev->addr_len;
2682 memcpy(&sip, arp_ptr, 4);
2683 arp_ptr += 4 + bond->dev->addr_len;
2684 memcpy(&tip, arp_ptr, 4);
2686 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2687 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2688 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2692 * Backup slaves won't see the ARP reply, but do come through
2693 * here for each ARP probe (so we swap the sip/tip to validate
2694 * the probe). In a "redundant switch, common router" type of
2695 * configuration, the ARP probe will (hopefully) travel from
2696 * the active, through one switch, the router, then the other
2697 * switch before reaching the backup.
2699 if (bond_is_active_slave(slave))
2700 bond_validate_arp(bond, slave, sip, tip);
2702 bond_validate_arp(bond, slave, tip, sip);
2705 read_unlock(&bond->lock);
2706 if (arp != (struct arphdr *)skb->data)
2708 return RX_HANDLER_ANOTHER;
2712 * this function is called regularly to monitor each slave's link
2713 * ensuring that traffic is being sent and received when arp monitoring
2714 * is used in load-balancing mode. if the adapter has been dormant, then an
2715 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2716 * arp monitoring in active backup mode.
2718 void bond_loadbalance_arp_mon(struct work_struct *work)
2720 struct bonding *bond = container_of(work, struct bonding,
2722 struct slave *slave, *oldcurrent;
2723 int do_failover = 0;
2724 int delta_in_ticks, extra_ticks;
2727 read_lock(&bond->lock);
2729 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2730 extra_ticks = delta_in_ticks / 2;
2732 if (bond->slave_cnt == 0)
2735 read_lock(&bond->curr_slave_lock);
2736 oldcurrent = bond->curr_active_slave;
2737 read_unlock(&bond->curr_slave_lock);
2739 /* see if any of the previous devices are up now (i.e. they have
2740 * xmt and rcv traffic). the curr_active_slave does not come into
2741 * the picture unless it is null. also, slave->jiffies is not needed
2742 * here because we send an arp on each slave and give a slave as
2743 * long as it needs to get the tx/rx within the delta.
2744 * TODO: what about up/down delay in arp mode? it wasn't here before
2747 bond_for_each_slave(bond, slave, i) {
2748 unsigned long trans_start = dev_trans_start(slave->dev);
2750 if (slave->link != BOND_LINK_UP) {
2751 if (time_in_range(jiffies,
2752 trans_start - delta_in_ticks,
2753 trans_start + delta_in_ticks + extra_ticks) &&
2754 time_in_range(jiffies,
2755 slave->dev->last_rx - delta_in_ticks,
2756 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2758 slave->link = BOND_LINK_UP;
2759 bond_set_active_slave(slave);
2761 /* primary_slave has no meaning in round-robin
2762 * mode. the window of a slave being up and
2763 * curr_active_slave being null after enslaving
2767 pr_info("%s: link status definitely up for interface %s, ",
2772 pr_info("%s: interface %s is now up\n",
2778 /* slave->link == BOND_LINK_UP */
2780 /* not all switches will respond to an arp request
2781 * when the source ip is 0, so don't take the link down
2782 * if we don't know our ip yet
2784 if (!time_in_range(jiffies,
2785 trans_start - delta_in_ticks,
2786 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2787 !time_in_range(jiffies,
2788 slave->dev->last_rx - delta_in_ticks,
2789 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2791 slave->link = BOND_LINK_DOWN;
2792 bond_set_backup_slave(slave);
2794 if (slave->link_failure_count < UINT_MAX)
2795 slave->link_failure_count++;
2797 pr_info("%s: interface %s is now down.\n",
2801 if (slave == oldcurrent)
2806 /* note: if switch is in round-robin mode, all links
2807 * must tx arp to ensure all links rx an arp - otherwise
2808 * links may oscillate or not come up at all; if switch is
2809 * in something like xor mode, there is nothing we can
2810 * do - all replies will be rx'ed on same link causing slaves
2811 * to be unstable during low/no traffic periods
2813 if (IS_UP(slave->dev))
2814 bond_arp_send_all(bond, slave);
2819 write_lock_bh(&bond->curr_slave_lock);
2821 bond_select_active_slave(bond);
2823 write_unlock_bh(&bond->curr_slave_lock);
2824 unblock_netpoll_tx();
2828 if (bond->params.arp_interval)
2829 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2831 read_unlock(&bond->lock);
2835 * Called to inspect slaves for active-backup mode ARP monitor link state
2836 * changes. Sets new_link in slaves to specify what action should take
2837 * place for the slave. Returns 0 if no changes are found, >0 if changes
2838 * to link states must be committed.
2840 * Called with bond->lock held for read.
2842 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2844 struct slave *slave;
2846 unsigned long trans_start;
2849 /* All the time comparisons below need some extra time. Otherwise, on
2850 * fast networks the ARP probe/reply may arrive within the same jiffy
2851 * as it was sent. Then, the next time the ARP monitor is run, one
2852 * arp_interval will already have passed in the comparisons.
2854 extra_ticks = delta_in_ticks / 2;
2856 bond_for_each_slave(bond, slave, i) {
2857 slave->new_link = BOND_LINK_NOCHANGE;
2859 if (slave->link != BOND_LINK_UP) {
2860 if (time_in_range(jiffies,
2861 slave_last_rx(bond, slave) - delta_in_ticks,
2862 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2864 slave->new_link = BOND_LINK_UP;
2872 * Give slaves 2*delta after being enslaved or made
2873 * active. This avoids bouncing, as the last receive
2874 * times need a full ARP monitor cycle to be updated.
2876 if (time_in_range(jiffies,
2877 slave->jiffies - delta_in_ticks,
2878 slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2882 * Backup slave is down if:
2883 * - No current_arp_slave AND
2884 * - more than 3*delta since last receive AND
2885 * - the bond has an IP address
2887 * Note: a non-null current_arp_slave indicates
2888 * the curr_active_slave went down and we are
2889 * searching for a new one; under this condition
2890 * we only take the curr_active_slave down - this
2891 * gives each slave a chance to tx/rx traffic
2892 * before being taken out
2894 if (!bond_is_active_slave(slave) &&
2895 !bond->current_arp_slave &&
2896 !time_in_range(jiffies,
2897 slave_last_rx(bond, slave) - delta_in_ticks,
2898 slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2900 slave->new_link = BOND_LINK_DOWN;
2905 * Active slave is down if:
2906 * - more than 2*delta since transmitting OR
2907 * - (more than 2*delta since receive AND
2908 * the bond has an IP address)
2910 trans_start = dev_trans_start(slave->dev);
2911 if (bond_is_active_slave(slave) &&
2912 (!time_in_range(jiffies,
2913 trans_start - delta_in_ticks,
2914 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2915 !time_in_range(jiffies,
2916 slave_last_rx(bond, slave) - delta_in_ticks,
2917 slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2919 slave->new_link = BOND_LINK_DOWN;
2928 * Called to commit link state changes noted by inspection step of
2929 * active-backup mode ARP monitor.
2931 * Called with RTNL and bond->lock for read.
2933 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2935 struct slave *slave;
2937 unsigned long trans_start;
2939 bond_for_each_slave(bond, slave, i) {
2940 switch (slave->new_link) {
2941 case BOND_LINK_NOCHANGE:
2945 trans_start = dev_trans_start(slave->dev);
2946 if ((!bond->curr_active_slave &&
2947 time_in_range(jiffies,
2948 trans_start - delta_in_ticks,
2949 trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2950 bond->curr_active_slave != slave) {
2951 slave->link = BOND_LINK_UP;
2952 if (bond->current_arp_slave) {
2953 bond_set_slave_inactive_flags(
2954 bond->current_arp_slave);
2955 bond->current_arp_slave = NULL;
2958 pr_info("%s: link status definitely up for interface %s.\n",
2959 bond->dev->name, slave->dev->name);
2961 if (!bond->curr_active_slave ||
2962 (slave == bond->primary_slave))
2969 case BOND_LINK_DOWN:
2970 if (slave->link_failure_count < UINT_MAX)
2971 slave->link_failure_count++;
2973 slave->link = BOND_LINK_DOWN;
2974 bond_set_slave_inactive_flags(slave);
2976 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2977 bond->dev->name, slave->dev->name);
2979 if (slave == bond->curr_active_slave) {
2980 bond->current_arp_slave = NULL;
2987 pr_err("%s: impossible: new_link %d on slave %s\n",
2988 bond->dev->name, slave->new_link,
2996 write_lock_bh(&bond->curr_slave_lock);
2997 bond_select_active_slave(bond);
2998 write_unlock_bh(&bond->curr_slave_lock);
2999 unblock_netpoll_tx();
3002 bond_set_carrier(bond);
3006 * Send ARP probes for active-backup mode ARP monitor.
3008 * Called with bond->lock held for read.
3010 static void bond_ab_arp_probe(struct bonding *bond)
3012 struct slave *slave;
3015 read_lock(&bond->curr_slave_lock);
3017 if (bond->current_arp_slave && bond->curr_active_slave)
3018 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3019 bond->current_arp_slave->dev->name,
3020 bond->curr_active_slave->dev->name);
3022 if (bond->curr_active_slave) {
3023 bond_arp_send_all(bond, bond->curr_active_slave);
3024 read_unlock(&bond->curr_slave_lock);
3028 read_unlock(&bond->curr_slave_lock);
3030 /* if we don't have a curr_active_slave, search for the next available
3031 * backup slave from the current_arp_slave and make it the candidate
3032 * for becoming the curr_active_slave
3035 if (!bond->current_arp_slave) {
3036 bond->current_arp_slave = bond->first_slave;
3037 if (!bond->current_arp_slave)
3041 bond_set_slave_inactive_flags(bond->current_arp_slave);
3043 /* search for next candidate */
3044 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3045 if (IS_UP(slave->dev)) {
3046 slave->link = BOND_LINK_BACK;
3047 bond_set_slave_active_flags(slave);
3048 bond_arp_send_all(bond, slave);
3049 slave->jiffies = jiffies;
3050 bond->current_arp_slave = slave;
3054 /* if the link state is up at this point, we
3055 * mark it down - this can happen if we have
3056 * simultaneous link failures and
3057 * reselect_active_interface doesn't make this
3058 * one the current slave so it is still marked
3059 * up when it is actually down
3061 if (slave->link == BOND_LINK_UP) {
3062 slave->link = BOND_LINK_DOWN;
3063 if (slave->link_failure_count < UINT_MAX)
3064 slave->link_failure_count++;
3066 bond_set_slave_inactive_flags(slave);
3068 pr_info("%s: backup interface %s is now down.\n",
3069 bond->dev->name, slave->dev->name);
3074 void bond_activebackup_arp_mon(struct work_struct *work)
3076 struct bonding *bond = container_of(work, struct bonding,
3078 bool should_notify_peers = false;
3081 read_lock(&bond->lock);
3083 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3085 if (bond->slave_cnt == 0)
3088 should_notify_peers = bond_should_notify_peers(bond);
3090 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3091 read_unlock(&bond->lock);
3093 /* Race avoidance with bond_close flush of workqueue */
3094 if (!rtnl_trylock()) {
3095 read_lock(&bond->lock);
3097 should_notify_peers = false;
3101 read_lock(&bond->lock);
3103 bond_ab_arp_commit(bond, delta_in_ticks);
3105 read_unlock(&bond->lock);
3107 read_lock(&bond->lock);
3110 bond_ab_arp_probe(bond);
3113 if (bond->params.arp_interval)
3114 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3116 read_unlock(&bond->lock);
3118 if (should_notify_peers) {
3119 if (!rtnl_trylock()) {
3120 read_lock(&bond->lock);
3121 bond->send_peer_notif++;
3122 read_unlock(&bond->lock);
3125 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3130 /*-------------------------- netdev event handling --------------------------*/
3133 * Change device name
3135 static int bond_event_changename(struct bonding *bond)
3137 bond_remove_proc_entry(bond);
3138 bond_create_proc_entry(bond);
3140 bond_debug_reregister(bond);
3145 static int bond_master_netdev_event(unsigned long event,
3146 struct net_device *bond_dev)
3148 struct bonding *event_bond = netdev_priv(bond_dev);
3151 case NETDEV_CHANGENAME:
3152 return bond_event_changename(event_bond);
3153 case NETDEV_UNREGISTER:
3154 bond_remove_proc_entry(event_bond);
3156 case NETDEV_REGISTER:
3157 bond_create_proc_entry(event_bond);
3166 static int bond_slave_netdev_event(unsigned long event,
3167 struct net_device *slave_dev)
3169 struct slave *slave = bond_slave_get_rtnl(slave_dev);
3170 struct bonding *bond = slave->bond;
3171 struct net_device *bond_dev = slave->bond->dev;
3176 case NETDEV_UNREGISTER:
3177 if (bond->setup_by_slave)
3178 bond_release_and_destroy(bond_dev, slave_dev);
3180 bond_release(bond_dev, slave_dev);
3184 old_speed = slave->speed;
3185 old_duplex = slave->duplex;
3187 bond_update_speed_duplex(slave);
3189 if (bond->params.mode == BOND_MODE_8023AD) {
3190 if (old_speed != slave->speed)
3191 bond_3ad_adapter_speed_changed(slave);
3192 if (old_duplex != slave->duplex)
3193 bond_3ad_adapter_duplex_changed(slave);
3198 * ... Or is it this?
3201 case NETDEV_CHANGEMTU:
3203 * TODO: Should slaves be allowed to
3204 * independently alter their MTU? For
3205 * an active-backup bond, slaves need
3206 * not be the same type of device, so
3207 * MTUs may vary. For other modes,
3208 * slaves arguably should have the
3209 * same MTUs. To do this, we'd need to
3210 * take over the slave's change_mtu
3211 * function for the duration of their
3215 case NETDEV_CHANGENAME:
3217 * TODO: handle changing the primary's name
3220 case NETDEV_FEAT_CHANGE:
3221 bond_compute_features(bond);
3231 * bond_netdev_event: handle netdev notifier chain events.
3233 * This function receives events for the netdev chain. The caller (an
3234 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3235 * locks for us to safely manipulate the slave devices (RTNL lock,
3238 static int bond_netdev_event(struct notifier_block *this,
3239 unsigned long event, void *ptr)
3241 struct net_device *event_dev = (struct net_device *)ptr;
3243 pr_debug("event_dev: %s, event: %lx\n",
3244 event_dev ? event_dev->name : "None",
3247 if (!(event_dev->priv_flags & IFF_BONDING))
3250 if (event_dev->flags & IFF_MASTER) {
3251 pr_debug("IFF_MASTER\n");
3252 return bond_master_netdev_event(event, event_dev);
3255 if (event_dev->flags & IFF_SLAVE) {
3256 pr_debug("IFF_SLAVE\n");
3257 return bond_slave_netdev_event(event, event_dev);
3263 static struct notifier_block bond_netdev_notifier = {
3264 .notifier_call = bond_netdev_event,
3267 /*---------------------------- Hashing Policies -----------------------------*/
3270 * Hash for the output device based upon layer 2 data
3272 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3274 struct ethhdr *data = (struct ethhdr *)skb->data;
3276 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3277 return (data->h_dest[5] ^ data->h_source[5]) % count;
3283 * Hash for the output device based upon layer 2 and layer 3 data. If
3284 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3286 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3288 struct ethhdr *data = (struct ethhdr *)skb->data;
3290 struct ipv6hdr *ipv6h;
3294 if (skb->protocol == htons(ETH_P_IP) &&
3295 skb_network_header_len(skb) >= sizeof(*iph)) {
3297 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3298 (data->h_dest[5] ^ data->h_source[5])) % count;
3299 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3300 skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3301 ipv6h = ipv6_hdr(skb);
3302 s = &ipv6h->saddr.s6_addr32[0];
3303 d = &ipv6h->daddr.s6_addr32[0];
3304 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3305 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3306 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3309 return bond_xmit_hash_policy_l2(skb, count);
3313 * Hash for the output device based upon layer 3 and layer 4 data. If
3314 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3315 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3317 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3321 struct ipv6hdr *ipv6h;
3325 if (skb->protocol == htons(ETH_P_IP) &&
3326 skb_network_header_len(skb) >= sizeof(*iph)) {
3328 if (!ip_is_fragment(iph) &&
3329 (iph->protocol == IPPROTO_TCP ||
3330 iph->protocol == IPPROTO_UDP) &&
3331 (skb_headlen(skb) - skb_network_offset(skb) >=
3332 iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3333 layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3334 layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3336 return (layer4_xor ^
3337 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3338 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3339 skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3340 ipv6h = ipv6_hdr(skb);
3341 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3342 ipv6h->nexthdr == IPPROTO_UDP) &&
3343 (skb_headlen(skb) - skb_network_offset(skb) >=
3344 sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3345 layer4hdr = (__be16 *)(ipv6h + 1);
3346 layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3348 s = &ipv6h->saddr.s6_addr32[0];
3349 d = &ipv6h->daddr.s6_addr32[0];
3350 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3351 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3353 return layer4_xor % count;
3356 return bond_xmit_hash_policy_l2(skb, count);
3359 /*-------------------------- Device entry points ----------------------------*/
3361 static void bond_work_init_all(struct bonding *bond)
3363 INIT_DELAYED_WORK(&bond->mcast_work,
3364 bond_resend_igmp_join_requests_delayed);
3365 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3366 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3367 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3368 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3370 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3371 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3374 static void bond_work_cancel_all(struct bonding *bond)
3376 cancel_delayed_work_sync(&bond->mii_work);
3377 cancel_delayed_work_sync(&bond->arp_work);
3378 cancel_delayed_work_sync(&bond->alb_work);
3379 cancel_delayed_work_sync(&bond->ad_work);
3380 cancel_delayed_work_sync(&bond->mcast_work);
3383 static int bond_open(struct net_device *bond_dev)
3385 struct bonding *bond = netdev_priv(bond_dev);
3386 struct slave *slave;
3389 /* reset slave->backup and slave->inactive */
3390 read_lock(&bond->lock);
3391 if (bond->slave_cnt > 0) {
3392 read_lock(&bond->curr_slave_lock);
3393 bond_for_each_slave(bond, slave, i) {
3394 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3395 && (slave != bond->curr_active_slave)) {
3396 bond_set_slave_inactive_flags(slave);
3398 bond_set_slave_active_flags(slave);
3401 read_unlock(&bond->curr_slave_lock);
3403 read_unlock(&bond->lock);
3405 bond_work_init_all(bond);
3407 if (bond_is_lb(bond)) {
3408 /* bond_alb_initialize must be called before the timer
3411 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3413 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3416 if (bond->params.miimon) /* link check interval, in milliseconds. */
3417 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3419 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3420 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3421 if (bond->params.arp_validate)
3422 bond->recv_probe = bond_arp_rcv;
3425 if (bond->params.mode == BOND_MODE_8023AD) {
3426 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3427 /* register to receive LACPDUs */
3428 bond->recv_probe = bond_3ad_lacpdu_recv;
3429 bond_3ad_initiate_agg_selection(bond, 1);
3435 static int bond_close(struct net_device *bond_dev)
3437 struct bonding *bond = netdev_priv(bond_dev);
3439 write_lock_bh(&bond->lock);
3440 bond->send_peer_notif = 0;
3441 write_unlock_bh(&bond->lock);
3443 bond_work_cancel_all(bond);
3444 if (bond_is_lb(bond)) {
3445 /* Must be called only after all
3446 * slaves have been released
3448 bond_alb_deinitialize(bond);
3450 bond->recv_probe = NULL;
3455 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3456 struct rtnl_link_stats64 *stats)
3458 struct bonding *bond = netdev_priv(bond_dev);
3459 struct rtnl_link_stats64 temp;
3460 struct slave *slave;
3463 memset(stats, 0, sizeof(*stats));
3465 read_lock_bh(&bond->lock);
3467 bond_for_each_slave(bond, slave, i) {
3468 const struct rtnl_link_stats64 *sstats =
3469 dev_get_stats(slave->dev, &temp);
3471 stats->rx_packets += sstats->rx_packets;
3472 stats->rx_bytes += sstats->rx_bytes;
3473 stats->rx_errors += sstats->rx_errors;
3474 stats->rx_dropped += sstats->rx_dropped;
3476 stats->tx_packets += sstats->tx_packets;
3477 stats->tx_bytes += sstats->tx_bytes;
3478 stats->tx_errors += sstats->tx_errors;
3479 stats->tx_dropped += sstats->tx_dropped;
3481 stats->multicast += sstats->multicast;
3482 stats->collisions += sstats->collisions;
3484 stats->rx_length_errors += sstats->rx_length_errors;
3485 stats->rx_over_errors += sstats->rx_over_errors;
3486 stats->rx_crc_errors += sstats->rx_crc_errors;
3487 stats->rx_frame_errors += sstats->rx_frame_errors;
3488 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3489 stats->rx_missed_errors += sstats->rx_missed_errors;
3491 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3492 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3493 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3494 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3495 stats->tx_window_errors += sstats->tx_window_errors;
3498 read_unlock_bh(&bond->lock);
3503 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3505 struct net_device *slave_dev = NULL;
3506 struct ifbond k_binfo;
3507 struct ifbond __user *u_binfo = NULL;
3508 struct ifslave k_sinfo;
3509 struct ifslave __user *u_sinfo = NULL;
3510 struct mii_ioctl_data *mii = NULL;
3514 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3526 * We do this again just in case we were called by SIOCGMIIREG
3527 * instead of SIOCGMIIPHY.
3534 if (mii->reg_num == 1) {
3535 struct bonding *bond = netdev_priv(bond_dev);
3537 read_lock(&bond->lock);
3538 read_lock(&bond->curr_slave_lock);
3539 if (netif_carrier_ok(bond->dev))
3540 mii->val_out = BMSR_LSTATUS;
3542 read_unlock(&bond->curr_slave_lock);
3543 read_unlock(&bond->lock);
3547 case BOND_INFO_QUERY_OLD:
3548 case SIOCBONDINFOQUERY:
3549 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3551 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3554 res = bond_info_query(bond_dev, &k_binfo);
3556 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3560 case BOND_SLAVE_INFO_QUERY_OLD:
3561 case SIOCBONDSLAVEINFOQUERY:
3562 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3564 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3567 res = bond_slave_info_query(bond_dev, &k_sinfo);
3569 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3578 net = dev_net(bond_dev);
3580 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3583 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3585 pr_debug("slave_dev=%p:\n", slave_dev);
3590 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3592 case BOND_ENSLAVE_OLD:
3593 case SIOCBONDENSLAVE:
3594 res = bond_enslave(bond_dev, slave_dev);
3596 case BOND_RELEASE_OLD:
3597 case SIOCBONDRELEASE:
3598 res = bond_release(bond_dev, slave_dev);
3600 case BOND_SETHWADDR_OLD:
3601 case SIOCBONDSETHWADDR:
3602 bond_set_dev_addr(bond_dev, slave_dev);
3605 case BOND_CHANGE_ACTIVE_OLD:
3606 case SIOCBONDCHANGEACTIVE:
3607 res = bond_ioctl_change_active(bond_dev, slave_dev);
3619 static bool bond_addr_in_mc_list(unsigned char *addr,
3620 struct netdev_hw_addr_list *list,
3623 struct netdev_hw_addr *ha;
3625 netdev_hw_addr_list_for_each(ha, list)
3626 if (!memcmp(ha->addr, addr, addrlen))
3632 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3634 struct bonding *bond = netdev_priv(bond_dev);
3636 if (change & IFF_PROMISC)
3637 bond_set_promiscuity(bond,
3638 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3640 if (change & IFF_ALLMULTI)
3641 bond_set_allmulti(bond,
3642 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3645 static void bond_set_multicast_list(struct net_device *bond_dev)
3647 struct bonding *bond = netdev_priv(bond_dev);
3648 struct netdev_hw_addr *ha;
3651 read_lock(&bond->lock);
3653 /* looking for addresses to add to slaves' mc list */
3654 netdev_for_each_mc_addr(ha, bond_dev) {
3655 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3656 bond_dev->addr_len);
3658 bond_mc_add(bond, ha->addr);
3661 /* looking for addresses to delete from slaves' list */
3662 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3663 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3664 bond_dev->addr_len);
3666 bond_mc_del(bond, ha->addr);
3669 /* save master's multicast list */
3670 __hw_addr_flush(&bond->mc_list);
3671 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3672 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3674 read_unlock(&bond->lock);
3677 static int bond_neigh_init(struct neighbour *n)
3679 struct bonding *bond = netdev_priv(n->dev);
3680 struct slave *slave = bond->first_slave;
3681 const struct net_device_ops *slave_ops;
3682 struct neigh_parms parms;
3688 slave_ops = slave->dev->netdev_ops;
3690 if (!slave_ops->ndo_neigh_setup)
3693 parms.neigh_setup = NULL;
3694 parms.neigh_cleanup = NULL;
3695 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3700 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3701 * after the last slave has been detached. Assumes that all slaves
3702 * utilize the same neigh_cleanup (true at this writing as only user
3705 n->parms->neigh_cleanup = parms.neigh_cleanup;
3707 if (!parms.neigh_setup)
3710 return parms.neigh_setup(n);
3714 * The bonding ndo_neigh_setup is called at init time beofre any
3715 * slave exists. So we must declare proxy setup function which will
3716 * be used at run time to resolve the actual slave neigh param setup.
3718 static int bond_neigh_setup(struct net_device *dev,
3719 struct neigh_parms *parms)
3721 parms->neigh_setup = bond_neigh_init;
3727 * Change the MTU of all of a master's slaves to match the master
3729 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3731 struct bonding *bond = netdev_priv(bond_dev);
3732 struct slave *slave, *stop_at;
3736 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3737 (bond_dev ? bond_dev->name : "None"), new_mtu);
3739 /* Can't hold bond->lock with bh disabled here since
3740 * some base drivers panic. On the other hand we can't
3741 * hold bond->lock without bh disabled because we'll
3742 * deadlock. The only solution is to rely on the fact
3743 * that we're under rtnl_lock here, and the slaves
3744 * list won't change. This doesn't solve the problem
3745 * of setting the slave's MTU while it is
3746 * transmitting, but the assumption is that the base
3747 * driver can handle that.
3749 * TODO: figure out a way to safely iterate the slaves
3750 * list, but without holding a lock around the actual
3751 * call to the base driver.
3754 bond_for_each_slave(bond, slave, i) {
3755 pr_debug("s %p s->p %p c_m %p\n",
3758 slave->dev->netdev_ops->ndo_change_mtu);
3760 res = dev_set_mtu(slave->dev, new_mtu);
3763 /* If we failed to set the slave's mtu to the new value
3764 * we must abort the operation even in ACTIVE_BACKUP
3765 * mode, because if we allow the backup slaves to have
3766 * different mtu values than the active slave we'll
3767 * need to change their mtu when doing a failover. That
3768 * means changing their mtu from timer context, which
3769 * is probably not a good idea.
3771 pr_debug("err %d %s\n", res, slave->dev->name);
3776 bond_dev->mtu = new_mtu;
3781 /* unwind from head to the slave that failed */
3783 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3786 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3788 pr_debug("unwind err %d dev %s\n",
3789 tmp_res, slave->dev->name);
3799 * Note that many devices must be down to change the HW address, and
3800 * downing the master releases all slaves. We can make bonds full of
3801 * bonding devices to test this, however.
3803 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3805 struct bonding *bond = netdev_priv(bond_dev);
3806 struct sockaddr *sa = addr, tmp_sa;
3807 struct slave *slave, *stop_at;
3811 if (bond->params.mode == BOND_MODE_ALB)
3812 return bond_alb_set_mac_address(bond_dev, addr);
3815 pr_debug("bond=%p, name=%s\n",
3816 bond, bond_dev ? bond_dev->name : "None");
3819 * If fail_over_mac is set to active, do nothing and return
3820 * success. Returning an error causes ifenslave to fail.
3822 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3825 if (!is_valid_ether_addr(sa->sa_data))
3826 return -EADDRNOTAVAIL;
3828 /* Can't hold bond->lock with bh disabled here since
3829 * some base drivers panic. On the other hand we can't
3830 * hold bond->lock without bh disabled because we'll
3831 * deadlock. The only solution is to rely on the fact
3832 * that we're under rtnl_lock here, and the slaves
3833 * list won't change. This doesn't solve the problem
3834 * of setting the slave's hw address while it is
3835 * transmitting, but the assumption is that the base
3836 * driver can handle that.
3838 * TODO: figure out a way to safely iterate the slaves
3839 * list, but without holding a lock around the actual
3840 * call to the base driver.
3843 bond_for_each_slave(bond, slave, i) {
3844 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3845 pr_debug("slave %p %s\n", slave, slave->dev->name);
3847 if (slave_ops->ndo_set_mac_address == NULL) {
3849 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3853 res = dev_set_mac_address(slave->dev, addr);
3855 /* TODO: consider downing the slave
3857 * User should expect communications
3858 * breakage anyway until ARP finish
3861 pr_debug("err %d %s\n", res, slave->dev->name);
3867 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3871 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3872 tmp_sa.sa_family = bond_dev->type;
3874 /* unwind from head to the slave that failed */
3876 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3879 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3881 pr_debug("unwind err %d dev %s\n",
3882 tmp_res, slave->dev->name);
3889 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3891 struct bonding *bond = netdev_priv(bond_dev);
3892 struct slave *slave, *start_at;
3893 int i, slave_no, res = 1;
3894 struct iphdr *iph = ip_hdr(skb);
3897 * Start with the curr_active_slave that joined the bond as the
3898 * default for sending IGMP traffic. For failover purposes one
3899 * needs to maintain some consistency for the interface that will
3900 * send the join/membership reports. The curr_active_slave found
3901 * will send all of this type of traffic.
3903 if ((iph->protocol == IPPROTO_IGMP) &&
3904 (skb->protocol == htons(ETH_P_IP))) {
3906 read_lock(&bond->curr_slave_lock);
3907 slave = bond->curr_active_slave;
3908 read_unlock(&bond->curr_slave_lock);
3914 * Concurrent TX may collide on rr_tx_counter; we accept
3915 * that as being rare enough not to justify using an
3918 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3920 bond_for_each_slave(bond, slave, i) {
3928 bond_for_each_slave_from(bond, slave, i, start_at) {
3929 if (IS_UP(slave->dev) &&
3930 (slave->link == BOND_LINK_UP) &&
3931 bond_is_active_slave(slave)) {
3932 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3939 /* no suitable interface, frame not sent */
3943 return NETDEV_TX_OK;
3948 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3949 * the bond has a usable interface.
3951 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3953 struct bonding *bond = netdev_priv(bond_dev);
3956 read_lock(&bond->curr_slave_lock);
3958 if (bond->curr_active_slave)
3959 res = bond_dev_queue_xmit(bond, skb,
3960 bond->curr_active_slave->dev);
3962 read_unlock(&bond->curr_slave_lock);
3965 /* no suitable interface, frame not sent */
3968 return NETDEV_TX_OK;
3972 * In bond_xmit_xor() , we determine the output device by using a pre-
3973 * determined xmit_hash_policy(), If the selected device is not enabled,
3974 * find the next active slave.
3976 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3978 struct bonding *bond = netdev_priv(bond_dev);
3979 struct slave *slave, *start_at;
3984 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3986 bond_for_each_slave(bond, slave, i) {
3994 bond_for_each_slave_from(bond, slave, i, start_at) {
3995 if (IS_UP(slave->dev) &&
3996 (slave->link == BOND_LINK_UP) &&
3997 bond_is_active_slave(slave)) {
3998 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4004 /* no suitable interface, frame not sent */
4008 return NETDEV_TX_OK;
4012 * in broadcast mode, we send everything to all usable interfaces.
4014 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4016 struct bonding *bond = netdev_priv(bond_dev);
4017 struct slave *slave, *start_at;
4018 struct net_device *tx_dev = NULL;
4022 read_lock(&bond->curr_slave_lock);
4023 start_at = bond->curr_active_slave;
4024 read_unlock(&bond->curr_slave_lock);
4029 bond_for_each_slave_from(bond, slave, i, start_at) {
4030 if (IS_UP(slave->dev) &&
4031 (slave->link == BOND_LINK_UP) &&
4032 bond_is_active_slave(slave)) {
4034 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4036 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4041 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4047 tx_dev = slave->dev;
4052 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4056 /* no suitable interface, frame not sent */
4059 /* frame sent to all suitable interfaces */
4060 return NETDEV_TX_OK;
4063 /*------------------------- Device initialization ---------------------------*/
4065 static void bond_set_xmit_hash_policy(struct bonding *bond)
4067 switch (bond->params.xmit_policy) {
4068 case BOND_XMIT_POLICY_LAYER23:
4069 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4071 case BOND_XMIT_POLICY_LAYER34:
4072 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4074 case BOND_XMIT_POLICY_LAYER2:
4076 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4082 * Lookup the slave that corresponds to a qid
4084 static inline int bond_slave_override(struct bonding *bond,
4085 struct sk_buff *skb)
4088 struct slave *slave = NULL;
4089 struct slave *check_slave;
4091 if (!skb->queue_mapping)
4094 /* Find out if any slaves have the same mapping as this skb. */
4095 bond_for_each_slave(bond, check_slave, i) {
4096 if (check_slave->queue_id == skb->queue_mapping) {
4097 slave = check_slave;
4102 /* If the slave isn't UP, use default transmit policy. */
4103 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4104 (slave->link == BOND_LINK_UP)) {
4105 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4112 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4115 * This helper function exists to help dev_pick_tx get the correct
4116 * destination queue. Using a helper function skips a call to
4117 * skb_tx_hash and will put the skbs in the queue we expect on their
4118 * way down to the bonding driver.
4120 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4123 * Save the original txq to restore before passing to the driver
4125 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4127 if (unlikely(txq >= dev->real_num_tx_queues)) {
4129 txq -= dev->real_num_tx_queues;
4130 } while (txq >= dev->real_num_tx_queues);
4135 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4137 struct bonding *bond = netdev_priv(dev);
4139 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4140 if (!bond_slave_override(bond, skb))
4141 return NETDEV_TX_OK;
4144 switch (bond->params.mode) {
4145 case BOND_MODE_ROUNDROBIN:
4146 return bond_xmit_roundrobin(skb, dev);
4147 case BOND_MODE_ACTIVEBACKUP:
4148 return bond_xmit_activebackup(skb, dev);
4150 return bond_xmit_xor(skb, dev);
4151 case BOND_MODE_BROADCAST:
4152 return bond_xmit_broadcast(skb, dev);
4153 case BOND_MODE_8023AD:
4154 return bond_3ad_xmit_xor(skb, dev);
4157 return bond_alb_xmit(skb, dev);
4159 /* Should never happen, mode already checked */
4160 pr_err("%s: Error: Unknown bonding mode %d\n",
4161 dev->name, bond->params.mode);
4164 return NETDEV_TX_OK;
4168 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4170 struct bonding *bond = netdev_priv(dev);
4171 netdev_tx_t ret = NETDEV_TX_OK;
4174 * If we risk deadlock from transmitting this in the
4175 * netpoll path, tell netpoll to queue the frame for later tx
4177 if (is_netpoll_tx_blocked(dev))
4178 return NETDEV_TX_BUSY;
4180 read_lock(&bond->lock);
4182 if (bond->slave_cnt)
4183 ret = __bond_start_xmit(skb, dev);
4187 read_unlock(&bond->lock);
4193 * set bond mode specific net device operations
4195 void bond_set_mode_ops(struct bonding *bond, int mode)
4197 struct net_device *bond_dev = bond->dev;
4200 case BOND_MODE_ROUNDROBIN:
4202 case BOND_MODE_ACTIVEBACKUP:
4205 bond_set_xmit_hash_policy(bond);
4207 case BOND_MODE_BROADCAST:
4209 case BOND_MODE_8023AD:
4210 bond_set_xmit_hash_policy(bond);
4217 /* Should never happen, mode already checked */
4218 pr_err("%s: Error: Unknown bonding mode %d\n",
4219 bond_dev->name, mode);
4224 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4225 struct ethtool_drvinfo *drvinfo)
4227 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4228 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4229 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4233 static const struct ethtool_ops bond_ethtool_ops = {
4234 .get_drvinfo = bond_ethtool_get_drvinfo,
4235 .get_link = ethtool_op_get_link,
4238 static const struct net_device_ops bond_netdev_ops = {
4239 .ndo_init = bond_init,
4240 .ndo_uninit = bond_uninit,
4241 .ndo_open = bond_open,
4242 .ndo_stop = bond_close,
4243 .ndo_start_xmit = bond_start_xmit,
4244 .ndo_select_queue = bond_select_queue,
4245 .ndo_get_stats64 = bond_get_stats,
4246 .ndo_do_ioctl = bond_do_ioctl,
4247 .ndo_change_rx_flags = bond_change_rx_flags,
4248 .ndo_set_rx_mode = bond_set_multicast_list,
4249 .ndo_change_mtu = bond_change_mtu,
4250 .ndo_set_mac_address = bond_set_mac_address,
4251 .ndo_neigh_setup = bond_neigh_setup,
4252 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4253 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4254 #ifdef CONFIG_NET_POLL_CONTROLLER
4255 .ndo_netpoll_setup = bond_netpoll_setup,
4256 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4257 .ndo_poll_controller = bond_poll_controller,
4259 .ndo_add_slave = bond_enslave,
4260 .ndo_del_slave = bond_release,
4261 .ndo_fix_features = bond_fix_features,
4264 static const struct device_type bond_type = {
4268 static void bond_destructor(struct net_device *bond_dev)
4270 struct bonding *bond = netdev_priv(bond_dev);
4272 destroy_workqueue(bond->wq);
4273 free_netdev(bond_dev);
4276 static void bond_setup(struct net_device *bond_dev)
4278 struct bonding *bond = netdev_priv(bond_dev);
4280 /* initialize rwlocks */
4281 rwlock_init(&bond->lock);
4282 rwlock_init(&bond->curr_slave_lock);
4284 bond->params = bonding_defaults;
4286 /* Initialize pointers */
4287 bond->dev = bond_dev;
4288 INIT_LIST_HEAD(&bond->vlan_list);
4290 /* Initialize the device entry points */
4291 ether_setup(bond_dev);
4292 bond_dev->netdev_ops = &bond_netdev_ops;
4293 bond_dev->ethtool_ops = &bond_ethtool_ops;
4294 bond_set_mode_ops(bond, bond->params.mode);
4296 bond_dev->destructor = bond_destructor;
4298 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4300 /* Initialize the device options */
4301 bond_dev->tx_queue_len = 0;
4302 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4303 bond_dev->priv_flags |= IFF_BONDING;
4304 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4306 /* At first, we block adding VLANs. That's the only way to
4307 * prevent problems that occur when adding VLANs over an
4308 * empty bond. The block will be removed once non-challenged
4309 * slaves are enslaved.
4311 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4313 /* don't acquire bond device's netif_tx_lock when
4315 bond_dev->features |= NETIF_F_LLTX;
4317 /* By default, we declare the bond to be fully
4318 * VLAN hardware accelerated capable. Special
4319 * care is taken in the various xmit functions
4320 * when there are slaves that are not hw accel
4324 bond_dev->hw_features = BOND_VLAN_FEATURES |
4325 NETIF_F_HW_VLAN_CTAG_TX |
4326 NETIF_F_HW_VLAN_CTAG_RX |
4327 NETIF_F_HW_VLAN_CTAG_FILTER;
4329 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4330 bond_dev->features |= bond_dev->hw_features;
4334 * Destroy a bonding device.
4335 * Must be under rtnl_lock when this function is called.
4337 static void bond_uninit(struct net_device *bond_dev)
4339 struct bonding *bond = netdev_priv(bond_dev);
4340 struct vlan_entry *vlan, *tmp;
4342 bond_netpoll_cleanup(bond_dev);
4344 /* Release the bonded slaves */
4345 while (bond->first_slave != NULL)
4346 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4347 pr_info("%s: released all slaves\n", bond_dev->name);
4349 list_del(&bond->bond_list);
4351 bond_debug_unregister(bond);
4353 __hw_addr_flush(&bond->mc_list);
4355 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4356 list_del(&vlan->vlan_list);
4361 /*------------------------- Module initialization ---------------------------*/
4364 * Convert string input module parms. Accept either the
4365 * number of the mode or its string name. A bit complicated because
4366 * some mode names are substrings of other names, and calls from sysfs
4367 * may have whitespace in the name (trailing newlines, for example).
4369 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4371 int modeint = -1, i, rv;
4372 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4374 for (p = (char *)buf; *p; p++)
4375 if (!(isdigit(*p) || isspace(*p)))
4379 rv = sscanf(buf, "%20s", modestr);
4381 rv = sscanf(buf, "%d", &modeint);
4386 for (i = 0; tbl[i].modename; i++) {
4387 if (modeint == tbl[i].mode)
4389 if (strcmp(modestr, tbl[i].modename) == 0)
4396 static int bond_check_params(struct bond_params *params)
4398 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4401 * Convert string parameters.
4404 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4405 if (bond_mode == -1) {
4406 pr_err("Error: Invalid bonding mode \"%s\"\n",
4407 mode == NULL ? "NULL" : mode);
4412 if (xmit_hash_policy) {
4413 if ((bond_mode != BOND_MODE_XOR) &&
4414 (bond_mode != BOND_MODE_8023AD)) {
4415 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4416 bond_mode_name(bond_mode));
4418 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4420 if (xmit_hashtype == -1) {
4421 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4422 xmit_hash_policy == NULL ? "NULL" :
4430 if (bond_mode != BOND_MODE_8023AD) {
4431 pr_info("lacp_rate param is irrelevant in mode %s\n",
4432 bond_mode_name(bond_mode));
4434 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4435 if (lacp_fast == -1) {
4436 pr_err("Error: Invalid lacp rate \"%s\"\n",
4437 lacp_rate == NULL ? "NULL" : lacp_rate);
4444 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4445 if (params->ad_select == -1) {
4446 pr_err("Error: Invalid ad_select \"%s\"\n",
4447 ad_select == NULL ? "NULL" : ad_select);
4451 if (bond_mode != BOND_MODE_8023AD) {
4452 pr_warning("ad_select param only affects 802.3ad mode\n");
4455 params->ad_select = BOND_AD_STABLE;
4458 if (max_bonds < 0) {
4459 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4460 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4461 max_bonds = BOND_DEFAULT_MAX_BONDS;
4465 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4466 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4467 miimon = BOND_LINK_MON_INTERV;
4471 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4476 if (downdelay < 0) {
4477 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4478 downdelay, INT_MAX);
4482 if ((use_carrier != 0) && (use_carrier != 1)) {
4483 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4488 if (num_peer_notif < 0 || num_peer_notif > 255) {
4489 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4494 /* reset values for 802.3ad */
4495 if (bond_mode == BOND_MODE_8023AD) {
4497 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");
4498 pr_warning("Forcing miimon to 100msec\n");
4503 if (tx_queues < 1 || tx_queues > 255) {
4504 pr_warning("Warning: tx_queues (%d) should be between "
4505 "1 and 255, resetting to %d\n",
4506 tx_queues, BOND_DEFAULT_TX_QUEUES);
4507 tx_queues = BOND_DEFAULT_TX_QUEUES;
4510 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4511 pr_warning("Warning: all_slaves_active module parameter (%d), "
4512 "not of valid value (0/1), so it was set to "
4513 "0\n", all_slaves_active);
4514 all_slaves_active = 0;
4517 if (resend_igmp < 0 || resend_igmp > 255) {
4518 pr_warning("Warning: resend_igmp (%d) should be between "
4519 "0 and 255, resetting to %d\n",
4520 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4521 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4524 /* reset values for TLB/ALB */
4525 if ((bond_mode == BOND_MODE_TLB) ||
4526 (bond_mode == BOND_MODE_ALB)) {
4528 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");
4529 pr_warning("Forcing miimon to 100msec\n");
4534 if (bond_mode == BOND_MODE_ALB) {
4535 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",
4540 if (updelay || downdelay) {
4541 /* just warn the user the up/down delay will have
4542 * no effect since miimon is zero...
4544 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",
4545 updelay, downdelay);
4548 /* don't allow arp monitoring */
4550 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4551 miimon, arp_interval);
4555 if ((updelay % miimon) != 0) {
4556 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4558 (updelay / miimon) * miimon);
4563 if ((downdelay % miimon) != 0) {
4564 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4566 (downdelay / miimon) * miimon);
4569 downdelay /= miimon;
4572 if (arp_interval < 0) {
4573 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4574 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4575 arp_interval = BOND_LINK_ARP_INTERV;
4578 for (arp_ip_count = 0;
4579 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4581 /* not complete check, but should be good enough to
4583 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4584 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4585 ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4586 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4587 arp_ip_target[arp_ip_count]);
4590 arp_target[arp_ip_count] = ip;
4594 if (arp_interval && !arp_ip_count) {
4595 /* don't allow arping if no arp_ip_target given... */
4596 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4602 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4603 pr_err("arp_validate only supported in active-backup mode\n");
4606 if (!arp_interval) {
4607 pr_err("arp_validate requires arp_interval\n");
4611 arp_validate_value = bond_parse_parm(arp_validate,
4613 if (arp_validate_value == -1) {
4614 pr_err("Error: invalid arp_validate \"%s\"\n",
4615 arp_validate == NULL ? "NULL" : arp_validate);
4619 arp_validate_value = 0;
4622 pr_info("MII link monitoring set to %d ms\n", miimon);
4623 } else if (arp_interval) {
4626 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4628 arp_validate_tbl[arp_validate_value].modename,
4631 for (i = 0; i < arp_ip_count; i++)
4632 pr_info(" %s", arp_ip_target[i]);
4636 } else if (max_bonds) {
4637 /* miimon and arp_interval not set, we need one so things
4638 * work as expected, see bonding.txt for details
4640 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");
4643 if (primary && !USES_PRIMARY(bond_mode)) {
4644 /* currently, using a primary only makes sense
4645 * in active backup, TLB or ALB modes
4647 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4648 primary, bond_mode_name(bond_mode));
4652 if (primary && primary_reselect) {
4653 primary_reselect_value = bond_parse_parm(primary_reselect,
4655 if (primary_reselect_value == -1) {
4656 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4658 NULL ? "NULL" : primary_reselect);
4662 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4665 if (fail_over_mac) {
4666 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4668 if (fail_over_mac_value == -1) {
4669 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4670 arp_validate == NULL ? "NULL" : arp_validate);
4674 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4675 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4677 fail_over_mac_value = BOND_FOM_NONE;
4680 /* fill params struct with the proper values */
4681 params->mode = bond_mode;
4682 params->xmit_policy = xmit_hashtype;
4683 params->miimon = miimon;
4684 params->num_peer_notif = num_peer_notif;
4685 params->arp_interval = arp_interval;
4686 params->arp_validate = arp_validate_value;
4687 params->updelay = updelay;
4688 params->downdelay = downdelay;
4689 params->use_carrier = use_carrier;
4690 params->lacp_fast = lacp_fast;
4691 params->primary[0] = 0;
4692 params->primary_reselect = primary_reselect_value;
4693 params->fail_over_mac = fail_over_mac_value;
4694 params->tx_queues = tx_queues;
4695 params->all_slaves_active = all_slaves_active;
4696 params->resend_igmp = resend_igmp;
4697 params->min_links = min_links;
4700 strncpy(params->primary, primary, IFNAMSIZ);
4701 params->primary[IFNAMSIZ - 1] = 0;
4704 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4709 static struct lock_class_key bonding_netdev_xmit_lock_key;
4710 static struct lock_class_key bonding_netdev_addr_lock_key;
4711 static struct lock_class_key bonding_tx_busylock_key;
4713 static void bond_set_lockdep_class_one(struct net_device *dev,
4714 struct netdev_queue *txq,
4717 lockdep_set_class(&txq->_xmit_lock,
4718 &bonding_netdev_xmit_lock_key);
4721 static void bond_set_lockdep_class(struct net_device *dev)
4723 lockdep_set_class(&dev->addr_list_lock,
4724 &bonding_netdev_addr_lock_key);
4725 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4726 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4730 * Called from registration process
4732 static int bond_init(struct net_device *bond_dev)
4734 struct bonding *bond = netdev_priv(bond_dev);
4735 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4736 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4738 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4741 * Initialize locks that may be required during
4742 * en/deslave operations. All of the bond_open work
4743 * (of which this is part) should really be moved to
4744 * a phase prior to dev_open
4746 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4747 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4749 bond->wq = create_singlethread_workqueue(bond_dev->name);
4753 bond_set_lockdep_class(bond_dev);
4755 list_add_tail(&bond->bond_list, &bn->dev_list);
4757 bond_prepare_sysfs_group(bond);
4759 bond_debug_register(bond);
4761 /* Ensure valid dev_addr */
4762 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4763 bond_dev->addr_assign_type == NET_ADDR_PERM) {
4764 eth_hw_addr_random(bond_dev);
4765 bond->dev_addr_from_first = true;
4768 __hw_addr_init(&bond->mc_list);
4772 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4774 if (tb[IFLA_ADDRESS]) {
4775 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4777 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4778 return -EADDRNOTAVAIL;
4783 static unsigned int bond_get_num_tx_queues(void)
4788 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4790 .priv_size = sizeof(struct bonding),
4791 .setup = bond_setup,
4792 .validate = bond_validate,
4793 .get_num_tx_queues = bond_get_num_tx_queues,
4794 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4798 /* Create a new bond based on the specified name and bonding parameters.
4799 * If name is NULL, obtain a suitable "bond%d" name for us.
4800 * Caller must NOT hold rtnl_lock; we need to release it here before we
4801 * set up our sysfs entries.
4803 int bond_create(struct net *net, const char *name)
4805 struct net_device *bond_dev;
4810 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4811 name ? name : "bond%d",
4812 bond_setup, tx_queues);
4814 pr_err("%s: eek! can't alloc netdev!\n", name);
4819 dev_net_set(bond_dev, net);
4820 bond_dev->rtnl_link_ops = &bond_link_ops;
4822 res = register_netdevice(bond_dev);
4824 netif_carrier_off(bond_dev);
4828 bond_destructor(bond_dev);
4832 static int __net_init bond_net_init(struct net *net)
4834 struct bond_net *bn = net_generic(net, bond_net_id);
4837 INIT_LIST_HEAD(&bn->dev_list);
4839 bond_create_proc_dir(bn);
4840 bond_create_sysfs(bn);
4845 static void __net_exit bond_net_exit(struct net *net)
4847 struct bond_net *bn = net_generic(net, bond_net_id);
4849 bond_destroy_sysfs(bn);
4850 bond_destroy_proc_dir(bn);
4853 static struct pernet_operations bond_net_ops = {
4854 .init = bond_net_init,
4855 .exit = bond_net_exit,
4857 .size = sizeof(struct bond_net),
4860 static int __init bonding_init(void)
4865 pr_info("%s", bond_version);
4867 res = bond_check_params(&bonding_defaults);
4871 res = register_pernet_subsys(&bond_net_ops);
4875 res = rtnl_link_register(&bond_link_ops);
4879 bond_create_debugfs();
4881 for (i = 0; i < max_bonds; i++) {
4882 res = bond_create(&init_net, NULL);
4887 register_netdevice_notifier(&bond_netdev_notifier);
4891 rtnl_link_unregister(&bond_link_ops);
4893 unregister_pernet_subsys(&bond_net_ops);
4898 static void __exit bonding_exit(void)
4900 unregister_netdevice_notifier(&bond_netdev_notifier);
4902 bond_destroy_debugfs();
4904 unregister_pernet_subsys(&bond_net_ops);
4905 rtnl_link_unregister(&bond_link_ops);
4907 #ifdef CONFIG_NET_POLL_CONTROLLER
4909 * Make sure we don't have an imbalance on our netpoll blocking
4911 WARN_ON(atomic_read(&netpoll_block_tx));
4915 module_init(bonding_init);
4916 module_exit(bonding_exit);
4917 MODULE_LICENSE("GPL");
4918 MODULE_VERSION(DRV_VERSION);
4919 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4920 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4921 MODULE_ALIAS_RTNL_LINK("bond");