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;
399 skb->queue_mapping = bond_queue_mapping(skb);
401 if (unlikely(netpoll_tx_running(slave_dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433 struct bonding *bond = netdev_priv(bond_dev);
437 bond_for_each_slave(bond, slave, i) {
438 struct net_device *slave_dev = slave->dev;
439 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
442 slave_ops->ndo_vlan_rx_add_vid) {
443 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
447 res = bond_add_vlan(bond, vid);
449 pr_err("%s: Error: Failed to add vlan id %d\n",
450 bond_dev->name, vid);
455 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
456 * @bond_dev: bonding net device that got called
457 * @vid: vlan id being removed
459 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
461 struct bonding *bond = netdev_priv(bond_dev);
465 bond_for_each_slave(bond, slave, i) {
466 struct net_device *slave_dev = slave->dev;
467 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
469 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
470 slave_ops->ndo_vlan_rx_kill_vid) {
471 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
475 res = bond_del_vlan(bond, vid);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev->name, vid);
482 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
484 struct vlan_entry *vlan;
485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
487 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
488 !(slave_ops->ndo_vlan_rx_add_vid))
491 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
492 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
495 static void bond_del_vlans_from_slave(struct bonding *bond,
496 struct net_device *slave_dev)
498 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
499 struct vlan_entry *vlan;
501 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
502 !(slave_ops->ndo_vlan_rx_kill_vid))
505 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
508 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
512 /*------------------------------- Link status -------------------------------*/
515 * Set the carrier state for the master according to the state of its
516 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
517 * do special 802.3ad magic.
519 * Returns zero if carrier state does not change, nonzero if it does.
521 static int bond_set_carrier(struct bonding *bond)
526 if (bond->slave_cnt == 0)
529 if (bond->params.mode == BOND_MODE_8023AD)
530 return bond_3ad_set_carrier(bond);
532 bond_for_each_slave(bond, slave, i) {
533 if (slave->link == BOND_LINK_UP) {
534 if (!netif_carrier_ok(bond->dev)) {
535 netif_carrier_on(bond->dev);
543 if (netif_carrier_ok(bond->dev)) {
544 netif_carrier_off(bond->dev);
551 * Get link speed and duplex from the slave's base driver
552 * using ethtool. If for some reason the call fails or the
553 * values are invalid, set speed and duplex to -1,
556 static int bond_update_speed_duplex(struct slave *slave)
558 struct net_device *slave_dev = slave->dev;
559 struct ethtool_cmd ecmd;
563 slave->speed = SPEED_UNKNOWN;
564 slave->duplex = DUPLEX_UNKNOWN;
566 res = __ethtool_get_settings(slave_dev, &ecmd);
570 slave_speed = ethtool_cmd_speed(&ecmd);
571 if (slave_speed == 0 || slave_speed == ((__u32) -1))
574 switch (ecmd.duplex) {
582 slave->speed = slave_speed;
583 slave->duplex = ecmd.duplex;
589 * if <dev> supports MII link status reporting, check its link status.
591 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
592 * depending upon the setting of the use_carrier parameter.
594 * Return either BMSR_LSTATUS, meaning that the link is up (or we
595 * can't tell and just pretend it is), or 0, meaning that the link is
598 * If reporting is non-zero, instead of faking link up, return -1 if
599 * both ETHTOOL and MII ioctls fail (meaning the device does not
600 * support them). If use_carrier is set, return whatever it says.
601 * It'd be nice if there was a good way to tell if a driver supports
602 * netif_carrier, but there really isn't.
604 static int bond_check_dev_link(struct bonding *bond,
605 struct net_device *slave_dev, int reporting)
607 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
608 int (*ioctl)(struct net_device *, struct ifreq *, int);
610 struct mii_ioctl_data *mii;
612 if (!reporting && !netif_running(slave_dev))
615 if (bond->params.use_carrier)
616 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
618 /* Try to get link status using Ethtool first. */
619 if (slave_dev->ethtool_ops) {
620 if (slave_dev->ethtool_ops->get_link) {
623 link = slave_dev->ethtool_ops->get_link(slave_dev);
625 return link ? BMSR_LSTATUS : 0;
629 /* Ethtool can't be used, fallback to MII ioctls. */
630 ioctl = slave_ops->ndo_do_ioctl;
632 /* TODO: set pointer to correct ioctl on a per team member */
633 /* bases to make this more efficient. that is, once */
634 /* we determine the correct ioctl, we will always */
635 /* call it and not the others for that team */
639 * We cannot assume that SIOCGMIIPHY will also read a
640 * register; not all network drivers (e.g., e100)
644 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
645 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
647 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
648 mii->reg_num = MII_BMSR;
649 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
650 return mii->val_out & BMSR_LSTATUS;
655 * If reporting, report that either there's no dev->do_ioctl,
656 * or both SIOCGMIIREG and get_link failed (meaning that we
657 * cannot report link status). If not reporting, pretend
660 return reporting ? -1 : BMSR_LSTATUS;
663 /*----------------------------- Multicast list ------------------------------*/
666 * Push the promiscuity flag down to appropriate slaves
668 static int bond_set_promiscuity(struct bonding *bond, int inc)
671 if (USES_PRIMARY(bond->params.mode)) {
672 /* write lock already acquired */
673 if (bond->curr_active_slave) {
674 err = dev_set_promiscuity(bond->curr_active_slave->dev,
680 bond_for_each_slave(bond, slave, i) {
681 err = dev_set_promiscuity(slave->dev, inc);
690 * Push the allmulti flag down to all slaves
692 static int bond_set_allmulti(struct bonding *bond, int inc)
695 if (USES_PRIMARY(bond->params.mode)) {
696 /* write lock already acquired */
697 if (bond->curr_active_slave) {
698 err = dev_set_allmulti(bond->curr_active_slave->dev,
704 bond_for_each_slave(bond, slave, i) {
705 err = dev_set_allmulti(slave->dev, inc);
714 * Add a Multicast address to slaves
717 static void bond_mc_add(struct bonding *bond, void *addr)
719 if (USES_PRIMARY(bond->params.mode)) {
720 /* write lock already acquired */
721 if (bond->curr_active_slave)
722 dev_mc_add(bond->curr_active_slave->dev, addr);
727 bond_for_each_slave(bond, slave, i)
728 dev_mc_add(slave->dev, addr);
733 * Remove a multicast address from slave
736 static void bond_mc_del(struct bonding *bond, void *addr)
738 if (USES_PRIMARY(bond->params.mode)) {
739 /* write lock already acquired */
740 if (bond->curr_active_slave)
741 dev_mc_del(bond->curr_active_slave->dev, addr);
745 bond_for_each_slave(bond, slave, i) {
746 dev_mc_del(slave->dev, addr);
752 static void __bond_resend_igmp_join_requests(struct net_device *dev)
754 struct in_device *in_dev;
757 in_dev = __in_dev_get_rcu(dev);
759 ip_mc_rejoin_groups(in_dev);
764 * Retrieve the list of registered multicast addresses for the bonding
765 * device and retransmit an IGMP JOIN request to the current active
768 static void bond_resend_igmp_join_requests(struct bonding *bond)
770 struct net_device *vlan_dev;
771 struct vlan_entry *vlan;
773 read_lock(&bond->lock);
775 /* rejoin all groups on bond device */
776 __bond_resend_igmp_join_requests(bond->dev);
778 /* rejoin all groups on vlan devices */
779 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
781 vlan_dev = __vlan_find_dev_deep(bond->dev,
785 __bond_resend_igmp_join_requests(vlan_dev);
788 if (--bond->igmp_retrans > 0)
789 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
791 read_unlock(&bond->lock);
794 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
796 struct bonding *bond = container_of(work, struct bonding,
798 bond_resend_igmp_join_requests(bond);
802 * flush all members of flush->mc_list from device dev->mc_list
804 static void bond_mc_list_flush(struct net_device *bond_dev,
805 struct net_device *slave_dev)
807 struct bonding *bond = netdev_priv(bond_dev);
808 struct netdev_hw_addr *ha;
810 netdev_for_each_mc_addr(ha, bond_dev)
811 dev_mc_del(slave_dev, ha->addr);
813 if (bond->params.mode == BOND_MODE_8023AD) {
814 /* del lacpdu mc addr from mc list */
815 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
817 dev_mc_del(slave_dev, lacpdu_multicast);
821 /*--------------------------- Active slave change ---------------------------*/
824 * Update the mc list and multicast-related flags for the new and
825 * old active slaves (if any) according to the multicast mode, and
826 * promiscuous flags unconditionally.
828 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
829 struct slave *old_active)
831 struct netdev_hw_addr *ha;
833 if (!USES_PRIMARY(bond->params.mode))
834 /* nothing to do - mc list is already up-to-date on
840 if (bond->dev->flags & IFF_PROMISC)
841 dev_set_promiscuity(old_active->dev, -1);
843 if (bond->dev->flags & IFF_ALLMULTI)
844 dev_set_allmulti(old_active->dev, -1);
846 netdev_for_each_mc_addr(ha, bond->dev)
847 dev_mc_del(old_active->dev, ha->addr);
851 /* FIXME: Signal errors upstream. */
852 if (bond->dev->flags & IFF_PROMISC)
853 dev_set_promiscuity(new_active->dev, 1);
855 if (bond->dev->flags & IFF_ALLMULTI)
856 dev_set_allmulti(new_active->dev, 1);
858 netdev_for_each_mc_addr(ha, bond->dev)
859 dev_mc_add(new_active->dev, ha->addr);
864 * bond_do_fail_over_mac
866 * Perform special MAC address swapping for fail_over_mac settings
868 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
870 static void bond_do_fail_over_mac(struct bonding *bond,
871 struct slave *new_active,
872 struct slave *old_active)
873 __releases(&bond->curr_slave_lock)
874 __releases(&bond->lock)
875 __acquires(&bond->lock)
876 __acquires(&bond->curr_slave_lock)
878 u8 tmp_mac[ETH_ALEN];
879 struct sockaddr saddr;
882 switch (bond->params.fail_over_mac) {
883 case BOND_FOM_ACTIVE:
885 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
886 new_active->dev->addr_len);
888 case BOND_FOM_FOLLOW:
890 * if new_active && old_active, swap them
891 * if just old_active, do nothing (going to no active slave)
892 * if just new_active, set new_active to bond's MAC
897 write_unlock_bh(&bond->curr_slave_lock);
898 read_unlock(&bond->lock);
901 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
902 memcpy(saddr.sa_data, old_active->dev->dev_addr,
904 saddr.sa_family = new_active->dev->type;
906 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
907 saddr.sa_family = bond->dev->type;
910 rv = dev_set_mac_address(new_active->dev, &saddr);
912 pr_err("%s: Error %d setting MAC of slave %s\n",
913 bond->dev->name, -rv, new_active->dev->name);
920 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
921 saddr.sa_family = old_active->dev->type;
923 rv = dev_set_mac_address(old_active->dev, &saddr);
925 pr_err("%s: Error %d setting MAC of slave %s\n",
926 bond->dev->name, -rv, new_active->dev->name);
928 read_lock(&bond->lock);
929 write_lock_bh(&bond->curr_slave_lock);
932 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
933 bond->dev->name, bond->params.fail_over_mac);
939 static bool bond_should_change_active(struct bonding *bond)
941 struct slave *prim = bond->primary_slave;
942 struct slave *curr = bond->curr_active_slave;
944 if (!prim || !curr || curr->link != BOND_LINK_UP)
946 if (bond->force_primary) {
947 bond->force_primary = false;
950 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
951 (prim->speed < curr->speed ||
952 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
954 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
960 * find_best_interface - select the best available slave to be the active one
961 * @bond: our bonding struct
963 * Warning: Caller must hold curr_slave_lock for writing.
965 static struct slave *bond_find_best_slave(struct bonding *bond)
967 struct slave *new_active, *old_active;
968 struct slave *bestslave = NULL;
969 int mintime = bond->params.updelay;
972 new_active = bond->curr_active_slave;
974 if (!new_active) { /* there were no active slaves left */
975 if (bond->slave_cnt > 0) /* found one slave */
976 new_active = bond->first_slave;
978 return NULL; /* still no slave, return NULL */
981 if ((bond->primary_slave) &&
982 bond->primary_slave->link == BOND_LINK_UP &&
983 bond_should_change_active(bond)) {
984 new_active = bond->primary_slave;
987 /* remember where to stop iterating over the slaves */
988 old_active = new_active;
990 bond_for_each_slave_from(bond, new_active, i, old_active) {
991 if (new_active->link == BOND_LINK_UP) {
993 } else if (new_active->link == BOND_LINK_BACK &&
994 IS_UP(new_active->dev)) {
995 /* link up, but waiting for stabilization */
996 if (new_active->delay < mintime) {
997 mintime = new_active->delay;
998 bestslave = new_active;
1006 static bool bond_should_notify_peers(struct bonding *bond)
1008 struct slave *slave = bond->curr_active_slave;
1010 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1011 bond->dev->name, slave ? slave->dev->name : "NULL");
1013 if (!slave || !bond->send_peer_notif ||
1014 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1017 bond->send_peer_notif--;
1022 * change_active_interface - change the active slave into the specified one
1023 * @bond: our bonding struct
1024 * @new: the new slave to make the active one
1026 * Set the new slave to the bond's settings and unset them on the old
1027 * curr_active_slave.
1028 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1030 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1031 * because it is apparently the best available slave we have, even though its
1032 * updelay hasn't timed out yet.
1034 * If new_active is not NULL, caller must hold bond->lock for read and
1035 * curr_slave_lock for write_bh.
1037 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1039 struct slave *old_active = bond->curr_active_slave;
1041 if (old_active == new_active)
1045 new_active->jiffies = jiffies;
1047 if (new_active->link == BOND_LINK_BACK) {
1048 if (USES_PRIMARY(bond->params.mode)) {
1049 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1050 bond->dev->name, new_active->dev->name,
1051 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1054 new_active->delay = 0;
1055 new_active->link = BOND_LINK_UP;
1057 if (bond->params.mode == BOND_MODE_8023AD)
1058 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1060 if (bond_is_lb(bond))
1061 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1063 if (USES_PRIMARY(bond->params.mode)) {
1064 pr_info("%s: making interface %s the new active one.\n",
1065 bond->dev->name, new_active->dev->name);
1070 if (USES_PRIMARY(bond->params.mode))
1071 bond_mc_swap(bond, new_active, old_active);
1073 if (bond_is_lb(bond)) {
1074 bond_alb_handle_active_change(bond, new_active);
1076 bond_set_slave_inactive_flags(old_active);
1078 bond_set_slave_active_flags(new_active);
1080 bond->curr_active_slave = new_active;
1083 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1085 bond_set_slave_inactive_flags(old_active);
1088 bool should_notify_peers = false;
1090 bond_set_slave_active_flags(new_active);
1092 if (bond->params.fail_over_mac)
1093 bond_do_fail_over_mac(bond, new_active,
1096 if (netif_running(bond->dev)) {
1097 bond->send_peer_notif =
1098 bond->params.num_peer_notif;
1099 should_notify_peers =
1100 bond_should_notify_peers(bond);
1103 write_unlock_bh(&bond->curr_slave_lock);
1104 read_unlock(&bond->lock);
1106 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1107 if (should_notify_peers)
1108 netdev_bonding_change(bond->dev,
1109 NETDEV_NOTIFY_PEERS);
1111 read_lock(&bond->lock);
1112 write_lock_bh(&bond->curr_slave_lock);
1116 /* resend IGMP joins since active slave has changed or
1117 * all were sent on curr_active_slave.
1118 * resend only if bond is brought up with the affected
1119 * bonding modes and the retransmission is enabled */
1120 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1121 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1122 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1123 bond->igmp_retrans = bond->params.resend_igmp;
1124 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1129 * bond_select_active_slave - select a new active slave, if needed
1130 * @bond: our bonding struct
1132 * This functions should be called when one of the following occurs:
1133 * - The old curr_active_slave has been released or lost its link.
1134 * - The primary_slave has got its link back.
1135 * - A slave has got its link back and there's no old curr_active_slave.
1137 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1139 void bond_select_active_slave(struct bonding *bond)
1141 struct slave *best_slave;
1144 best_slave = bond_find_best_slave(bond);
1145 if (best_slave != bond->curr_active_slave) {
1146 bond_change_active_slave(bond, best_slave);
1147 rv = bond_set_carrier(bond);
1151 if (netif_carrier_ok(bond->dev)) {
1152 pr_info("%s: first active interface up!\n",
1155 pr_info("%s: now running without any active interface !\n",
1161 /*--------------------------- slave list handling ---------------------------*/
1164 * This function attaches the slave to the end of list.
1166 * bond->lock held for writing by caller.
1168 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1170 if (bond->first_slave == NULL) { /* attaching the first slave */
1171 new_slave->next = new_slave;
1172 new_slave->prev = new_slave;
1173 bond->first_slave = new_slave;
1175 new_slave->next = bond->first_slave;
1176 new_slave->prev = bond->first_slave->prev;
1177 new_slave->next->prev = new_slave;
1178 new_slave->prev->next = new_slave;
1185 * This function detaches the slave from the list.
1186 * WARNING: no check is made to verify if the slave effectively
1187 * belongs to <bond>.
1188 * Nothing is freed on return, structures are just unchained.
1189 * If any slave pointer in bond was pointing to <slave>,
1190 * it should be changed by the calling function.
1192 * bond->lock held for writing by caller.
1194 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1197 slave->next->prev = slave->prev;
1200 slave->prev->next = slave->next;
1202 if (bond->first_slave == slave) { /* slave is the first slave */
1203 if (bond->slave_cnt > 1) { /* there are more slave */
1204 bond->first_slave = slave->next;
1206 bond->first_slave = NULL; /* slave was the last one */
1215 #ifdef CONFIG_NET_POLL_CONTROLLER
1216 static inline int slave_enable_netpoll(struct slave *slave)
1221 np = kzalloc(sizeof(*np), GFP_KERNEL);
1226 np->dev = slave->dev;
1227 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1228 err = __netpoll_setup(np);
1237 static inline void slave_disable_netpoll(struct slave *slave)
1239 struct netpoll *np = slave->np;
1245 synchronize_rcu_bh();
1246 __netpoll_cleanup(np);
1249 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1251 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1253 if (!slave_dev->netdev_ops->ndo_poll_controller)
1258 static void bond_poll_controller(struct net_device *bond_dev)
1262 static void __bond_netpoll_cleanup(struct bonding *bond)
1264 struct slave *slave;
1267 bond_for_each_slave(bond, slave, i)
1268 if (IS_UP(slave->dev))
1269 slave_disable_netpoll(slave);
1271 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1273 struct bonding *bond = netdev_priv(bond_dev);
1275 read_lock(&bond->lock);
1276 __bond_netpoll_cleanup(bond);
1277 read_unlock(&bond->lock);
1280 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1282 struct bonding *bond = netdev_priv(dev);
1283 struct slave *slave;
1286 read_lock(&bond->lock);
1287 bond_for_each_slave(bond, slave, i) {
1288 err = slave_enable_netpoll(slave);
1290 __bond_netpoll_cleanup(bond);
1294 read_unlock(&bond->lock);
1298 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1300 return bond->dev->npinfo;
1304 static inline int slave_enable_netpoll(struct slave *slave)
1308 static inline void slave_disable_netpoll(struct slave *slave)
1311 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1316 /*---------------------------------- IOCTL ----------------------------------*/
1318 static int bond_sethwaddr(struct net_device *bond_dev,
1319 struct net_device *slave_dev)
1321 pr_debug("bond_dev=%p\n", bond_dev);
1322 pr_debug("slave_dev=%p\n", slave_dev);
1323 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1324 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1328 static netdev_features_t bond_fix_features(struct net_device *dev,
1329 netdev_features_t features)
1331 struct slave *slave;
1332 struct bonding *bond = netdev_priv(dev);
1333 netdev_features_t mask;
1336 read_lock(&bond->lock);
1338 if (!bond->first_slave) {
1339 /* Disable adding VLANs to empty bond. But why? --mq */
1340 features |= NETIF_F_VLAN_CHALLENGED;
1345 features &= ~NETIF_F_ONE_FOR_ALL;
1346 features |= NETIF_F_ALL_FOR_ALL;
1348 bond_for_each_slave(bond, slave, i) {
1349 features = netdev_increment_features(features,
1350 slave->dev->features,
1355 read_unlock(&bond->lock);
1359 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1360 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1361 NETIF_F_HIGHDMA | NETIF_F_LRO)
1363 static void bond_compute_features(struct bonding *bond)
1365 struct slave *slave;
1366 struct net_device *bond_dev = bond->dev;
1367 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1368 unsigned short max_hard_header_len = ETH_HLEN;
1371 read_lock(&bond->lock);
1373 if (!bond->first_slave)
1376 bond_for_each_slave(bond, slave, i) {
1377 vlan_features = netdev_increment_features(vlan_features,
1378 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1380 if (slave->dev->hard_header_len > max_hard_header_len)
1381 max_hard_header_len = slave->dev->hard_header_len;
1385 bond_dev->vlan_features = vlan_features;
1386 bond_dev->hard_header_len = max_hard_header_len;
1388 read_unlock(&bond->lock);
1390 netdev_change_features(bond_dev);
1393 static void bond_setup_by_slave(struct net_device *bond_dev,
1394 struct net_device *slave_dev)
1396 struct bonding *bond = netdev_priv(bond_dev);
1398 bond_dev->header_ops = slave_dev->header_ops;
1400 bond_dev->type = slave_dev->type;
1401 bond_dev->hard_header_len = slave_dev->hard_header_len;
1402 bond_dev->addr_len = slave_dev->addr_len;
1404 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1405 slave_dev->addr_len);
1406 bond->setup_by_slave = 1;
1409 /* On bonding slaves other than the currently active slave, suppress
1410 * duplicates except for alb non-mcast/bcast.
1412 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1413 struct slave *slave,
1414 struct bonding *bond)
1416 if (bond_is_slave_inactive(slave)) {
1417 if (bond->params.mode == BOND_MODE_ALB &&
1418 skb->pkt_type != PACKET_BROADCAST &&
1419 skb->pkt_type != PACKET_MULTICAST)
1426 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1428 struct sk_buff *skb = *pskb;
1429 struct slave *slave;
1430 struct bonding *bond;
1431 void (*recv_probe)(struct sk_buff *, struct bonding *,
1434 skb = skb_share_check(skb, GFP_ATOMIC);
1436 return RX_HANDLER_CONSUMED;
1440 slave = bond_slave_get_rcu(skb->dev);
1443 if (bond->params.arp_interval)
1444 slave->dev->last_rx = jiffies;
1446 recv_probe = ACCESS_ONCE(bond->recv_probe);
1448 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1451 recv_probe(nskb, bond, slave);
1452 dev_kfree_skb(nskb);
1456 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1457 return RX_HANDLER_EXACT;
1460 skb->dev = bond->dev;
1462 if (bond->params.mode == BOND_MODE_ALB &&
1463 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1464 skb->pkt_type == PACKET_HOST) {
1466 if (unlikely(skb_cow_head(skb,
1467 skb->data - skb_mac_header(skb)))) {
1469 return RX_HANDLER_CONSUMED;
1471 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1474 return RX_HANDLER_ANOTHER;
1477 /* enslave device <slave> to bond device <master> */
1478 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1480 struct bonding *bond = netdev_priv(bond_dev);
1481 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1482 struct slave *new_slave = NULL;
1483 struct netdev_hw_addr *ha;
1484 struct sockaddr addr;
1488 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1489 slave_ops->ndo_do_ioctl == NULL) {
1490 pr_warning("%s: Warning: no link monitoring support for %s\n",
1491 bond_dev->name, slave_dev->name);
1494 /* already enslaved */
1495 if (slave_dev->flags & IFF_SLAVE) {
1496 pr_debug("Error, Device was already enslaved\n");
1500 /* vlan challenged mutual exclusion */
1501 /* no need to lock since we're protected by rtnl_lock */
1502 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1503 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1504 if (bond_vlan_used(bond)) {
1505 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1506 bond_dev->name, slave_dev->name, bond_dev->name);
1509 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1510 bond_dev->name, slave_dev->name,
1511 slave_dev->name, bond_dev->name);
1514 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1518 * Old ifenslave binaries are no longer supported. These can
1519 * be identified with moderate accuracy by the state of the slave:
1520 * the current ifenslave will set the interface down prior to
1521 * enslaving it; the old ifenslave will not.
1523 if ((slave_dev->flags & IFF_UP)) {
1524 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1527 goto err_undo_flags;
1530 /* set bonding device ether type by slave - bonding netdevices are
1531 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1532 * there is a need to override some of the type dependent attribs/funcs.
1534 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1535 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1537 if (bond->slave_cnt == 0) {
1538 if (bond_dev->type != slave_dev->type) {
1539 pr_debug("%s: change device type from %d to %d\n",
1541 bond_dev->type, slave_dev->type);
1543 res = netdev_bonding_change(bond_dev,
1544 NETDEV_PRE_TYPE_CHANGE);
1545 res = notifier_to_errno(res);
1547 pr_err("%s: refused to change device type\n",
1550 goto err_undo_flags;
1553 /* Flush unicast and multicast addresses */
1554 dev_uc_flush(bond_dev);
1555 dev_mc_flush(bond_dev);
1557 if (slave_dev->type != ARPHRD_ETHER)
1558 bond_setup_by_slave(bond_dev, slave_dev);
1560 ether_setup(bond_dev);
1561 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1564 netdev_bonding_change(bond_dev,
1565 NETDEV_POST_TYPE_CHANGE);
1567 } else if (bond_dev->type != slave_dev->type) {
1568 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1570 slave_dev->type, bond_dev->type);
1572 goto err_undo_flags;
1575 if (slave_ops->ndo_set_mac_address == NULL) {
1576 if (bond->slave_cnt == 0) {
1577 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1579 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1580 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1581 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",
1584 goto err_undo_flags;
1588 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1590 /* If this is the first slave, then we need to set the master's hardware
1591 * address to be the same as the slave's. */
1592 if (is_zero_ether_addr(bond->dev->dev_addr))
1593 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1594 slave_dev->addr_len);
1597 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1600 goto err_undo_flags;
1604 * Set the new_slave's queue_id to be zero. Queue ID mapping
1605 * is set via sysfs or module option if desired.
1607 new_slave->queue_id = 0;
1609 /* Save slave's original mtu and then set it to match the bond */
1610 new_slave->original_mtu = slave_dev->mtu;
1611 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1613 pr_debug("Error %d calling dev_set_mtu\n", res);
1618 * Save slave's original ("permanent") mac address for modes
1619 * that need it, and for restoring it upon release, and then
1620 * set it to the master's address
1622 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1624 if (!bond->params.fail_over_mac) {
1626 * Set slave to master's mac address. The application already
1627 * set the master's mac address to that of the first slave
1629 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1630 addr.sa_family = slave_dev->type;
1631 res = dev_set_mac_address(slave_dev, &addr);
1633 pr_debug("Error %d calling set_mac_address\n", res);
1634 goto err_restore_mtu;
1638 res = netdev_set_bond_master(slave_dev, bond_dev);
1640 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1641 goto err_restore_mac;
1644 /* open the slave since the application closed it */
1645 res = dev_open(slave_dev);
1647 pr_debug("Opening slave %s failed\n", slave_dev->name);
1648 goto err_unset_master;
1651 new_slave->bond = bond;
1652 new_slave->dev = slave_dev;
1653 slave_dev->priv_flags |= IFF_BONDING;
1655 if (bond_is_lb(bond)) {
1656 /* bond_alb_init_slave() must be called before all other stages since
1657 * it might fail and we do not want to have to undo everything
1659 res = bond_alb_init_slave(bond, new_slave);
1664 /* If the mode USES_PRIMARY, then the new slave gets the
1665 * master's promisc (and mc) settings only if it becomes the
1666 * curr_active_slave, and that is taken care of later when calling
1667 * bond_change_active()
1669 if (!USES_PRIMARY(bond->params.mode)) {
1670 /* set promiscuity level to new slave */
1671 if (bond_dev->flags & IFF_PROMISC) {
1672 res = dev_set_promiscuity(slave_dev, 1);
1677 /* set allmulti level to new slave */
1678 if (bond_dev->flags & IFF_ALLMULTI) {
1679 res = dev_set_allmulti(slave_dev, 1);
1684 netif_addr_lock_bh(bond_dev);
1685 /* upload master's mc_list to new slave */
1686 netdev_for_each_mc_addr(ha, bond_dev)
1687 dev_mc_add(slave_dev, ha->addr);
1688 netif_addr_unlock_bh(bond_dev);
1691 if (bond->params.mode == BOND_MODE_8023AD) {
1692 /* add lacpdu mc addr to mc list */
1693 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1695 dev_mc_add(slave_dev, lacpdu_multicast);
1698 bond_add_vlans_on_slave(bond, slave_dev);
1700 write_lock_bh(&bond->lock);
1702 bond_attach_slave(bond, new_slave);
1704 new_slave->delay = 0;
1705 new_slave->link_failure_count = 0;
1707 write_unlock_bh(&bond->lock);
1709 bond_compute_features(bond);
1711 read_lock(&bond->lock);
1713 new_slave->last_arp_rx = jiffies;
1715 if (bond->params.miimon && !bond->params.use_carrier) {
1716 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1718 if ((link_reporting == -1) && !bond->params.arp_interval) {
1720 * miimon is set but a bonded network driver
1721 * does not support ETHTOOL/MII and
1722 * arp_interval is not set. Note: if
1723 * use_carrier is enabled, we will never go
1724 * here (because netif_carrier is always
1725 * supported); thus, we don't need to change
1726 * the messages for netif_carrier.
1728 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",
1729 bond_dev->name, slave_dev->name);
1730 } else if (link_reporting == -1) {
1731 /* unable get link status using mii/ethtool */
1732 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",
1733 bond_dev->name, slave_dev->name);
1737 /* check for initial state */
1738 if (!bond->params.miimon ||
1739 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1740 if (bond->params.updelay) {
1741 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1742 new_slave->link = BOND_LINK_BACK;
1743 new_slave->delay = bond->params.updelay;
1745 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1746 new_slave->link = BOND_LINK_UP;
1748 new_slave->jiffies = jiffies;
1750 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1751 new_slave->link = BOND_LINK_DOWN;
1754 bond_update_speed_duplex(new_slave);
1756 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1757 /* if there is a primary slave, remember it */
1758 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1759 bond->primary_slave = new_slave;
1760 bond->force_primary = true;
1764 write_lock_bh(&bond->curr_slave_lock);
1766 switch (bond->params.mode) {
1767 case BOND_MODE_ACTIVEBACKUP:
1768 bond_set_slave_inactive_flags(new_slave);
1769 bond_select_active_slave(bond);
1771 case BOND_MODE_8023AD:
1772 /* in 802.3ad mode, the internal mechanism
1773 * will activate the slaves in the selected
1776 bond_set_slave_inactive_flags(new_slave);
1777 /* if this is the first slave */
1778 if (bond->slave_cnt == 1) {
1779 SLAVE_AD_INFO(new_slave).id = 1;
1780 /* Initialize AD with the number of times that the AD timer is called in 1 second
1781 * can be called only after the mac address of the bond is set
1783 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1785 SLAVE_AD_INFO(new_slave).id =
1786 SLAVE_AD_INFO(new_slave->prev).id + 1;
1789 bond_3ad_bind_slave(new_slave);
1793 bond_set_active_slave(new_slave);
1794 bond_set_slave_inactive_flags(new_slave);
1795 bond_select_active_slave(bond);
1798 pr_debug("This slave is always active in trunk mode\n");
1800 /* always active in trunk mode */
1801 bond_set_active_slave(new_slave);
1803 /* In trunking mode there is little meaning to curr_active_slave
1804 * anyway (it holds no special properties of the bond device),
1805 * so we can change it without calling change_active_interface()
1807 if (!bond->curr_active_slave)
1808 bond->curr_active_slave = new_slave;
1811 } /* switch(bond_mode) */
1813 write_unlock_bh(&bond->curr_slave_lock);
1815 bond_set_carrier(bond);
1817 #ifdef CONFIG_NET_POLL_CONTROLLER
1818 slave_dev->npinfo = bond_netpoll_info(bond);
1819 if (slave_dev->npinfo) {
1820 if (slave_enable_netpoll(new_slave)) {
1821 read_unlock(&bond->lock);
1822 pr_info("Error, %s: master_dev is using netpoll, "
1823 "but new slave device does not support netpoll.\n",
1831 read_unlock(&bond->lock);
1833 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1837 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1840 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1841 goto err_dest_symlinks;
1844 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1845 bond_dev->name, slave_dev->name,
1846 bond_is_active_slave(new_slave) ? "n active" : " backup",
1847 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1849 /* enslave is successful */
1852 /* Undo stages on error */
1854 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1857 dev_close(slave_dev);
1860 netdev_set_bond_master(slave_dev, NULL);
1863 if (!bond->params.fail_over_mac) {
1864 /* XXX TODO - fom follow mode needs to change master's
1865 * MAC if this slave's MAC is in use by the bond, or at
1866 * least print a warning.
1868 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1869 addr.sa_family = slave_dev->type;
1870 dev_set_mac_address(slave_dev, &addr);
1874 dev_set_mtu(slave_dev, new_slave->original_mtu);
1880 bond_compute_features(bond);
1886 * Try to release the slave device <slave> from the bond device <master>
1887 * It is legal to access curr_active_slave without a lock because all the function
1890 * The rules for slave state should be:
1891 * for Active/Backup:
1892 * Active stays on all backups go down
1893 * for Bonded connections:
1894 * The first up interface should be left on and all others downed.
1896 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1898 struct bonding *bond = netdev_priv(bond_dev);
1899 struct slave *slave, *oldcurrent;
1900 struct sockaddr addr;
1901 netdev_features_t old_features = bond_dev->features;
1903 /* slave is not a slave or master is not master of this slave */
1904 if (!(slave_dev->flags & IFF_SLAVE) ||
1905 (slave_dev->master != bond_dev)) {
1906 pr_err("%s: Error: cannot release %s.\n",
1907 bond_dev->name, slave_dev->name);
1912 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1913 write_lock_bh(&bond->lock);
1915 slave = bond_get_slave_by_dev(bond, slave_dev);
1917 /* not a slave of this bond */
1918 pr_info("%s: %s not enslaved\n",
1919 bond_dev->name, slave_dev->name);
1920 write_unlock_bh(&bond->lock);
1921 unblock_netpoll_tx();
1925 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1926 * for this slave anymore.
1928 netdev_rx_handler_unregister(slave_dev);
1929 write_unlock_bh(&bond->lock);
1931 write_lock_bh(&bond->lock);
1933 if (!bond->params.fail_over_mac) {
1934 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1935 bond->slave_cnt > 1)
1936 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",
1937 bond_dev->name, slave_dev->name,
1939 bond_dev->name, slave_dev->name);
1942 /* Inform AD package of unbinding of slave. */
1943 if (bond->params.mode == BOND_MODE_8023AD) {
1944 /* must be called before the slave is
1945 * detached from the list
1947 bond_3ad_unbind_slave(slave);
1950 pr_info("%s: releasing %s interface %s\n",
1952 bond_is_active_slave(slave) ? "active" : "backup",
1955 oldcurrent = bond->curr_active_slave;
1957 bond->current_arp_slave = NULL;
1959 /* release the slave from its bond */
1960 bond_detach_slave(bond, slave);
1962 if (bond->primary_slave == slave)
1963 bond->primary_slave = NULL;
1965 if (oldcurrent == slave)
1966 bond_change_active_slave(bond, NULL);
1968 if (bond_is_lb(bond)) {
1969 /* Must be called only after the slave has been
1970 * detached from the list and the curr_active_slave
1971 * has been cleared (if our_slave == old_current),
1972 * but before a new active slave is selected.
1974 write_unlock_bh(&bond->lock);
1975 bond_alb_deinit_slave(bond, slave);
1976 write_lock_bh(&bond->lock);
1979 if (oldcurrent == slave) {
1981 * Note that we hold RTNL over this sequence, so there
1982 * is no concern that another slave add/remove event
1985 write_unlock_bh(&bond->lock);
1986 read_lock(&bond->lock);
1987 write_lock_bh(&bond->curr_slave_lock);
1989 bond_select_active_slave(bond);
1991 write_unlock_bh(&bond->curr_slave_lock);
1992 read_unlock(&bond->lock);
1993 write_lock_bh(&bond->lock);
1996 if (bond->slave_cnt == 0) {
1997 bond_set_carrier(bond);
1999 /* if the last slave was removed, zero the mac address
2000 * of the master so it will be set by the application
2001 * to the mac address of the first slave
2003 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2005 if (bond_vlan_used(bond)) {
2006 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2007 bond_dev->name, bond_dev->name);
2008 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2013 write_unlock_bh(&bond->lock);
2014 unblock_netpoll_tx();
2016 bond_compute_features(bond);
2017 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2018 (old_features & NETIF_F_VLAN_CHALLENGED))
2019 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2020 bond_dev->name, slave_dev->name, bond_dev->name);
2022 /* must do this from outside any spinlocks */
2023 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2025 bond_del_vlans_from_slave(bond, slave_dev);
2027 /* If the mode USES_PRIMARY, then we should only remove its
2028 * promisc and mc settings if it was the curr_active_slave, but that was
2029 * already taken care of above when we detached the slave
2031 if (!USES_PRIMARY(bond->params.mode)) {
2032 /* unset promiscuity level from slave */
2033 if (bond_dev->flags & IFF_PROMISC)
2034 dev_set_promiscuity(slave_dev, -1);
2036 /* unset allmulti level from slave */
2037 if (bond_dev->flags & IFF_ALLMULTI)
2038 dev_set_allmulti(slave_dev, -1);
2040 /* flush master's mc_list from slave */
2041 netif_addr_lock_bh(bond_dev);
2042 bond_mc_list_flush(bond_dev, slave_dev);
2043 netif_addr_unlock_bh(bond_dev);
2046 netdev_set_bond_master(slave_dev, NULL);
2048 slave_disable_netpoll(slave);
2050 /* close slave before restoring its mac address */
2051 dev_close(slave_dev);
2053 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2054 /* restore original ("permanent") mac address */
2055 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2056 addr.sa_family = slave_dev->type;
2057 dev_set_mac_address(slave_dev, &addr);
2060 dev_set_mtu(slave_dev, slave->original_mtu);
2062 slave_dev->priv_flags &= ~IFF_BONDING;
2066 return 0; /* deletion OK */
2070 * First release a slave and then destroy the bond if no more slaves are left.
2071 * Must be under rtnl_lock when this function is called.
2073 static int bond_release_and_destroy(struct net_device *bond_dev,
2074 struct net_device *slave_dev)
2076 struct bonding *bond = netdev_priv(bond_dev);
2079 ret = bond_release(bond_dev, slave_dev);
2080 if ((ret == 0) && (bond->slave_cnt == 0)) {
2081 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2082 pr_info("%s: destroying bond %s.\n",
2083 bond_dev->name, bond_dev->name);
2084 unregister_netdevice(bond_dev);
2090 * This function releases all slaves.
2092 static int bond_release_all(struct net_device *bond_dev)
2094 struct bonding *bond = netdev_priv(bond_dev);
2095 struct slave *slave;
2096 struct net_device *slave_dev;
2097 struct sockaddr addr;
2099 write_lock_bh(&bond->lock);
2101 netif_carrier_off(bond_dev);
2103 if (bond->slave_cnt == 0)
2106 bond->current_arp_slave = NULL;
2107 bond->primary_slave = NULL;
2108 bond_change_active_slave(bond, NULL);
2110 while ((slave = bond->first_slave) != NULL) {
2111 /* Inform AD package of unbinding of slave
2112 * before slave is detached from the list.
2114 if (bond->params.mode == BOND_MODE_8023AD)
2115 bond_3ad_unbind_slave(slave);
2117 slave_dev = slave->dev;
2118 bond_detach_slave(bond, slave);
2120 /* now that the slave is detached, unlock and perform
2121 * all the undo steps that should not be called from
2124 write_unlock_bh(&bond->lock);
2126 /* unregister rx_handler early so bond_handle_frame wouldn't
2127 * be called for this slave anymore.
2129 netdev_rx_handler_unregister(slave_dev);
2132 if (bond_is_lb(bond)) {
2133 /* must be called only after the slave
2134 * has been detached from the list
2136 bond_alb_deinit_slave(bond, slave);
2139 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2140 bond_del_vlans_from_slave(bond, slave_dev);
2142 /* If the mode USES_PRIMARY, then we should only remove its
2143 * promisc and mc settings if it was the curr_active_slave, but that was
2144 * already taken care of above when we detached the slave
2146 if (!USES_PRIMARY(bond->params.mode)) {
2147 /* unset promiscuity level from slave */
2148 if (bond_dev->flags & IFF_PROMISC)
2149 dev_set_promiscuity(slave_dev, -1);
2151 /* unset allmulti level from slave */
2152 if (bond_dev->flags & IFF_ALLMULTI)
2153 dev_set_allmulti(slave_dev, -1);
2155 /* flush master's mc_list from slave */
2156 netif_addr_lock_bh(bond_dev);
2157 bond_mc_list_flush(bond_dev, slave_dev);
2158 netif_addr_unlock_bh(bond_dev);
2161 netdev_set_bond_master(slave_dev, NULL);
2163 slave_disable_netpoll(slave);
2165 /* close slave before restoring its mac address */
2166 dev_close(slave_dev);
2168 if (!bond->params.fail_over_mac) {
2169 /* restore original ("permanent") mac address*/
2170 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2171 addr.sa_family = slave_dev->type;
2172 dev_set_mac_address(slave_dev, &addr);
2177 /* re-acquire the lock before getting the next slave */
2178 write_lock_bh(&bond->lock);
2181 /* zero the mac address of the master so it will be
2182 * set by the application to the mac address of the
2185 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2187 if (bond_vlan_used(bond)) {
2188 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2189 bond_dev->name, bond_dev->name);
2190 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2194 pr_info("%s: released all slaves\n", bond_dev->name);
2197 write_unlock_bh(&bond->lock);
2199 bond_compute_features(bond);
2205 * This function changes the active slave to slave <slave_dev>.
2206 * It returns -EINVAL in the following cases.
2207 * - <slave_dev> is not found in the list.
2208 * - There is not active slave now.
2209 * - <slave_dev> is already active.
2210 * - The link state of <slave_dev> is not BOND_LINK_UP.
2211 * - <slave_dev> is not running.
2212 * In these cases, this function does nothing.
2213 * In the other cases, current_slave pointer is changed and 0 is returned.
2215 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2217 struct bonding *bond = netdev_priv(bond_dev);
2218 struct slave *old_active = NULL;
2219 struct slave *new_active = NULL;
2222 if (!USES_PRIMARY(bond->params.mode))
2225 /* Verify that master_dev is indeed the master of slave_dev */
2226 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2229 read_lock(&bond->lock);
2231 read_lock(&bond->curr_slave_lock);
2232 old_active = bond->curr_active_slave;
2233 read_unlock(&bond->curr_slave_lock);
2235 new_active = bond_get_slave_by_dev(bond, slave_dev);
2238 * Changing to the current active: do nothing; return success.
2240 if (new_active && (new_active == old_active)) {
2241 read_unlock(&bond->lock);
2247 (new_active->link == BOND_LINK_UP) &&
2248 IS_UP(new_active->dev)) {
2250 write_lock_bh(&bond->curr_slave_lock);
2251 bond_change_active_slave(bond, new_active);
2252 write_unlock_bh(&bond->curr_slave_lock);
2253 unblock_netpoll_tx();
2257 read_unlock(&bond->lock);
2262 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2264 struct bonding *bond = netdev_priv(bond_dev);
2266 info->bond_mode = bond->params.mode;
2267 info->miimon = bond->params.miimon;
2269 read_lock(&bond->lock);
2270 info->num_slaves = bond->slave_cnt;
2271 read_unlock(&bond->lock);
2276 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2278 struct bonding *bond = netdev_priv(bond_dev);
2279 struct slave *slave;
2280 int i, res = -ENODEV;
2282 read_lock(&bond->lock);
2284 bond_for_each_slave(bond, slave, i) {
2285 if (i == (int)info->slave_id) {
2287 strcpy(info->slave_name, slave->dev->name);
2288 info->link = slave->link;
2289 info->state = bond_slave_state(slave);
2290 info->link_failure_count = slave->link_failure_count;
2295 read_unlock(&bond->lock);
2300 /*-------------------------------- Monitoring -------------------------------*/
2303 static int bond_miimon_inspect(struct bonding *bond)
2305 struct slave *slave;
2306 int i, link_state, commit = 0;
2307 bool ignore_updelay;
2309 ignore_updelay = !bond->curr_active_slave ? true : false;
2311 bond_for_each_slave(bond, slave, i) {
2312 slave->new_link = BOND_LINK_NOCHANGE;
2314 link_state = bond_check_dev_link(bond, slave->dev, 0);
2316 switch (slave->link) {
2321 slave->link = BOND_LINK_FAIL;
2322 slave->delay = bond->params.downdelay;
2324 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2326 (bond->params.mode ==
2327 BOND_MODE_ACTIVEBACKUP) ?
2328 (bond_is_active_slave(slave) ?
2329 "active " : "backup ") : "",
2331 bond->params.downdelay * bond->params.miimon);
2334 case BOND_LINK_FAIL:
2337 * recovered before downdelay expired
2339 slave->link = BOND_LINK_UP;
2340 slave->jiffies = jiffies;
2341 pr_info("%s: link status up again after %d ms for interface %s.\n",
2343 (bond->params.downdelay - slave->delay) *
2344 bond->params.miimon,
2349 if (slave->delay <= 0) {
2350 slave->new_link = BOND_LINK_DOWN;
2358 case BOND_LINK_DOWN:
2362 slave->link = BOND_LINK_BACK;
2363 slave->delay = bond->params.updelay;
2366 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2367 bond->dev->name, slave->dev->name,
2368 ignore_updelay ? 0 :
2369 bond->params.updelay *
2370 bond->params.miimon);
2373 case BOND_LINK_BACK:
2375 slave->link = BOND_LINK_DOWN;
2376 pr_info("%s: link status down again after %d ms for interface %s.\n",
2378 (bond->params.updelay - slave->delay) *
2379 bond->params.miimon,
2388 if (slave->delay <= 0) {
2389 slave->new_link = BOND_LINK_UP;
2391 ignore_updelay = false;
2403 static void bond_miimon_commit(struct bonding *bond)
2405 struct slave *slave;
2408 bond_for_each_slave(bond, slave, i) {
2409 switch (slave->new_link) {
2410 case BOND_LINK_NOCHANGE:
2414 slave->link = BOND_LINK_UP;
2415 slave->jiffies = jiffies;
2417 if (bond->params.mode == BOND_MODE_8023AD) {
2418 /* prevent it from being the active one */
2419 bond_set_backup_slave(slave);
2420 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2421 /* make it immediately active */
2422 bond_set_active_slave(slave);
2423 } else if (slave != bond->primary_slave) {
2424 /* prevent it from being the active one */
2425 bond_set_backup_slave(slave);
2428 bond_update_speed_duplex(slave);
2430 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2431 bond->dev->name, slave->dev->name,
2432 slave->speed, slave->duplex ? "full" : "half");
2434 /* notify ad that the link status has changed */
2435 if (bond->params.mode == BOND_MODE_8023AD)
2436 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2438 if (bond_is_lb(bond))
2439 bond_alb_handle_link_change(bond, slave,
2442 if (!bond->curr_active_slave ||
2443 (slave == bond->primary_slave))
2448 case BOND_LINK_DOWN:
2449 if (slave->link_failure_count < UINT_MAX)
2450 slave->link_failure_count++;
2452 slave->link = BOND_LINK_DOWN;
2454 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2455 bond->params.mode == BOND_MODE_8023AD)
2456 bond_set_slave_inactive_flags(slave);
2458 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2459 bond->dev->name, slave->dev->name);
2461 if (bond->params.mode == BOND_MODE_8023AD)
2462 bond_3ad_handle_link_change(slave,
2465 if (bond_is_lb(bond))
2466 bond_alb_handle_link_change(bond, slave,
2469 if (slave == bond->curr_active_slave)
2475 pr_err("%s: invalid new link %d on slave %s\n",
2476 bond->dev->name, slave->new_link,
2478 slave->new_link = BOND_LINK_NOCHANGE;
2486 write_lock_bh(&bond->curr_slave_lock);
2487 bond_select_active_slave(bond);
2488 write_unlock_bh(&bond->curr_slave_lock);
2489 unblock_netpoll_tx();
2492 bond_set_carrier(bond);
2498 * Really a wrapper that splits the mii monitor into two phases: an
2499 * inspection, then (if inspection indicates something needs to be done)
2500 * an acquisition of appropriate locks followed by a commit phase to
2501 * implement whatever link state changes are indicated.
2503 void bond_mii_monitor(struct work_struct *work)
2505 struct bonding *bond = container_of(work, struct bonding,
2507 bool should_notify_peers = false;
2508 unsigned long delay;
2510 read_lock(&bond->lock);
2512 delay = msecs_to_jiffies(bond->params.miimon);
2514 if (bond->slave_cnt == 0)
2517 should_notify_peers = bond_should_notify_peers(bond);
2519 if (bond_miimon_inspect(bond)) {
2520 read_unlock(&bond->lock);
2522 /* Race avoidance with bond_close cancel of workqueue */
2523 if (!rtnl_trylock()) {
2524 read_lock(&bond->lock);
2526 should_notify_peers = false;
2530 read_lock(&bond->lock);
2532 bond_miimon_commit(bond);
2534 read_unlock(&bond->lock);
2535 rtnl_unlock(); /* might sleep, hold no other locks */
2536 read_lock(&bond->lock);
2540 if (bond->params.miimon)
2541 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2543 read_unlock(&bond->lock);
2545 if (should_notify_peers) {
2546 if (!rtnl_trylock()) {
2547 read_lock(&bond->lock);
2548 bond->send_peer_notif++;
2549 read_unlock(&bond->lock);
2552 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2557 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2559 struct vlan_entry *vlan;
2561 if (ip == bond->master_ip)
2564 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2565 if (ip == vlan->vlan_ip)
2573 * We go to the (large) trouble of VLAN tagging ARP frames because
2574 * switches in VLAN mode (especially if ports are configured as
2575 * "native" to a VLAN) might not pass non-tagged frames.
2577 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2579 struct sk_buff *skb;
2581 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2582 slave_dev->name, dest_ip, src_ip, vlan_id);
2584 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2585 NULL, slave_dev->dev_addr, NULL);
2588 pr_err("ARP packet allocation failed\n");
2592 skb = vlan_put_tag(skb, vlan_id);
2594 pr_err("failed to insert VLAN tag\n");
2602 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2605 __be32 *targets = bond->params.arp_targets;
2606 struct vlan_entry *vlan;
2607 struct net_device *vlan_dev;
2610 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2613 pr_debug("basa: target %x\n", targets[i]);
2614 if (!bond_vlan_used(bond)) {
2615 pr_debug("basa: empty vlan: arp_send\n");
2616 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617 bond->master_ip, 0);
2622 * If VLANs are configured, we do a route lookup to
2623 * determine which VLAN interface would be used, so we
2624 * can tag the ARP with the proper VLAN tag.
2626 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2629 if (net_ratelimit()) {
2630 pr_warning("%s: no route to arp_ip_target %pI4\n",
2631 bond->dev->name, &targets[i]);
2637 * This target is not on a VLAN
2639 if (rt->dst.dev == bond->dev) {
2641 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2642 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2643 bond->master_ip, 0);
2648 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2650 vlan_dev = __vlan_find_dev_deep(bond->dev,
2653 if (vlan_dev == rt->dst.dev) {
2654 vlan_id = vlan->vlan_id;
2655 pr_debug("basa: vlan match on %s %d\n",
2656 vlan_dev->name, vlan_id);
2663 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2664 vlan->vlan_ip, vlan_id);
2668 if (net_ratelimit()) {
2669 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2670 bond->dev->name, &targets[i],
2671 rt->dst.dev ? rt->dst.dev->name : "NULL");
2677 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2680 __be32 *targets = bond->params.arp_targets;
2682 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2683 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2684 &sip, &tip, i, &targets[i],
2685 bond_has_this_ip(bond, tip));
2686 if (sip == targets[i]) {
2687 if (bond_has_this_ip(bond, tip))
2688 slave->last_arp_rx = jiffies;
2694 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2695 struct slave *slave)
2698 unsigned char *arp_ptr;
2701 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2704 read_lock(&bond->lock);
2706 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2707 bond->dev->name, skb->dev->name);
2709 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2713 if (arp->ar_hln != bond->dev->addr_len ||
2714 skb->pkt_type == PACKET_OTHERHOST ||
2715 skb->pkt_type == PACKET_LOOPBACK ||
2716 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2717 arp->ar_pro != htons(ETH_P_IP) ||
2721 arp_ptr = (unsigned char *)(arp + 1);
2722 arp_ptr += bond->dev->addr_len;
2723 memcpy(&sip, arp_ptr, 4);
2724 arp_ptr += 4 + bond->dev->addr_len;
2725 memcpy(&tip, arp_ptr, 4);
2727 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2728 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2729 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2733 * Backup slaves won't see the ARP reply, but do come through
2734 * here for each ARP probe (so we swap the sip/tip to validate
2735 * the probe). In a "redundant switch, common router" type of
2736 * configuration, the ARP probe will (hopefully) travel from
2737 * the active, through one switch, the router, then the other
2738 * switch before reaching the backup.
2740 if (bond_is_active_slave(slave))
2741 bond_validate_arp(bond, slave, sip, tip);
2743 bond_validate_arp(bond, slave, tip, sip);
2746 read_unlock(&bond->lock);
2750 * this function is called regularly to monitor each slave's link
2751 * ensuring that traffic is being sent and received when arp monitoring
2752 * is used in load-balancing mode. if the adapter has been dormant, then an
2753 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2754 * arp monitoring in active backup mode.
2756 void bond_loadbalance_arp_mon(struct work_struct *work)
2758 struct bonding *bond = container_of(work, struct bonding,
2760 struct slave *slave, *oldcurrent;
2761 int do_failover = 0;
2765 read_lock(&bond->lock);
2767 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2769 if (bond->slave_cnt == 0)
2772 read_lock(&bond->curr_slave_lock);
2773 oldcurrent = bond->curr_active_slave;
2774 read_unlock(&bond->curr_slave_lock);
2776 /* see if any of the previous devices are up now (i.e. they have
2777 * xmt and rcv traffic). the curr_active_slave does not come into
2778 * the picture unless it is null. also, slave->jiffies is not needed
2779 * here because we send an arp on each slave and give a slave as
2780 * long as it needs to get the tx/rx within the delta.
2781 * TODO: what about up/down delay in arp mode? it wasn't here before
2784 bond_for_each_slave(bond, slave, i) {
2785 unsigned long trans_start = dev_trans_start(slave->dev);
2787 if (slave->link != BOND_LINK_UP) {
2788 if (time_in_range(jiffies,
2789 trans_start - delta_in_ticks,
2790 trans_start + delta_in_ticks) &&
2791 time_in_range(jiffies,
2792 slave->dev->last_rx - delta_in_ticks,
2793 slave->dev->last_rx + delta_in_ticks)) {
2795 slave->link = BOND_LINK_UP;
2796 bond_set_active_slave(slave);
2798 /* primary_slave has no meaning in round-robin
2799 * mode. the window of a slave being up and
2800 * curr_active_slave being null after enslaving
2804 pr_info("%s: link status definitely up for interface %s, ",
2809 pr_info("%s: interface %s is now up\n",
2815 /* slave->link == BOND_LINK_UP */
2817 /* not all switches will respond to an arp request
2818 * when the source ip is 0, so don't take the link down
2819 * if we don't know our ip yet
2821 if (!time_in_range(jiffies,
2822 trans_start - delta_in_ticks,
2823 trans_start + 2 * delta_in_ticks) ||
2824 !time_in_range(jiffies,
2825 slave->dev->last_rx - delta_in_ticks,
2826 slave->dev->last_rx + 2 * delta_in_ticks)) {
2828 slave->link = BOND_LINK_DOWN;
2829 bond_set_backup_slave(slave);
2831 if (slave->link_failure_count < UINT_MAX)
2832 slave->link_failure_count++;
2834 pr_info("%s: interface %s is now down.\n",
2838 if (slave == oldcurrent)
2843 /* note: if switch is in round-robin mode, all links
2844 * must tx arp to ensure all links rx an arp - otherwise
2845 * links may oscillate or not come up at all; if switch is
2846 * in something like xor mode, there is nothing we can
2847 * do - all replies will be rx'ed on same link causing slaves
2848 * to be unstable during low/no traffic periods
2850 if (IS_UP(slave->dev))
2851 bond_arp_send_all(bond, slave);
2856 write_lock_bh(&bond->curr_slave_lock);
2858 bond_select_active_slave(bond);
2860 write_unlock_bh(&bond->curr_slave_lock);
2861 unblock_netpoll_tx();
2865 if (bond->params.arp_interval)
2866 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2868 read_unlock(&bond->lock);
2872 * Called to inspect slaves for active-backup mode ARP monitor link state
2873 * changes. Sets new_link in slaves to specify what action should take
2874 * place for the slave. Returns 0 if no changes are found, >0 if changes
2875 * to link states must be committed.
2877 * Called with bond->lock held for read.
2879 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2881 struct slave *slave;
2883 unsigned long trans_start;
2885 bond_for_each_slave(bond, slave, i) {
2886 slave->new_link = BOND_LINK_NOCHANGE;
2888 if (slave->link != BOND_LINK_UP) {
2889 if (time_in_range(jiffies,
2890 slave_last_rx(bond, slave) - delta_in_ticks,
2891 slave_last_rx(bond, slave) + delta_in_ticks)) {
2893 slave->new_link = BOND_LINK_UP;
2901 * Give slaves 2*delta after being enslaved or made
2902 * active. This avoids bouncing, as the last receive
2903 * times need a full ARP monitor cycle to be updated.
2905 if (time_in_range(jiffies,
2906 slave->jiffies - delta_in_ticks,
2907 slave->jiffies + 2 * delta_in_ticks))
2911 * Backup slave is down if:
2912 * - No current_arp_slave AND
2913 * - more than 3*delta since last receive AND
2914 * - the bond has an IP address
2916 * Note: a non-null current_arp_slave indicates
2917 * the curr_active_slave went down and we are
2918 * searching for a new one; under this condition
2919 * we only take the curr_active_slave down - this
2920 * gives each slave a chance to tx/rx traffic
2921 * before being taken out
2923 if (!bond_is_active_slave(slave) &&
2924 !bond->current_arp_slave &&
2925 !time_in_range(jiffies,
2926 slave_last_rx(bond, slave) - delta_in_ticks,
2927 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2929 slave->new_link = BOND_LINK_DOWN;
2934 * Active slave is down if:
2935 * - more than 2*delta since transmitting OR
2936 * - (more than 2*delta since receive AND
2937 * the bond has an IP address)
2939 trans_start = dev_trans_start(slave->dev);
2940 if (bond_is_active_slave(slave) &&
2941 (!time_in_range(jiffies,
2942 trans_start - delta_in_ticks,
2943 trans_start + 2 * delta_in_ticks) ||
2944 !time_in_range(jiffies,
2945 slave_last_rx(bond, slave) - delta_in_ticks,
2946 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2948 slave->new_link = BOND_LINK_DOWN;
2957 * Called to commit link state changes noted by inspection step of
2958 * active-backup mode ARP monitor.
2960 * Called with RTNL and bond->lock for read.
2962 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2964 struct slave *slave;
2966 unsigned long trans_start;
2968 bond_for_each_slave(bond, slave, i) {
2969 switch (slave->new_link) {
2970 case BOND_LINK_NOCHANGE:
2974 trans_start = dev_trans_start(slave->dev);
2975 if ((!bond->curr_active_slave &&
2976 time_in_range(jiffies,
2977 trans_start - delta_in_ticks,
2978 trans_start + delta_in_ticks)) ||
2979 bond->curr_active_slave != slave) {
2980 slave->link = BOND_LINK_UP;
2981 bond->current_arp_slave = NULL;
2983 pr_info("%s: link status definitely up for interface %s.\n",
2984 bond->dev->name, slave->dev->name);
2986 if (!bond->curr_active_slave ||
2987 (slave == bond->primary_slave))
2994 case BOND_LINK_DOWN:
2995 if (slave->link_failure_count < UINT_MAX)
2996 slave->link_failure_count++;
2998 slave->link = BOND_LINK_DOWN;
2999 bond_set_slave_inactive_flags(slave);
3001 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3002 bond->dev->name, slave->dev->name);
3004 if (slave == bond->curr_active_slave) {
3005 bond->current_arp_slave = NULL;
3012 pr_err("%s: impossible: new_link %d on slave %s\n",
3013 bond->dev->name, slave->new_link,
3021 write_lock_bh(&bond->curr_slave_lock);
3022 bond_select_active_slave(bond);
3023 write_unlock_bh(&bond->curr_slave_lock);
3024 unblock_netpoll_tx();
3027 bond_set_carrier(bond);
3031 * Send ARP probes for active-backup mode ARP monitor.
3033 * Called with bond->lock held for read.
3035 static void bond_ab_arp_probe(struct bonding *bond)
3037 struct slave *slave;
3040 read_lock(&bond->curr_slave_lock);
3042 if (bond->current_arp_slave && bond->curr_active_slave)
3043 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3044 bond->current_arp_slave->dev->name,
3045 bond->curr_active_slave->dev->name);
3047 if (bond->curr_active_slave) {
3048 bond_arp_send_all(bond, bond->curr_active_slave);
3049 read_unlock(&bond->curr_slave_lock);
3053 read_unlock(&bond->curr_slave_lock);
3055 /* if we don't have a curr_active_slave, search for the next available
3056 * backup slave from the current_arp_slave and make it the candidate
3057 * for becoming the curr_active_slave
3060 if (!bond->current_arp_slave) {
3061 bond->current_arp_slave = bond->first_slave;
3062 if (!bond->current_arp_slave)
3066 bond_set_slave_inactive_flags(bond->current_arp_slave);
3068 /* search for next candidate */
3069 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3070 if (IS_UP(slave->dev)) {
3071 slave->link = BOND_LINK_BACK;
3072 bond_set_slave_active_flags(slave);
3073 bond_arp_send_all(bond, slave);
3074 slave->jiffies = jiffies;
3075 bond->current_arp_slave = slave;
3079 /* if the link state is up at this point, we
3080 * mark it down - this can happen if we have
3081 * simultaneous link failures and
3082 * reselect_active_interface doesn't make this
3083 * one the current slave so it is still marked
3084 * up when it is actually down
3086 if (slave->link == BOND_LINK_UP) {
3087 slave->link = BOND_LINK_DOWN;
3088 if (slave->link_failure_count < UINT_MAX)
3089 slave->link_failure_count++;
3091 bond_set_slave_inactive_flags(slave);
3093 pr_info("%s: backup interface %s is now down.\n",
3094 bond->dev->name, slave->dev->name);
3099 void bond_activebackup_arp_mon(struct work_struct *work)
3101 struct bonding *bond = container_of(work, struct bonding,
3103 bool should_notify_peers = false;
3106 read_lock(&bond->lock);
3108 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3110 if (bond->slave_cnt == 0)
3113 should_notify_peers = bond_should_notify_peers(bond);
3115 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3116 read_unlock(&bond->lock);
3118 /* Race avoidance with bond_close flush of workqueue */
3119 if (!rtnl_trylock()) {
3120 read_lock(&bond->lock);
3122 should_notify_peers = false;
3126 read_lock(&bond->lock);
3128 bond_ab_arp_commit(bond, delta_in_ticks);
3130 read_unlock(&bond->lock);
3132 read_lock(&bond->lock);
3135 bond_ab_arp_probe(bond);
3138 if (bond->params.arp_interval)
3139 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3141 read_unlock(&bond->lock);
3143 if (should_notify_peers) {
3144 if (!rtnl_trylock()) {
3145 read_lock(&bond->lock);
3146 bond->send_peer_notif++;
3147 read_unlock(&bond->lock);
3150 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3155 /*-------------------------- netdev event handling --------------------------*/
3158 * Change device name
3160 static int bond_event_changename(struct bonding *bond)
3162 bond_remove_proc_entry(bond);
3163 bond_create_proc_entry(bond);
3165 bond_debug_reregister(bond);
3170 static int bond_master_netdev_event(unsigned long event,
3171 struct net_device *bond_dev)
3173 struct bonding *event_bond = netdev_priv(bond_dev);
3176 case NETDEV_CHANGENAME:
3177 return bond_event_changename(event_bond);
3185 static int bond_slave_netdev_event(unsigned long event,
3186 struct net_device *slave_dev)
3188 struct net_device *bond_dev = slave_dev->master;
3189 struct bonding *bond = netdev_priv(bond_dev);
3190 struct slave *slave = NULL;
3193 case NETDEV_UNREGISTER:
3195 if (bond->setup_by_slave)
3196 bond_release_and_destroy(bond_dev, slave_dev);
3198 bond_release(bond_dev, slave_dev);
3203 slave = bond_get_slave_by_dev(bond, slave_dev);
3205 u32 old_speed = slave->speed;
3206 u8 old_duplex = slave->duplex;
3208 bond_update_speed_duplex(slave);
3210 if (bond->params.mode == BOND_MODE_8023AD) {
3211 if (old_speed != slave->speed)
3212 bond_3ad_adapter_speed_changed(slave);
3213 if (old_duplex != slave->duplex)
3214 bond_3ad_adapter_duplex_changed(slave);
3221 * ... Or is it this?
3224 case NETDEV_CHANGEMTU:
3226 * TODO: Should slaves be allowed to
3227 * independently alter their MTU? For
3228 * an active-backup bond, slaves need
3229 * not be the same type of device, so
3230 * MTUs may vary. For other modes,
3231 * slaves arguably should have the
3232 * same MTUs. To do this, we'd need to
3233 * take over the slave's change_mtu
3234 * function for the duration of their
3238 case NETDEV_CHANGENAME:
3240 * TODO: handle changing the primary's name
3243 case NETDEV_FEAT_CHANGE:
3244 bond_compute_features(bond);
3254 * bond_netdev_event: handle netdev notifier chain events.
3256 * This function receives events for the netdev chain. The caller (an
3257 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3258 * locks for us to safely manipulate the slave devices (RTNL lock,
3261 static int bond_netdev_event(struct notifier_block *this,
3262 unsigned long event, void *ptr)
3264 struct net_device *event_dev = (struct net_device *)ptr;
3266 pr_debug("event_dev: %s, event: %lx\n",
3267 event_dev ? event_dev->name : "None",
3270 if (!(event_dev->priv_flags & IFF_BONDING))
3273 if (event_dev->flags & IFF_MASTER) {
3274 pr_debug("IFF_MASTER\n");
3275 return bond_master_netdev_event(event, event_dev);
3278 if (event_dev->flags & IFF_SLAVE) {
3279 pr_debug("IFF_SLAVE\n");
3280 return bond_slave_netdev_event(event, event_dev);
3287 * bond_inetaddr_event: handle inetaddr notifier chain events.
3289 * We keep track of device IPs primarily to use as source addresses in
3290 * ARP monitor probes (rather than spewing out broadcasts all the time).
3292 * We track one IP for the main device (if it has one), plus one per VLAN.
3294 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3296 struct in_ifaddr *ifa = ptr;
3297 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3298 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3299 struct bonding *bond;
3300 struct vlan_entry *vlan;
3302 /* we only care about primary address */
3303 if(ifa->ifa_flags & IFA_F_SECONDARY)
3306 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3307 if (bond->dev == event_dev) {
3310 bond->master_ip = ifa->ifa_local;
3313 bond->master_ip = 0;
3320 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3321 vlan_dev = __vlan_find_dev_deep(bond->dev,
3323 if (vlan_dev == event_dev) {
3326 vlan->vlan_ip = ifa->ifa_local;
3340 static struct notifier_block bond_netdev_notifier = {
3341 .notifier_call = bond_netdev_event,
3344 static struct notifier_block bond_inetaddr_notifier = {
3345 .notifier_call = bond_inetaddr_event,
3348 /*---------------------------- Hashing Policies -----------------------------*/
3351 * Hash for the output device based upon layer 2 and layer 3 data. If
3352 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3354 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3356 struct ethhdr *data = (struct ethhdr *)skb->data;
3357 struct iphdr *iph = ip_hdr(skb);
3359 if (skb->protocol == htons(ETH_P_IP)) {
3360 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3361 (data->h_dest[5] ^ data->h_source[5])) % count;
3364 return (data->h_dest[5] ^ data->h_source[5]) % count;
3368 * Hash for the output device based upon layer 3 and layer 4 data. If
3369 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3370 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3372 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3374 struct ethhdr *data = (struct ethhdr *)skb->data;
3375 struct iphdr *iph = ip_hdr(skb);
3376 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3379 if (skb->protocol == htons(ETH_P_IP)) {
3380 if (!ip_is_fragment(iph) &&
3381 (iph->protocol == IPPROTO_TCP ||
3382 iph->protocol == IPPROTO_UDP)) {
3383 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3385 return (layer4_xor ^
3386 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3390 return (data->h_dest[5] ^ data->h_source[5]) % count;
3394 * Hash for the output device based upon layer 2 data
3396 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3398 struct ethhdr *data = (struct ethhdr *)skb->data;
3400 return (data->h_dest[5] ^ data->h_source[5]) % count;
3403 /*-------------------------- Device entry points ----------------------------*/
3405 static int bond_open(struct net_device *bond_dev)
3407 struct bonding *bond = netdev_priv(bond_dev);
3408 struct slave *slave;
3411 /* reset slave->backup and slave->inactive */
3412 read_lock(&bond->lock);
3413 if (bond->slave_cnt > 0) {
3414 read_lock(&bond->curr_slave_lock);
3415 bond_for_each_slave(bond, slave, i) {
3416 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3417 && (slave != bond->curr_active_slave)) {
3418 bond_set_slave_inactive_flags(slave);
3420 bond_set_slave_active_flags(slave);
3423 read_unlock(&bond->curr_slave_lock);
3425 read_unlock(&bond->lock);
3427 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3429 if (bond_is_lb(bond)) {
3430 /* bond_alb_initialize must be called before the timer
3433 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3434 /* something went wrong - fail the open operation */
3438 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3439 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3442 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3443 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3444 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3447 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3448 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3449 INIT_DELAYED_WORK(&bond->arp_work,
3450 bond_activebackup_arp_mon);
3452 INIT_DELAYED_WORK(&bond->arp_work,
3453 bond_loadbalance_arp_mon);
3455 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3456 if (bond->params.arp_validate)
3457 bond->recv_probe = bond_arp_rcv;
3460 if (bond->params.mode == BOND_MODE_8023AD) {
3461 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3462 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3463 /* register to receive LACPDUs */
3464 bond->recv_probe = bond_3ad_lacpdu_recv;
3465 bond_3ad_initiate_agg_selection(bond, 1);
3471 static int bond_close(struct net_device *bond_dev)
3473 struct bonding *bond = netdev_priv(bond_dev);
3475 write_lock_bh(&bond->lock);
3477 bond->send_peer_notif = 0;
3479 write_unlock_bh(&bond->lock);
3481 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3482 cancel_delayed_work_sync(&bond->mii_work);
3485 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3486 cancel_delayed_work_sync(&bond->arp_work);
3489 switch (bond->params.mode) {
3490 case BOND_MODE_8023AD:
3491 cancel_delayed_work_sync(&bond->ad_work);
3495 cancel_delayed_work_sync(&bond->alb_work);
3501 if (delayed_work_pending(&bond->mcast_work))
3502 cancel_delayed_work_sync(&bond->mcast_work);
3504 if (bond_is_lb(bond)) {
3505 /* Must be called only after all
3506 * slaves have been released
3508 bond_alb_deinitialize(bond);
3510 bond->recv_probe = NULL;
3515 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3516 struct rtnl_link_stats64 *stats)
3518 struct bonding *bond = netdev_priv(bond_dev);
3519 struct rtnl_link_stats64 temp;
3520 struct slave *slave;
3523 memset(stats, 0, sizeof(*stats));
3525 read_lock_bh(&bond->lock);
3527 bond_for_each_slave(bond, slave, i) {
3528 const struct rtnl_link_stats64 *sstats =
3529 dev_get_stats(slave->dev, &temp);
3531 stats->rx_packets += sstats->rx_packets;
3532 stats->rx_bytes += sstats->rx_bytes;
3533 stats->rx_errors += sstats->rx_errors;
3534 stats->rx_dropped += sstats->rx_dropped;
3536 stats->tx_packets += sstats->tx_packets;
3537 stats->tx_bytes += sstats->tx_bytes;
3538 stats->tx_errors += sstats->tx_errors;
3539 stats->tx_dropped += sstats->tx_dropped;
3541 stats->multicast += sstats->multicast;
3542 stats->collisions += sstats->collisions;
3544 stats->rx_length_errors += sstats->rx_length_errors;
3545 stats->rx_over_errors += sstats->rx_over_errors;
3546 stats->rx_crc_errors += sstats->rx_crc_errors;
3547 stats->rx_frame_errors += sstats->rx_frame_errors;
3548 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3549 stats->rx_missed_errors += sstats->rx_missed_errors;
3551 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3552 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3553 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3554 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3555 stats->tx_window_errors += sstats->tx_window_errors;
3558 read_unlock_bh(&bond->lock);
3563 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3565 struct net_device *slave_dev = NULL;
3566 struct ifbond k_binfo;
3567 struct ifbond __user *u_binfo = NULL;
3568 struct ifslave k_sinfo;
3569 struct ifslave __user *u_sinfo = NULL;
3570 struct mii_ioctl_data *mii = NULL;
3573 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3585 * We do this again just in case we were called by SIOCGMIIREG
3586 * instead of SIOCGMIIPHY.
3593 if (mii->reg_num == 1) {
3594 struct bonding *bond = netdev_priv(bond_dev);
3596 read_lock(&bond->lock);
3597 read_lock(&bond->curr_slave_lock);
3598 if (netif_carrier_ok(bond->dev))
3599 mii->val_out = BMSR_LSTATUS;
3601 read_unlock(&bond->curr_slave_lock);
3602 read_unlock(&bond->lock);
3606 case BOND_INFO_QUERY_OLD:
3607 case SIOCBONDINFOQUERY:
3608 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3610 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3613 res = bond_info_query(bond_dev, &k_binfo);
3615 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3619 case BOND_SLAVE_INFO_QUERY_OLD:
3620 case SIOCBONDSLAVEINFOQUERY:
3621 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3623 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3626 res = bond_slave_info_query(bond_dev, &k_sinfo);
3628 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3637 if (!capable(CAP_NET_ADMIN))
3640 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3642 pr_debug("slave_dev=%p:\n", slave_dev);
3647 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3649 case BOND_ENSLAVE_OLD:
3650 case SIOCBONDENSLAVE:
3651 res = bond_enslave(bond_dev, slave_dev);
3653 case BOND_RELEASE_OLD:
3654 case SIOCBONDRELEASE:
3655 res = bond_release(bond_dev, slave_dev);
3657 case BOND_SETHWADDR_OLD:
3658 case SIOCBONDSETHWADDR:
3659 res = bond_sethwaddr(bond_dev, slave_dev);
3661 case BOND_CHANGE_ACTIVE_OLD:
3662 case SIOCBONDCHANGEACTIVE:
3663 res = bond_ioctl_change_active(bond_dev, slave_dev);
3675 static bool bond_addr_in_mc_list(unsigned char *addr,
3676 struct netdev_hw_addr_list *list,
3679 struct netdev_hw_addr *ha;
3681 netdev_hw_addr_list_for_each(ha, list)
3682 if (!memcmp(ha->addr, addr, addrlen))
3688 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3690 struct bonding *bond = netdev_priv(bond_dev);
3692 if (change & IFF_PROMISC)
3693 bond_set_promiscuity(bond,
3694 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3696 if (change & IFF_ALLMULTI)
3697 bond_set_allmulti(bond,
3698 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3701 static void bond_set_multicast_list(struct net_device *bond_dev)
3703 struct bonding *bond = netdev_priv(bond_dev);
3704 struct netdev_hw_addr *ha;
3707 read_lock(&bond->lock);
3709 /* looking for addresses to add to slaves' mc list */
3710 netdev_for_each_mc_addr(ha, bond_dev) {
3711 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3712 bond_dev->addr_len);
3714 bond_mc_add(bond, ha->addr);
3717 /* looking for addresses to delete from slaves' list */
3718 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3719 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3720 bond_dev->addr_len);
3722 bond_mc_del(bond, ha->addr);
3725 /* save master's multicast list */
3726 __hw_addr_flush(&bond->mc_list);
3727 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3728 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3730 read_unlock(&bond->lock);
3733 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3735 struct bonding *bond = netdev_priv(dev);
3736 struct slave *slave = bond->first_slave;
3739 const struct net_device_ops *slave_ops
3740 = slave->dev->netdev_ops;
3741 if (slave_ops->ndo_neigh_setup)
3742 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3748 * Change the MTU of all of a master's slaves to match the master
3750 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3752 struct bonding *bond = netdev_priv(bond_dev);
3753 struct slave *slave, *stop_at;
3757 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3758 (bond_dev ? bond_dev->name : "None"), new_mtu);
3760 /* Can't hold bond->lock with bh disabled here since
3761 * some base drivers panic. On the other hand we can't
3762 * hold bond->lock without bh disabled because we'll
3763 * deadlock. The only solution is to rely on the fact
3764 * that we're under rtnl_lock here, and the slaves
3765 * list won't change. This doesn't solve the problem
3766 * of setting the slave's MTU while it is
3767 * transmitting, but the assumption is that the base
3768 * driver can handle that.
3770 * TODO: figure out a way to safely iterate the slaves
3771 * list, but without holding a lock around the actual
3772 * call to the base driver.
3775 bond_for_each_slave(bond, slave, i) {
3776 pr_debug("s %p s->p %p c_m %p\n",
3779 slave->dev->netdev_ops->ndo_change_mtu);
3781 res = dev_set_mtu(slave->dev, new_mtu);
3784 /* If we failed to set the slave's mtu to the new value
3785 * we must abort the operation even in ACTIVE_BACKUP
3786 * mode, because if we allow the backup slaves to have
3787 * different mtu values than the active slave we'll
3788 * need to change their mtu when doing a failover. That
3789 * means changing their mtu from timer context, which
3790 * is probably not a good idea.
3792 pr_debug("err %d %s\n", res, slave->dev->name);
3797 bond_dev->mtu = new_mtu;
3802 /* unwind from head to the slave that failed */
3804 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3807 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3809 pr_debug("unwind err %d dev %s\n",
3810 tmp_res, slave->dev->name);
3820 * Note that many devices must be down to change the HW address, and
3821 * downing the master releases all slaves. We can make bonds full of
3822 * bonding devices to test this, however.
3824 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3826 struct bonding *bond = netdev_priv(bond_dev);
3827 struct sockaddr *sa = addr, tmp_sa;
3828 struct slave *slave, *stop_at;
3832 if (bond->params.mode == BOND_MODE_ALB)
3833 return bond_alb_set_mac_address(bond_dev, addr);
3836 pr_debug("bond=%p, name=%s\n",
3837 bond, bond_dev ? bond_dev->name : "None");
3840 * If fail_over_mac is set to active, do nothing and return
3841 * success. Returning an error causes ifenslave to fail.
3843 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3846 if (!is_valid_ether_addr(sa->sa_data))
3847 return -EADDRNOTAVAIL;
3849 /* Can't hold bond->lock with bh disabled here since
3850 * some base drivers panic. On the other hand we can't
3851 * hold bond->lock without bh disabled because we'll
3852 * deadlock. The only solution is to rely on the fact
3853 * that we're under rtnl_lock here, and the slaves
3854 * list won't change. This doesn't solve the problem
3855 * of setting the slave's hw address while it is
3856 * transmitting, but the assumption is that the base
3857 * driver can handle that.
3859 * TODO: figure out a way to safely iterate the slaves
3860 * list, but without holding a lock around the actual
3861 * call to the base driver.
3864 bond_for_each_slave(bond, slave, i) {
3865 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3866 pr_debug("slave %p %s\n", slave, slave->dev->name);
3868 if (slave_ops->ndo_set_mac_address == NULL) {
3870 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3874 res = dev_set_mac_address(slave->dev, addr);
3876 /* TODO: consider downing the slave
3878 * User should expect communications
3879 * breakage anyway until ARP finish
3882 pr_debug("err %d %s\n", res, slave->dev->name);
3888 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3892 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3893 tmp_sa.sa_family = bond_dev->type;
3895 /* unwind from head to the slave that failed */
3897 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3900 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3902 pr_debug("unwind err %d dev %s\n",
3903 tmp_res, slave->dev->name);
3910 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3912 struct bonding *bond = netdev_priv(bond_dev);
3913 struct slave *slave, *start_at;
3914 int i, slave_no, res = 1;
3915 struct iphdr *iph = ip_hdr(skb);
3918 * Start with the curr_active_slave that joined the bond as the
3919 * default for sending IGMP traffic. For failover purposes one
3920 * needs to maintain some consistency for the interface that will
3921 * send the join/membership reports. The curr_active_slave found
3922 * will send all of this type of traffic.
3924 if ((iph->protocol == IPPROTO_IGMP) &&
3925 (skb->protocol == htons(ETH_P_IP))) {
3927 read_lock(&bond->curr_slave_lock);
3928 slave = bond->curr_active_slave;
3929 read_unlock(&bond->curr_slave_lock);
3935 * Concurrent TX may collide on rr_tx_counter; we accept
3936 * that as being rare enough not to justify using an
3939 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3941 bond_for_each_slave(bond, slave, i) {
3949 bond_for_each_slave_from(bond, slave, i, start_at) {
3950 if (IS_UP(slave->dev) &&
3951 (slave->link == BOND_LINK_UP) &&
3952 bond_is_active_slave(slave)) {
3953 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3960 /* no suitable interface, frame not sent */
3964 return NETDEV_TX_OK;
3969 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3970 * the bond has a usable interface.
3972 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3974 struct bonding *bond = netdev_priv(bond_dev);
3977 read_lock(&bond->curr_slave_lock);
3979 if (bond->curr_active_slave)
3980 res = bond_dev_queue_xmit(bond, skb,
3981 bond->curr_active_slave->dev);
3984 /* no suitable interface, frame not sent */
3987 read_unlock(&bond->curr_slave_lock);
3989 return NETDEV_TX_OK;
3993 * In bond_xmit_xor() , we determine the output device by using a pre-
3994 * determined xmit_hash_policy(), If the selected device is not enabled,
3995 * find the next active slave.
3997 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3999 struct bonding *bond = netdev_priv(bond_dev);
4000 struct slave *slave, *start_at;
4005 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4007 bond_for_each_slave(bond, slave, i) {
4015 bond_for_each_slave_from(bond, slave, i, start_at) {
4016 if (IS_UP(slave->dev) &&
4017 (slave->link == BOND_LINK_UP) &&
4018 bond_is_active_slave(slave)) {
4019 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4025 /* no suitable interface, frame not sent */
4029 return NETDEV_TX_OK;
4033 * in broadcast mode, we send everything to all usable interfaces.
4035 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4037 struct bonding *bond = netdev_priv(bond_dev);
4038 struct slave *slave, *start_at;
4039 struct net_device *tx_dev = NULL;
4043 read_lock(&bond->curr_slave_lock);
4044 start_at = bond->curr_active_slave;
4045 read_unlock(&bond->curr_slave_lock);
4050 bond_for_each_slave_from(bond, slave, i, start_at) {
4051 if (IS_UP(slave->dev) &&
4052 (slave->link == BOND_LINK_UP) &&
4053 bond_is_active_slave(slave)) {
4055 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4057 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4062 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4064 dev_kfree_skb(skb2);
4068 tx_dev = slave->dev;
4073 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4077 /* no suitable interface, frame not sent */
4080 /* frame sent to all suitable interfaces */
4081 return NETDEV_TX_OK;
4084 /*------------------------- Device initialization ---------------------------*/
4086 static void bond_set_xmit_hash_policy(struct bonding *bond)
4088 switch (bond->params.xmit_policy) {
4089 case BOND_XMIT_POLICY_LAYER23:
4090 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4092 case BOND_XMIT_POLICY_LAYER34:
4093 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4095 case BOND_XMIT_POLICY_LAYER2:
4097 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4103 * Lookup the slave that corresponds to a qid
4105 static inline int bond_slave_override(struct bonding *bond,
4106 struct sk_buff *skb)
4109 struct slave *slave = NULL;
4110 struct slave *check_slave;
4112 if (!skb->queue_mapping)
4115 /* Find out if any slaves have the same mapping as this skb. */
4116 bond_for_each_slave(bond, check_slave, i) {
4117 if (check_slave->queue_id == skb->queue_mapping) {
4118 slave = check_slave;
4123 /* If the slave isn't UP, use default transmit policy. */
4124 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4125 (slave->link == BOND_LINK_UP)) {
4126 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4133 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4136 * This helper function exists to help dev_pick_tx get the correct
4137 * destination queue. Using a helper function skips a call to
4138 * skb_tx_hash and will put the skbs in the queue we expect on their
4139 * way down to the bonding driver.
4141 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4144 * Save the original txq to restore before passing to the driver
4146 bond_queue_mapping(skb) = skb->queue_mapping;
4148 if (unlikely(txq >= dev->real_num_tx_queues)) {
4150 txq -= dev->real_num_tx_queues;
4151 } while (txq >= dev->real_num_tx_queues);
4156 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4158 struct bonding *bond = netdev_priv(dev);
4160 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4161 if (!bond_slave_override(bond, skb))
4162 return NETDEV_TX_OK;
4165 switch (bond->params.mode) {
4166 case BOND_MODE_ROUNDROBIN:
4167 return bond_xmit_roundrobin(skb, dev);
4168 case BOND_MODE_ACTIVEBACKUP:
4169 return bond_xmit_activebackup(skb, dev);
4171 return bond_xmit_xor(skb, dev);
4172 case BOND_MODE_BROADCAST:
4173 return bond_xmit_broadcast(skb, dev);
4174 case BOND_MODE_8023AD:
4175 return bond_3ad_xmit_xor(skb, dev);
4178 return bond_alb_xmit(skb, dev);
4180 /* Should never happen, mode already checked */
4181 pr_err("%s: Error: Unknown bonding mode %d\n",
4182 dev->name, bond->params.mode);
4185 return NETDEV_TX_OK;
4189 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4191 struct bonding *bond = netdev_priv(dev);
4192 netdev_tx_t ret = NETDEV_TX_OK;
4195 * If we risk deadlock from transmitting this in the
4196 * netpoll path, tell netpoll to queue the frame for later tx
4198 if (is_netpoll_tx_blocked(dev))
4199 return NETDEV_TX_BUSY;
4201 read_lock(&bond->lock);
4203 if (bond->slave_cnt)
4204 ret = __bond_start_xmit(skb, dev);
4208 read_unlock(&bond->lock);
4214 * set bond mode specific net device operations
4216 void bond_set_mode_ops(struct bonding *bond, int mode)
4218 struct net_device *bond_dev = bond->dev;
4221 case BOND_MODE_ROUNDROBIN:
4223 case BOND_MODE_ACTIVEBACKUP:
4226 bond_set_xmit_hash_policy(bond);
4228 case BOND_MODE_BROADCAST:
4230 case BOND_MODE_8023AD:
4231 bond_set_xmit_hash_policy(bond);
4238 /* Should never happen, mode already checked */
4239 pr_err("%s: Error: Unknown bonding mode %d\n",
4240 bond_dev->name, mode);
4245 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4246 struct ethtool_drvinfo *drvinfo)
4248 strncpy(drvinfo->driver, DRV_NAME, 32);
4249 strncpy(drvinfo->version, DRV_VERSION, 32);
4250 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4253 static const struct ethtool_ops bond_ethtool_ops = {
4254 .get_drvinfo = bond_ethtool_get_drvinfo,
4255 .get_link = ethtool_op_get_link,
4258 static const struct net_device_ops bond_netdev_ops = {
4259 .ndo_init = bond_init,
4260 .ndo_uninit = bond_uninit,
4261 .ndo_open = bond_open,
4262 .ndo_stop = bond_close,
4263 .ndo_start_xmit = bond_start_xmit,
4264 .ndo_select_queue = bond_select_queue,
4265 .ndo_get_stats64 = bond_get_stats,
4266 .ndo_do_ioctl = bond_do_ioctl,
4267 .ndo_change_rx_flags = bond_change_rx_flags,
4268 .ndo_set_rx_mode = bond_set_multicast_list,
4269 .ndo_change_mtu = bond_change_mtu,
4270 .ndo_set_mac_address = bond_set_mac_address,
4271 .ndo_neigh_setup = bond_neigh_setup,
4272 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4273 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4274 #ifdef CONFIG_NET_POLL_CONTROLLER
4275 .ndo_netpoll_setup = bond_netpoll_setup,
4276 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4277 .ndo_poll_controller = bond_poll_controller,
4279 .ndo_add_slave = bond_enslave,
4280 .ndo_del_slave = bond_release,
4281 .ndo_fix_features = bond_fix_features,
4284 static void bond_destructor(struct net_device *bond_dev)
4286 struct bonding *bond = netdev_priv(bond_dev);
4288 destroy_workqueue(bond->wq);
4289 free_netdev(bond_dev);
4292 static void bond_setup(struct net_device *bond_dev)
4294 struct bonding *bond = netdev_priv(bond_dev);
4296 /* initialize rwlocks */
4297 rwlock_init(&bond->lock);
4298 rwlock_init(&bond->curr_slave_lock);
4300 bond->params = bonding_defaults;
4302 /* Initialize pointers */
4303 bond->dev = bond_dev;
4304 INIT_LIST_HEAD(&bond->vlan_list);
4306 /* Initialize the device entry points */
4307 ether_setup(bond_dev);
4308 bond_dev->netdev_ops = &bond_netdev_ops;
4309 bond_dev->ethtool_ops = &bond_ethtool_ops;
4310 bond_set_mode_ops(bond, bond->params.mode);
4312 bond_dev->destructor = bond_destructor;
4314 /* Initialize the device options */
4315 bond_dev->tx_queue_len = 0;
4316 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4317 bond_dev->priv_flags |= IFF_BONDING;
4318 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4320 /* At first, we block adding VLANs. That's the only way to
4321 * prevent problems that occur when adding VLANs over an
4322 * empty bond. The block will be removed once non-challenged
4323 * slaves are enslaved.
4325 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4327 /* don't acquire bond device's netif_tx_lock when
4329 bond_dev->features |= NETIF_F_LLTX;
4331 /* By default, we declare the bond to be fully
4332 * VLAN hardware accelerated capable. Special
4333 * care is taken in the various xmit functions
4334 * when there are slaves that are not hw accel
4338 bond_dev->hw_features = BOND_VLAN_FEATURES |
4339 NETIF_F_HW_VLAN_TX |
4340 NETIF_F_HW_VLAN_RX |
4341 NETIF_F_HW_VLAN_FILTER;
4343 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4344 bond_dev->features |= bond_dev->hw_features;
4347 static void bond_work_cancel_all(struct bonding *bond)
4349 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4350 cancel_delayed_work_sync(&bond->mii_work);
4352 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4353 cancel_delayed_work_sync(&bond->arp_work);
4355 if (bond->params.mode == BOND_MODE_ALB &&
4356 delayed_work_pending(&bond->alb_work))
4357 cancel_delayed_work_sync(&bond->alb_work);
4359 if (bond->params.mode == BOND_MODE_8023AD &&
4360 delayed_work_pending(&bond->ad_work))
4361 cancel_delayed_work_sync(&bond->ad_work);
4363 if (delayed_work_pending(&bond->mcast_work))
4364 cancel_delayed_work_sync(&bond->mcast_work);
4368 * Destroy a bonding device.
4369 * Must be under rtnl_lock when this function is called.
4371 static void bond_uninit(struct net_device *bond_dev)
4373 struct bonding *bond = netdev_priv(bond_dev);
4374 struct vlan_entry *vlan, *tmp;
4376 bond_netpoll_cleanup(bond_dev);
4378 /* Release the bonded slaves */
4379 bond_release_all(bond_dev);
4381 list_del(&bond->bond_list);
4383 bond_work_cancel_all(bond);
4385 bond_remove_proc_entry(bond);
4387 bond_debug_unregister(bond);
4389 __hw_addr_flush(&bond->mc_list);
4391 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4392 list_del(&vlan->vlan_list);
4397 /*------------------------- Module initialization ---------------------------*/
4400 * Convert string input module parms. Accept either the
4401 * number of the mode or its string name. A bit complicated because
4402 * some mode names are substrings of other names, and calls from sysfs
4403 * may have whitespace in the name (trailing newlines, for example).
4405 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4407 int modeint = -1, i, rv;
4408 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4410 for (p = (char *)buf; *p; p++)
4411 if (!(isdigit(*p) || isspace(*p)))
4415 rv = sscanf(buf, "%20s", modestr);
4417 rv = sscanf(buf, "%d", &modeint);
4422 for (i = 0; tbl[i].modename; i++) {
4423 if (modeint == tbl[i].mode)
4425 if (strcmp(modestr, tbl[i].modename) == 0)
4432 static int bond_check_params(struct bond_params *params)
4434 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4437 * Convert string parameters.
4440 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4441 if (bond_mode == -1) {
4442 pr_err("Error: Invalid bonding mode \"%s\"\n",
4443 mode == NULL ? "NULL" : mode);
4448 if (xmit_hash_policy) {
4449 if ((bond_mode != BOND_MODE_XOR) &&
4450 (bond_mode != BOND_MODE_8023AD)) {
4451 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4452 bond_mode_name(bond_mode));
4454 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4456 if (xmit_hashtype == -1) {
4457 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4458 xmit_hash_policy == NULL ? "NULL" :
4466 if (bond_mode != BOND_MODE_8023AD) {
4467 pr_info("lacp_rate param is irrelevant in mode %s\n",
4468 bond_mode_name(bond_mode));
4470 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4471 if (lacp_fast == -1) {
4472 pr_err("Error: Invalid lacp rate \"%s\"\n",
4473 lacp_rate == NULL ? "NULL" : lacp_rate);
4480 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4481 if (params->ad_select == -1) {
4482 pr_err("Error: Invalid ad_select \"%s\"\n",
4483 ad_select == NULL ? "NULL" : ad_select);
4487 if (bond_mode != BOND_MODE_8023AD) {
4488 pr_warning("ad_select param only affects 802.3ad mode\n");
4491 params->ad_select = BOND_AD_STABLE;
4494 if (max_bonds < 0) {
4495 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4496 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4497 max_bonds = BOND_DEFAULT_MAX_BONDS;
4501 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4502 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4503 miimon = BOND_LINK_MON_INTERV;
4507 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4512 if (downdelay < 0) {
4513 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4514 downdelay, INT_MAX);
4518 if ((use_carrier != 0) && (use_carrier != 1)) {
4519 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4524 if (num_peer_notif < 0 || num_peer_notif > 255) {
4525 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4530 /* reset values for 802.3ad */
4531 if (bond_mode == BOND_MODE_8023AD) {
4533 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");
4534 pr_warning("Forcing miimon to 100msec\n");
4539 if (tx_queues < 1 || tx_queues > 255) {
4540 pr_warning("Warning: tx_queues (%d) should be between "
4541 "1 and 255, resetting to %d\n",
4542 tx_queues, BOND_DEFAULT_TX_QUEUES);
4543 tx_queues = BOND_DEFAULT_TX_QUEUES;
4546 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4547 pr_warning("Warning: all_slaves_active module parameter (%d), "
4548 "not of valid value (0/1), so it was set to "
4549 "0\n", all_slaves_active);
4550 all_slaves_active = 0;
4553 if (resend_igmp < 0 || resend_igmp > 255) {
4554 pr_warning("Warning: resend_igmp (%d) should be between "
4555 "0 and 255, resetting to %d\n",
4556 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4557 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4560 /* reset values for TLB/ALB */
4561 if ((bond_mode == BOND_MODE_TLB) ||
4562 (bond_mode == BOND_MODE_ALB)) {
4564 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");
4565 pr_warning("Forcing miimon to 100msec\n");
4570 if (bond_mode == BOND_MODE_ALB) {
4571 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",
4576 if (updelay || downdelay) {
4577 /* just warn the user the up/down delay will have
4578 * no effect since miimon is zero...
4580 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",
4581 updelay, downdelay);
4584 /* don't allow arp monitoring */
4586 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4587 miimon, arp_interval);
4591 if ((updelay % miimon) != 0) {
4592 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4594 (updelay / miimon) * miimon);
4599 if ((downdelay % miimon) != 0) {
4600 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4602 (downdelay / miimon) * miimon);
4605 downdelay /= miimon;
4608 if (arp_interval < 0) {
4609 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4610 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4611 arp_interval = BOND_LINK_ARP_INTERV;
4614 for (arp_ip_count = 0;
4615 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4617 /* not complete check, but should be good enough to
4619 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4620 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4621 arp_ip_target[arp_ip_count]);
4624 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4625 arp_target[arp_ip_count] = ip;
4629 if (arp_interval && !arp_ip_count) {
4630 /* don't allow arping if no arp_ip_target given... */
4631 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4637 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4638 pr_err("arp_validate only supported in active-backup mode\n");
4641 if (!arp_interval) {
4642 pr_err("arp_validate requires arp_interval\n");
4646 arp_validate_value = bond_parse_parm(arp_validate,
4648 if (arp_validate_value == -1) {
4649 pr_err("Error: invalid arp_validate \"%s\"\n",
4650 arp_validate == NULL ? "NULL" : arp_validate);
4654 arp_validate_value = 0;
4657 pr_info("MII link monitoring set to %d ms\n", miimon);
4658 } else if (arp_interval) {
4661 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4663 arp_validate_tbl[arp_validate_value].modename,
4666 for (i = 0; i < arp_ip_count; i++)
4667 pr_info(" %s", arp_ip_target[i]);
4671 } else if (max_bonds) {
4672 /* miimon and arp_interval not set, we need one so things
4673 * work as expected, see bonding.txt for details
4675 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");
4678 if (primary && !USES_PRIMARY(bond_mode)) {
4679 /* currently, using a primary only makes sense
4680 * in active backup, TLB or ALB modes
4682 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4683 primary, bond_mode_name(bond_mode));
4687 if (primary && primary_reselect) {
4688 primary_reselect_value = bond_parse_parm(primary_reselect,
4690 if (primary_reselect_value == -1) {
4691 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4693 NULL ? "NULL" : primary_reselect);
4697 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4700 if (fail_over_mac) {
4701 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4703 if (fail_over_mac_value == -1) {
4704 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4705 arp_validate == NULL ? "NULL" : arp_validate);
4709 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4710 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4712 fail_over_mac_value = BOND_FOM_NONE;
4715 /* fill params struct with the proper values */
4716 params->mode = bond_mode;
4717 params->xmit_policy = xmit_hashtype;
4718 params->miimon = miimon;
4719 params->num_peer_notif = num_peer_notif;
4720 params->arp_interval = arp_interval;
4721 params->arp_validate = arp_validate_value;
4722 params->updelay = updelay;
4723 params->downdelay = downdelay;
4724 params->use_carrier = use_carrier;
4725 params->lacp_fast = lacp_fast;
4726 params->primary[0] = 0;
4727 params->primary_reselect = primary_reselect_value;
4728 params->fail_over_mac = fail_over_mac_value;
4729 params->tx_queues = tx_queues;
4730 params->all_slaves_active = all_slaves_active;
4731 params->resend_igmp = resend_igmp;
4732 params->min_links = min_links;
4735 strncpy(params->primary, primary, IFNAMSIZ);
4736 params->primary[IFNAMSIZ - 1] = 0;
4739 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4744 static struct lock_class_key bonding_netdev_xmit_lock_key;
4745 static struct lock_class_key bonding_netdev_addr_lock_key;
4747 static void bond_set_lockdep_class_one(struct net_device *dev,
4748 struct netdev_queue *txq,
4751 lockdep_set_class(&txq->_xmit_lock,
4752 &bonding_netdev_xmit_lock_key);
4755 static void bond_set_lockdep_class(struct net_device *dev)
4757 lockdep_set_class(&dev->addr_list_lock,
4758 &bonding_netdev_addr_lock_key);
4759 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4763 * Called from registration process
4765 static int bond_init(struct net_device *bond_dev)
4767 struct bonding *bond = netdev_priv(bond_dev);
4768 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4769 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4771 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4774 * Initialize locks that may be required during
4775 * en/deslave operations. All of the bond_open work
4776 * (of which this is part) should really be moved to
4777 * a phase prior to dev_open
4779 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4780 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4782 bond->wq = create_singlethread_workqueue(bond_dev->name);
4786 bond_set_lockdep_class(bond_dev);
4788 bond_create_proc_entry(bond);
4789 list_add_tail(&bond->bond_list, &bn->dev_list);
4791 bond_prepare_sysfs_group(bond);
4793 bond_debug_register(bond);
4795 __hw_addr_init(&bond->mc_list);
4799 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4801 if (tb[IFLA_ADDRESS]) {
4802 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4804 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4805 return -EADDRNOTAVAIL;
4810 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4811 unsigned int *num_queues,
4812 unsigned int *real_num_queues)
4814 *num_queues = tx_queues;
4818 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4820 .priv_size = sizeof(struct bonding),
4821 .setup = bond_setup,
4822 .validate = bond_validate,
4823 .get_tx_queues = bond_get_tx_queues,
4826 /* Create a new bond based on the specified name and bonding parameters.
4827 * If name is NULL, obtain a suitable "bond%d" name for us.
4828 * Caller must NOT hold rtnl_lock; we need to release it here before we
4829 * set up our sysfs entries.
4831 int bond_create(struct net *net, const char *name)
4833 struct net_device *bond_dev;
4838 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4839 name ? name : "bond%d",
4840 bond_setup, tx_queues);
4842 pr_err("%s: eek! can't alloc netdev!\n", name);
4847 dev_net_set(bond_dev, net);
4848 bond_dev->rtnl_link_ops = &bond_link_ops;
4850 res = register_netdevice(bond_dev);
4852 netif_carrier_off(bond_dev);
4856 bond_destructor(bond_dev);
4860 static int __net_init bond_net_init(struct net *net)
4862 struct bond_net *bn = net_generic(net, bond_net_id);
4865 INIT_LIST_HEAD(&bn->dev_list);
4867 bond_create_proc_dir(bn);
4868 bond_create_sysfs(bn);
4873 static void __net_exit bond_net_exit(struct net *net)
4875 struct bond_net *bn = net_generic(net, bond_net_id);
4877 bond_destroy_sysfs(bn);
4878 bond_destroy_proc_dir(bn);
4881 static struct pernet_operations bond_net_ops = {
4882 .init = bond_net_init,
4883 .exit = bond_net_exit,
4885 .size = sizeof(struct bond_net),
4888 static int __init bonding_init(void)
4893 pr_info("%s", bond_version);
4895 res = bond_check_params(&bonding_defaults);
4899 res = register_pernet_subsys(&bond_net_ops);
4903 res = rtnl_link_register(&bond_link_ops);
4907 bond_create_debugfs();
4909 for (i = 0; i < max_bonds; i++) {
4910 res = bond_create(&init_net, NULL);
4915 register_netdevice_notifier(&bond_netdev_notifier);
4916 register_inetaddr_notifier(&bond_inetaddr_notifier);
4920 rtnl_link_unregister(&bond_link_ops);
4922 unregister_pernet_subsys(&bond_net_ops);
4927 static void __exit bonding_exit(void)
4929 unregister_netdevice_notifier(&bond_netdev_notifier);
4930 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4932 bond_destroy_debugfs();
4934 rtnl_link_unregister(&bond_link_ops);
4935 unregister_pernet_subsys(&bond_net_ops);
4937 #ifdef CONFIG_NET_POLL_CONTROLLER
4939 * Make sure we don't have an imbalance on our netpoll blocking
4941 WARN_ON(atomic_read(&netpoll_block_tx));
4945 module_init(bonding_init);
4946 module_exit(bonding_exit);
4947 MODULE_LICENSE("GPL");
4948 MODULE_VERSION(DRV_VERSION);
4949 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4950 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4951 MODULE_ALIAS_RTNL_LINK("bond");