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>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <linux/preempt.h>
77 #include <net/route.h>
78 #include <net/net_namespace.h>
79 #include <net/netns/generic.h>
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);
385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
388 * bond_dev_queue_xmit - Prepare skb for xmit.
390 * @bond: bond device that got this skb for tx.
391 * @skb: hw accel VLAN tagged skb to transmit
392 * @slave_dev: slave that is supposed to xmit this skbuff
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
395 struct net_device *slave_dev)
397 skb->dev = slave_dev;
400 skb->queue_mapping = bond_queue_mapping(skb);
402 if (unlikely(netpoll_tx_running(slave_dev)))
403 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
411 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412 * We don't protect the slave list iteration with a lock because:
413 * a. This operation is performed in IOCTL context,
414 * b. The operation is protected by the RTNL semaphore in the 8021q code,
415 * c. Holding a lock with BH disabled while directly calling a base driver
416 * entry point is generally a BAD idea.
418 * The design of synchronization/protection for this operation in the 8021q
419 * module is good for one or more VLAN devices over a single physical device
420 * and cannot be extended for a teaming solution like bonding, so there is a
421 * potential race condition here where a net device from the vlan group might
422 * be referenced (either by a base driver or the 8021q code) while it is being
423 * removed from the system. However, it turns out we're not making matters
424 * worse, and if it works for regular VLAN usage it will work here too.
428 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429 * @bond_dev: bonding net device that got called
430 * @vid: vlan id being added
432 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
434 struct bonding *bond = netdev_priv(bond_dev);
438 bond_for_each_slave(bond, slave, i) {
439 struct net_device *slave_dev = slave->dev;
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
442 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
443 slave_ops->ndo_vlan_rx_add_vid) {
444 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
448 res = bond_add_vlan(bond, vid);
450 pr_err("%s: Error: Failed to add vlan id %d\n",
451 bond_dev->name, vid);
456 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being removed
460 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
462 struct bonding *bond = netdev_priv(bond_dev);
466 bond_for_each_slave(bond, slave, i) {
467 struct net_device *slave_dev = slave->dev;
468 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
470 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
471 slave_ops->ndo_vlan_rx_kill_vid) {
472 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
476 res = bond_del_vlan(bond, vid);
478 pr_err("%s: Error: Failed to remove vlan id %d\n",
479 bond_dev->name, vid);
483 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
485 struct vlan_entry *vlan;
486 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
488 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
489 !(slave_ops->ndo_vlan_rx_add_vid))
492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
493 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
496 static void bond_del_vlans_from_slave(struct bonding *bond,
497 struct net_device *slave_dev)
499 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
500 struct vlan_entry *vlan;
502 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
503 !(slave_ops->ndo_vlan_rx_kill_vid))
506 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
509 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
513 /*------------------------------- Link status -------------------------------*/
516 * Set the carrier state for the master according to the state of its
517 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
518 * do special 802.3ad magic.
520 * Returns zero if carrier state does not change, nonzero if it does.
522 static int bond_set_carrier(struct bonding *bond)
527 if (bond->slave_cnt == 0)
530 if (bond->params.mode == BOND_MODE_8023AD)
531 return bond_3ad_set_carrier(bond);
533 bond_for_each_slave(bond, slave, i) {
534 if (slave->link == BOND_LINK_UP) {
535 if (!netif_carrier_ok(bond->dev)) {
536 netif_carrier_on(bond->dev);
544 if (netif_carrier_ok(bond->dev)) {
545 netif_carrier_off(bond->dev);
552 * Get link speed and duplex from the slave's base driver
553 * using ethtool. If for some reason the call fails or the
554 * values are invalid, fake speed and duplex to 100/Full
557 static int bond_update_speed_duplex(struct slave *slave)
559 struct net_device *slave_dev = slave->dev;
560 struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
564 /* Fake speed and duplex */
565 slave->speed = SPEED_100;
566 slave->duplex = DUPLEX_FULL;
568 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
571 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
575 slave_speed = ethtool_cmd_speed(&etool);
576 if (slave_speed == 0 || slave_speed == ((__u32) -1))
579 switch (etool.duplex) {
587 slave->speed = slave_speed;
588 slave->duplex = etool.duplex;
594 * if <dev> supports MII link status reporting, check its link status.
596 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
597 * depending upon the setting of the use_carrier parameter.
599 * Return either BMSR_LSTATUS, meaning that the link is up (or we
600 * can't tell and just pretend it is), or 0, meaning that the link is
603 * If reporting is non-zero, instead of faking link up, return -1 if
604 * both ETHTOOL and MII ioctls fail (meaning the device does not
605 * support them). If use_carrier is set, return whatever it says.
606 * It'd be nice if there was a good way to tell if a driver supports
607 * netif_carrier, but there really isn't.
609 static int bond_check_dev_link(struct bonding *bond,
610 struct net_device *slave_dev, int reporting)
612 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
613 int (*ioctl)(struct net_device *, struct ifreq *, int);
615 struct mii_ioctl_data *mii;
617 if (!reporting && !netif_running(slave_dev))
620 if (bond->params.use_carrier)
621 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
623 /* Try to get link status using Ethtool first. */
624 if (slave_dev->ethtool_ops) {
625 if (slave_dev->ethtool_ops->get_link) {
628 link = slave_dev->ethtool_ops->get_link(slave_dev);
630 return link ? BMSR_LSTATUS : 0;
634 /* Ethtool can't be used, fallback to MII ioctls. */
635 ioctl = slave_ops->ndo_do_ioctl;
637 /* TODO: set pointer to correct ioctl on a per team member */
638 /* bases to make this more efficient. that is, once */
639 /* we determine the correct ioctl, we will always */
640 /* call it and not the others for that team */
644 * We cannot assume that SIOCGMIIPHY will also read a
645 * register; not all network drivers (e.g., e100)
649 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
650 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
652 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
653 mii->reg_num = MII_BMSR;
654 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
655 return mii->val_out & BMSR_LSTATUS;
660 * If reporting, report that either there's no dev->do_ioctl,
661 * or both SIOCGMIIREG and get_link failed (meaning that we
662 * cannot report link status). If not reporting, pretend
665 return reporting ? -1 : BMSR_LSTATUS;
668 /*----------------------------- Multicast list ------------------------------*/
671 * Push the promiscuity flag down to appropriate slaves
673 static int bond_set_promiscuity(struct bonding *bond, int inc)
676 if (USES_PRIMARY(bond->params.mode)) {
677 /* write lock already acquired */
678 if (bond->curr_active_slave) {
679 err = dev_set_promiscuity(bond->curr_active_slave->dev,
685 bond_for_each_slave(bond, slave, i) {
686 err = dev_set_promiscuity(slave->dev, inc);
695 * Push the allmulti flag down to all slaves
697 static int bond_set_allmulti(struct bonding *bond, int inc)
700 if (USES_PRIMARY(bond->params.mode)) {
701 /* write lock already acquired */
702 if (bond->curr_active_slave) {
703 err = dev_set_allmulti(bond->curr_active_slave->dev,
709 bond_for_each_slave(bond, slave, i) {
710 err = dev_set_allmulti(slave->dev, inc);
719 * Add a Multicast address to slaves
722 static void bond_mc_add(struct bonding *bond, void *addr)
724 if (USES_PRIMARY(bond->params.mode)) {
725 /* write lock already acquired */
726 if (bond->curr_active_slave)
727 dev_mc_add(bond->curr_active_slave->dev, addr);
732 bond_for_each_slave(bond, slave, i)
733 dev_mc_add(slave->dev, addr);
738 * Remove a multicast address from slave
741 static void bond_mc_del(struct bonding *bond, void *addr)
743 if (USES_PRIMARY(bond->params.mode)) {
744 /* write lock already acquired */
745 if (bond->curr_active_slave)
746 dev_mc_del(bond->curr_active_slave->dev, addr);
750 bond_for_each_slave(bond, slave, i) {
751 dev_mc_del(slave->dev, addr);
757 static void __bond_resend_igmp_join_requests(struct net_device *dev)
759 struct in_device *in_dev;
762 in_dev = __in_dev_get_rcu(dev);
764 ip_mc_rejoin_groups(in_dev);
769 * Retrieve the list of registered multicast addresses for the bonding
770 * device and retransmit an IGMP JOIN request to the current active
773 static void bond_resend_igmp_join_requests(struct bonding *bond)
775 struct net_device *vlan_dev;
776 struct vlan_entry *vlan;
778 read_lock(&bond->lock);
780 if (bond->kill_timers)
783 /* rejoin all groups on bond device */
784 __bond_resend_igmp_join_requests(bond->dev);
786 /* rejoin all groups on vlan devices */
787 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
789 vlan_dev = __vlan_find_dev_deep(bond->dev,
793 __bond_resend_igmp_join_requests(vlan_dev);
796 if ((--bond->igmp_retrans > 0) && !bond->kill_timers)
797 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
799 read_unlock(&bond->lock);
802 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
804 struct bonding *bond = container_of(work, struct bonding,
806 bond_resend_igmp_join_requests(bond);
810 * flush all members of flush->mc_list from device dev->mc_list
812 static void bond_mc_list_flush(struct net_device *bond_dev,
813 struct net_device *slave_dev)
815 struct bonding *bond = netdev_priv(bond_dev);
816 struct netdev_hw_addr *ha;
818 netdev_for_each_mc_addr(ha, bond_dev)
819 dev_mc_del(slave_dev, ha->addr);
821 if (bond->params.mode == BOND_MODE_8023AD) {
822 /* del lacpdu mc addr from mc list */
823 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
825 dev_mc_del(slave_dev, lacpdu_multicast);
829 /*--------------------------- Active slave change ---------------------------*/
832 * Update the mc list and multicast-related flags for the new and
833 * old active slaves (if any) according to the multicast mode, and
834 * promiscuous flags unconditionally.
836 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
837 struct slave *old_active)
839 struct netdev_hw_addr *ha;
841 if (!USES_PRIMARY(bond->params.mode))
842 /* nothing to do - mc list is already up-to-date on
848 if (bond->dev->flags & IFF_PROMISC)
849 dev_set_promiscuity(old_active->dev, -1);
851 if (bond->dev->flags & IFF_ALLMULTI)
852 dev_set_allmulti(old_active->dev, -1);
854 netdev_for_each_mc_addr(ha, bond->dev)
855 dev_mc_del(old_active->dev, ha->addr);
859 /* FIXME: Signal errors upstream. */
860 if (bond->dev->flags & IFF_PROMISC)
861 dev_set_promiscuity(new_active->dev, 1);
863 if (bond->dev->flags & IFF_ALLMULTI)
864 dev_set_allmulti(new_active->dev, 1);
866 netdev_for_each_mc_addr(ha, bond->dev)
867 dev_mc_add(new_active->dev, ha->addr);
872 * bond_do_fail_over_mac
874 * Perform special MAC address swapping for fail_over_mac settings
876 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
878 static void bond_do_fail_over_mac(struct bonding *bond,
879 struct slave *new_active,
880 struct slave *old_active)
881 __releases(&bond->curr_slave_lock)
882 __releases(&bond->lock)
883 __acquires(&bond->lock)
884 __acquires(&bond->curr_slave_lock)
886 u8 tmp_mac[ETH_ALEN];
887 struct sockaddr saddr;
890 switch (bond->params.fail_over_mac) {
891 case BOND_FOM_ACTIVE:
893 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
894 new_active->dev->addr_len);
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 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1115 if (should_notify_peers)
1116 netdev_bonding_change(bond->dev,
1117 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_KERNEL);
1234 np->dev = slave->dev;
1235 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1236 err = __netpoll_setup(np);
1245 static inline void slave_disable_netpoll(struct slave *slave)
1247 struct netpoll *np = slave->np;
1253 synchronize_rcu_bh();
1254 __netpoll_cleanup(np);
1257 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1259 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1261 if (!slave_dev->netdev_ops->ndo_poll_controller)
1266 static void bond_poll_controller(struct net_device *bond_dev)
1270 static void __bond_netpoll_cleanup(struct bonding *bond)
1272 struct slave *slave;
1275 bond_for_each_slave(bond, slave, i)
1276 if (IS_UP(slave->dev))
1277 slave_disable_netpoll(slave);
1279 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1281 struct bonding *bond = netdev_priv(bond_dev);
1283 read_lock(&bond->lock);
1284 __bond_netpoll_cleanup(bond);
1285 read_unlock(&bond->lock);
1288 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1290 struct bonding *bond = netdev_priv(dev);
1291 struct slave *slave;
1294 read_lock(&bond->lock);
1295 bond_for_each_slave(bond, slave, i) {
1296 err = slave_enable_netpoll(slave);
1298 __bond_netpoll_cleanup(bond);
1302 read_unlock(&bond->lock);
1306 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1308 return bond->dev->npinfo;
1312 static inline int slave_enable_netpoll(struct slave *slave)
1316 static inline void slave_disable_netpoll(struct slave *slave)
1319 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1324 /*---------------------------------- IOCTL ----------------------------------*/
1326 static int bond_sethwaddr(struct net_device *bond_dev,
1327 struct net_device *slave_dev)
1329 pr_debug("bond_dev=%p\n", bond_dev);
1330 pr_debug("slave_dev=%p\n", slave_dev);
1331 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1332 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1336 static u32 bond_fix_features(struct net_device *dev, u32 features)
1338 struct slave *slave;
1339 struct bonding *bond = netdev_priv(dev);
1343 read_lock(&bond->lock);
1345 if (!bond->first_slave) {
1346 /* Disable adding VLANs to empty bond. But why? --mq */
1347 features |= NETIF_F_VLAN_CHALLENGED;
1352 features &= ~NETIF_F_ONE_FOR_ALL;
1353 features |= NETIF_F_ALL_FOR_ALL;
1355 bond_for_each_slave(bond, slave, i) {
1356 features = netdev_increment_features(features,
1357 slave->dev->features,
1362 read_unlock(&bond->lock);
1366 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1367 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1368 NETIF_F_HIGHDMA | NETIF_F_LRO)
1370 static void bond_compute_features(struct bonding *bond)
1372 struct slave *slave;
1373 struct net_device *bond_dev = bond->dev;
1374 u32 vlan_features = BOND_VLAN_FEATURES;
1375 unsigned short max_hard_header_len = ETH_HLEN;
1378 read_lock(&bond->lock);
1380 if (!bond->first_slave)
1383 bond_for_each_slave(bond, slave, i) {
1384 vlan_features = netdev_increment_features(vlan_features,
1385 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1387 if (slave->dev->hard_header_len > max_hard_header_len)
1388 max_hard_header_len = slave->dev->hard_header_len;
1392 bond_dev->vlan_features = vlan_features;
1393 bond_dev->hard_header_len = max_hard_header_len;
1395 read_unlock(&bond->lock);
1397 netdev_change_features(bond_dev);
1400 static void bond_setup_by_slave(struct net_device *bond_dev,
1401 struct net_device *slave_dev)
1403 struct bonding *bond = netdev_priv(bond_dev);
1405 bond_dev->header_ops = slave_dev->header_ops;
1407 bond_dev->type = slave_dev->type;
1408 bond_dev->hard_header_len = slave_dev->hard_header_len;
1409 bond_dev->addr_len = slave_dev->addr_len;
1411 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1412 slave_dev->addr_len);
1413 bond->setup_by_slave = 1;
1416 /* On bonding slaves other than the currently active slave, suppress
1417 * duplicates except for alb non-mcast/bcast.
1419 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1420 struct slave *slave,
1421 struct bonding *bond)
1423 if (bond_is_slave_inactive(slave)) {
1424 if (bond->params.mode == BOND_MODE_ALB &&
1425 skb->pkt_type != PACKET_BROADCAST &&
1426 skb->pkt_type != PACKET_MULTICAST)
1433 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1435 struct sk_buff *skb = *pskb;
1436 struct slave *slave;
1437 struct bonding *bond;
1439 skb = skb_share_check(skb, GFP_ATOMIC);
1441 return RX_HANDLER_CONSUMED;
1445 slave = bond_slave_get_rcu(skb->dev);
1448 if (bond->params.arp_interval)
1449 slave->dev->last_rx = jiffies;
1451 if (bond->recv_probe) {
1452 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1455 bond->recv_probe(nskb, bond, slave);
1456 dev_kfree_skb(nskb);
1460 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1461 return RX_HANDLER_EXACT;
1464 skb->dev = bond->dev;
1466 if (bond->params.mode == BOND_MODE_ALB &&
1467 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1468 skb->pkt_type == PACKET_HOST) {
1470 if (unlikely(skb_cow_head(skb,
1471 skb->data - skb_mac_header(skb)))) {
1473 return RX_HANDLER_CONSUMED;
1475 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1478 return RX_HANDLER_ANOTHER;
1481 /* enslave device <slave> to bond device <master> */
1482 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1484 struct bonding *bond = netdev_priv(bond_dev);
1485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1486 struct slave *new_slave = NULL;
1487 struct netdev_hw_addr *ha;
1488 struct sockaddr addr;
1492 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1493 slave_ops->ndo_do_ioctl == NULL) {
1494 pr_warning("%s: Warning: no link monitoring support for %s\n",
1495 bond_dev->name, slave_dev->name);
1498 /* already enslaved */
1499 if (slave_dev->flags & IFF_SLAVE) {
1500 pr_debug("Error, Device was already enslaved\n");
1504 /* vlan challenged mutual exclusion */
1505 /* no need to lock since we're protected by rtnl_lock */
1506 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1507 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1508 if (bond_vlan_used(bond)) {
1509 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1510 bond_dev->name, slave_dev->name, bond_dev->name);
1513 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1514 bond_dev->name, slave_dev->name,
1515 slave_dev->name, bond_dev->name);
1518 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1522 * Old ifenslave binaries are no longer supported. These can
1523 * be identified with moderate accuracy by the state of the slave:
1524 * the current ifenslave will set the interface down prior to
1525 * enslaving it; the old ifenslave will not.
1527 if ((slave_dev->flags & IFF_UP)) {
1528 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1531 goto err_undo_flags;
1534 /* set bonding device ether type by slave - bonding netdevices are
1535 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1536 * there is a need to override some of the type dependent attribs/funcs.
1538 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1539 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1541 if (bond->slave_cnt == 0) {
1542 if (bond_dev->type != slave_dev->type) {
1543 pr_debug("%s: change device type from %d to %d\n",
1545 bond_dev->type, slave_dev->type);
1547 res = netdev_bonding_change(bond_dev,
1548 NETDEV_PRE_TYPE_CHANGE);
1549 res = notifier_to_errno(res);
1551 pr_err("%s: refused to change device type\n",
1554 goto err_undo_flags;
1557 /* Flush unicast and multicast addresses */
1558 dev_uc_flush(bond_dev);
1559 dev_mc_flush(bond_dev);
1561 if (slave_dev->type != ARPHRD_ETHER)
1562 bond_setup_by_slave(bond_dev, slave_dev);
1564 ether_setup(bond_dev);
1565 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1568 netdev_bonding_change(bond_dev,
1569 NETDEV_POST_TYPE_CHANGE);
1571 } else if (bond_dev->type != slave_dev->type) {
1572 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1574 slave_dev->type, bond_dev->type);
1576 goto err_undo_flags;
1579 if (slave_ops->ndo_set_mac_address == NULL) {
1580 if (bond->slave_cnt == 0) {
1581 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1583 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1584 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1585 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",
1588 goto err_undo_flags;
1592 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1594 /* If this is the first slave, then we need to set the master's hardware
1595 * address to be the same as the slave's. */
1596 if (is_zero_ether_addr(bond->dev->dev_addr))
1597 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1598 slave_dev->addr_len);
1601 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1604 goto err_undo_flags;
1608 * Set the new_slave's queue_id to be zero. Queue ID mapping
1609 * is set via sysfs or module option if desired.
1611 new_slave->queue_id = 0;
1613 /* Save slave's original mtu and then set it to match the bond */
1614 new_slave->original_mtu = slave_dev->mtu;
1615 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1617 pr_debug("Error %d calling dev_set_mtu\n", res);
1622 * Save slave's original ("permanent") mac address for modes
1623 * that need it, and for restoring it upon release, and then
1624 * set it to the master's address
1626 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1628 if (!bond->params.fail_over_mac) {
1630 * Set slave to master's mac address. The application already
1631 * set the master's mac address to that of the first slave
1633 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1634 addr.sa_family = slave_dev->type;
1635 res = dev_set_mac_address(slave_dev, &addr);
1637 pr_debug("Error %d calling set_mac_address\n", res);
1638 goto err_restore_mtu;
1642 res = netdev_set_bond_master(slave_dev, bond_dev);
1644 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1645 goto err_restore_mac;
1648 /* open the slave since the application closed it */
1649 res = dev_open(slave_dev);
1651 pr_debug("Opening slave %s failed\n", slave_dev->name);
1652 goto err_unset_master;
1655 new_slave->bond = bond;
1656 new_slave->dev = slave_dev;
1657 slave_dev->priv_flags |= IFF_BONDING;
1659 if (bond_is_lb(bond)) {
1660 /* bond_alb_init_slave() must be called before all other stages since
1661 * it might fail and we do not want to have to undo everything
1663 res = bond_alb_init_slave(bond, new_slave);
1668 /* If the mode USES_PRIMARY, then the new slave gets the
1669 * master's promisc (and mc) settings only if it becomes the
1670 * curr_active_slave, and that is taken care of later when calling
1671 * bond_change_active()
1673 if (!USES_PRIMARY(bond->params.mode)) {
1674 /* set promiscuity level to new slave */
1675 if (bond_dev->flags & IFF_PROMISC) {
1676 res = dev_set_promiscuity(slave_dev, 1);
1681 /* set allmulti level to new slave */
1682 if (bond_dev->flags & IFF_ALLMULTI) {
1683 res = dev_set_allmulti(slave_dev, 1);
1688 netif_addr_lock_bh(bond_dev);
1689 /* upload master's mc_list to new slave */
1690 netdev_for_each_mc_addr(ha, bond_dev)
1691 dev_mc_add(slave_dev, ha->addr);
1692 netif_addr_unlock_bh(bond_dev);
1695 if (bond->params.mode == BOND_MODE_8023AD) {
1696 /* add lacpdu mc addr to mc list */
1697 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1699 dev_mc_add(slave_dev, lacpdu_multicast);
1702 bond_add_vlans_on_slave(bond, slave_dev);
1704 write_lock_bh(&bond->lock);
1706 bond_attach_slave(bond, new_slave);
1708 new_slave->delay = 0;
1709 new_slave->link_failure_count = 0;
1711 write_unlock_bh(&bond->lock);
1713 bond_compute_features(bond);
1715 read_lock(&bond->lock);
1717 new_slave->last_arp_rx = jiffies;
1719 if (bond->params.miimon && !bond->params.use_carrier) {
1720 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1722 if ((link_reporting == -1) && !bond->params.arp_interval) {
1724 * miimon is set but a bonded network driver
1725 * does not support ETHTOOL/MII and
1726 * arp_interval is not set. Note: if
1727 * use_carrier is enabled, we will never go
1728 * here (because netif_carrier is always
1729 * supported); thus, we don't need to change
1730 * the messages for netif_carrier.
1732 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",
1733 bond_dev->name, slave_dev->name);
1734 } else if (link_reporting == -1) {
1735 /* unable get link status using mii/ethtool */
1736 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",
1737 bond_dev->name, slave_dev->name);
1741 /* check for initial state */
1742 if (!bond->params.miimon ||
1743 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1744 if (bond->params.updelay) {
1745 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1746 new_slave->link = BOND_LINK_BACK;
1747 new_slave->delay = bond->params.updelay;
1749 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1750 new_slave->link = BOND_LINK_UP;
1752 new_slave->jiffies = jiffies;
1754 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1755 new_slave->link = BOND_LINK_DOWN;
1758 if (bond_update_speed_duplex(new_slave) &&
1759 (new_slave->link != BOND_LINK_DOWN)) {
1760 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1761 bond_dev->name, new_slave->dev->name);
1763 if (bond->params.mode == BOND_MODE_8023AD) {
1764 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1769 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1770 /* if there is a primary slave, remember it */
1771 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1772 bond->primary_slave = new_slave;
1773 bond->force_primary = true;
1777 write_lock_bh(&bond->curr_slave_lock);
1779 switch (bond->params.mode) {
1780 case BOND_MODE_ACTIVEBACKUP:
1781 bond_set_slave_inactive_flags(new_slave);
1782 bond_select_active_slave(bond);
1784 case BOND_MODE_8023AD:
1785 /* in 802.3ad mode, the internal mechanism
1786 * will activate the slaves in the selected
1789 bond_set_slave_inactive_flags(new_slave);
1790 /* if this is the first slave */
1791 if (bond->slave_cnt == 1) {
1792 SLAVE_AD_INFO(new_slave).id = 1;
1793 /* Initialize AD with the number of times that the AD timer is called in 1 second
1794 * can be called only after the mac address of the bond is set
1796 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1798 SLAVE_AD_INFO(new_slave).id =
1799 SLAVE_AD_INFO(new_slave->prev).id + 1;
1802 bond_3ad_bind_slave(new_slave);
1806 bond_set_active_slave(new_slave);
1807 bond_set_slave_inactive_flags(new_slave);
1808 bond_select_active_slave(bond);
1811 pr_debug("This slave is always active in trunk mode\n");
1813 /* always active in trunk mode */
1814 bond_set_active_slave(new_slave);
1816 /* In trunking mode there is little meaning to curr_active_slave
1817 * anyway (it holds no special properties of the bond device),
1818 * so we can change it without calling change_active_interface()
1820 if (!bond->curr_active_slave)
1821 bond->curr_active_slave = new_slave;
1824 } /* switch(bond_mode) */
1826 write_unlock_bh(&bond->curr_slave_lock);
1828 bond_set_carrier(bond);
1830 #ifdef CONFIG_NET_POLL_CONTROLLER
1831 slave_dev->npinfo = bond_netpoll_info(bond);
1832 if (slave_dev->npinfo) {
1833 if (slave_enable_netpoll(new_slave)) {
1834 read_unlock(&bond->lock);
1835 pr_info("Error, %s: master_dev is using netpoll, "
1836 "but new slave device does not support netpoll.\n",
1844 read_unlock(&bond->lock);
1846 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1850 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1853 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1854 goto err_dest_symlinks;
1857 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1858 bond_dev->name, slave_dev->name,
1859 bond_is_active_slave(new_slave) ? "n active" : " backup",
1860 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1862 /* enslave is successful */
1865 /* Undo stages on error */
1867 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1870 dev_close(slave_dev);
1873 netdev_set_bond_master(slave_dev, NULL);
1876 if (!bond->params.fail_over_mac) {
1877 /* XXX TODO - fom follow mode needs to change master's
1878 * MAC if this slave's MAC is in use by the bond, or at
1879 * least print a warning.
1881 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1882 addr.sa_family = slave_dev->type;
1883 dev_set_mac_address(slave_dev, &addr);
1887 dev_set_mtu(slave_dev, new_slave->original_mtu);
1893 bond_compute_features(bond);
1899 * Try to release the slave device <slave> from the bond device <master>
1900 * It is legal to access curr_active_slave without a lock because all the function
1903 * The rules for slave state should be:
1904 * for Active/Backup:
1905 * Active stays on all backups go down
1906 * for Bonded connections:
1907 * The first up interface should be left on and all others downed.
1909 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1911 struct bonding *bond = netdev_priv(bond_dev);
1912 struct slave *slave, *oldcurrent;
1913 struct sockaddr addr;
1914 u32 old_features = bond_dev->features;
1916 /* slave is not a slave or master is not master of this slave */
1917 if (!(slave_dev->flags & IFF_SLAVE) ||
1918 (slave_dev->master != bond_dev)) {
1919 pr_err("%s: Error: cannot release %s.\n",
1920 bond_dev->name, slave_dev->name);
1925 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1926 write_lock_bh(&bond->lock);
1928 slave = bond_get_slave_by_dev(bond, slave_dev);
1930 /* not a slave of this bond */
1931 pr_info("%s: %s not enslaved\n",
1932 bond_dev->name, slave_dev->name);
1933 write_unlock_bh(&bond->lock);
1934 unblock_netpoll_tx();
1938 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1939 * for this slave anymore.
1941 netdev_rx_handler_unregister(slave_dev);
1942 write_unlock_bh(&bond->lock);
1944 write_lock_bh(&bond->lock);
1946 if (!bond->params.fail_over_mac) {
1947 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1948 bond->slave_cnt > 1)
1949 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",
1950 bond_dev->name, slave_dev->name,
1952 bond_dev->name, slave_dev->name);
1955 /* Inform AD package of unbinding of slave. */
1956 if (bond->params.mode == BOND_MODE_8023AD) {
1957 /* must be called before the slave is
1958 * detached from the list
1960 bond_3ad_unbind_slave(slave);
1963 pr_info("%s: releasing %s interface %s\n",
1965 bond_is_active_slave(slave) ? "active" : "backup",
1968 oldcurrent = bond->curr_active_slave;
1970 bond->current_arp_slave = NULL;
1972 /* release the slave from its bond */
1973 bond_detach_slave(bond, slave);
1975 if (bond->primary_slave == slave)
1976 bond->primary_slave = NULL;
1978 if (oldcurrent == slave)
1979 bond_change_active_slave(bond, NULL);
1981 if (bond_is_lb(bond)) {
1982 /* Must be called only after the slave has been
1983 * detached from the list and the curr_active_slave
1984 * has been cleared (if our_slave == old_current),
1985 * but before a new active slave is selected.
1987 write_unlock_bh(&bond->lock);
1988 bond_alb_deinit_slave(bond, slave);
1989 write_lock_bh(&bond->lock);
1992 if (oldcurrent == slave) {
1994 * Note that we hold RTNL over this sequence, so there
1995 * is no concern that another slave add/remove event
1998 write_unlock_bh(&bond->lock);
1999 read_lock(&bond->lock);
2000 write_lock_bh(&bond->curr_slave_lock);
2002 bond_select_active_slave(bond);
2004 write_unlock_bh(&bond->curr_slave_lock);
2005 read_unlock(&bond->lock);
2006 write_lock_bh(&bond->lock);
2009 if (bond->slave_cnt == 0) {
2010 bond_set_carrier(bond);
2012 /* if the last slave was removed, zero the mac address
2013 * of the master so it will be set by the application
2014 * to the mac address of the first slave
2016 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2018 if (bond_vlan_used(bond)) {
2019 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2020 bond_dev->name, bond_dev->name);
2021 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2026 write_unlock_bh(&bond->lock);
2027 unblock_netpoll_tx();
2029 bond_compute_features(bond);
2030 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2031 (old_features & NETIF_F_VLAN_CHALLENGED))
2032 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2033 bond_dev->name, slave_dev->name, bond_dev->name);
2035 /* must do this from outside any spinlocks */
2036 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2038 bond_del_vlans_from_slave(bond, slave_dev);
2040 /* If the mode USES_PRIMARY, then we should only remove its
2041 * promisc and mc settings if it was the curr_active_slave, but that was
2042 * already taken care of above when we detached the slave
2044 if (!USES_PRIMARY(bond->params.mode)) {
2045 /* unset promiscuity level from slave */
2046 if (bond_dev->flags & IFF_PROMISC)
2047 dev_set_promiscuity(slave_dev, -1);
2049 /* unset allmulti level from slave */
2050 if (bond_dev->flags & IFF_ALLMULTI)
2051 dev_set_allmulti(slave_dev, -1);
2053 /* flush master's mc_list from slave */
2054 netif_addr_lock_bh(bond_dev);
2055 bond_mc_list_flush(bond_dev, slave_dev);
2056 netif_addr_unlock_bh(bond_dev);
2059 netdev_set_bond_master(slave_dev, NULL);
2061 slave_disable_netpoll(slave);
2063 /* close slave before restoring its mac address */
2064 dev_close(slave_dev);
2066 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2067 /* restore original ("permanent") mac address */
2068 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2069 addr.sa_family = slave_dev->type;
2070 dev_set_mac_address(slave_dev, &addr);
2073 dev_set_mtu(slave_dev, slave->original_mtu);
2075 slave_dev->priv_flags &= ~IFF_BONDING;
2079 return 0; /* deletion OK */
2083 * First release a slave and then destroy the bond if no more slaves are left.
2084 * Must be under rtnl_lock when this function is called.
2086 static int bond_release_and_destroy(struct net_device *bond_dev,
2087 struct net_device *slave_dev)
2089 struct bonding *bond = netdev_priv(bond_dev);
2092 ret = bond_release(bond_dev, slave_dev);
2093 if ((ret == 0) && (bond->slave_cnt == 0)) {
2094 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2095 pr_info("%s: destroying bond %s.\n",
2096 bond_dev->name, bond_dev->name);
2097 unregister_netdevice(bond_dev);
2103 * This function releases all slaves.
2105 static int bond_release_all(struct net_device *bond_dev)
2107 struct bonding *bond = netdev_priv(bond_dev);
2108 struct slave *slave;
2109 struct net_device *slave_dev;
2110 struct sockaddr addr;
2112 write_lock_bh(&bond->lock);
2114 netif_carrier_off(bond_dev);
2116 if (bond->slave_cnt == 0)
2119 bond->current_arp_slave = NULL;
2120 bond->primary_slave = NULL;
2121 bond_change_active_slave(bond, NULL);
2123 while ((slave = bond->first_slave) != NULL) {
2124 /* Inform AD package of unbinding of slave
2125 * before slave is detached from the list.
2127 if (bond->params.mode == BOND_MODE_8023AD)
2128 bond_3ad_unbind_slave(slave);
2130 slave_dev = slave->dev;
2131 bond_detach_slave(bond, slave);
2133 /* now that the slave is detached, unlock and perform
2134 * all the undo steps that should not be called from
2137 write_unlock_bh(&bond->lock);
2139 /* unregister rx_handler early so bond_handle_frame wouldn't
2140 * be called for this slave anymore.
2142 netdev_rx_handler_unregister(slave_dev);
2145 if (bond_is_lb(bond)) {
2146 /* must be called only after the slave
2147 * has been detached from the list
2149 bond_alb_deinit_slave(bond, slave);
2152 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2153 bond_del_vlans_from_slave(bond, slave_dev);
2155 /* If the mode USES_PRIMARY, then we should only remove its
2156 * promisc and mc settings if it was the curr_active_slave, but that was
2157 * already taken care of above when we detached the slave
2159 if (!USES_PRIMARY(bond->params.mode)) {
2160 /* unset promiscuity level from slave */
2161 if (bond_dev->flags & IFF_PROMISC)
2162 dev_set_promiscuity(slave_dev, -1);
2164 /* unset allmulti level from slave */
2165 if (bond_dev->flags & IFF_ALLMULTI)
2166 dev_set_allmulti(slave_dev, -1);
2168 /* flush master's mc_list from slave */
2169 netif_addr_lock_bh(bond_dev);
2170 bond_mc_list_flush(bond_dev, slave_dev);
2171 netif_addr_unlock_bh(bond_dev);
2174 netdev_set_bond_master(slave_dev, NULL);
2176 slave_disable_netpoll(slave);
2178 /* close slave before restoring its mac address */
2179 dev_close(slave_dev);
2181 if (!bond->params.fail_over_mac) {
2182 /* restore original ("permanent") mac address*/
2183 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2184 addr.sa_family = slave_dev->type;
2185 dev_set_mac_address(slave_dev, &addr);
2190 /* re-acquire the lock before getting the next slave */
2191 write_lock_bh(&bond->lock);
2194 /* zero the mac address of the master so it will be
2195 * set by the application to the mac address of the
2198 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2200 if (bond_vlan_used(bond)) {
2201 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2202 bond_dev->name, bond_dev->name);
2203 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2207 pr_info("%s: released all slaves\n", bond_dev->name);
2210 write_unlock_bh(&bond->lock);
2212 bond_compute_features(bond);
2218 * This function changes the active slave to slave <slave_dev>.
2219 * It returns -EINVAL in the following cases.
2220 * - <slave_dev> is not found in the list.
2221 * - There is not active slave now.
2222 * - <slave_dev> is already active.
2223 * - The link state of <slave_dev> is not BOND_LINK_UP.
2224 * - <slave_dev> is not running.
2225 * In these cases, this function does nothing.
2226 * In the other cases, current_slave pointer is changed and 0 is returned.
2228 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2230 struct bonding *bond = netdev_priv(bond_dev);
2231 struct slave *old_active = NULL;
2232 struct slave *new_active = NULL;
2235 if (!USES_PRIMARY(bond->params.mode))
2238 /* Verify that master_dev is indeed the master of slave_dev */
2239 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2242 read_lock(&bond->lock);
2244 read_lock(&bond->curr_slave_lock);
2245 old_active = bond->curr_active_slave;
2246 read_unlock(&bond->curr_slave_lock);
2248 new_active = bond_get_slave_by_dev(bond, slave_dev);
2251 * Changing to the current active: do nothing; return success.
2253 if (new_active && (new_active == old_active)) {
2254 read_unlock(&bond->lock);
2260 (new_active->link == BOND_LINK_UP) &&
2261 IS_UP(new_active->dev)) {
2263 write_lock_bh(&bond->curr_slave_lock);
2264 bond_change_active_slave(bond, new_active);
2265 write_unlock_bh(&bond->curr_slave_lock);
2266 unblock_netpoll_tx();
2270 read_unlock(&bond->lock);
2275 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2277 struct bonding *bond = netdev_priv(bond_dev);
2279 info->bond_mode = bond->params.mode;
2280 info->miimon = bond->params.miimon;
2282 read_lock(&bond->lock);
2283 info->num_slaves = bond->slave_cnt;
2284 read_unlock(&bond->lock);
2289 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2291 struct bonding *bond = netdev_priv(bond_dev);
2292 struct slave *slave;
2293 int i, res = -ENODEV;
2295 read_lock(&bond->lock);
2297 bond_for_each_slave(bond, slave, i) {
2298 if (i == (int)info->slave_id) {
2300 strcpy(info->slave_name, slave->dev->name);
2301 info->link = slave->link;
2302 info->state = bond_slave_state(slave);
2303 info->link_failure_count = slave->link_failure_count;
2308 read_unlock(&bond->lock);
2313 /*-------------------------------- Monitoring -------------------------------*/
2316 static int bond_miimon_inspect(struct bonding *bond)
2318 struct slave *slave;
2319 int i, link_state, commit = 0;
2320 bool ignore_updelay;
2322 ignore_updelay = !bond->curr_active_slave ? true : false;
2324 bond_for_each_slave(bond, slave, i) {
2325 slave->new_link = BOND_LINK_NOCHANGE;
2327 link_state = bond_check_dev_link(bond, slave->dev, 0);
2329 switch (slave->link) {
2334 slave->link = BOND_LINK_FAIL;
2335 slave->delay = bond->params.downdelay;
2337 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2339 (bond->params.mode ==
2340 BOND_MODE_ACTIVEBACKUP) ?
2341 (bond_is_active_slave(slave) ?
2342 "active " : "backup ") : "",
2344 bond->params.downdelay * bond->params.miimon);
2347 case BOND_LINK_FAIL:
2350 * recovered before downdelay expired
2352 slave->link = BOND_LINK_UP;
2353 slave->jiffies = jiffies;
2354 pr_info("%s: link status up again after %d ms for interface %s.\n",
2356 (bond->params.downdelay - slave->delay) *
2357 bond->params.miimon,
2362 if (slave->delay <= 0) {
2363 slave->new_link = BOND_LINK_DOWN;
2371 case BOND_LINK_DOWN:
2375 slave->link = BOND_LINK_BACK;
2376 slave->delay = bond->params.updelay;
2379 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2380 bond->dev->name, slave->dev->name,
2381 ignore_updelay ? 0 :
2382 bond->params.updelay *
2383 bond->params.miimon);
2386 case BOND_LINK_BACK:
2388 slave->link = BOND_LINK_DOWN;
2389 pr_info("%s: link status down again after %d ms for interface %s.\n",
2391 (bond->params.updelay - slave->delay) *
2392 bond->params.miimon,
2401 if (slave->delay <= 0) {
2402 slave->new_link = BOND_LINK_UP;
2404 ignore_updelay = false;
2416 static void bond_miimon_commit(struct bonding *bond)
2418 struct slave *slave;
2421 bond_for_each_slave(bond, slave, i) {
2422 switch (slave->new_link) {
2423 case BOND_LINK_NOCHANGE:
2427 slave->link = BOND_LINK_UP;
2428 slave->jiffies = jiffies;
2430 if (bond->params.mode == BOND_MODE_8023AD) {
2431 /* prevent it from being the active one */
2432 bond_set_backup_slave(slave);
2433 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2434 /* make it immediately active */
2435 bond_set_active_slave(slave);
2436 } else if (slave != bond->primary_slave) {
2437 /* prevent it from being the active one */
2438 bond_set_backup_slave(slave);
2441 bond_update_speed_duplex(slave);
2443 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2444 bond->dev->name, slave->dev->name,
2445 slave->speed, slave->duplex ? "full" : "half");
2447 /* notify ad that the link status has changed */
2448 if (bond->params.mode == BOND_MODE_8023AD)
2449 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2451 if (bond_is_lb(bond))
2452 bond_alb_handle_link_change(bond, slave,
2455 if (!bond->curr_active_slave ||
2456 (slave == bond->primary_slave))
2461 case BOND_LINK_DOWN:
2462 if (slave->link_failure_count < UINT_MAX)
2463 slave->link_failure_count++;
2465 slave->link = BOND_LINK_DOWN;
2467 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2468 bond->params.mode == BOND_MODE_8023AD)
2469 bond_set_slave_inactive_flags(slave);
2471 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2472 bond->dev->name, slave->dev->name);
2474 if (bond->params.mode == BOND_MODE_8023AD)
2475 bond_3ad_handle_link_change(slave,
2478 if (bond_is_lb(bond))
2479 bond_alb_handle_link_change(bond, slave,
2482 if (slave == bond->curr_active_slave)
2488 pr_err("%s: invalid new link %d on slave %s\n",
2489 bond->dev->name, slave->new_link,
2491 slave->new_link = BOND_LINK_NOCHANGE;
2499 write_lock_bh(&bond->curr_slave_lock);
2500 bond_select_active_slave(bond);
2501 write_unlock_bh(&bond->curr_slave_lock);
2502 unblock_netpoll_tx();
2505 bond_set_carrier(bond);
2511 * Really a wrapper that splits the mii monitor into two phases: an
2512 * inspection, then (if inspection indicates something needs to be done)
2513 * an acquisition of appropriate locks followed by a commit phase to
2514 * implement whatever link state changes are indicated.
2516 void bond_mii_monitor(struct work_struct *work)
2518 struct bonding *bond = container_of(work, struct bonding,
2520 bool should_notify_peers = false;
2522 read_lock(&bond->lock);
2523 if (bond->kill_timers)
2526 if (bond->slave_cnt == 0)
2529 should_notify_peers = bond_should_notify_peers(bond);
2531 if (bond_miimon_inspect(bond)) {
2532 read_unlock(&bond->lock);
2534 read_lock(&bond->lock);
2536 bond_miimon_commit(bond);
2538 read_unlock(&bond->lock);
2539 rtnl_unlock(); /* might sleep, hold no other locks */
2540 read_lock(&bond->lock);
2544 if (bond->params.miimon && !bond->kill_timers)
2545 queue_delayed_work(bond->wq, &bond->mii_work,
2546 msecs_to_jiffies(bond->params.miimon));
2548 read_unlock(&bond->lock);
2550 if (should_notify_peers) {
2552 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2557 static __be32 bond_glean_dev_ip(struct net_device *dev)
2559 struct in_device *idev;
2560 struct in_ifaddr *ifa;
2567 idev = __in_dev_get_rcu(dev);
2571 ifa = idev->ifa_list;
2575 addr = ifa->ifa_local;
2581 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2583 struct vlan_entry *vlan;
2585 if (ip == bond->master_ip)
2588 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2589 if (ip == vlan->vlan_ip)
2597 * We go to the (large) trouble of VLAN tagging ARP frames because
2598 * switches in VLAN mode (especially if ports are configured as
2599 * "native" to a VLAN) might not pass non-tagged frames.
2601 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2603 struct sk_buff *skb;
2605 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2606 slave_dev->name, dest_ip, src_ip, vlan_id);
2608 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2609 NULL, slave_dev->dev_addr, NULL);
2612 pr_err("ARP packet allocation failed\n");
2616 skb = vlan_put_tag(skb, vlan_id);
2618 pr_err("failed to insert VLAN tag\n");
2626 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2629 __be32 *targets = bond->params.arp_targets;
2630 struct vlan_entry *vlan;
2631 struct net_device *vlan_dev;
2634 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2637 pr_debug("basa: target %x\n", targets[i]);
2638 if (!bond_vlan_used(bond)) {
2639 pr_debug("basa: empty vlan: arp_send\n");
2640 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2641 bond->master_ip, 0);
2646 * If VLANs are configured, we do a route lookup to
2647 * determine which VLAN interface would be used, so we
2648 * can tag the ARP with the proper VLAN tag.
2650 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2653 if (net_ratelimit()) {
2654 pr_warning("%s: no route to arp_ip_target %pI4\n",
2655 bond->dev->name, &targets[i]);
2661 * This target is not on a VLAN
2663 if (rt->dst.dev == bond->dev) {
2665 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2666 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2667 bond->master_ip, 0);
2672 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2674 vlan_dev = __vlan_find_dev_deep(bond->dev,
2677 if (vlan_dev == rt->dst.dev) {
2678 vlan_id = vlan->vlan_id;
2679 pr_debug("basa: vlan match on %s %d\n",
2680 vlan_dev->name, vlan_id);
2687 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2688 vlan->vlan_ip, vlan_id);
2692 if (net_ratelimit()) {
2693 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2694 bond->dev->name, &targets[i],
2695 rt->dst.dev ? rt->dst.dev->name : "NULL");
2701 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2704 __be32 *targets = bond->params.arp_targets;
2706 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2707 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2708 &sip, &tip, i, &targets[i],
2709 bond_has_this_ip(bond, tip));
2710 if (sip == targets[i]) {
2711 if (bond_has_this_ip(bond, tip))
2712 slave->last_arp_rx = jiffies;
2718 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2719 struct slave *slave)
2722 unsigned char *arp_ptr;
2725 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2728 read_lock(&bond->lock);
2730 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2731 bond->dev->name, skb->dev->name);
2733 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2737 if (arp->ar_hln != bond->dev->addr_len ||
2738 skb->pkt_type == PACKET_OTHERHOST ||
2739 skb->pkt_type == PACKET_LOOPBACK ||
2740 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2741 arp->ar_pro != htons(ETH_P_IP) ||
2745 arp_ptr = (unsigned char *)(arp + 1);
2746 arp_ptr += bond->dev->addr_len;
2747 memcpy(&sip, arp_ptr, 4);
2748 arp_ptr += 4 + bond->dev->addr_len;
2749 memcpy(&tip, arp_ptr, 4);
2751 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2752 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2753 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2757 * Backup slaves won't see the ARP reply, but do come through
2758 * here for each ARP probe (so we swap the sip/tip to validate
2759 * the probe). In a "redundant switch, common router" type of
2760 * configuration, the ARP probe will (hopefully) travel from
2761 * the active, through one switch, the router, then the other
2762 * switch before reaching the backup.
2764 if (bond_is_active_slave(slave))
2765 bond_validate_arp(bond, slave, sip, tip);
2767 bond_validate_arp(bond, slave, tip, sip);
2770 read_unlock(&bond->lock);
2774 * this function is called regularly to monitor each slave's link
2775 * ensuring that traffic is being sent and received when arp monitoring
2776 * is used in load-balancing mode. if the adapter has been dormant, then an
2777 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2778 * arp monitoring in active backup mode.
2780 void bond_loadbalance_arp_mon(struct work_struct *work)
2782 struct bonding *bond = container_of(work, struct bonding,
2784 struct slave *slave, *oldcurrent;
2785 int do_failover = 0;
2789 read_lock(&bond->lock);
2791 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2793 if (bond->kill_timers)
2796 if (bond->slave_cnt == 0)
2799 read_lock(&bond->curr_slave_lock);
2800 oldcurrent = bond->curr_active_slave;
2801 read_unlock(&bond->curr_slave_lock);
2803 /* see if any of the previous devices are up now (i.e. they have
2804 * xmt and rcv traffic). the curr_active_slave does not come into
2805 * the picture unless it is null. also, slave->jiffies is not needed
2806 * here because we send an arp on each slave and give a slave as
2807 * long as it needs to get the tx/rx within the delta.
2808 * TODO: what about up/down delay in arp mode? it wasn't here before
2811 bond_for_each_slave(bond, slave, i) {
2812 unsigned long trans_start = dev_trans_start(slave->dev);
2814 if (slave->link != BOND_LINK_UP) {
2815 if (time_in_range(jiffies,
2816 trans_start - delta_in_ticks,
2817 trans_start + delta_in_ticks) &&
2818 time_in_range(jiffies,
2819 slave->dev->last_rx - delta_in_ticks,
2820 slave->dev->last_rx + delta_in_ticks)) {
2822 slave->link = BOND_LINK_UP;
2823 bond_set_active_slave(slave);
2825 /* primary_slave has no meaning in round-robin
2826 * mode. the window of a slave being up and
2827 * curr_active_slave being null after enslaving
2831 pr_info("%s: link status definitely up for interface %s, ",
2836 pr_info("%s: interface %s is now up\n",
2842 /* slave->link == BOND_LINK_UP */
2844 /* not all switches will respond to an arp request
2845 * when the source ip is 0, so don't take the link down
2846 * if we don't know our ip yet
2848 if (!time_in_range(jiffies,
2849 trans_start - delta_in_ticks,
2850 trans_start + 2 * delta_in_ticks) ||
2851 !time_in_range(jiffies,
2852 slave->dev->last_rx - delta_in_ticks,
2853 slave->dev->last_rx + 2 * delta_in_ticks)) {
2855 slave->link = BOND_LINK_DOWN;
2856 bond_set_backup_slave(slave);
2858 if (slave->link_failure_count < UINT_MAX)
2859 slave->link_failure_count++;
2861 pr_info("%s: interface %s is now down.\n",
2865 if (slave == oldcurrent)
2870 /* note: if switch is in round-robin mode, all links
2871 * must tx arp to ensure all links rx an arp - otherwise
2872 * links may oscillate or not come up at all; if switch is
2873 * in something like xor mode, there is nothing we can
2874 * do - all replies will be rx'ed on same link causing slaves
2875 * to be unstable during low/no traffic periods
2877 if (IS_UP(slave->dev))
2878 bond_arp_send_all(bond, slave);
2883 write_lock_bh(&bond->curr_slave_lock);
2885 bond_select_active_slave(bond);
2887 write_unlock_bh(&bond->curr_slave_lock);
2888 unblock_netpoll_tx();
2892 if (bond->params.arp_interval && !bond->kill_timers)
2893 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2895 read_unlock(&bond->lock);
2899 * Called to inspect slaves for active-backup mode ARP monitor link state
2900 * changes. Sets new_link in slaves to specify what action should take
2901 * place for the slave. Returns 0 if no changes are found, >0 if changes
2902 * to link states must be committed.
2904 * Called with bond->lock held for read.
2906 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2908 struct slave *slave;
2910 unsigned long trans_start;
2912 bond_for_each_slave(bond, slave, i) {
2913 slave->new_link = BOND_LINK_NOCHANGE;
2915 if (slave->link != BOND_LINK_UP) {
2916 if (time_in_range(jiffies,
2917 slave_last_rx(bond, slave) - delta_in_ticks,
2918 slave_last_rx(bond, slave) + delta_in_ticks)) {
2920 slave->new_link = BOND_LINK_UP;
2928 * Give slaves 2*delta after being enslaved or made
2929 * active. This avoids bouncing, as the last receive
2930 * times need a full ARP monitor cycle to be updated.
2932 if (time_in_range(jiffies,
2933 slave->jiffies - delta_in_ticks,
2934 slave->jiffies + 2 * delta_in_ticks))
2938 * Backup slave is down if:
2939 * - No current_arp_slave AND
2940 * - more than 3*delta since last receive AND
2941 * - the bond has an IP address
2943 * Note: a non-null current_arp_slave indicates
2944 * the curr_active_slave went down and we are
2945 * searching for a new one; under this condition
2946 * we only take the curr_active_slave down - this
2947 * gives each slave a chance to tx/rx traffic
2948 * before being taken out
2950 if (!bond_is_active_slave(slave) &&
2951 !bond->current_arp_slave &&
2952 !time_in_range(jiffies,
2953 slave_last_rx(bond, slave) - delta_in_ticks,
2954 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2956 slave->new_link = BOND_LINK_DOWN;
2961 * Active slave is down if:
2962 * - more than 2*delta since transmitting OR
2963 * - (more than 2*delta since receive AND
2964 * the bond has an IP address)
2966 trans_start = dev_trans_start(slave->dev);
2967 if (bond_is_active_slave(slave) &&
2968 (!time_in_range(jiffies,
2969 trans_start - delta_in_ticks,
2970 trans_start + 2 * delta_in_ticks) ||
2971 !time_in_range(jiffies,
2972 slave_last_rx(bond, slave) - delta_in_ticks,
2973 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2975 slave->new_link = BOND_LINK_DOWN;
2984 * Called to commit link state changes noted by inspection step of
2985 * active-backup mode ARP monitor.
2987 * Called with RTNL and bond->lock for read.
2989 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2991 struct slave *slave;
2993 unsigned long trans_start;
2995 bond_for_each_slave(bond, slave, i) {
2996 switch (slave->new_link) {
2997 case BOND_LINK_NOCHANGE:
3001 trans_start = dev_trans_start(slave->dev);
3002 if ((!bond->curr_active_slave &&
3003 time_in_range(jiffies,
3004 trans_start - delta_in_ticks,
3005 trans_start + delta_in_ticks)) ||
3006 bond->curr_active_slave != slave) {
3007 slave->link = BOND_LINK_UP;
3008 bond->current_arp_slave = NULL;
3010 pr_info("%s: link status definitely up for interface %s.\n",
3011 bond->dev->name, slave->dev->name);
3013 if (!bond->curr_active_slave ||
3014 (slave == bond->primary_slave))
3021 case BOND_LINK_DOWN:
3022 if (slave->link_failure_count < UINT_MAX)
3023 slave->link_failure_count++;
3025 slave->link = BOND_LINK_DOWN;
3026 bond_set_slave_inactive_flags(slave);
3028 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3029 bond->dev->name, slave->dev->name);
3031 if (slave == bond->curr_active_slave) {
3032 bond->current_arp_slave = NULL;
3039 pr_err("%s: impossible: new_link %d on slave %s\n",
3040 bond->dev->name, slave->new_link,
3048 write_lock_bh(&bond->curr_slave_lock);
3049 bond_select_active_slave(bond);
3050 write_unlock_bh(&bond->curr_slave_lock);
3051 unblock_netpoll_tx();
3054 bond_set_carrier(bond);
3058 * Send ARP probes for active-backup mode ARP monitor.
3060 * Called with bond->lock held for read.
3062 static void bond_ab_arp_probe(struct bonding *bond)
3064 struct slave *slave;
3067 read_lock(&bond->curr_slave_lock);
3069 if (bond->current_arp_slave && bond->curr_active_slave)
3070 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3071 bond->current_arp_slave->dev->name,
3072 bond->curr_active_slave->dev->name);
3074 if (bond->curr_active_slave) {
3075 bond_arp_send_all(bond, bond->curr_active_slave);
3076 read_unlock(&bond->curr_slave_lock);
3080 read_unlock(&bond->curr_slave_lock);
3082 /* if we don't have a curr_active_slave, search for the next available
3083 * backup slave from the current_arp_slave and make it the candidate
3084 * for becoming the curr_active_slave
3087 if (!bond->current_arp_slave) {
3088 bond->current_arp_slave = bond->first_slave;
3089 if (!bond->current_arp_slave)
3093 bond_set_slave_inactive_flags(bond->current_arp_slave);
3095 /* search for next candidate */
3096 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3097 if (IS_UP(slave->dev)) {
3098 slave->link = BOND_LINK_BACK;
3099 bond_set_slave_active_flags(slave);
3100 bond_arp_send_all(bond, slave);
3101 slave->jiffies = jiffies;
3102 bond->current_arp_slave = slave;
3106 /* if the link state is up at this point, we
3107 * mark it down - this can happen if we have
3108 * simultaneous link failures and
3109 * reselect_active_interface doesn't make this
3110 * one the current slave so it is still marked
3111 * up when it is actually down
3113 if (slave->link == BOND_LINK_UP) {
3114 slave->link = BOND_LINK_DOWN;
3115 if (slave->link_failure_count < UINT_MAX)
3116 slave->link_failure_count++;
3118 bond_set_slave_inactive_flags(slave);
3120 pr_info("%s: backup interface %s is now down.\n",
3121 bond->dev->name, slave->dev->name);
3126 void bond_activebackup_arp_mon(struct work_struct *work)
3128 struct bonding *bond = container_of(work, struct bonding,
3130 bool should_notify_peers = false;
3133 read_lock(&bond->lock);
3135 if (bond->kill_timers)
3138 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3140 if (bond->slave_cnt == 0)
3143 should_notify_peers = bond_should_notify_peers(bond);
3145 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3146 read_unlock(&bond->lock);
3148 read_lock(&bond->lock);
3150 bond_ab_arp_commit(bond, delta_in_ticks);
3152 read_unlock(&bond->lock);
3154 read_lock(&bond->lock);
3157 bond_ab_arp_probe(bond);
3160 if (bond->params.arp_interval && !bond->kill_timers)
3161 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3163 read_unlock(&bond->lock);
3165 if (should_notify_peers) {
3167 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3172 /*-------------------------- netdev event handling --------------------------*/
3175 * Change device name
3177 static int bond_event_changename(struct bonding *bond)
3179 bond_remove_proc_entry(bond);
3180 bond_create_proc_entry(bond);
3182 bond_debug_reregister(bond);
3187 static int bond_master_netdev_event(unsigned long event,
3188 struct net_device *bond_dev)
3190 struct bonding *event_bond = netdev_priv(bond_dev);
3193 case NETDEV_CHANGENAME:
3194 return bond_event_changename(event_bond);
3202 static int bond_slave_netdev_event(unsigned long event,
3203 struct net_device *slave_dev)
3205 struct net_device *bond_dev = slave_dev->master;
3206 struct bonding *bond = netdev_priv(bond_dev);
3209 case NETDEV_UNREGISTER:
3211 if (bond->setup_by_slave)
3212 bond_release_and_destroy(bond_dev, slave_dev);
3214 bond_release(bond_dev, slave_dev);
3218 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3219 struct slave *slave;
3221 slave = bond_get_slave_by_dev(bond, slave_dev);
3223 u32 old_speed = slave->speed;
3224 u8 old_duplex = slave->duplex;
3226 bond_update_speed_duplex(slave);
3228 if (bond_is_lb(bond))
3231 if (old_speed != slave->speed)
3232 bond_3ad_adapter_speed_changed(slave);
3233 if (old_duplex != slave->duplex)
3234 bond_3ad_adapter_duplex_changed(slave);
3241 * ... Or is it this?
3244 case NETDEV_CHANGEMTU:
3246 * TODO: Should slaves be allowed to
3247 * independently alter their MTU? For
3248 * an active-backup bond, slaves need
3249 * not be the same type of device, so
3250 * MTUs may vary. For other modes,
3251 * slaves arguably should have the
3252 * same MTUs. To do this, we'd need to
3253 * take over the slave's change_mtu
3254 * function for the duration of their
3258 case NETDEV_CHANGENAME:
3260 * TODO: handle changing the primary's name
3263 case NETDEV_FEAT_CHANGE:
3264 bond_compute_features(bond);
3274 * bond_netdev_event: handle netdev notifier chain events.
3276 * This function receives events for the netdev chain. The caller (an
3277 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3278 * locks for us to safely manipulate the slave devices (RTNL lock,
3281 static int bond_netdev_event(struct notifier_block *this,
3282 unsigned long event, void *ptr)
3284 struct net_device *event_dev = (struct net_device *)ptr;
3286 pr_debug("event_dev: %s, event: %lx\n",
3287 event_dev ? event_dev->name : "None",
3290 if (!(event_dev->priv_flags & IFF_BONDING))
3293 if (event_dev->flags & IFF_MASTER) {
3294 pr_debug("IFF_MASTER\n");
3295 return bond_master_netdev_event(event, event_dev);
3298 if (event_dev->flags & IFF_SLAVE) {
3299 pr_debug("IFF_SLAVE\n");
3300 return bond_slave_netdev_event(event, event_dev);
3307 * bond_inetaddr_event: handle inetaddr notifier chain events.
3309 * We keep track of device IPs primarily to use as source addresses in
3310 * ARP monitor probes (rather than spewing out broadcasts all the time).
3312 * We track one IP for the main device (if it has one), plus one per VLAN.
3314 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3316 struct in_ifaddr *ifa = ptr;
3317 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3318 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3319 struct bonding *bond;
3320 struct vlan_entry *vlan;
3322 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3323 if (bond->dev == event_dev) {
3326 bond->master_ip = ifa->ifa_local;
3329 bond->master_ip = bond_glean_dev_ip(bond->dev);
3336 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3337 vlan_dev = __vlan_find_dev_deep(bond->dev,
3339 if (vlan_dev == event_dev) {
3342 vlan->vlan_ip = ifa->ifa_local;
3346 bond_glean_dev_ip(vlan_dev);
3357 static struct notifier_block bond_netdev_notifier = {
3358 .notifier_call = bond_netdev_event,
3361 static struct notifier_block bond_inetaddr_notifier = {
3362 .notifier_call = bond_inetaddr_event,
3365 /*---------------------------- Hashing Policies -----------------------------*/
3368 * Hash for the output device based upon layer 2 and layer 3 data. If
3369 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3371 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3373 struct ethhdr *data = (struct ethhdr *)skb->data;
3374 struct iphdr *iph = ip_hdr(skb);
3376 if (skb->protocol == htons(ETH_P_IP)) {
3377 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3378 (data->h_dest[5] ^ data->h_source[5])) % count;
3381 return (data->h_dest[5] ^ data->h_source[5]) % count;
3385 * Hash for the output device based upon layer 3 and layer 4 data. If
3386 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3387 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3389 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3391 struct ethhdr *data = (struct ethhdr *)skb->data;
3392 struct iphdr *iph = ip_hdr(skb);
3393 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3396 if (skb->protocol == htons(ETH_P_IP)) {
3397 if (!ip_is_fragment(iph) &&
3398 (iph->protocol == IPPROTO_TCP ||
3399 iph->protocol == IPPROTO_UDP)) {
3400 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3402 return (layer4_xor ^
3403 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3407 return (data->h_dest[5] ^ data->h_source[5]) % count;
3411 * Hash for the output device based upon layer 2 data
3413 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3415 struct ethhdr *data = (struct ethhdr *)skb->data;
3417 return (data->h_dest[5] ^ data->h_source[5]) % count;
3420 /*-------------------------- Device entry points ----------------------------*/
3422 static int bond_open(struct net_device *bond_dev)
3424 struct bonding *bond = netdev_priv(bond_dev);
3425 struct slave *slave;
3428 bond->kill_timers = 0;
3430 /* reset slave->backup and slave->inactive */
3431 read_lock(&bond->lock);
3432 if (bond->slave_cnt > 0) {
3433 read_lock(&bond->curr_slave_lock);
3434 bond_for_each_slave(bond, slave, i) {
3435 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3436 && (slave != bond->curr_active_slave)) {
3437 bond_set_slave_inactive_flags(slave);
3439 bond_set_slave_active_flags(slave);
3442 read_unlock(&bond->curr_slave_lock);
3444 read_unlock(&bond->lock);
3446 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3448 if (bond_is_lb(bond)) {
3449 /* bond_alb_initialize must be called before the timer
3452 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3453 /* something went wrong - fail the open operation */
3457 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3458 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3461 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3462 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3463 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3466 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3467 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3468 INIT_DELAYED_WORK(&bond->arp_work,
3469 bond_activebackup_arp_mon);
3471 INIT_DELAYED_WORK(&bond->arp_work,
3472 bond_loadbalance_arp_mon);
3474 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3475 if (bond->params.arp_validate)
3476 bond->recv_probe = bond_arp_rcv;
3479 if (bond->params.mode == BOND_MODE_8023AD) {
3480 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3481 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3482 /* register to receive LACPDUs */
3483 bond->recv_probe = bond_3ad_lacpdu_recv;
3484 bond_3ad_initiate_agg_selection(bond, 1);
3490 static int bond_close(struct net_device *bond_dev)
3492 struct bonding *bond = netdev_priv(bond_dev);
3494 write_lock_bh(&bond->lock);
3496 bond->send_peer_notif = 0;
3498 /* signal timers not to re-arm */
3499 bond->kill_timers = 1;
3501 write_unlock_bh(&bond->lock);
3503 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3504 cancel_delayed_work(&bond->mii_work);
3507 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3508 cancel_delayed_work(&bond->arp_work);
3511 switch (bond->params.mode) {
3512 case BOND_MODE_8023AD:
3513 cancel_delayed_work(&bond->ad_work);
3517 cancel_delayed_work(&bond->alb_work);
3523 if (delayed_work_pending(&bond->mcast_work))
3524 cancel_delayed_work(&bond->mcast_work);
3526 if (bond_is_lb(bond)) {
3527 /* Must be called only after all
3528 * slaves have been released
3530 bond_alb_deinitialize(bond);
3532 bond->recv_probe = NULL;
3537 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3538 struct rtnl_link_stats64 *stats)
3540 struct bonding *bond = netdev_priv(bond_dev);
3541 struct rtnl_link_stats64 temp;
3542 struct slave *slave;
3545 memset(stats, 0, sizeof(*stats));
3547 read_lock_bh(&bond->lock);
3549 bond_for_each_slave(bond, slave, i) {
3550 const struct rtnl_link_stats64 *sstats =
3551 dev_get_stats(slave->dev, &temp);
3553 stats->rx_packets += sstats->rx_packets;
3554 stats->rx_bytes += sstats->rx_bytes;
3555 stats->rx_errors += sstats->rx_errors;
3556 stats->rx_dropped += sstats->rx_dropped;
3558 stats->tx_packets += sstats->tx_packets;
3559 stats->tx_bytes += sstats->tx_bytes;
3560 stats->tx_errors += sstats->tx_errors;
3561 stats->tx_dropped += sstats->tx_dropped;
3563 stats->multicast += sstats->multicast;
3564 stats->collisions += sstats->collisions;
3566 stats->rx_length_errors += sstats->rx_length_errors;
3567 stats->rx_over_errors += sstats->rx_over_errors;
3568 stats->rx_crc_errors += sstats->rx_crc_errors;
3569 stats->rx_frame_errors += sstats->rx_frame_errors;
3570 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3571 stats->rx_missed_errors += sstats->rx_missed_errors;
3573 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3574 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3575 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3576 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3577 stats->tx_window_errors += sstats->tx_window_errors;
3580 read_unlock_bh(&bond->lock);
3585 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3587 struct net_device *slave_dev = NULL;
3588 struct ifbond k_binfo;
3589 struct ifbond __user *u_binfo = NULL;
3590 struct ifslave k_sinfo;
3591 struct ifslave __user *u_sinfo = NULL;
3592 struct mii_ioctl_data *mii = NULL;
3595 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3607 * We do this again just in case we were called by SIOCGMIIREG
3608 * instead of SIOCGMIIPHY.
3615 if (mii->reg_num == 1) {
3616 struct bonding *bond = netdev_priv(bond_dev);
3618 read_lock(&bond->lock);
3619 read_lock(&bond->curr_slave_lock);
3620 if (netif_carrier_ok(bond->dev))
3621 mii->val_out = BMSR_LSTATUS;
3623 read_unlock(&bond->curr_slave_lock);
3624 read_unlock(&bond->lock);
3628 case BOND_INFO_QUERY_OLD:
3629 case SIOCBONDINFOQUERY:
3630 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3632 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3635 res = bond_info_query(bond_dev, &k_binfo);
3637 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3641 case BOND_SLAVE_INFO_QUERY_OLD:
3642 case SIOCBONDSLAVEINFOQUERY:
3643 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3645 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3648 res = bond_slave_info_query(bond_dev, &k_sinfo);
3650 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3659 if (!capable(CAP_NET_ADMIN))
3662 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3664 pr_debug("slave_dev=%p:\n", slave_dev);
3669 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3671 case BOND_ENSLAVE_OLD:
3672 case SIOCBONDENSLAVE:
3673 res = bond_enslave(bond_dev, slave_dev);
3675 case BOND_RELEASE_OLD:
3676 case SIOCBONDRELEASE:
3677 res = bond_release(bond_dev, slave_dev);
3679 case BOND_SETHWADDR_OLD:
3680 case SIOCBONDSETHWADDR:
3681 res = bond_sethwaddr(bond_dev, slave_dev);
3683 case BOND_CHANGE_ACTIVE_OLD:
3684 case SIOCBONDCHANGEACTIVE:
3685 res = bond_ioctl_change_active(bond_dev, slave_dev);
3697 static bool bond_addr_in_mc_list(unsigned char *addr,
3698 struct netdev_hw_addr_list *list,
3701 struct netdev_hw_addr *ha;
3703 netdev_hw_addr_list_for_each(ha, list)
3704 if (!memcmp(ha->addr, addr, addrlen))
3710 static void bond_set_multicast_list(struct net_device *bond_dev)
3712 struct bonding *bond = netdev_priv(bond_dev);
3713 struct netdev_hw_addr *ha;
3717 * Do promisc before checking multicast_mode
3719 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3721 * FIXME: Need to handle the error when one of the multi-slaves
3724 bond_set_promiscuity(bond, 1);
3727 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3728 bond_set_promiscuity(bond, -1);
3731 /* set allmulti flag to slaves */
3732 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3734 * FIXME: Need to handle the error when one of the multi-slaves
3737 bond_set_allmulti(bond, 1);
3740 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3741 bond_set_allmulti(bond, -1);
3744 read_lock(&bond->lock);
3746 bond->flags = bond_dev->flags;
3748 /* looking for addresses to add to slaves' mc list */
3749 netdev_for_each_mc_addr(ha, bond_dev) {
3750 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3751 bond_dev->addr_len);
3753 bond_mc_add(bond, ha->addr);
3756 /* looking for addresses to delete from slaves' list */
3757 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3758 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3759 bond_dev->addr_len);
3761 bond_mc_del(bond, ha->addr);
3764 /* save master's multicast list */
3765 __hw_addr_flush(&bond->mc_list);
3766 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3767 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3769 read_unlock(&bond->lock);
3772 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3774 struct bonding *bond = netdev_priv(dev);
3775 struct slave *slave = bond->first_slave;
3778 const struct net_device_ops *slave_ops
3779 = slave->dev->netdev_ops;
3780 if (slave_ops->ndo_neigh_setup)
3781 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3787 * Change the MTU of all of a master's slaves to match the master
3789 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3791 struct bonding *bond = netdev_priv(bond_dev);
3792 struct slave *slave, *stop_at;
3796 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3797 (bond_dev ? bond_dev->name : "None"), new_mtu);
3799 /* Can't hold bond->lock with bh disabled here since
3800 * some base drivers panic. On the other hand we can't
3801 * hold bond->lock without bh disabled because we'll
3802 * deadlock. The only solution is to rely on the fact
3803 * that we're under rtnl_lock here, and the slaves
3804 * list won't change. This doesn't solve the problem
3805 * of setting the slave's MTU while it is
3806 * transmitting, but the assumption is that the base
3807 * driver can handle that.
3809 * TODO: figure out a way to safely iterate the slaves
3810 * list, but without holding a lock around the actual
3811 * call to the base driver.
3814 bond_for_each_slave(bond, slave, i) {
3815 pr_debug("s %p s->p %p c_m %p\n",
3818 slave->dev->netdev_ops->ndo_change_mtu);
3820 res = dev_set_mtu(slave->dev, new_mtu);
3823 /* If we failed to set the slave's mtu to the new value
3824 * we must abort the operation even in ACTIVE_BACKUP
3825 * mode, because if we allow the backup slaves to have
3826 * different mtu values than the active slave we'll
3827 * need to change their mtu when doing a failover. That
3828 * means changing their mtu from timer context, which
3829 * is probably not a good idea.
3831 pr_debug("err %d %s\n", res, slave->dev->name);
3836 bond_dev->mtu = new_mtu;
3841 /* unwind from head to the slave that failed */
3843 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3846 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3848 pr_debug("unwind err %d dev %s\n",
3849 tmp_res, slave->dev->name);
3859 * Note that many devices must be down to change the HW address, and
3860 * downing the master releases all slaves. We can make bonds full of
3861 * bonding devices to test this, however.
3863 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3865 struct bonding *bond = netdev_priv(bond_dev);
3866 struct sockaddr *sa = addr, tmp_sa;
3867 struct slave *slave, *stop_at;
3871 if (bond->params.mode == BOND_MODE_ALB)
3872 return bond_alb_set_mac_address(bond_dev, addr);
3875 pr_debug("bond=%p, name=%s\n",
3876 bond, bond_dev ? bond_dev->name : "None");
3879 * If fail_over_mac is set to active, do nothing and return
3880 * success. Returning an error causes ifenslave to fail.
3882 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3885 if (!is_valid_ether_addr(sa->sa_data))
3886 return -EADDRNOTAVAIL;
3888 /* Can't hold bond->lock with bh disabled here since
3889 * some base drivers panic. On the other hand we can't
3890 * hold bond->lock without bh disabled because we'll
3891 * deadlock. The only solution is to rely on the fact
3892 * that we're under rtnl_lock here, and the slaves
3893 * list won't change. This doesn't solve the problem
3894 * of setting the slave's hw address while it is
3895 * transmitting, but the assumption is that the base
3896 * driver can handle that.
3898 * TODO: figure out a way to safely iterate the slaves
3899 * list, but without holding a lock around the actual
3900 * call to the base driver.
3903 bond_for_each_slave(bond, slave, i) {
3904 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3905 pr_debug("slave %p %s\n", slave, slave->dev->name);
3907 if (slave_ops->ndo_set_mac_address == NULL) {
3909 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3913 res = dev_set_mac_address(slave->dev, addr);
3915 /* TODO: consider downing the slave
3917 * User should expect communications
3918 * breakage anyway until ARP finish
3921 pr_debug("err %d %s\n", res, slave->dev->name);
3927 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3931 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3932 tmp_sa.sa_family = bond_dev->type;
3934 /* unwind from head to the slave that failed */
3936 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3939 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3941 pr_debug("unwind err %d dev %s\n",
3942 tmp_res, slave->dev->name);
3949 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3951 struct bonding *bond = netdev_priv(bond_dev);
3952 struct slave *slave, *start_at;
3953 int i, slave_no, res = 1;
3954 struct iphdr *iph = ip_hdr(skb);
3957 * Start with the curr_active_slave that joined the bond as the
3958 * default for sending IGMP traffic. For failover purposes one
3959 * needs to maintain some consistency for the interface that will
3960 * send the join/membership reports. The curr_active_slave found
3961 * will send all of this type of traffic.
3963 if ((iph->protocol == IPPROTO_IGMP) &&
3964 (skb->protocol == htons(ETH_P_IP))) {
3966 read_lock(&bond->curr_slave_lock);
3967 slave = bond->curr_active_slave;
3968 read_unlock(&bond->curr_slave_lock);
3974 * Concurrent TX may collide on rr_tx_counter; we accept
3975 * that as being rare enough not to justify using an
3978 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3980 bond_for_each_slave(bond, slave, i) {
3988 bond_for_each_slave_from(bond, slave, i, start_at) {
3989 if (IS_UP(slave->dev) &&
3990 (slave->link == BOND_LINK_UP) &&
3991 bond_is_active_slave(slave)) {
3992 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3999 /* no suitable interface, frame not sent */
4003 return NETDEV_TX_OK;
4008 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4009 * the bond has a usable interface.
4011 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4013 struct bonding *bond = netdev_priv(bond_dev);
4016 read_lock(&bond->curr_slave_lock);
4018 if (bond->curr_active_slave)
4019 res = bond_dev_queue_xmit(bond, skb,
4020 bond->curr_active_slave->dev);
4023 /* no suitable interface, frame not sent */
4026 read_unlock(&bond->curr_slave_lock);
4028 return NETDEV_TX_OK;
4032 * In bond_xmit_xor() , we determine the output device by using a pre-
4033 * determined xmit_hash_policy(), If the selected device is not enabled,
4034 * find the next active slave.
4036 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4038 struct bonding *bond = netdev_priv(bond_dev);
4039 struct slave *slave, *start_at;
4044 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4046 bond_for_each_slave(bond, slave, i) {
4054 bond_for_each_slave_from(bond, slave, i, start_at) {
4055 if (IS_UP(slave->dev) &&
4056 (slave->link == BOND_LINK_UP) &&
4057 bond_is_active_slave(slave)) {
4058 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4064 /* no suitable interface, frame not sent */
4068 return NETDEV_TX_OK;
4072 * in broadcast mode, we send everything to all usable interfaces.
4074 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4076 struct bonding *bond = netdev_priv(bond_dev);
4077 struct slave *slave, *start_at;
4078 struct net_device *tx_dev = NULL;
4082 read_lock(&bond->curr_slave_lock);
4083 start_at = bond->curr_active_slave;
4084 read_unlock(&bond->curr_slave_lock);
4089 bond_for_each_slave_from(bond, slave, i, start_at) {
4090 if (IS_UP(slave->dev) &&
4091 (slave->link == BOND_LINK_UP) &&
4092 bond_is_active_slave(slave)) {
4094 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4096 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4101 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4103 dev_kfree_skb(skb2);
4107 tx_dev = slave->dev;
4112 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4116 /* no suitable interface, frame not sent */
4119 /* frame sent to all suitable interfaces */
4120 return NETDEV_TX_OK;
4123 /*------------------------- Device initialization ---------------------------*/
4125 static void bond_set_xmit_hash_policy(struct bonding *bond)
4127 switch (bond->params.xmit_policy) {
4128 case BOND_XMIT_POLICY_LAYER23:
4129 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4131 case BOND_XMIT_POLICY_LAYER34:
4132 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4134 case BOND_XMIT_POLICY_LAYER2:
4136 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4142 * Lookup the slave that corresponds to a qid
4144 static inline int bond_slave_override(struct bonding *bond,
4145 struct sk_buff *skb)
4148 struct slave *slave = NULL;
4149 struct slave *check_slave;
4151 if (!skb->queue_mapping)
4154 /* Find out if any slaves have the same mapping as this skb. */
4155 bond_for_each_slave(bond, check_slave, i) {
4156 if (check_slave->queue_id == skb->queue_mapping) {
4157 slave = check_slave;
4162 /* If the slave isn't UP, use default transmit policy. */
4163 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4164 (slave->link == BOND_LINK_UP)) {
4165 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4172 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4175 * This helper function exists to help dev_pick_tx get the correct
4176 * destination queue. Using a helper function skips a call to
4177 * skb_tx_hash and will put the skbs in the queue we expect on their
4178 * way down to the bonding driver.
4180 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4183 * Save the original txq to restore before passing to the driver
4185 bond_queue_mapping(skb) = skb->queue_mapping;
4187 if (unlikely(txq >= dev->real_num_tx_queues)) {
4189 txq -= dev->real_num_tx_queues;
4190 } while (txq >= dev->real_num_tx_queues);
4195 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4197 struct bonding *bond = netdev_priv(dev);
4199 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4200 if (!bond_slave_override(bond, skb))
4201 return NETDEV_TX_OK;
4204 switch (bond->params.mode) {
4205 case BOND_MODE_ROUNDROBIN:
4206 return bond_xmit_roundrobin(skb, dev);
4207 case BOND_MODE_ACTIVEBACKUP:
4208 return bond_xmit_activebackup(skb, dev);
4210 return bond_xmit_xor(skb, dev);
4211 case BOND_MODE_BROADCAST:
4212 return bond_xmit_broadcast(skb, dev);
4213 case BOND_MODE_8023AD:
4214 return bond_3ad_xmit_xor(skb, dev);
4217 return bond_alb_xmit(skb, dev);
4219 /* Should never happen, mode already checked */
4220 pr_err("%s: Error: Unknown bonding mode %d\n",
4221 dev->name, bond->params.mode);
4224 return NETDEV_TX_OK;
4228 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4230 struct bonding *bond = netdev_priv(dev);
4231 netdev_tx_t ret = NETDEV_TX_OK;
4234 * If we risk deadlock from transmitting this in the
4235 * netpoll path, tell netpoll to queue the frame for later tx
4237 if (is_netpoll_tx_blocked(dev))
4238 return NETDEV_TX_BUSY;
4240 read_lock(&bond->lock);
4242 if (bond->slave_cnt)
4243 ret = __bond_start_xmit(skb, dev);
4247 read_unlock(&bond->lock);
4253 * set bond mode specific net device operations
4255 void bond_set_mode_ops(struct bonding *bond, int mode)
4257 struct net_device *bond_dev = bond->dev;
4260 case BOND_MODE_ROUNDROBIN:
4262 case BOND_MODE_ACTIVEBACKUP:
4265 bond_set_xmit_hash_policy(bond);
4267 case BOND_MODE_BROADCAST:
4269 case BOND_MODE_8023AD:
4270 bond_set_xmit_hash_policy(bond);
4277 /* Should never happen, mode already checked */
4278 pr_err("%s: Error: Unknown bonding mode %d\n",
4279 bond_dev->name, mode);
4284 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4285 struct ethtool_drvinfo *drvinfo)
4287 strncpy(drvinfo->driver, DRV_NAME, 32);
4288 strncpy(drvinfo->version, DRV_VERSION, 32);
4289 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4292 static const struct ethtool_ops bond_ethtool_ops = {
4293 .get_drvinfo = bond_ethtool_get_drvinfo,
4294 .get_link = ethtool_op_get_link,
4297 static const struct net_device_ops bond_netdev_ops = {
4298 .ndo_init = bond_init,
4299 .ndo_uninit = bond_uninit,
4300 .ndo_open = bond_open,
4301 .ndo_stop = bond_close,
4302 .ndo_start_xmit = bond_start_xmit,
4303 .ndo_select_queue = bond_select_queue,
4304 .ndo_get_stats64 = bond_get_stats,
4305 .ndo_do_ioctl = bond_do_ioctl,
4306 .ndo_set_multicast_list = bond_set_multicast_list,
4307 .ndo_change_mtu = bond_change_mtu,
4308 .ndo_set_mac_address = bond_set_mac_address,
4309 .ndo_neigh_setup = bond_neigh_setup,
4310 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4311 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4312 #ifdef CONFIG_NET_POLL_CONTROLLER
4313 .ndo_netpoll_setup = bond_netpoll_setup,
4314 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4315 .ndo_poll_controller = bond_poll_controller,
4317 .ndo_add_slave = bond_enslave,
4318 .ndo_del_slave = bond_release,
4319 .ndo_fix_features = bond_fix_features,
4322 static void bond_destructor(struct net_device *bond_dev)
4324 struct bonding *bond = netdev_priv(bond_dev);
4326 destroy_workqueue(bond->wq);
4327 free_netdev(bond_dev);
4330 static void bond_setup(struct net_device *bond_dev)
4332 struct bonding *bond = netdev_priv(bond_dev);
4334 /* initialize rwlocks */
4335 rwlock_init(&bond->lock);
4336 rwlock_init(&bond->curr_slave_lock);
4338 bond->params = bonding_defaults;
4340 /* Initialize pointers */
4341 bond->dev = bond_dev;
4342 INIT_LIST_HEAD(&bond->vlan_list);
4344 /* Initialize the device entry points */
4345 ether_setup(bond_dev);
4346 bond_dev->netdev_ops = &bond_netdev_ops;
4347 bond_dev->ethtool_ops = &bond_ethtool_ops;
4348 bond_set_mode_ops(bond, bond->params.mode);
4350 bond_dev->destructor = bond_destructor;
4352 /* Initialize the device options */
4353 bond_dev->tx_queue_len = 0;
4354 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4355 bond_dev->priv_flags |= IFF_BONDING;
4356 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4358 /* At first, we block adding VLANs. That's the only way to
4359 * prevent problems that occur when adding VLANs over an
4360 * empty bond. The block will be removed once non-challenged
4361 * slaves are enslaved.
4363 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4365 /* don't acquire bond device's netif_tx_lock when
4367 bond_dev->features |= NETIF_F_LLTX;
4369 /* By default, we declare the bond to be fully
4370 * VLAN hardware accelerated capable. Special
4371 * care is taken in the various xmit functions
4372 * when there are slaves that are not hw accel
4376 bond_dev->hw_features = BOND_VLAN_FEATURES |
4377 NETIF_F_HW_VLAN_TX |
4378 NETIF_F_HW_VLAN_RX |
4379 NETIF_F_HW_VLAN_FILTER;
4381 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4382 bond_dev->features |= bond_dev->hw_features;
4385 static void bond_work_cancel_all(struct bonding *bond)
4387 write_lock_bh(&bond->lock);
4388 bond->kill_timers = 1;
4389 write_unlock_bh(&bond->lock);
4391 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4392 cancel_delayed_work(&bond->mii_work);
4394 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4395 cancel_delayed_work(&bond->arp_work);
4397 if (bond->params.mode == BOND_MODE_ALB &&
4398 delayed_work_pending(&bond->alb_work))
4399 cancel_delayed_work(&bond->alb_work);
4401 if (bond->params.mode == BOND_MODE_8023AD &&
4402 delayed_work_pending(&bond->ad_work))
4403 cancel_delayed_work(&bond->ad_work);
4405 if (delayed_work_pending(&bond->mcast_work))
4406 cancel_delayed_work(&bond->mcast_work);
4410 * Destroy a bonding device.
4411 * Must be under rtnl_lock when this function is called.
4413 static void bond_uninit(struct net_device *bond_dev)
4415 struct bonding *bond = netdev_priv(bond_dev);
4416 struct vlan_entry *vlan, *tmp;
4418 bond_netpoll_cleanup(bond_dev);
4420 /* Release the bonded slaves */
4421 bond_release_all(bond_dev);
4423 list_del(&bond->bond_list);
4425 bond_work_cancel_all(bond);
4427 bond_remove_proc_entry(bond);
4429 bond_debug_unregister(bond);
4431 __hw_addr_flush(&bond->mc_list);
4433 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4434 list_del(&vlan->vlan_list);
4439 /*------------------------- Module initialization ---------------------------*/
4442 * Convert string input module parms. Accept either the
4443 * number of the mode or its string name. A bit complicated because
4444 * some mode names are substrings of other names, and calls from sysfs
4445 * may have whitespace in the name (trailing newlines, for example).
4447 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4449 int modeint = -1, i, rv;
4450 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4452 for (p = (char *)buf; *p; p++)
4453 if (!(isdigit(*p) || isspace(*p)))
4457 rv = sscanf(buf, "%20s", modestr);
4459 rv = sscanf(buf, "%d", &modeint);
4464 for (i = 0; tbl[i].modename; i++) {
4465 if (modeint == tbl[i].mode)
4467 if (strcmp(modestr, tbl[i].modename) == 0)
4474 static int bond_check_params(struct bond_params *params)
4476 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4479 * Convert string parameters.
4482 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4483 if (bond_mode == -1) {
4484 pr_err("Error: Invalid bonding mode \"%s\"\n",
4485 mode == NULL ? "NULL" : mode);
4490 if (xmit_hash_policy) {
4491 if ((bond_mode != BOND_MODE_XOR) &&
4492 (bond_mode != BOND_MODE_8023AD)) {
4493 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4494 bond_mode_name(bond_mode));
4496 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4498 if (xmit_hashtype == -1) {
4499 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4500 xmit_hash_policy == NULL ? "NULL" :
4508 if (bond_mode != BOND_MODE_8023AD) {
4509 pr_info("lacp_rate param is irrelevant in mode %s\n",
4510 bond_mode_name(bond_mode));
4512 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4513 if (lacp_fast == -1) {
4514 pr_err("Error: Invalid lacp rate \"%s\"\n",
4515 lacp_rate == NULL ? "NULL" : lacp_rate);
4522 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4523 if (params->ad_select == -1) {
4524 pr_err("Error: Invalid ad_select \"%s\"\n",
4525 ad_select == NULL ? "NULL" : ad_select);
4529 if (bond_mode != BOND_MODE_8023AD) {
4530 pr_warning("ad_select param only affects 802.3ad mode\n");
4533 params->ad_select = BOND_AD_STABLE;
4536 if (max_bonds < 0) {
4537 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4538 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4539 max_bonds = BOND_DEFAULT_MAX_BONDS;
4543 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4544 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4545 miimon = BOND_LINK_MON_INTERV;
4549 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4554 if (downdelay < 0) {
4555 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4556 downdelay, INT_MAX);
4560 if ((use_carrier != 0) && (use_carrier != 1)) {
4561 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4566 if (num_peer_notif < 0 || num_peer_notif > 255) {
4567 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4572 /* reset values for 802.3ad */
4573 if (bond_mode == BOND_MODE_8023AD) {
4575 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");
4576 pr_warning("Forcing miimon to 100msec\n");
4581 if (tx_queues < 1 || tx_queues > 255) {
4582 pr_warning("Warning: tx_queues (%d) should be between "
4583 "1 and 255, resetting to %d\n",
4584 tx_queues, BOND_DEFAULT_TX_QUEUES);
4585 tx_queues = BOND_DEFAULT_TX_QUEUES;
4588 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4589 pr_warning("Warning: all_slaves_active module parameter (%d), "
4590 "not of valid value (0/1), so it was set to "
4591 "0\n", all_slaves_active);
4592 all_slaves_active = 0;
4595 if (resend_igmp < 0 || resend_igmp > 255) {
4596 pr_warning("Warning: resend_igmp (%d) should be between "
4597 "0 and 255, resetting to %d\n",
4598 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4599 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4602 /* reset values for TLB/ALB */
4603 if ((bond_mode == BOND_MODE_TLB) ||
4604 (bond_mode == BOND_MODE_ALB)) {
4606 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");
4607 pr_warning("Forcing miimon to 100msec\n");
4612 if (bond_mode == BOND_MODE_ALB) {
4613 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",
4618 if (updelay || downdelay) {
4619 /* just warn the user the up/down delay will have
4620 * no effect since miimon is zero...
4622 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",
4623 updelay, downdelay);
4626 /* don't allow arp monitoring */
4628 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4629 miimon, arp_interval);
4633 if ((updelay % miimon) != 0) {
4634 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4636 (updelay / miimon) * miimon);
4641 if ((downdelay % miimon) != 0) {
4642 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4644 (downdelay / miimon) * miimon);
4647 downdelay /= miimon;
4650 if (arp_interval < 0) {
4651 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4652 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4653 arp_interval = BOND_LINK_ARP_INTERV;
4656 for (arp_ip_count = 0;
4657 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4659 /* not complete check, but should be good enough to
4661 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4662 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4663 arp_ip_target[arp_ip_count]);
4666 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4667 arp_target[arp_ip_count] = ip;
4671 if (arp_interval && !arp_ip_count) {
4672 /* don't allow arping if no arp_ip_target given... */
4673 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4679 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4680 pr_err("arp_validate only supported in active-backup mode\n");
4683 if (!arp_interval) {
4684 pr_err("arp_validate requires arp_interval\n");
4688 arp_validate_value = bond_parse_parm(arp_validate,
4690 if (arp_validate_value == -1) {
4691 pr_err("Error: invalid arp_validate \"%s\"\n",
4692 arp_validate == NULL ? "NULL" : arp_validate);
4696 arp_validate_value = 0;
4699 pr_info("MII link monitoring set to %d ms\n", miimon);
4700 } else if (arp_interval) {
4703 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4705 arp_validate_tbl[arp_validate_value].modename,
4708 for (i = 0; i < arp_ip_count; i++)
4709 pr_info(" %s", arp_ip_target[i]);
4713 } else if (max_bonds) {
4714 /* miimon and arp_interval not set, we need one so things
4715 * work as expected, see bonding.txt for details
4717 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");
4720 if (primary && !USES_PRIMARY(bond_mode)) {
4721 /* currently, using a primary only makes sense
4722 * in active backup, TLB or ALB modes
4724 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4725 primary, bond_mode_name(bond_mode));
4729 if (primary && primary_reselect) {
4730 primary_reselect_value = bond_parse_parm(primary_reselect,
4732 if (primary_reselect_value == -1) {
4733 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4735 NULL ? "NULL" : primary_reselect);
4739 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4742 if (fail_over_mac) {
4743 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4745 if (fail_over_mac_value == -1) {
4746 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4747 arp_validate == NULL ? "NULL" : arp_validate);
4751 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4752 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4754 fail_over_mac_value = BOND_FOM_NONE;
4757 /* fill params struct with the proper values */
4758 params->mode = bond_mode;
4759 params->xmit_policy = xmit_hashtype;
4760 params->miimon = miimon;
4761 params->num_peer_notif = num_peer_notif;
4762 params->arp_interval = arp_interval;
4763 params->arp_validate = arp_validate_value;
4764 params->updelay = updelay;
4765 params->downdelay = downdelay;
4766 params->use_carrier = use_carrier;
4767 params->lacp_fast = lacp_fast;
4768 params->primary[0] = 0;
4769 params->primary_reselect = primary_reselect_value;
4770 params->fail_over_mac = fail_over_mac_value;
4771 params->tx_queues = tx_queues;
4772 params->all_slaves_active = all_slaves_active;
4773 params->resend_igmp = resend_igmp;
4774 params->min_links = min_links;
4777 strncpy(params->primary, primary, IFNAMSIZ);
4778 params->primary[IFNAMSIZ - 1] = 0;
4781 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4786 static struct lock_class_key bonding_netdev_xmit_lock_key;
4787 static struct lock_class_key bonding_netdev_addr_lock_key;
4789 static void bond_set_lockdep_class_one(struct net_device *dev,
4790 struct netdev_queue *txq,
4793 lockdep_set_class(&txq->_xmit_lock,
4794 &bonding_netdev_xmit_lock_key);
4797 static void bond_set_lockdep_class(struct net_device *dev)
4799 lockdep_set_class(&dev->addr_list_lock,
4800 &bonding_netdev_addr_lock_key);
4801 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4805 * Called from registration process
4807 static int bond_init(struct net_device *bond_dev)
4809 struct bonding *bond = netdev_priv(bond_dev);
4810 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4811 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4813 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4816 * Initialize locks that may be required during
4817 * en/deslave operations. All of the bond_open work
4818 * (of which this is part) should really be moved to
4819 * a phase prior to dev_open
4821 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4822 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4824 bond->wq = create_singlethread_workqueue(bond_dev->name);
4828 bond_set_lockdep_class(bond_dev);
4830 bond_create_proc_entry(bond);
4831 list_add_tail(&bond->bond_list, &bn->dev_list);
4833 bond_prepare_sysfs_group(bond);
4835 bond_debug_register(bond);
4837 __hw_addr_init(&bond->mc_list);
4841 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4843 if (tb[IFLA_ADDRESS]) {
4844 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4846 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4847 return -EADDRNOTAVAIL;
4852 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4854 .priv_size = sizeof(struct bonding),
4855 .setup = bond_setup,
4856 .validate = bond_validate,
4859 /* Create a new bond based on the specified name and bonding parameters.
4860 * If name is NULL, obtain a suitable "bond%d" name for us.
4861 * Caller must NOT hold rtnl_lock; we need to release it here before we
4862 * set up our sysfs entries.
4864 int bond_create(struct net *net, const char *name)
4866 struct net_device *bond_dev;
4871 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4872 name ? name : "bond%d",
4873 bond_setup, tx_queues);
4875 pr_err("%s: eek! can't alloc netdev!\n", name);
4880 dev_net_set(bond_dev, net);
4881 bond_dev->rtnl_link_ops = &bond_link_ops;
4883 res = register_netdevice(bond_dev);
4885 netif_carrier_off(bond_dev);
4889 bond_destructor(bond_dev);
4893 static int __net_init bond_net_init(struct net *net)
4895 struct bond_net *bn = net_generic(net, bond_net_id);
4898 INIT_LIST_HEAD(&bn->dev_list);
4900 bond_create_proc_dir(bn);
4905 static void __net_exit bond_net_exit(struct net *net)
4907 struct bond_net *bn = net_generic(net, bond_net_id);
4909 bond_destroy_proc_dir(bn);
4912 static struct pernet_operations bond_net_ops = {
4913 .init = bond_net_init,
4914 .exit = bond_net_exit,
4916 .size = sizeof(struct bond_net),
4919 static int __init bonding_init(void)
4924 pr_info("%s", bond_version);
4926 res = bond_check_params(&bonding_defaults);
4930 res = register_pernet_subsys(&bond_net_ops);
4934 res = rtnl_link_register(&bond_link_ops);
4938 bond_create_debugfs();
4940 for (i = 0; i < max_bonds; i++) {
4941 res = bond_create(&init_net, NULL);
4946 res = bond_create_sysfs();
4950 register_netdevice_notifier(&bond_netdev_notifier);
4951 register_inetaddr_notifier(&bond_inetaddr_notifier);
4955 rtnl_link_unregister(&bond_link_ops);
4957 unregister_pernet_subsys(&bond_net_ops);
4962 static void __exit bonding_exit(void)
4964 unregister_netdevice_notifier(&bond_netdev_notifier);
4965 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4967 bond_destroy_sysfs();
4968 bond_destroy_debugfs();
4970 rtnl_link_unregister(&bond_link_ops);
4971 unregister_pernet_subsys(&bond_net_ops);
4973 #ifdef CONFIG_NET_POLL_CONTROLLER
4975 * Make sure we don't have an imbalance on our netpoll blocking
4977 WARN_ON(atomic_read(&netpoll_block_tx));
4981 module_init(bonding_init);
4982 module_exit(bonding_exit);
4983 MODULE_LICENSE("GPL");
4984 MODULE_VERSION(DRV_VERSION);
4985 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4986 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4987 MODULE_ALIAS_RTNL_LINK("bond");