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;
1438 void (*recv_probe)(struct sk_buff *, struct bonding *,
1441 skb = skb_share_check(skb, GFP_ATOMIC);
1443 return RX_HANDLER_CONSUMED;
1447 slave = bond_slave_get_rcu(skb->dev);
1450 if (bond->params.arp_interval)
1451 slave->dev->last_rx = jiffies;
1453 recv_probe = ACCESS_ONCE(bond->recv_probe);
1455 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1458 recv_probe(nskb, bond, slave);
1459 dev_kfree_skb(nskb);
1463 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1464 return RX_HANDLER_EXACT;
1467 skb->dev = bond->dev;
1469 if (bond->params.mode == BOND_MODE_ALB &&
1470 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1471 skb->pkt_type == PACKET_HOST) {
1473 if (unlikely(skb_cow_head(skb,
1474 skb->data - skb_mac_header(skb)))) {
1476 return RX_HANDLER_CONSUMED;
1478 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1481 return RX_HANDLER_ANOTHER;
1484 /* enslave device <slave> to bond device <master> */
1485 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1487 struct bonding *bond = netdev_priv(bond_dev);
1488 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1489 struct slave *new_slave = NULL;
1490 struct netdev_hw_addr *ha;
1491 struct sockaddr addr;
1495 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1496 slave_ops->ndo_do_ioctl == NULL) {
1497 pr_warning("%s: Warning: no link monitoring support for %s\n",
1498 bond_dev->name, slave_dev->name);
1501 /* already enslaved */
1502 if (slave_dev->flags & IFF_SLAVE) {
1503 pr_debug("Error, Device was already enslaved\n");
1507 /* vlan challenged mutual exclusion */
1508 /* no need to lock since we're protected by rtnl_lock */
1509 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1510 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1511 if (bond_vlan_used(bond)) {
1512 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1513 bond_dev->name, slave_dev->name, bond_dev->name);
1516 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1517 bond_dev->name, slave_dev->name,
1518 slave_dev->name, bond_dev->name);
1521 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1525 * Old ifenslave binaries are no longer supported. These can
1526 * be identified with moderate accuracy by the state of the slave:
1527 * the current ifenslave will set the interface down prior to
1528 * enslaving it; the old ifenslave will not.
1530 if ((slave_dev->flags & IFF_UP)) {
1531 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1534 goto err_undo_flags;
1537 /* set bonding device ether type by slave - bonding netdevices are
1538 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1539 * there is a need to override some of the type dependent attribs/funcs.
1541 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1542 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1544 if (bond->slave_cnt == 0) {
1545 if (bond_dev->type != slave_dev->type) {
1546 pr_debug("%s: change device type from %d to %d\n",
1548 bond_dev->type, slave_dev->type);
1550 res = netdev_bonding_change(bond_dev,
1551 NETDEV_PRE_TYPE_CHANGE);
1552 res = notifier_to_errno(res);
1554 pr_err("%s: refused to change device type\n",
1557 goto err_undo_flags;
1560 /* Flush unicast and multicast addresses */
1561 dev_uc_flush(bond_dev);
1562 dev_mc_flush(bond_dev);
1564 if (slave_dev->type != ARPHRD_ETHER)
1565 bond_setup_by_slave(bond_dev, slave_dev);
1567 ether_setup(bond_dev);
1568 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1571 netdev_bonding_change(bond_dev,
1572 NETDEV_POST_TYPE_CHANGE);
1574 } else if (bond_dev->type != slave_dev->type) {
1575 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1577 slave_dev->type, bond_dev->type);
1579 goto err_undo_flags;
1582 if (slave_ops->ndo_set_mac_address == NULL) {
1583 if (bond->slave_cnt == 0) {
1584 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1586 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1587 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1588 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1591 goto err_undo_flags;
1595 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1597 /* If this is the first slave, then we need to set the master's hardware
1598 * address to be the same as the slave's. */
1599 if (is_zero_ether_addr(bond->dev->dev_addr))
1600 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1601 slave_dev->addr_len);
1604 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1607 goto err_undo_flags;
1611 * Set the new_slave's queue_id to be zero. Queue ID mapping
1612 * is set via sysfs or module option if desired.
1614 new_slave->queue_id = 0;
1616 /* Save slave's original mtu and then set it to match the bond */
1617 new_slave->original_mtu = slave_dev->mtu;
1618 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1620 pr_debug("Error %d calling dev_set_mtu\n", res);
1625 * Save slave's original ("permanent") mac address for modes
1626 * that need it, and for restoring it upon release, and then
1627 * set it to the master's address
1629 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1631 if (!bond->params.fail_over_mac) {
1633 * Set slave to master's mac address. The application already
1634 * set the master's mac address to that of the first slave
1636 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1637 addr.sa_family = slave_dev->type;
1638 res = dev_set_mac_address(slave_dev, &addr);
1640 pr_debug("Error %d calling set_mac_address\n", res);
1641 goto err_restore_mtu;
1645 res = netdev_set_bond_master(slave_dev, bond_dev);
1647 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1648 goto err_restore_mac;
1651 /* open the slave since the application closed it */
1652 res = dev_open(slave_dev);
1654 pr_debug("Opening slave %s failed\n", slave_dev->name);
1655 goto err_unset_master;
1658 new_slave->bond = bond;
1659 new_slave->dev = slave_dev;
1660 slave_dev->priv_flags |= IFF_BONDING;
1662 if (bond_is_lb(bond)) {
1663 /* bond_alb_init_slave() must be called before all other stages since
1664 * it might fail and we do not want to have to undo everything
1666 res = bond_alb_init_slave(bond, new_slave);
1671 /* If the mode USES_PRIMARY, then the new slave gets the
1672 * master's promisc (and mc) settings only if it becomes the
1673 * curr_active_slave, and that is taken care of later when calling
1674 * bond_change_active()
1676 if (!USES_PRIMARY(bond->params.mode)) {
1677 /* set promiscuity level to new slave */
1678 if (bond_dev->flags & IFF_PROMISC) {
1679 res = dev_set_promiscuity(slave_dev, 1);
1684 /* set allmulti level to new slave */
1685 if (bond_dev->flags & IFF_ALLMULTI) {
1686 res = dev_set_allmulti(slave_dev, 1);
1691 netif_addr_lock_bh(bond_dev);
1692 /* upload master's mc_list to new slave */
1693 netdev_for_each_mc_addr(ha, bond_dev)
1694 dev_mc_add(slave_dev, ha->addr);
1695 netif_addr_unlock_bh(bond_dev);
1698 if (bond->params.mode == BOND_MODE_8023AD) {
1699 /* add lacpdu mc addr to mc list */
1700 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1702 dev_mc_add(slave_dev, lacpdu_multicast);
1705 bond_add_vlans_on_slave(bond, slave_dev);
1707 write_lock_bh(&bond->lock);
1709 bond_attach_slave(bond, new_slave);
1711 new_slave->delay = 0;
1712 new_slave->link_failure_count = 0;
1714 write_unlock_bh(&bond->lock);
1716 bond_compute_features(bond);
1718 read_lock(&bond->lock);
1720 new_slave->last_arp_rx = jiffies;
1722 if (bond->params.miimon && !bond->params.use_carrier) {
1723 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1725 if ((link_reporting == -1) && !bond->params.arp_interval) {
1727 * miimon is set but a bonded network driver
1728 * does not support ETHTOOL/MII and
1729 * arp_interval is not set. Note: if
1730 * use_carrier is enabled, we will never go
1731 * here (because netif_carrier is always
1732 * supported); thus, we don't need to change
1733 * the messages for netif_carrier.
1735 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1736 bond_dev->name, slave_dev->name);
1737 } else if (link_reporting == -1) {
1738 /* unable get link status using mii/ethtool */
1739 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1740 bond_dev->name, slave_dev->name);
1744 /* check for initial state */
1745 if (!bond->params.miimon ||
1746 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1747 if (bond->params.updelay) {
1748 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1749 new_slave->link = BOND_LINK_BACK;
1750 new_slave->delay = bond->params.updelay;
1752 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1753 new_slave->link = BOND_LINK_UP;
1755 new_slave->jiffies = jiffies;
1757 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1758 new_slave->link = BOND_LINK_DOWN;
1761 if (bond_update_speed_duplex(new_slave) &&
1762 (new_slave->link != BOND_LINK_DOWN)) {
1763 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1764 bond_dev->name, new_slave->dev->name);
1766 if (bond->params.mode == BOND_MODE_8023AD) {
1767 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1772 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1773 /* if there is a primary slave, remember it */
1774 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1775 bond->primary_slave = new_slave;
1776 bond->force_primary = true;
1780 write_lock_bh(&bond->curr_slave_lock);
1782 switch (bond->params.mode) {
1783 case BOND_MODE_ACTIVEBACKUP:
1784 bond_set_slave_inactive_flags(new_slave);
1785 bond_select_active_slave(bond);
1787 case BOND_MODE_8023AD:
1788 /* in 802.3ad mode, the internal mechanism
1789 * will activate the slaves in the selected
1792 bond_set_slave_inactive_flags(new_slave);
1793 /* if this is the first slave */
1794 if (bond->slave_cnt == 1) {
1795 SLAVE_AD_INFO(new_slave).id = 1;
1796 /* Initialize AD with the number of times that the AD timer is called in 1 second
1797 * can be called only after the mac address of the bond is set
1799 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1801 SLAVE_AD_INFO(new_slave).id =
1802 SLAVE_AD_INFO(new_slave->prev).id + 1;
1805 bond_3ad_bind_slave(new_slave);
1809 bond_set_active_slave(new_slave);
1810 bond_set_slave_inactive_flags(new_slave);
1811 bond_select_active_slave(bond);
1814 pr_debug("This slave is always active in trunk mode\n");
1816 /* always active in trunk mode */
1817 bond_set_active_slave(new_slave);
1819 /* In trunking mode there is little meaning to curr_active_slave
1820 * anyway (it holds no special properties of the bond device),
1821 * so we can change it without calling change_active_interface()
1823 if (!bond->curr_active_slave)
1824 bond->curr_active_slave = new_slave;
1827 } /* switch(bond_mode) */
1829 write_unlock_bh(&bond->curr_slave_lock);
1831 bond_set_carrier(bond);
1833 #ifdef CONFIG_NET_POLL_CONTROLLER
1834 slave_dev->npinfo = bond_netpoll_info(bond);
1835 if (slave_dev->npinfo) {
1836 if (slave_enable_netpoll(new_slave)) {
1837 read_unlock(&bond->lock);
1838 pr_info("Error, %s: master_dev is using netpoll, "
1839 "but new slave device does not support netpoll.\n",
1847 read_unlock(&bond->lock);
1849 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1853 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1856 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1857 goto err_dest_symlinks;
1860 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1861 bond_dev->name, slave_dev->name,
1862 bond_is_active_slave(new_slave) ? "n active" : " backup",
1863 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1865 /* enslave is successful */
1868 /* Undo stages on error */
1870 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1873 dev_close(slave_dev);
1876 netdev_set_bond_master(slave_dev, NULL);
1879 if (!bond->params.fail_over_mac) {
1880 /* XXX TODO - fom follow mode needs to change master's
1881 * MAC if this slave's MAC is in use by the bond, or at
1882 * least print a warning.
1884 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1885 addr.sa_family = slave_dev->type;
1886 dev_set_mac_address(slave_dev, &addr);
1890 dev_set_mtu(slave_dev, new_slave->original_mtu);
1896 bond_compute_features(bond);
1902 * Try to release the slave device <slave> from the bond device <master>
1903 * It is legal to access curr_active_slave without a lock because all the function
1906 * The rules for slave state should be:
1907 * for Active/Backup:
1908 * Active stays on all backups go down
1909 * for Bonded connections:
1910 * The first up interface should be left on and all others downed.
1912 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1914 struct bonding *bond = netdev_priv(bond_dev);
1915 struct slave *slave, *oldcurrent;
1916 struct sockaddr addr;
1917 u32 old_features = bond_dev->features;
1919 /* slave is not a slave or master is not master of this slave */
1920 if (!(slave_dev->flags & IFF_SLAVE) ||
1921 (slave_dev->master != bond_dev)) {
1922 pr_err("%s: Error: cannot release %s.\n",
1923 bond_dev->name, slave_dev->name);
1928 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1929 write_lock_bh(&bond->lock);
1931 slave = bond_get_slave_by_dev(bond, slave_dev);
1933 /* not a slave of this bond */
1934 pr_info("%s: %s not enslaved\n",
1935 bond_dev->name, slave_dev->name);
1936 write_unlock_bh(&bond->lock);
1937 unblock_netpoll_tx();
1941 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1942 * for this slave anymore.
1944 netdev_rx_handler_unregister(slave_dev);
1945 write_unlock_bh(&bond->lock);
1947 write_lock_bh(&bond->lock);
1949 if (!bond->params.fail_over_mac) {
1950 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1951 bond->slave_cnt > 1)
1952 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1953 bond_dev->name, slave_dev->name,
1955 bond_dev->name, slave_dev->name);
1958 /* Inform AD package of unbinding of slave. */
1959 if (bond->params.mode == BOND_MODE_8023AD) {
1960 /* must be called before the slave is
1961 * detached from the list
1963 bond_3ad_unbind_slave(slave);
1966 pr_info("%s: releasing %s interface %s\n",
1968 bond_is_active_slave(slave) ? "active" : "backup",
1971 oldcurrent = bond->curr_active_slave;
1973 bond->current_arp_slave = NULL;
1975 /* release the slave from its bond */
1976 bond_detach_slave(bond, slave);
1978 if (bond->primary_slave == slave)
1979 bond->primary_slave = NULL;
1981 if (oldcurrent == slave)
1982 bond_change_active_slave(bond, NULL);
1984 if (bond_is_lb(bond)) {
1985 /* Must be called only after the slave has been
1986 * detached from the list and the curr_active_slave
1987 * has been cleared (if our_slave == old_current),
1988 * but before a new active slave is selected.
1990 write_unlock_bh(&bond->lock);
1991 bond_alb_deinit_slave(bond, slave);
1992 write_lock_bh(&bond->lock);
1995 if (oldcurrent == slave) {
1997 * Note that we hold RTNL over this sequence, so there
1998 * is no concern that another slave add/remove event
2001 write_unlock_bh(&bond->lock);
2002 read_lock(&bond->lock);
2003 write_lock_bh(&bond->curr_slave_lock);
2005 bond_select_active_slave(bond);
2007 write_unlock_bh(&bond->curr_slave_lock);
2008 read_unlock(&bond->lock);
2009 write_lock_bh(&bond->lock);
2012 if (bond->slave_cnt == 0) {
2013 bond_set_carrier(bond);
2015 /* if the last slave was removed, zero the mac address
2016 * of the master so it will be set by the application
2017 * to the mac address of the first slave
2019 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2021 if (bond_vlan_used(bond)) {
2022 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2023 bond_dev->name, bond_dev->name);
2024 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2029 write_unlock_bh(&bond->lock);
2030 unblock_netpoll_tx();
2032 bond_compute_features(bond);
2033 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2034 (old_features & NETIF_F_VLAN_CHALLENGED))
2035 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2036 bond_dev->name, slave_dev->name, bond_dev->name);
2038 /* must do this from outside any spinlocks */
2039 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2041 bond_del_vlans_from_slave(bond, slave_dev);
2043 /* If the mode USES_PRIMARY, then we should only remove its
2044 * promisc and mc settings if it was the curr_active_slave, but that was
2045 * already taken care of above when we detached the slave
2047 if (!USES_PRIMARY(bond->params.mode)) {
2048 /* unset promiscuity level from slave */
2049 if (bond_dev->flags & IFF_PROMISC)
2050 dev_set_promiscuity(slave_dev, -1);
2052 /* unset allmulti level from slave */
2053 if (bond_dev->flags & IFF_ALLMULTI)
2054 dev_set_allmulti(slave_dev, -1);
2056 /* flush master's mc_list from slave */
2057 netif_addr_lock_bh(bond_dev);
2058 bond_mc_list_flush(bond_dev, slave_dev);
2059 netif_addr_unlock_bh(bond_dev);
2062 netdev_set_bond_master(slave_dev, NULL);
2064 slave_disable_netpoll(slave);
2066 /* close slave before restoring its mac address */
2067 dev_close(slave_dev);
2069 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2070 /* restore original ("permanent") mac address */
2071 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2072 addr.sa_family = slave_dev->type;
2073 dev_set_mac_address(slave_dev, &addr);
2076 dev_set_mtu(slave_dev, slave->original_mtu);
2078 slave_dev->priv_flags &= ~IFF_BONDING;
2082 return 0; /* deletion OK */
2086 * First release a slave and then destroy the bond if no more slaves are left.
2087 * Must be under rtnl_lock when this function is called.
2089 static int bond_release_and_destroy(struct net_device *bond_dev,
2090 struct net_device *slave_dev)
2092 struct bonding *bond = netdev_priv(bond_dev);
2095 ret = bond_release(bond_dev, slave_dev);
2096 if ((ret == 0) && (bond->slave_cnt == 0)) {
2097 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2098 pr_info("%s: destroying bond %s.\n",
2099 bond_dev->name, bond_dev->name);
2100 unregister_netdevice(bond_dev);
2106 * This function releases all slaves.
2108 static int bond_release_all(struct net_device *bond_dev)
2110 struct bonding *bond = netdev_priv(bond_dev);
2111 struct slave *slave;
2112 struct net_device *slave_dev;
2113 struct sockaddr addr;
2115 write_lock_bh(&bond->lock);
2117 netif_carrier_off(bond_dev);
2119 if (bond->slave_cnt == 0)
2122 bond->current_arp_slave = NULL;
2123 bond->primary_slave = NULL;
2124 bond_change_active_slave(bond, NULL);
2126 while ((slave = bond->first_slave) != NULL) {
2127 /* Inform AD package of unbinding of slave
2128 * before slave is detached from the list.
2130 if (bond->params.mode == BOND_MODE_8023AD)
2131 bond_3ad_unbind_slave(slave);
2133 slave_dev = slave->dev;
2134 bond_detach_slave(bond, slave);
2136 /* now that the slave is detached, unlock and perform
2137 * all the undo steps that should not be called from
2140 write_unlock_bh(&bond->lock);
2142 /* unregister rx_handler early so bond_handle_frame wouldn't
2143 * be called for this slave anymore.
2145 netdev_rx_handler_unregister(slave_dev);
2148 if (bond_is_lb(bond)) {
2149 /* must be called only after the slave
2150 * has been detached from the list
2152 bond_alb_deinit_slave(bond, slave);
2155 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2156 bond_del_vlans_from_slave(bond, slave_dev);
2158 /* If the mode USES_PRIMARY, then we should only remove its
2159 * promisc and mc settings if it was the curr_active_slave, but that was
2160 * already taken care of above when we detached the slave
2162 if (!USES_PRIMARY(bond->params.mode)) {
2163 /* unset promiscuity level from slave */
2164 if (bond_dev->flags & IFF_PROMISC)
2165 dev_set_promiscuity(slave_dev, -1);
2167 /* unset allmulti level from slave */
2168 if (bond_dev->flags & IFF_ALLMULTI)
2169 dev_set_allmulti(slave_dev, -1);
2171 /* flush master's mc_list from slave */
2172 netif_addr_lock_bh(bond_dev);
2173 bond_mc_list_flush(bond_dev, slave_dev);
2174 netif_addr_unlock_bh(bond_dev);
2177 netdev_set_bond_master(slave_dev, NULL);
2179 slave_disable_netpoll(slave);
2181 /* close slave before restoring its mac address */
2182 dev_close(slave_dev);
2184 if (!bond->params.fail_over_mac) {
2185 /* restore original ("permanent") mac address*/
2186 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2187 addr.sa_family = slave_dev->type;
2188 dev_set_mac_address(slave_dev, &addr);
2193 /* re-acquire the lock before getting the next slave */
2194 write_lock_bh(&bond->lock);
2197 /* zero the mac address of the master so it will be
2198 * set by the application to the mac address of the
2201 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2203 if (bond_vlan_used(bond)) {
2204 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2205 bond_dev->name, bond_dev->name);
2206 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2210 pr_info("%s: released all slaves\n", bond_dev->name);
2213 write_unlock_bh(&bond->lock);
2215 bond_compute_features(bond);
2221 * This function changes the active slave to slave <slave_dev>.
2222 * It returns -EINVAL in the following cases.
2223 * - <slave_dev> is not found in the list.
2224 * - There is not active slave now.
2225 * - <slave_dev> is already active.
2226 * - The link state of <slave_dev> is not BOND_LINK_UP.
2227 * - <slave_dev> is not running.
2228 * In these cases, this function does nothing.
2229 * In the other cases, current_slave pointer is changed and 0 is returned.
2231 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2233 struct bonding *bond = netdev_priv(bond_dev);
2234 struct slave *old_active = NULL;
2235 struct slave *new_active = NULL;
2238 if (!USES_PRIMARY(bond->params.mode))
2241 /* Verify that master_dev is indeed the master of slave_dev */
2242 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2245 read_lock(&bond->lock);
2247 read_lock(&bond->curr_slave_lock);
2248 old_active = bond->curr_active_slave;
2249 read_unlock(&bond->curr_slave_lock);
2251 new_active = bond_get_slave_by_dev(bond, slave_dev);
2254 * Changing to the current active: do nothing; return success.
2256 if (new_active && (new_active == old_active)) {
2257 read_unlock(&bond->lock);
2263 (new_active->link == BOND_LINK_UP) &&
2264 IS_UP(new_active->dev)) {
2266 write_lock_bh(&bond->curr_slave_lock);
2267 bond_change_active_slave(bond, new_active);
2268 write_unlock_bh(&bond->curr_slave_lock);
2269 unblock_netpoll_tx();
2273 read_unlock(&bond->lock);
2278 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2280 struct bonding *bond = netdev_priv(bond_dev);
2282 info->bond_mode = bond->params.mode;
2283 info->miimon = bond->params.miimon;
2285 read_lock(&bond->lock);
2286 info->num_slaves = bond->slave_cnt;
2287 read_unlock(&bond->lock);
2292 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2294 struct bonding *bond = netdev_priv(bond_dev);
2295 struct slave *slave;
2296 int i, res = -ENODEV;
2298 read_lock(&bond->lock);
2300 bond_for_each_slave(bond, slave, i) {
2301 if (i == (int)info->slave_id) {
2303 strcpy(info->slave_name, slave->dev->name);
2304 info->link = slave->link;
2305 info->state = bond_slave_state(slave);
2306 info->link_failure_count = slave->link_failure_count;
2311 read_unlock(&bond->lock);
2316 /*-------------------------------- Monitoring -------------------------------*/
2319 static int bond_miimon_inspect(struct bonding *bond)
2321 struct slave *slave;
2322 int i, link_state, commit = 0;
2323 bool ignore_updelay;
2325 ignore_updelay = !bond->curr_active_slave ? true : false;
2327 bond_for_each_slave(bond, slave, i) {
2328 slave->new_link = BOND_LINK_NOCHANGE;
2330 link_state = bond_check_dev_link(bond, slave->dev, 0);
2332 switch (slave->link) {
2337 slave->link = BOND_LINK_FAIL;
2338 slave->delay = bond->params.downdelay;
2340 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2342 (bond->params.mode ==
2343 BOND_MODE_ACTIVEBACKUP) ?
2344 (bond_is_active_slave(slave) ?
2345 "active " : "backup ") : "",
2347 bond->params.downdelay * bond->params.miimon);
2350 case BOND_LINK_FAIL:
2353 * recovered before downdelay expired
2355 slave->link = BOND_LINK_UP;
2356 slave->jiffies = jiffies;
2357 pr_info("%s: link status up again after %d ms for interface %s.\n",
2359 (bond->params.downdelay - slave->delay) *
2360 bond->params.miimon,
2365 if (slave->delay <= 0) {
2366 slave->new_link = BOND_LINK_DOWN;
2374 case BOND_LINK_DOWN:
2378 slave->link = BOND_LINK_BACK;
2379 slave->delay = bond->params.updelay;
2382 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2383 bond->dev->name, slave->dev->name,
2384 ignore_updelay ? 0 :
2385 bond->params.updelay *
2386 bond->params.miimon);
2389 case BOND_LINK_BACK:
2391 slave->link = BOND_LINK_DOWN;
2392 pr_info("%s: link status down again after %d ms for interface %s.\n",
2394 (bond->params.updelay - slave->delay) *
2395 bond->params.miimon,
2404 if (slave->delay <= 0) {
2405 slave->new_link = BOND_LINK_UP;
2407 ignore_updelay = false;
2419 static void bond_miimon_commit(struct bonding *bond)
2421 struct slave *slave;
2424 bond_for_each_slave(bond, slave, i) {
2425 switch (slave->new_link) {
2426 case BOND_LINK_NOCHANGE:
2430 slave->link = BOND_LINK_UP;
2431 slave->jiffies = jiffies;
2433 if (bond->params.mode == BOND_MODE_8023AD) {
2434 /* prevent it from being the active one */
2435 bond_set_backup_slave(slave);
2436 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2437 /* make it immediately active */
2438 bond_set_active_slave(slave);
2439 } else if (slave != bond->primary_slave) {
2440 /* prevent it from being the active one */
2441 bond_set_backup_slave(slave);
2444 bond_update_speed_duplex(slave);
2446 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2447 bond->dev->name, slave->dev->name,
2448 slave->speed, slave->duplex ? "full" : "half");
2450 /* notify ad that the link status has changed */
2451 if (bond->params.mode == BOND_MODE_8023AD)
2452 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2454 if (bond_is_lb(bond))
2455 bond_alb_handle_link_change(bond, slave,
2458 if (!bond->curr_active_slave ||
2459 (slave == bond->primary_slave))
2464 case BOND_LINK_DOWN:
2465 if (slave->link_failure_count < UINT_MAX)
2466 slave->link_failure_count++;
2468 slave->link = BOND_LINK_DOWN;
2470 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2471 bond->params.mode == BOND_MODE_8023AD)
2472 bond_set_slave_inactive_flags(slave);
2474 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2475 bond->dev->name, slave->dev->name);
2477 if (bond->params.mode == BOND_MODE_8023AD)
2478 bond_3ad_handle_link_change(slave,
2481 if (bond_is_lb(bond))
2482 bond_alb_handle_link_change(bond, slave,
2485 if (slave == bond->curr_active_slave)
2491 pr_err("%s: invalid new link %d on slave %s\n",
2492 bond->dev->name, slave->new_link,
2494 slave->new_link = BOND_LINK_NOCHANGE;
2502 write_lock_bh(&bond->curr_slave_lock);
2503 bond_select_active_slave(bond);
2504 write_unlock_bh(&bond->curr_slave_lock);
2505 unblock_netpoll_tx();
2508 bond_set_carrier(bond);
2514 * Really a wrapper that splits the mii monitor into two phases: an
2515 * inspection, then (if inspection indicates something needs to be done)
2516 * an acquisition of appropriate locks followed by a commit phase to
2517 * implement whatever link state changes are indicated.
2519 void bond_mii_monitor(struct work_struct *work)
2521 struct bonding *bond = container_of(work, struct bonding,
2523 bool should_notify_peers = false;
2525 read_lock(&bond->lock);
2526 if (bond->kill_timers)
2529 if (bond->slave_cnt == 0)
2532 should_notify_peers = bond_should_notify_peers(bond);
2534 if (bond_miimon_inspect(bond)) {
2535 read_unlock(&bond->lock);
2537 read_lock(&bond->lock);
2539 bond_miimon_commit(bond);
2541 read_unlock(&bond->lock);
2542 rtnl_unlock(); /* might sleep, hold no other locks */
2543 read_lock(&bond->lock);
2547 if (bond->params.miimon && !bond->kill_timers)
2548 queue_delayed_work(bond->wq, &bond->mii_work,
2549 msecs_to_jiffies(bond->params.miimon));
2551 read_unlock(&bond->lock);
2553 if (should_notify_peers) {
2555 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2560 static __be32 bond_glean_dev_ip(struct net_device *dev)
2562 struct in_device *idev;
2563 struct in_ifaddr *ifa;
2570 idev = __in_dev_get_rcu(dev);
2574 ifa = idev->ifa_list;
2578 addr = ifa->ifa_local;
2584 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2586 struct vlan_entry *vlan;
2588 if (ip == bond->master_ip)
2591 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2592 if (ip == vlan->vlan_ip)
2600 * We go to the (large) trouble of VLAN tagging ARP frames because
2601 * switches in VLAN mode (especially if ports are configured as
2602 * "native" to a VLAN) might not pass non-tagged frames.
2604 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2606 struct sk_buff *skb;
2608 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2609 slave_dev->name, dest_ip, src_ip, vlan_id);
2611 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2612 NULL, slave_dev->dev_addr, NULL);
2615 pr_err("ARP packet allocation failed\n");
2619 skb = vlan_put_tag(skb, vlan_id);
2621 pr_err("failed to insert VLAN tag\n");
2629 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2632 __be32 *targets = bond->params.arp_targets;
2633 struct vlan_entry *vlan;
2634 struct net_device *vlan_dev;
2637 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2640 pr_debug("basa: target %x\n", targets[i]);
2641 if (!bond_vlan_used(bond)) {
2642 pr_debug("basa: empty vlan: arp_send\n");
2643 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2644 bond->master_ip, 0);
2649 * If VLANs are configured, we do a route lookup to
2650 * determine which VLAN interface would be used, so we
2651 * can tag the ARP with the proper VLAN tag.
2653 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2656 if (net_ratelimit()) {
2657 pr_warning("%s: no route to arp_ip_target %pI4\n",
2658 bond->dev->name, &targets[i]);
2664 * This target is not on a VLAN
2666 if (rt->dst.dev == bond->dev) {
2668 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2669 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2670 bond->master_ip, 0);
2675 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2677 vlan_dev = __vlan_find_dev_deep(bond->dev,
2680 if (vlan_dev == rt->dst.dev) {
2681 vlan_id = vlan->vlan_id;
2682 pr_debug("basa: vlan match on %s %d\n",
2683 vlan_dev->name, vlan_id);
2690 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2691 vlan->vlan_ip, vlan_id);
2695 if (net_ratelimit()) {
2696 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2697 bond->dev->name, &targets[i],
2698 rt->dst.dev ? rt->dst.dev->name : "NULL");
2704 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2707 __be32 *targets = bond->params.arp_targets;
2709 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2710 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2711 &sip, &tip, i, &targets[i],
2712 bond_has_this_ip(bond, tip));
2713 if (sip == targets[i]) {
2714 if (bond_has_this_ip(bond, tip))
2715 slave->last_arp_rx = jiffies;
2721 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2722 struct slave *slave)
2725 unsigned char *arp_ptr;
2728 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2731 read_lock(&bond->lock);
2733 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2734 bond->dev->name, skb->dev->name);
2736 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2740 if (arp->ar_hln != bond->dev->addr_len ||
2741 skb->pkt_type == PACKET_OTHERHOST ||
2742 skb->pkt_type == PACKET_LOOPBACK ||
2743 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2744 arp->ar_pro != htons(ETH_P_IP) ||
2748 arp_ptr = (unsigned char *)(arp + 1);
2749 arp_ptr += bond->dev->addr_len;
2750 memcpy(&sip, arp_ptr, 4);
2751 arp_ptr += 4 + bond->dev->addr_len;
2752 memcpy(&tip, arp_ptr, 4);
2754 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2755 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2756 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2760 * Backup slaves won't see the ARP reply, but do come through
2761 * here for each ARP probe (so we swap the sip/tip to validate
2762 * the probe). In a "redundant switch, common router" type of
2763 * configuration, the ARP probe will (hopefully) travel from
2764 * the active, through one switch, the router, then the other
2765 * switch before reaching the backup.
2767 if (bond_is_active_slave(slave))
2768 bond_validate_arp(bond, slave, sip, tip);
2770 bond_validate_arp(bond, slave, tip, sip);
2773 read_unlock(&bond->lock);
2777 * this function is called regularly to monitor each slave's link
2778 * ensuring that traffic is being sent and received when arp monitoring
2779 * is used in load-balancing mode. if the adapter has been dormant, then an
2780 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2781 * arp monitoring in active backup mode.
2783 void bond_loadbalance_arp_mon(struct work_struct *work)
2785 struct bonding *bond = container_of(work, struct bonding,
2787 struct slave *slave, *oldcurrent;
2788 int do_failover = 0;
2792 read_lock(&bond->lock);
2794 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2796 if (bond->kill_timers)
2799 if (bond->slave_cnt == 0)
2802 read_lock(&bond->curr_slave_lock);
2803 oldcurrent = bond->curr_active_slave;
2804 read_unlock(&bond->curr_slave_lock);
2806 /* see if any of the previous devices are up now (i.e. they have
2807 * xmt and rcv traffic). the curr_active_slave does not come into
2808 * the picture unless it is null. also, slave->jiffies is not needed
2809 * here because we send an arp on each slave and give a slave as
2810 * long as it needs to get the tx/rx within the delta.
2811 * TODO: what about up/down delay in arp mode? it wasn't here before
2814 bond_for_each_slave(bond, slave, i) {
2815 unsigned long trans_start = dev_trans_start(slave->dev);
2817 if (slave->link != BOND_LINK_UP) {
2818 if (time_in_range(jiffies,
2819 trans_start - delta_in_ticks,
2820 trans_start + delta_in_ticks) &&
2821 time_in_range(jiffies,
2822 slave->dev->last_rx - delta_in_ticks,
2823 slave->dev->last_rx + delta_in_ticks)) {
2825 slave->link = BOND_LINK_UP;
2826 bond_set_active_slave(slave);
2828 /* primary_slave has no meaning in round-robin
2829 * mode. the window of a slave being up and
2830 * curr_active_slave being null after enslaving
2834 pr_info("%s: link status definitely up for interface %s, ",
2839 pr_info("%s: interface %s is now up\n",
2845 /* slave->link == BOND_LINK_UP */
2847 /* not all switches will respond to an arp request
2848 * when the source ip is 0, so don't take the link down
2849 * if we don't know our ip yet
2851 if (!time_in_range(jiffies,
2852 trans_start - delta_in_ticks,
2853 trans_start + 2 * delta_in_ticks) ||
2854 !time_in_range(jiffies,
2855 slave->dev->last_rx - delta_in_ticks,
2856 slave->dev->last_rx + 2 * delta_in_ticks)) {
2858 slave->link = BOND_LINK_DOWN;
2859 bond_set_backup_slave(slave);
2861 if (slave->link_failure_count < UINT_MAX)
2862 slave->link_failure_count++;
2864 pr_info("%s: interface %s is now down.\n",
2868 if (slave == oldcurrent)
2873 /* note: if switch is in round-robin mode, all links
2874 * must tx arp to ensure all links rx an arp - otherwise
2875 * links may oscillate or not come up at all; if switch is
2876 * in something like xor mode, there is nothing we can
2877 * do - all replies will be rx'ed on same link causing slaves
2878 * to be unstable during low/no traffic periods
2880 if (IS_UP(slave->dev))
2881 bond_arp_send_all(bond, slave);
2886 write_lock_bh(&bond->curr_slave_lock);
2888 bond_select_active_slave(bond);
2890 write_unlock_bh(&bond->curr_slave_lock);
2891 unblock_netpoll_tx();
2895 if (bond->params.arp_interval && !bond->kill_timers)
2896 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2898 read_unlock(&bond->lock);
2902 * Called to inspect slaves for active-backup mode ARP monitor link state
2903 * changes. Sets new_link in slaves to specify what action should take
2904 * place for the slave. Returns 0 if no changes are found, >0 if changes
2905 * to link states must be committed.
2907 * Called with bond->lock held for read.
2909 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2911 struct slave *slave;
2913 unsigned long trans_start;
2915 bond_for_each_slave(bond, slave, i) {
2916 slave->new_link = BOND_LINK_NOCHANGE;
2918 if (slave->link != BOND_LINK_UP) {
2919 if (time_in_range(jiffies,
2920 slave_last_rx(bond, slave) - delta_in_ticks,
2921 slave_last_rx(bond, slave) + delta_in_ticks)) {
2923 slave->new_link = BOND_LINK_UP;
2931 * Give slaves 2*delta after being enslaved or made
2932 * active. This avoids bouncing, as the last receive
2933 * times need a full ARP monitor cycle to be updated.
2935 if (time_in_range(jiffies,
2936 slave->jiffies - delta_in_ticks,
2937 slave->jiffies + 2 * delta_in_ticks))
2941 * Backup slave is down if:
2942 * - No current_arp_slave AND
2943 * - more than 3*delta since last receive AND
2944 * - the bond has an IP address
2946 * Note: a non-null current_arp_slave indicates
2947 * the curr_active_slave went down and we are
2948 * searching for a new one; under this condition
2949 * we only take the curr_active_slave down - this
2950 * gives each slave a chance to tx/rx traffic
2951 * before being taken out
2953 if (!bond_is_active_slave(slave) &&
2954 !bond->current_arp_slave &&
2955 !time_in_range(jiffies,
2956 slave_last_rx(bond, slave) - delta_in_ticks,
2957 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2959 slave->new_link = BOND_LINK_DOWN;
2964 * Active slave is down if:
2965 * - more than 2*delta since transmitting OR
2966 * - (more than 2*delta since receive AND
2967 * the bond has an IP address)
2969 trans_start = dev_trans_start(slave->dev);
2970 if (bond_is_active_slave(slave) &&
2971 (!time_in_range(jiffies,
2972 trans_start - delta_in_ticks,
2973 trans_start + 2 * delta_in_ticks) ||
2974 !time_in_range(jiffies,
2975 slave_last_rx(bond, slave) - delta_in_ticks,
2976 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2978 slave->new_link = BOND_LINK_DOWN;
2987 * Called to commit link state changes noted by inspection step of
2988 * active-backup mode ARP monitor.
2990 * Called with RTNL and bond->lock for read.
2992 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2994 struct slave *slave;
2996 unsigned long trans_start;
2998 bond_for_each_slave(bond, slave, i) {
2999 switch (slave->new_link) {
3000 case BOND_LINK_NOCHANGE:
3004 trans_start = dev_trans_start(slave->dev);
3005 if ((!bond->curr_active_slave &&
3006 time_in_range(jiffies,
3007 trans_start - delta_in_ticks,
3008 trans_start + delta_in_ticks)) ||
3009 bond->curr_active_slave != slave) {
3010 slave->link = BOND_LINK_UP;
3011 bond->current_arp_slave = NULL;
3013 pr_info("%s: link status definitely up for interface %s.\n",
3014 bond->dev->name, slave->dev->name);
3016 if (!bond->curr_active_slave ||
3017 (slave == bond->primary_slave))
3024 case BOND_LINK_DOWN:
3025 if (slave->link_failure_count < UINT_MAX)
3026 slave->link_failure_count++;
3028 slave->link = BOND_LINK_DOWN;
3029 bond_set_slave_inactive_flags(slave);
3031 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3032 bond->dev->name, slave->dev->name);
3034 if (slave == bond->curr_active_slave) {
3035 bond->current_arp_slave = NULL;
3042 pr_err("%s: impossible: new_link %d on slave %s\n",
3043 bond->dev->name, slave->new_link,
3051 write_lock_bh(&bond->curr_slave_lock);
3052 bond_select_active_slave(bond);
3053 write_unlock_bh(&bond->curr_slave_lock);
3054 unblock_netpoll_tx();
3057 bond_set_carrier(bond);
3061 * Send ARP probes for active-backup mode ARP monitor.
3063 * Called with bond->lock held for read.
3065 static void bond_ab_arp_probe(struct bonding *bond)
3067 struct slave *slave;
3070 read_lock(&bond->curr_slave_lock);
3072 if (bond->current_arp_slave && bond->curr_active_slave)
3073 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3074 bond->current_arp_slave->dev->name,
3075 bond->curr_active_slave->dev->name);
3077 if (bond->curr_active_slave) {
3078 bond_arp_send_all(bond, bond->curr_active_slave);
3079 read_unlock(&bond->curr_slave_lock);
3083 read_unlock(&bond->curr_slave_lock);
3085 /* if we don't have a curr_active_slave, search for the next available
3086 * backup slave from the current_arp_slave and make it the candidate
3087 * for becoming the curr_active_slave
3090 if (!bond->current_arp_slave) {
3091 bond->current_arp_slave = bond->first_slave;
3092 if (!bond->current_arp_slave)
3096 bond_set_slave_inactive_flags(bond->current_arp_slave);
3098 /* search for next candidate */
3099 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3100 if (IS_UP(slave->dev)) {
3101 slave->link = BOND_LINK_BACK;
3102 bond_set_slave_active_flags(slave);
3103 bond_arp_send_all(bond, slave);
3104 slave->jiffies = jiffies;
3105 bond->current_arp_slave = slave;
3109 /* if the link state is up at this point, we
3110 * mark it down - this can happen if we have
3111 * simultaneous link failures and
3112 * reselect_active_interface doesn't make this
3113 * one the current slave so it is still marked
3114 * up when it is actually down
3116 if (slave->link == BOND_LINK_UP) {
3117 slave->link = BOND_LINK_DOWN;
3118 if (slave->link_failure_count < UINT_MAX)
3119 slave->link_failure_count++;
3121 bond_set_slave_inactive_flags(slave);
3123 pr_info("%s: backup interface %s is now down.\n",
3124 bond->dev->name, slave->dev->name);
3129 void bond_activebackup_arp_mon(struct work_struct *work)
3131 struct bonding *bond = container_of(work, struct bonding,
3133 bool should_notify_peers = false;
3136 read_lock(&bond->lock);
3138 if (bond->kill_timers)
3141 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3143 if (bond->slave_cnt == 0)
3146 should_notify_peers = bond_should_notify_peers(bond);
3148 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3149 read_unlock(&bond->lock);
3151 read_lock(&bond->lock);
3153 bond_ab_arp_commit(bond, delta_in_ticks);
3155 read_unlock(&bond->lock);
3157 read_lock(&bond->lock);
3160 bond_ab_arp_probe(bond);
3163 if (bond->params.arp_interval && !bond->kill_timers)
3164 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3166 read_unlock(&bond->lock);
3168 if (should_notify_peers) {
3170 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3175 /*-------------------------- netdev event handling --------------------------*/
3178 * Change device name
3180 static int bond_event_changename(struct bonding *bond)
3182 bond_remove_proc_entry(bond);
3183 bond_create_proc_entry(bond);
3185 bond_debug_reregister(bond);
3190 static int bond_master_netdev_event(unsigned long event,
3191 struct net_device *bond_dev)
3193 struct bonding *event_bond = netdev_priv(bond_dev);
3196 case NETDEV_CHANGENAME:
3197 return bond_event_changename(event_bond);
3205 static int bond_slave_netdev_event(unsigned long event,
3206 struct net_device *slave_dev)
3208 struct net_device *bond_dev = slave_dev->master;
3209 struct bonding *bond = netdev_priv(bond_dev);
3212 case NETDEV_UNREGISTER:
3214 if (bond->setup_by_slave)
3215 bond_release_and_destroy(bond_dev, slave_dev);
3217 bond_release(bond_dev, slave_dev);
3221 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3222 struct slave *slave;
3224 slave = bond_get_slave_by_dev(bond, slave_dev);
3226 u32 old_speed = slave->speed;
3227 u8 old_duplex = slave->duplex;
3229 bond_update_speed_duplex(slave);
3231 if (bond_is_lb(bond))
3234 if (old_speed != slave->speed)
3235 bond_3ad_adapter_speed_changed(slave);
3236 if (old_duplex != slave->duplex)
3237 bond_3ad_adapter_duplex_changed(slave);
3244 * ... Or is it this?
3247 case NETDEV_CHANGEMTU:
3249 * TODO: Should slaves be allowed to
3250 * independently alter their MTU? For
3251 * an active-backup bond, slaves need
3252 * not be the same type of device, so
3253 * MTUs may vary. For other modes,
3254 * slaves arguably should have the
3255 * same MTUs. To do this, we'd need to
3256 * take over the slave's change_mtu
3257 * function for the duration of their
3261 case NETDEV_CHANGENAME:
3263 * TODO: handle changing the primary's name
3266 case NETDEV_FEAT_CHANGE:
3267 bond_compute_features(bond);
3277 * bond_netdev_event: handle netdev notifier chain events.
3279 * This function receives events for the netdev chain. The caller (an
3280 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3281 * locks for us to safely manipulate the slave devices (RTNL lock,
3284 static int bond_netdev_event(struct notifier_block *this,
3285 unsigned long event, void *ptr)
3287 struct net_device *event_dev = (struct net_device *)ptr;
3289 pr_debug("event_dev: %s, event: %lx\n",
3290 event_dev ? event_dev->name : "None",
3293 if (!(event_dev->priv_flags & IFF_BONDING))
3296 if (event_dev->flags & IFF_MASTER) {
3297 pr_debug("IFF_MASTER\n");
3298 return bond_master_netdev_event(event, event_dev);
3301 if (event_dev->flags & IFF_SLAVE) {
3302 pr_debug("IFF_SLAVE\n");
3303 return bond_slave_netdev_event(event, event_dev);
3310 * bond_inetaddr_event: handle inetaddr notifier chain events.
3312 * We keep track of device IPs primarily to use as source addresses in
3313 * ARP monitor probes (rather than spewing out broadcasts all the time).
3315 * We track one IP for the main device (if it has one), plus one per VLAN.
3317 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3319 struct in_ifaddr *ifa = ptr;
3320 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3321 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3322 struct bonding *bond;
3323 struct vlan_entry *vlan;
3325 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3326 if (bond->dev == event_dev) {
3329 bond->master_ip = ifa->ifa_local;
3332 bond->master_ip = bond_glean_dev_ip(bond->dev);
3339 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3340 vlan_dev = __vlan_find_dev_deep(bond->dev,
3342 if (vlan_dev == event_dev) {
3345 vlan->vlan_ip = ifa->ifa_local;
3349 bond_glean_dev_ip(vlan_dev);
3360 static struct notifier_block bond_netdev_notifier = {
3361 .notifier_call = bond_netdev_event,
3364 static struct notifier_block bond_inetaddr_notifier = {
3365 .notifier_call = bond_inetaddr_event,
3368 /*---------------------------- Hashing Policies -----------------------------*/
3371 * Hash for the output device based upon layer 2 and layer 3 data. If
3372 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3374 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3376 struct ethhdr *data = (struct ethhdr *)skb->data;
3377 struct iphdr *iph = ip_hdr(skb);
3379 if (skb->protocol == htons(ETH_P_IP)) {
3380 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3381 (data->h_dest[5] ^ data->h_source[5])) % count;
3384 return (data->h_dest[5] ^ data->h_source[5]) % count;
3388 * Hash for the output device based upon layer 3 and layer 4 data. If
3389 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3390 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3392 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3394 struct ethhdr *data = (struct ethhdr *)skb->data;
3395 struct iphdr *iph = ip_hdr(skb);
3396 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3399 if (skb->protocol == htons(ETH_P_IP)) {
3400 if (!ip_is_fragment(iph) &&
3401 (iph->protocol == IPPROTO_TCP ||
3402 iph->protocol == IPPROTO_UDP)) {
3403 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3405 return (layer4_xor ^
3406 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3410 return (data->h_dest[5] ^ data->h_source[5]) % count;
3414 * Hash for the output device based upon layer 2 data
3416 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3418 struct ethhdr *data = (struct ethhdr *)skb->data;
3420 return (data->h_dest[5] ^ data->h_source[5]) % count;
3423 /*-------------------------- Device entry points ----------------------------*/
3425 static int bond_open(struct net_device *bond_dev)
3427 struct bonding *bond = netdev_priv(bond_dev);
3428 struct slave *slave;
3431 bond->kill_timers = 0;
3433 /* reset slave->backup and slave->inactive */
3434 read_lock(&bond->lock);
3435 if (bond->slave_cnt > 0) {
3436 read_lock(&bond->curr_slave_lock);
3437 bond_for_each_slave(bond, slave, i) {
3438 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3439 && (slave != bond->curr_active_slave)) {
3440 bond_set_slave_inactive_flags(slave);
3442 bond_set_slave_active_flags(slave);
3445 read_unlock(&bond->curr_slave_lock);
3447 read_unlock(&bond->lock);
3449 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3451 if (bond_is_lb(bond)) {
3452 /* bond_alb_initialize must be called before the timer
3455 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3456 /* something went wrong - fail the open operation */
3460 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3461 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3464 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3465 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3466 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3469 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3470 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3471 INIT_DELAYED_WORK(&bond->arp_work,
3472 bond_activebackup_arp_mon);
3474 INIT_DELAYED_WORK(&bond->arp_work,
3475 bond_loadbalance_arp_mon);
3477 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3478 if (bond->params.arp_validate)
3479 bond->recv_probe = bond_arp_rcv;
3482 if (bond->params.mode == BOND_MODE_8023AD) {
3483 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3484 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3485 /* register to receive LACPDUs */
3486 bond->recv_probe = bond_3ad_lacpdu_recv;
3487 bond_3ad_initiate_agg_selection(bond, 1);
3493 static int bond_close(struct net_device *bond_dev)
3495 struct bonding *bond = netdev_priv(bond_dev);
3497 write_lock_bh(&bond->lock);
3499 bond->send_peer_notif = 0;
3501 /* signal timers not to re-arm */
3502 bond->kill_timers = 1;
3504 write_unlock_bh(&bond->lock);
3506 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3507 cancel_delayed_work(&bond->mii_work);
3510 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3511 cancel_delayed_work(&bond->arp_work);
3514 switch (bond->params.mode) {
3515 case BOND_MODE_8023AD:
3516 cancel_delayed_work(&bond->ad_work);
3520 cancel_delayed_work(&bond->alb_work);
3526 if (delayed_work_pending(&bond->mcast_work))
3527 cancel_delayed_work(&bond->mcast_work);
3529 if (bond_is_lb(bond)) {
3530 /* Must be called only after all
3531 * slaves have been released
3533 bond_alb_deinitialize(bond);
3535 bond->recv_probe = NULL;
3540 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3541 struct rtnl_link_stats64 *stats)
3543 struct bonding *bond = netdev_priv(bond_dev);
3544 struct rtnl_link_stats64 temp;
3545 struct slave *slave;
3548 memset(stats, 0, sizeof(*stats));
3550 read_lock_bh(&bond->lock);
3552 bond_for_each_slave(bond, slave, i) {
3553 const struct rtnl_link_stats64 *sstats =
3554 dev_get_stats(slave->dev, &temp);
3556 stats->rx_packets += sstats->rx_packets;
3557 stats->rx_bytes += sstats->rx_bytes;
3558 stats->rx_errors += sstats->rx_errors;
3559 stats->rx_dropped += sstats->rx_dropped;
3561 stats->tx_packets += sstats->tx_packets;
3562 stats->tx_bytes += sstats->tx_bytes;
3563 stats->tx_errors += sstats->tx_errors;
3564 stats->tx_dropped += sstats->tx_dropped;
3566 stats->multicast += sstats->multicast;
3567 stats->collisions += sstats->collisions;
3569 stats->rx_length_errors += sstats->rx_length_errors;
3570 stats->rx_over_errors += sstats->rx_over_errors;
3571 stats->rx_crc_errors += sstats->rx_crc_errors;
3572 stats->rx_frame_errors += sstats->rx_frame_errors;
3573 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3574 stats->rx_missed_errors += sstats->rx_missed_errors;
3576 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3577 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3578 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3579 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3580 stats->tx_window_errors += sstats->tx_window_errors;
3583 read_unlock_bh(&bond->lock);
3588 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3590 struct net_device *slave_dev = NULL;
3591 struct ifbond k_binfo;
3592 struct ifbond __user *u_binfo = NULL;
3593 struct ifslave k_sinfo;
3594 struct ifslave __user *u_sinfo = NULL;
3595 struct mii_ioctl_data *mii = NULL;
3598 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3610 * We do this again just in case we were called by SIOCGMIIREG
3611 * instead of SIOCGMIIPHY.
3618 if (mii->reg_num == 1) {
3619 struct bonding *bond = netdev_priv(bond_dev);
3621 read_lock(&bond->lock);
3622 read_lock(&bond->curr_slave_lock);
3623 if (netif_carrier_ok(bond->dev))
3624 mii->val_out = BMSR_LSTATUS;
3626 read_unlock(&bond->curr_slave_lock);
3627 read_unlock(&bond->lock);
3631 case BOND_INFO_QUERY_OLD:
3632 case SIOCBONDINFOQUERY:
3633 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3635 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3638 res = bond_info_query(bond_dev, &k_binfo);
3640 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3644 case BOND_SLAVE_INFO_QUERY_OLD:
3645 case SIOCBONDSLAVEINFOQUERY:
3646 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3648 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3651 res = bond_slave_info_query(bond_dev, &k_sinfo);
3653 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3662 if (!capable(CAP_NET_ADMIN))
3665 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3667 pr_debug("slave_dev=%p:\n", slave_dev);
3672 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3674 case BOND_ENSLAVE_OLD:
3675 case SIOCBONDENSLAVE:
3676 res = bond_enslave(bond_dev, slave_dev);
3678 case BOND_RELEASE_OLD:
3679 case SIOCBONDRELEASE:
3680 res = bond_release(bond_dev, slave_dev);
3682 case BOND_SETHWADDR_OLD:
3683 case SIOCBONDSETHWADDR:
3684 res = bond_sethwaddr(bond_dev, slave_dev);
3686 case BOND_CHANGE_ACTIVE_OLD:
3687 case SIOCBONDCHANGEACTIVE:
3688 res = bond_ioctl_change_active(bond_dev, slave_dev);
3700 static bool bond_addr_in_mc_list(unsigned char *addr,
3701 struct netdev_hw_addr_list *list,
3704 struct netdev_hw_addr *ha;
3706 netdev_hw_addr_list_for_each(ha, list)
3707 if (!memcmp(ha->addr, addr, addrlen))
3713 static void bond_set_multicast_list(struct net_device *bond_dev)
3715 struct bonding *bond = netdev_priv(bond_dev);
3716 struct netdev_hw_addr *ha;
3720 * Do promisc before checking multicast_mode
3722 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3724 * FIXME: Need to handle the error when one of the multi-slaves
3727 bond_set_promiscuity(bond, 1);
3730 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3731 bond_set_promiscuity(bond, -1);
3734 /* set allmulti flag to slaves */
3735 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3737 * FIXME: Need to handle the error when one of the multi-slaves
3740 bond_set_allmulti(bond, 1);
3743 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3744 bond_set_allmulti(bond, -1);
3747 read_lock(&bond->lock);
3749 bond->flags = bond_dev->flags;
3751 /* looking for addresses to add to slaves' mc list */
3752 netdev_for_each_mc_addr(ha, bond_dev) {
3753 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3754 bond_dev->addr_len);
3756 bond_mc_add(bond, ha->addr);
3759 /* looking for addresses to delete from slaves' list */
3760 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3761 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3762 bond_dev->addr_len);
3764 bond_mc_del(bond, ha->addr);
3767 /* save master's multicast list */
3768 __hw_addr_flush(&bond->mc_list);
3769 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3770 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3772 read_unlock(&bond->lock);
3775 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3777 struct bonding *bond = netdev_priv(dev);
3778 struct slave *slave = bond->first_slave;
3781 const struct net_device_ops *slave_ops
3782 = slave->dev->netdev_ops;
3783 if (slave_ops->ndo_neigh_setup)
3784 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3790 * Change the MTU of all of a master's slaves to match the master
3792 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3794 struct bonding *bond = netdev_priv(bond_dev);
3795 struct slave *slave, *stop_at;
3799 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3800 (bond_dev ? bond_dev->name : "None"), new_mtu);
3802 /* Can't hold bond->lock with bh disabled here since
3803 * some base drivers panic. On the other hand we can't
3804 * hold bond->lock without bh disabled because we'll
3805 * deadlock. The only solution is to rely on the fact
3806 * that we're under rtnl_lock here, and the slaves
3807 * list won't change. This doesn't solve the problem
3808 * of setting the slave's MTU while it is
3809 * transmitting, but the assumption is that the base
3810 * driver can handle that.
3812 * TODO: figure out a way to safely iterate the slaves
3813 * list, but without holding a lock around the actual
3814 * call to the base driver.
3817 bond_for_each_slave(bond, slave, i) {
3818 pr_debug("s %p s->p %p c_m %p\n",
3821 slave->dev->netdev_ops->ndo_change_mtu);
3823 res = dev_set_mtu(slave->dev, new_mtu);
3826 /* If we failed to set the slave's mtu to the new value
3827 * we must abort the operation even in ACTIVE_BACKUP
3828 * mode, because if we allow the backup slaves to have
3829 * different mtu values than the active slave we'll
3830 * need to change their mtu when doing a failover. That
3831 * means changing their mtu from timer context, which
3832 * is probably not a good idea.
3834 pr_debug("err %d %s\n", res, slave->dev->name);
3839 bond_dev->mtu = new_mtu;
3844 /* unwind from head to the slave that failed */
3846 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3849 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3851 pr_debug("unwind err %d dev %s\n",
3852 tmp_res, slave->dev->name);
3862 * Note that many devices must be down to change the HW address, and
3863 * downing the master releases all slaves. We can make bonds full of
3864 * bonding devices to test this, however.
3866 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3868 struct bonding *bond = netdev_priv(bond_dev);
3869 struct sockaddr *sa = addr, tmp_sa;
3870 struct slave *slave, *stop_at;
3874 if (bond->params.mode == BOND_MODE_ALB)
3875 return bond_alb_set_mac_address(bond_dev, addr);
3878 pr_debug("bond=%p, name=%s\n",
3879 bond, bond_dev ? bond_dev->name : "None");
3882 * If fail_over_mac is set to active, do nothing and return
3883 * success. Returning an error causes ifenslave to fail.
3885 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3888 if (!is_valid_ether_addr(sa->sa_data))
3889 return -EADDRNOTAVAIL;
3891 /* Can't hold bond->lock with bh disabled here since
3892 * some base drivers panic. On the other hand we can't
3893 * hold bond->lock without bh disabled because we'll
3894 * deadlock. The only solution is to rely on the fact
3895 * that we're under rtnl_lock here, and the slaves
3896 * list won't change. This doesn't solve the problem
3897 * of setting the slave's hw address while it is
3898 * transmitting, but the assumption is that the base
3899 * driver can handle that.
3901 * TODO: figure out a way to safely iterate the slaves
3902 * list, but without holding a lock around the actual
3903 * call to the base driver.
3906 bond_for_each_slave(bond, slave, i) {
3907 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3908 pr_debug("slave %p %s\n", slave, slave->dev->name);
3910 if (slave_ops->ndo_set_mac_address == NULL) {
3912 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3916 res = dev_set_mac_address(slave->dev, addr);
3918 /* TODO: consider downing the slave
3920 * User should expect communications
3921 * breakage anyway until ARP finish
3924 pr_debug("err %d %s\n", res, slave->dev->name);
3930 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3934 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3935 tmp_sa.sa_family = bond_dev->type;
3937 /* unwind from head to the slave that failed */
3939 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3942 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3944 pr_debug("unwind err %d dev %s\n",
3945 tmp_res, slave->dev->name);
3952 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3954 struct bonding *bond = netdev_priv(bond_dev);
3955 struct slave *slave, *start_at;
3956 int i, slave_no, res = 1;
3957 struct iphdr *iph = ip_hdr(skb);
3960 * Start with the curr_active_slave that joined the bond as the
3961 * default for sending IGMP traffic. For failover purposes one
3962 * needs to maintain some consistency for the interface that will
3963 * send the join/membership reports. The curr_active_slave found
3964 * will send all of this type of traffic.
3966 if ((iph->protocol == IPPROTO_IGMP) &&
3967 (skb->protocol == htons(ETH_P_IP))) {
3969 read_lock(&bond->curr_slave_lock);
3970 slave = bond->curr_active_slave;
3971 read_unlock(&bond->curr_slave_lock);
3977 * Concurrent TX may collide on rr_tx_counter; we accept
3978 * that as being rare enough not to justify using an
3981 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3983 bond_for_each_slave(bond, slave, i) {
3991 bond_for_each_slave_from(bond, slave, i, start_at) {
3992 if (IS_UP(slave->dev) &&
3993 (slave->link == BOND_LINK_UP) &&
3994 bond_is_active_slave(slave)) {
3995 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4002 /* no suitable interface, frame not sent */
4006 return NETDEV_TX_OK;
4011 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4012 * the bond has a usable interface.
4014 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4016 struct bonding *bond = netdev_priv(bond_dev);
4019 read_lock(&bond->curr_slave_lock);
4021 if (bond->curr_active_slave)
4022 res = bond_dev_queue_xmit(bond, skb,
4023 bond->curr_active_slave->dev);
4026 /* no suitable interface, frame not sent */
4029 read_unlock(&bond->curr_slave_lock);
4031 return NETDEV_TX_OK;
4035 * In bond_xmit_xor() , we determine the output device by using a pre-
4036 * determined xmit_hash_policy(), If the selected device is not enabled,
4037 * find the next active slave.
4039 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4041 struct bonding *bond = netdev_priv(bond_dev);
4042 struct slave *slave, *start_at;
4047 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4049 bond_for_each_slave(bond, slave, i) {
4057 bond_for_each_slave_from(bond, slave, i, start_at) {
4058 if (IS_UP(slave->dev) &&
4059 (slave->link == BOND_LINK_UP) &&
4060 bond_is_active_slave(slave)) {
4061 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4067 /* no suitable interface, frame not sent */
4071 return NETDEV_TX_OK;
4075 * in broadcast mode, we send everything to all usable interfaces.
4077 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4079 struct bonding *bond = netdev_priv(bond_dev);
4080 struct slave *slave, *start_at;
4081 struct net_device *tx_dev = NULL;
4085 read_lock(&bond->curr_slave_lock);
4086 start_at = bond->curr_active_slave;
4087 read_unlock(&bond->curr_slave_lock);
4092 bond_for_each_slave_from(bond, slave, i, start_at) {
4093 if (IS_UP(slave->dev) &&
4094 (slave->link == BOND_LINK_UP) &&
4095 bond_is_active_slave(slave)) {
4097 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4099 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4104 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4106 dev_kfree_skb(skb2);
4110 tx_dev = slave->dev;
4115 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4119 /* no suitable interface, frame not sent */
4122 /* frame sent to all suitable interfaces */
4123 return NETDEV_TX_OK;
4126 /*------------------------- Device initialization ---------------------------*/
4128 static void bond_set_xmit_hash_policy(struct bonding *bond)
4130 switch (bond->params.xmit_policy) {
4131 case BOND_XMIT_POLICY_LAYER23:
4132 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4134 case BOND_XMIT_POLICY_LAYER34:
4135 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4137 case BOND_XMIT_POLICY_LAYER2:
4139 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4145 * Lookup the slave that corresponds to a qid
4147 static inline int bond_slave_override(struct bonding *bond,
4148 struct sk_buff *skb)
4151 struct slave *slave = NULL;
4152 struct slave *check_slave;
4154 if (!skb->queue_mapping)
4157 /* Find out if any slaves have the same mapping as this skb. */
4158 bond_for_each_slave(bond, check_slave, i) {
4159 if (check_slave->queue_id == skb->queue_mapping) {
4160 slave = check_slave;
4165 /* If the slave isn't UP, use default transmit policy. */
4166 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4167 (slave->link == BOND_LINK_UP)) {
4168 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4175 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4178 * This helper function exists to help dev_pick_tx get the correct
4179 * destination queue. Using a helper function skips a call to
4180 * skb_tx_hash and will put the skbs in the queue we expect on their
4181 * way down to the bonding driver.
4183 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4186 * Save the original txq to restore before passing to the driver
4188 bond_queue_mapping(skb) = skb->queue_mapping;
4190 if (unlikely(txq >= dev->real_num_tx_queues)) {
4192 txq -= dev->real_num_tx_queues;
4193 } while (txq >= dev->real_num_tx_queues);
4198 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4200 struct bonding *bond = netdev_priv(dev);
4202 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4203 if (!bond_slave_override(bond, skb))
4204 return NETDEV_TX_OK;
4207 switch (bond->params.mode) {
4208 case BOND_MODE_ROUNDROBIN:
4209 return bond_xmit_roundrobin(skb, dev);
4210 case BOND_MODE_ACTIVEBACKUP:
4211 return bond_xmit_activebackup(skb, dev);
4213 return bond_xmit_xor(skb, dev);
4214 case BOND_MODE_BROADCAST:
4215 return bond_xmit_broadcast(skb, dev);
4216 case BOND_MODE_8023AD:
4217 return bond_3ad_xmit_xor(skb, dev);
4220 return bond_alb_xmit(skb, dev);
4222 /* Should never happen, mode already checked */
4223 pr_err("%s: Error: Unknown bonding mode %d\n",
4224 dev->name, bond->params.mode);
4227 return NETDEV_TX_OK;
4231 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4233 struct bonding *bond = netdev_priv(dev);
4234 netdev_tx_t ret = NETDEV_TX_OK;
4237 * If we risk deadlock from transmitting this in the
4238 * netpoll path, tell netpoll to queue the frame for later tx
4240 if (is_netpoll_tx_blocked(dev))
4241 return NETDEV_TX_BUSY;
4243 read_lock(&bond->lock);
4245 if (bond->slave_cnt)
4246 ret = __bond_start_xmit(skb, dev);
4250 read_unlock(&bond->lock);
4256 * set bond mode specific net device operations
4258 void bond_set_mode_ops(struct bonding *bond, int mode)
4260 struct net_device *bond_dev = bond->dev;
4263 case BOND_MODE_ROUNDROBIN:
4265 case BOND_MODE_ACTIVEBACKUP:
4268 bond_set_xmit_hash_policy(bond);
4270 case BOND_MODE_BROADCAST:
4272 case BOND_MODE_8023AD:
4273 bond_set_xmit_hash_policy(bond);
4280 /* Should never happen, mode already checked */
4281 pr_err("%s: Error: Unknown bonding mode %d\n",
4282 bond_dev->name, mode);
4287 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4288 struct ethtool_drvinfo *drvinfo)
4290 strncpy(drvinfo->driver, DRV_NAME, 32);
4291 strncpy(drvinfo->version, DRV_VERSION, 32);
4292 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4295 static const struct ethtool_ops bond_ethtool_ops = {
4296 .get_drvinfo = bond_ethtool_get_drvinfo,
4297 .get_link = ethtool_op_get_link,
4300 static const struct net_device_ops bond_netdev_ops = {
4301 .ndo_init = bond_init,
4302 .ndo_uninit = bond_uninit,
4303 .ndo_open = bond_open,
4304 .ndo_stop = bond_close,
4305 .ndo_start_xmit = bond_start_xmit,
4306 .ndo_select_queue = bond_select_queue,
4307 .ndo_get_stats64 = bond_get_stats,
4308 .ndo_do_ioctl = bond_do_ioctl,
4309 .ndo_set_multicast_list = bond_set_multicast_list,
4310 .ndo_change_mtu = bond_change_mtu,
4311 .ndo_set_mac_address = bond_set_mac_address,
4312 .ndo_neigh_setup = bond_neigh_setup,
4313 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4314 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4315 #ifdef CONFIG_NET_POLL_CONTROLLER
4316 .ndo_netpoll_setup = bond_netpoll_setup,
4317 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4318 .ndo_poll_controller = bond_poll_controller,
4320 .ndo_add_slave = bond_enslave,
4321 .ndo_del_slave = bond_release,
4322 .ndo_fix_features = bond_fix_features,
4325 static void bond_destructor(struct net_device *bond_dev)
4327 struct bonding *bond = netdev_priv(bond_dev);
4329 destroy_workqueue(bond->wq);
4330 free_netdev(bond_dev);
4333 static void bond_setup(struct net_device *bond_dev)
4335 struct bonding *bond = netdev_priv(bond_dev);
4337 /* initialize rwlocks */
4338 rwlock_init(&bond->lock);
4339 rwlock_init(&bond->curr_slave_lock);
4341 bond->params = bonding_defaults;
4343 /* Initialize pointers */
4344 bond->dev = bond_dev;
4345 INIT_LIST_HEAD(&bond->vlan_list);
4347 /* Initialize the device entry points */
4348 ether_setup(bond_dev);
4349 bond_dev->netdev_ops = &bond_netdev_ops;
4350 bond_dev->ethtool_ops = &bond_ethtool_ops;
4351 bond_set_mode_ops(bond, bond->params.mode);
4353 bond_dev->destructor = bond_destructor;
4355 /* Initialize the device options */
4356 bond_dev->tx_queue_len = 0;
4357 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4358 bond_dev->priv_flags |= IFF_BONDING;
4359 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4361 /* At first, we block adding VLANs. That's the only way to
4362 * prevent problems that occur when adding VLANs over an
4363 * empty bond. The block will be removed once non-challenged
4364 * slaves are enslaved.
4366 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4368 /* don't acquire bond device's netif_tx_lock when
4370 bond_dev->features |= NETIF_F_LLTX;
4372 /* By default, we declare the bond to be fully
4373 * VLAN hardware accelerated capable. Special
4374 * care is taken in the various xmit functions
4375 * when there are slaves that are not hw accel
4379 bond_dev->hw_features = BOND_VLAN_FEATURES |
4380 NETIF_F_HW_VLAN_TX |
4381 NETIF_F_HW_VLAN_RX |
4382 NETIF_F_HW_VLAN_FILTER;
4384 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4385 bond_dev->features |= bond_dev->hw_features;
4388 static void bond_work_cancel_all(struct bonding *bond)
4390 write_lock_bh(&bond->lock);
4391 bond->kill_timers = 1;
4392 write_unlock_bh(&bond->lock);
4394 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4395 cancel_delayed_work(&bond->mii_work);
4397 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4398 cancel_delayed_work(&bond->arp_work);
4400 if (bond->params.mode == BOND_MODE_ALB &&
4401 delayed_work_pending(&bond->alb_work))
4402 cancel_delayed_work(&bond->alb_work);
4404 if (bond->params.mode == BOND_MODE_8023AD &&
4405 delayed_work_pending(&bond->ad_work))
4406 cancel_delayed_work(&bond->ad_work);
4408 if (delayed_work_pending(&bond->mcast_work))
4409 cancel_delayed_work(&bond->mcast_work);
4413 * Destroy a bonding device.
4414 * Must be under rtnl_lock when this function is called.
4416 static void bond_uninit(struct net_device *bond_dev)
4418 struct bonding *bond = netdev_priv(bond_dev);
4419 struct vlan_entry *vlan, *tmp;
4421 bond_netpoll_cleanup(bond_dev);
4423 /* Release the bonded slaves */
4424 bond_release_all(bond_dev);
4426 list_del(&bond->bond_list);
4428 bond_work_cancel_all(bond);
4430 bond_remove_proc_entry(bond);
4432 bond_debug_unregister(bond);
4434 __hw_addr_flush(&bond->mc_list);
4436 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4437 list_del(&vlan->vlan_list);
4442 /*------------------------- Module initialization ---------------------------*/
4445 * Convert string input module parms. Accept either the
4446 * number of the mode or its string name. A bit complicated because
4447 * some mode names are substrings of other names, and calls from sysfs
4448 * may have whitespace in the name (trailing newlines, for example).
4450 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4452 int modeint = -1, i, rv;
4453 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4455 for (p = (char *)buf; *p; p++)
4456 if (!(isdigit(*p) || isspace(*p)))
4460 rv = sscanf(buf, "%20s", modestr);
4462 rv = sscanf(buf, "%d", &modeint);
4467 for (i = 0; tbl[i].modename; i++) {
4468 if (modeint == tbl[i].mode)
4470 if (strcmp(modestr, tbl[i].modename) == 0)
4477 static int bond_check_params(struct bond_params *params)
4479 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4482 * Convert string parameters.
4485 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4486 if (bond_mode == -1) {
4487 pr_err("Error: Invalid bonding mode \"%s\"\n",
4488 mode == NULL ? "NULL" : mode);
4493 if (xmit_hash_policy) {
4494 if ((bond_mode != BOND_MODE_XOR) &&
4495 (bond_mode != BOND_MODE_8023AD)) {
4496 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4497 bond_mode_name(bond_mode));
4499 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4501 if (xmit_hashtype == -1) {
4502 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4503 xmit_hash_policy == NULL ? "NULL" :
4511 if (bond_mode != BOND_MODE_8023AD) {
4512 pr_info("lacp_rate param is irrelevant in mode %s\n",
4513 bond_mode_name(bond_mode));
4515 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4516 if (lacp_fast == -1) {
4517 pr_err("Error: Invalid lacp rate \"%s\"\n",
4518 lacp_rate == NULL ? "NULL" : lacp_rate);
4525 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4526 if (params->ad_select == -1) {
4527 pr_err("Error: Invalid ad_select \"%s\"\n",
4528 ad_select == NULL ? "NULL" : ad_select);
4532 if (bond_mode != BOND_MODE_8023AD) {
4533 pr_warning("ad_select param only affects 802.3ad mode\n");
4536 params->ad_select = BOND_AD_STABLE;
4539 if (max_bonds < 0) {
4540 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4541 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4542 max_bonds = BOND_DEFAULT_MAX_BONDS;
4546 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4547 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4548 miimon = BOND_LINK_MON_INTERV;
4552 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4557 if (downdelay < 0) {
4558 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4559 downdelay, INT_MAX);
4563 if ((use_carrier != 0) && (use_carrier != 1)) {
4564 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4569 if (num_peer_notif < 0 || num_peer_notif > 255) {
4570 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4575 /* reset values for 802.3ad */
4576 if (bond_mode == BOND_MODE_8023AD) {
4578 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4579 pr_warning("Forcing miimon to 100msec\n");
4584 if (tx_queues < 1 || tx_queues > 255) {
4585 pr_warning("Warning: tx_queues (%d) should be between "
4586 "1 and 255, resetting to %d\n",
4587 tx_queues, BOND_DEFAULT_TX_QUEUES);
4588 tx_queues = BOND_DEFAULT_TX_QUEUES;
4591 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4592 pr_warning("Warning: all_slaves_active module parameter (%d), "
4593 "not of valid value (0/1), so it was set to "
4594 "0\n", all_slaves_active);
4595 all_slaves_active = 0;
4598 if (resend_igmp < 0 || resend_igmp > 255) {
4599 pr_warning("Warning: resend_igmp (%d) should be between "
4600 "0 and 255, resetting to %d\n",
4601 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4602 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4605 /* reset values for TLB/ALB */
4606 if ((bond_mode == BOND_MODE_TLB) ||
4607 (bond_mode == BOND_MODE_ALB)) {
4609 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4610 pr_warning("Forcing miimon to 100msec\n");
4615 if (bond_mode == BOND_MODE_ALB) {
4616 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4621 if (updelay || downdelay) {
4622 /* just warn the user the up/down delay will have
4623 * no effect since miimon is zero...
4625 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4626 updelay, downdelay);
4629 /* don't allow arp monitoring */
4631 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4632 miimon, arp_interval);
4636 if ((updelay % miimon) != 0) {
4637 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4639 (updelay / miimon) * miimon);
4644 if ((downdelay % miimon) != 0) {
4645 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4647 (downdelay / miimon) * miimon);
4650 downdelay /= miimon;
4653 if (arp_interval < 0) {
4654 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4655 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4656 arp_interval = BOND_LINK_ARP_INTERV;
4659 for (arp_ip_count = 0;
4660 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4662 /* not complete check, but should be good enough to
4664 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4665 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4666 arp_ip_target[arp_ip_count]);
4669 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4670 arp_target[arp_ip_count] = ip;
4674 if (arp_interval && !arp_ip_count) {
4675 /* don't allow arping if no arp_ip_target given... */
4676 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4682 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4683 pr_err("arp_validate only supported in active-backup mode\n");
4686 if (!arp_interval) {
4687 pr_err("arp_validate requires arp_interval\n");
4691 arp_validate_value = bond_parse_parm(arp_validate,
4693 if (arp_validate_value == -1) {
4694 pr_err("Error: invalid arp_validate \"%s\"\n",
4695 arp_validate == NULL ? "NULL" : arp_validate);
4699 arp_validate_value = 0;
4702 pr_info("MII link monitoring set to %d ms\n", miimon);
4703 } else if (arp_interval) {
4706 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4708 arp_validate_tbl[arp_validate_value].modename,
4711 for (i = 0; i < arp_ip_count; i++)
4712 pr_info(" %s", arp_ip_target[i]);
4716 } else if (max_bonds) {
4717 /* miimon and arp_interval not set, we need one so things
4718 * work as expected, see bonding.txt for details
4720 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4723 if (primary && !USES_PRIMARY(bond_mode)) {
4724 /* currently, using a primary only makes sense
4725 * in active backup, TLB or ALB modes
4727 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4728 primary, bond_mode_name(bond_mode));
4732 if (primary && primary_reselect) {
4733 primary_reselect_value = bond_parse_parm(primary_reselect,
4735 if (primary_reselect_value == -1) {
4736 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4738 NULL ? "NULL" : primary_reselect);
4742 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4745 if (fail_over_mac) {
4746 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4748 if (fail_over_mac_value == -1) {
4749 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4750 arp_validate == NULL ? "NULL" : arp_validate);
4754 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4755 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4757 fail_over_mac_value = BOND_FOM_NONE;
4760 /* fill params struct with the proper values */
4761 params->mode = bond_mode;
4762 params->xmit_policy = xmit_hashtype;
4763 params->miimon = miimon;
4764 params->num_peer_notif = num_peer_notif;
4765 params->arp_interval = arp_interval;
4766 params->arp_validate = arp_validate_value;
4767 params->updelay = updelay;
4768 params->downdelay = downdelay;
4769 params->use_carrier = use_carrier;
4770 params->lacp_fast = lacp_fast;
4771 params->primary[0] = 0;
4772 params->primary_reselect = primary_reselect_value;
4773 params->fail_over_mac = fail_over_mac_value;
4774 params->tx_queues = tx_queues;
4775 params->all_slaves_active = all_slaves_active;
4776 params->resend_igmp = resend_igmp;
4777 params->min_links = min_links;
4780 strncpy(params->primary, primary, IFNAMSIZ);
4781 params->primary[IFNAMSIZ - 1] = 0;
4784 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4789 static struct lock_class_key bonding_netdev_xmit_lock_key;
4790 static struct lock_class_key bonding_netdev_addr_lock_key;
4792 static void bond_set_lockdep_class_one(struct net_device *dev,
4793 struct netdev_queue *txq,
4796 lockdep_set_class(&txq->_xmit_lock,
4797 &bonding_netdev_xmit_lock_key);
4800 static void bond_set_lockdep_class(struct net_device *dev)
4802 lockdep_set_class(&dev->addr_list_lock,
4803 &bonding_netdev_addr_lock_key);
4804 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4808 * Called from registration process
4810 static int bond_init(struct net_device *bond_dev)
4812 struct bonding *bond = netdev_priv(bond_dev);
4813 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4814 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4816 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4819 * Initialize locks that may be required during
4820 * en/deslave operations. All of the bond_open work
4821 * (of which this is part) should really be moved to
4822 * a phase prior to dev_open
4824 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4825 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4827 bond->wq = create_singlethread_workqueue(bond_dev->name);
4831 bond_set_lockdep_class(bond_dev);
4833 bond_create_proc_entry(bond);
4834 list_add_tail(&bond->bond_list, &bn->dev_list);
4836 bond_prepare_sysfs_group(bond);
4838 bond_debug_register(bond);
4840 __hw_addr_init(&bond->mc_list);
4844 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4846 if (tb[IFLA_ADDRESS]) {
4847 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4849 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4850 return -EADDRNOTAVAIL;
4855 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4857 .priv_size = sizeof(struct bonding),
4858 .setup = bond_setup,
4859 .validate = bond_validate,
4862 /* Create a new bond based on the specified name and bonding parameters.
4863 * If name is NULL, obtain a suitable "bond%d" name for us.
4864 * Caller must NOT hold rtnl_lock; we need to release it here before we
4865 * set up our sysfs entries.
4867 int bond_create(struct net *net, const char *name)
4869 struct net_device *bond_dev;
4874 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4875 name ? name : "bond%d",
4876 bond_setup, tx_queues);
4878 pr_err("%s: eek! can't alloc netdev!\n", name);
4883 dev_net_set(bond_dev, net);
4884 bond_dev->rtnl_link_ops = &bond_link_ops;
4886 res = register_netdevice(bond_dev);
4888 netif_carrier_off(bond_dev);
4892 bond_destructor(bond_dev);
4896 static int __net_init bond_net_init(struct net *net)
4898 struct bond_net *bn = net_generic(net, bond_net_id);
4901 INIT_LIST_HEAD(&bn->dev_list);
4903 bond_create_proc_dir(bn);
4908 static void __net_exit bond_net_exit(struct net *net)
4910 struct bond_net *bn = net_generic(net, bond_net_id);
4912 bond_destroy_proc_dir(bn);
4915 static struct pernet_operations bond_net_ops = {
4916 .init = bond_net_init,
4917 .exit = bond_net_exit,
4919 .size = sizeof(struct bond_net),
4922 static int __init bonding_init(void)
4927 pr_info("%s", bond_version);
4929 res = bond_check_params(&bonding_defaults);
4933 res = register_pernet_subsys(&bond_net_ops);
4937 res = rtnl_link_register(&bond_link_ops);
4941 bond_create_debugfs();
4943 for (i = 0; i < max_bonds; i++) {
4944 res = bond_create(&init_net, NULL);
4949 res = bond_create_sysfs();
4953 register_netdevice_notifier(&bond_netdev_notifier);
4954 register_inetaddr_notifier(&bond_inetaddr_notifier);
4958 rtnl_link_unregister(&bond_link_ops);
4960 unregister_pernet_subsys(&bond_net_ops);
4965 static void __exit bonding_exit(void)
4967 unregister_netdevice_notifier(&bond_netdev_notifier);
4968 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4970 bond_destroy_sysfs();
4971 bond_destroy_debugfs();
4973 rtnl_link_unregister(&bond_link_ops);
4974 unregister_pernet_subsys(&bond_net_ops);
4976 #ifdef CONFIG_NET_POLL_CONTROLLER
4978 * Make sure we don't have an imbalance on our netpoll blocking
4980 WARN_ON(atomic_read(&netpoll_block_tx));
4984 module_init(bonding_init);
4985 module_exit(bonding_exit);
4986 MODULE_LICENSE("GPL");
4987 MODULE_VERSION(DRV_VERSION);
4988 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4989 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4990 MODULE_ALIAS_RTNL_LINK("bond");