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>
80 #include <net/pkt_sched.h>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon = BOND_LINK_MON_INTERV;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval = BOND_LINK_ARP_INTERV;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *fail_over_mac;
109 static int all_slaves_active = 0;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
155 module_param(min_links, int, 0);
156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
158 module_param(xmit_hash_policy, charp, 0);
159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
162 module_param(arp_interval, int, 0);
163 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
164 module_param_array(arp_ip_target, charp, NULL, 0);
165 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
166 module_param(arp_validate, charp, 0);
167 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
168 "0 for none (default), 1 for active, "
169 "2 for backup, 3 for all");
170 module_param(fail_over_mac, charp, 0);
171 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
172 "the same MAC; 0 for none (default), "
173 "1 for active, 2 for follow");
174 module_param(all_slaves_active, int, 0);
175 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
176 "by setting active flag for all slaves; "
177 "0 for never (default), 1 for always.");
178 module_param(resend_igmp, int, 0);
179 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
182 /*----------------------------- Global variables ----------------------------*/
184 #ifdef CONFIG_NET_POLL_CONTROLLER
185 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
188 int bond_net_id __read_mostly;
190 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
191 static int arp_ip_count;
192 static int bond_mode = BOND_MODE_ROUNDROBIN;
193 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
194 static int lacp_fast;
196 const struct bond_parm_tbl bond_lacp_tbl[] = {
197 { "slow", AD_LACP_SLOW},
198 { "fast", AD_LACP_FAST},
202 const struct bond_parm_tbl bond_mode_tbl[] = {
203 { "balance-rr", BOND_MODE_ROUNDROBIN},
204 { "active-backup", BOND_MODE_ACTIVEBACKUP},
205 { "balance-xor", BOND_MODE_XOR},
206 { "broadcast", BOND_MODE_BROADCAST},
207 { "802.3ad", BOND_MODE_8023AD},
208 { "balance-tlb", BOND_MODE_TLB},
209 { "balance-alb", BOND_MODE_ALB},
213 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
214 { "layer2", BOND_XMIT_POLICY_LAYER2},
215 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
216 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
220 const struct bond_parm_tbl arp_validate_tbl[] = {
221 { "none", BOND_ARP_VALIDATE_NONE},
222 { "active", BOND_ARP_VALIDATE_ACTIVE},
223 { "backup", BOND_ARP_VALIDATE_BACKUP},
224 { "all", BOND_ARP_VALIDATE_ALL},
228 const struct bond_parm_tbl fail_over_mac_tbl[] = {
229 { "none", BOND_FOM_NONE},
230 { "active", BOND_FOM_ACTIVE},
231 { "follow", BOND_FOM_FOLLOW},
235 const struct bond_parm_tbl pri_reselect_tbl[] = {
236 { "always", BOND_PRI_RESELECT_ALWAYS},
237 { "better", BOND_PRI_RESELECT_BETTER},
238 { "failure", BOND_PRI_RESELECT_FAILURE},
242 struct bond_parm_tbl ad_select_tbl[] = {
243 { "stable", BOND_AD_STABLE},
244 { "bandwidth", BOND_AD_BANDWIDTH},
245 { "count", BOND_AD_COUNT},
249 /*-------------------------- Forward declarations ---------------------------*/
251 static int bond_init(struct net_device *bond_dev);
252 static void bond_uninit(struct net_device *bond_dev);
254 /*---------------------------- General routines -----------------------------*/
256 const char *bond_mode_name(int mode)
258 static const char *names[] = {
259 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
260 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
261 [BOND_MODE_XOR] = "load balancing (xor)",
262 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
263 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
264 [BOND_MODE_TLB] = "transmit load balancing",
265 [BOND_MODE_ALB] = "adaptive load balancing",
268 if (mode < 0 || mode > BOND_MODE_ALB)
274 /*---------------------------------- VLAN -----------------------------------*/
277 * bond_add_vlan - add a new vlan id on bond
278 * @bond: bond that got the notification
279 * @vlan_id: the vlan id to add
281 * Returns -ENOMEM if allocation failed.
283 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
285 struct vlan_entry *vlan;
287 pr_debug("bond: %s, vlan id %d\n",
288 (bond ? bond->dev->name : "None"), vlan_id);
290 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
294 INIT_LIST_HEAD(&vlan->vlan_list);
295 vlan->vlan_id = vlan_id;
297 write_lock_bh(&bond->lock);
299 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
301 write_unlock_bh(&bond->lock);
303 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
309 * bond_del_vlan - delete a vlan id from bond
310 * @bond: bond that got the notification
311 * @vlan_id: the vlan id to delete
313 * returns -ENODEV if @vlan_id was not found in @bond.
315 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
317 struct vlan_entry *vlan;
320 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
323 write_lock_bh(&bond->lock);
325 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
326 if (vlan->vlan_id == vlan_id) {
327 list_del(&vlan->vlan_list);
329 if (bond_is_lb(bond))
330 bond_alb_clear_vlan(bond, vlan_id);
332 pr_debug("removed VLAN ID %d from bond %s\n",
333 vlan_id, bond->dev->name);
342 pr_debug("couldn't find VLAN ID %d in bond %s\n",
343 vlan_id, bond->dev->name);
346 write_unlock_bh(&bond->lock);
347 unblock_netpoll_tx();
352 * bond_next_vlan - safely skip to the next item in the vlans list.
353 * @bond: the bond we're working on
354 * @curr: item we're advancing from
356 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
357 * or @curr->next otherwise (even if it is @curr itself again).
359 * Caller must hold bond->lock
361 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
363 struct vlan_entry *next, *last;
365 if (list_empty(&bond->vlan_list))
369 next = list_entry(bond->vlan_list.next,
370 struct vlan_entry, vlan_list);
372 last = list_entry(bond->vlan_list.prev,
373 struct vlan_entry, vlan_list);
375 next = list_entry(bond->vlan_list.next,
376 struct vlan_entry, vlan_list);
378 next = list_entry(curr->vlan_list.next,
379 struct vlan_entry, vlan_list);
387 * bond_dev_queue_xmit - Prepare skb for xmit.
389 * @bond: bond device that got this skb for tx.
390 * @skb: hw accel VLAN tagged skb to transmit
391 * @slave_dev: slave that is supposed to xmit this skbuff
393 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
394 struct net_device *slave_dev)
396 skb->dev = slave_dev;
398 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
399 sizeof(qdisc_skb_cb(skb)->bond_queue_mapping));
400 skb->queue_mapping = qdisc_skb_cb(skb)->bond_queue_mapping;
402 if (unlikely(netpoll_tx_running(slave_dev)))
403 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
411 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
412 * We don't protect the slave list iteration with a lock because:
413 * a. This operation is performed in IOCTL context,
414 * b. The operation is protected by the RTNL semaphore in the 8021q code,
415 * c. Holding a lock with BH disabled while directly calling a base driver
416 * entry point is generally a BAD idea.
418 * The design of synchronization/protection for this operation in the 8021q
419 * module is good for one or more VLAN devices over a single physical device
420 * and cannot be extended for a teaming solution like bonding, so there is a
421 * potential race condition here where a net device from the vlan group might
422 * be referenced (either by a base driver or the 8021q code) while it is being
423 * removed from the system. However, it turns out we're not making matters
424 * worse, and if it works for regular VLAN usage it will work here too.
428 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
429 * @bond_dev: bonding net device that got called
430 * @vid: vlan id being added
432 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
434 struct bonding *bond = netdev_priv(bond_dev);
438 bond_for_each_slave(bond, slave, i) {
439 struct net_device *slave_dev = slave->dev;
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
442 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
443 slave_ops->ndo_vlan_rx_add_vid) {
444 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
448 res = bond_add_vlan(bond, vid);
450 pr_err("%s: Error: Failed to add vlan id %d\n",
451 bond_dev->name, vid);
456 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
457 * @bond_dev: bonding net device that got called
458 * @vid: vlan id being removed
460 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
462 struct bonding *bond = netdev_priv(bond_dev);
466 bond_for_each_slave(bond, slave, i) {
467 struct net_device *slave_dev = slave->dev;
468 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
470 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
471 slave_ops->ndo_vlan_rx_kill_vid) {
472 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
476 res = bond_del_vlan(bond, vid);
478 pr_err("%s: Error: Failed to remove vlan id %d\n",
479 bond_dev->name, vid);
483 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
485 struct vlan_entry *vlan;
486 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
488 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
489 !(slave_ops->ndo_vlan_rx_add_vid))
492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
493 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
496 static void bond_del_vlans_from_slave(struct bonding *bond,
497 struct net_device *slave_dev)
499 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
500 struct vlan_entry *vlan;
502 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
503 !(slave_ops->ndo_vlan_rx_kill_vid))
506 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
509 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
513 /*------------------------------- Link status -------------------------------*/
516 * Set the carrier state for the master according to the state of its
517 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
518 * do special 802.3ad magic.
520 * Returns zero if carrier state does not change, nonzero if it does.
522 static int bond_set_carrier(struct bonding *bond)
527 if (bond->slave_cnt == 0)
530 if (bond->params.mode == BOND_MODE_8023AD)
531 return bond_3ad_set_carrier(bond);
533 bond_for_each_slave(bond, slave, i) {
534 if (slave->link == BOND_LINK_UP) {
535 if (!netif_carrier_ok(bond->dev)) {
536 netif_carrier_on(bond->dev);
544 if (netif_carrier_ok(bond->dev)) {
545 netif_carrier_off(bond->dev);
552 * Get link speed and duplex from the slave's base driver
553 * using ethtool. If for some reason the call fails or the
554 * values are invalid, set speed and duplex to -1,
557 static int bond_update_speed_duplex(struct slave *slave)
559 struct net_device *slave_dev = slave->dev;
560 struct ethtool_cmd ecmd;
564 slave->speed = SPEED_UNKNOWN;
565 slave->duplex = DUPLEX_UNKNOWN;
567 res = __ethtool_get_settings(slave_dev, &ecmd);
571 slave_speed = ethtool_cmd_speed(&ecmd);
572 if (slave_speed == 0 || slave_speed == ((__u32) -1))
575 switch (ecmd.duplex) {
583 slave->speed = slave_speed;
584 slave->duplex = ecmd.duplex;
590 * if <dev> supports MII link status reporting, check its link status.
592 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
593 * depending upon the setting of the use_carrier parameter.
595 * Return either BMSR_LSTATUS, meaning that the link is up (or we
596 * can't tell and just pretend it is), or 0, meaning that the link is
599 * If reporting is non-zero, instead of faking link up, return -1 if
600 * both ETHTOOL and MII ioctls fail (meaning the device does not
601 * support them). If use_carrier is set, return whatever it says.
602 * It'd be nice if there was a good way to tell if a driver supports
603 * netif_carrier, but there really isn't.
605 static int bond_check_dev_link(struct bonding *bond,
606 struct net_device *slave_dev, int reporting)
608 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
609 int (*ioctl)(struct net_device *, struct ifreq *, int);
611 struct mii_ioctl_data *mii;
613 if (!reporting && !netif_running(slave_dev))
616 if (bond->params.use_carrier)
617 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
619 /* Try to get link status using Ethtool first. */
620 if (slave_dev->ethtool_ops) {
621 if (slave_dev->ethtool_ops->get_link) {
624 link = slave_dev->ethtool_ops->get_link(slave_dev);
626 return link ? BMSR_LSTATUS : 0;
630 /* Ethtool can't be used, fallback to MII ioctls. */
631 ioctl = slave_ops->ndo_do_ioctl;
633 /* TODO: set pointer to correct ioctl on a per team member */
634 /* bases to make this more efficient. that is, once */
635 /* we determine the correct ioctl, we will always */
636 /* call it and not the others for that team */
640 * We cannot assume that SIOCGMIIPHY will also read a
641 * register; not all network drivers (e.g., e100)
645 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
646 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
648 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
649 mii->reg_num = MII_BMSR;
650 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
651 return mii->val_out & BMSR_LSTATUS;
656 * If reporting, report that either there's no dev->do_ioctl,
657 * or both SIOCGMIIREG and get_link failed (meaning that we
658 * cannot report link status). If not reporting, pretend
661 return reporting ? -1 : BMSR_LSTATUS;
664 /*----------------------------- Multicast list ------------------------------*/
667 * Push the promiscuity flag down to appropriate slaves
669 static int bond_set_promiscuity(struct bonding *bond, int inc)
672 if (USES_PRIMARY(bond->params.mode)) {
673 /* write lock already acquired */
674 if (bond->curr_active_slave) {
675 err = dev_set_promiscuity(bond->curr_active_slave->dev,
681 bond_for_each_slave(bond, slave, i) {
682 err = dev_set_promiscuity(slave->dev, inc);
691 * Push the allmulti flag down to all slaves
693 static int bond_set_allmulti(struct bonding *bond, int inc)
696 if (USES_PRIMARY(bond->params.mode)) {
697 /* write lock already acquired */
698 if (bond->curr_active_slave) {
699 err = dev_set_allmulti(bond->curr_active_slave->dev,
705 bond_for_each_slave(bond, slave, i) {
706 err = dev_set_allmulti(slave->dev, inc);
715 * Add a Multicast address to slaves
718 static void bond_mc_add(struct bonding *bond, void *addr)
720 if (USES_PRIMARY(bond->params.mode)) {
721 /* write lock already acquired */
722 if (bond->curr_active_slave)
723 dev_mc_add(bond->curr_active_slave->dev, addr);
728 bond_for_each_slave(bond, slave, i)
729 dev_mc_add(slave->dev, addr);
734 * Remove a multicast address from slave
737 static void bond_mc_del(struct bonding *bond, void *addr)
739 if (USES_PRIMARY(bond->params.mode)) {
740 /* write lock already acquired */
741 if (bond->curr_active_slave)
742 dev_mc_del(bond->curr_active_slave->dev, addr);
746 bond_for_each_slave(bond, slave, i) {
747 dev_mc_del(slave->dev, addr);
753 static void __bond_resend_igmp_join_requests(struct net_device *dev)
755 struct in_device *in_dev;
758 in_dev = __in_dev_get_rcu(dev);
760 ip_mc_rejoin_groups(in_dev);
765 * Retrieve the list of registered multicast addresses for the bonding
766 * device and retransmit an IGMP JOIN request to the current active
769 static void bond_resend_igmp_join_requests(struct bonding *bond)
771 struct net_device *vlan_dev;
772 struct vlan_entry *vlan;
774 read_lock(&bond->lock);
776 /* rejoin all groups on bond device */
777 __bond_resend_igmp_join_requests(bond->dev);
779 /* rejoin all groups on vlan devices */
780 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
782 vlan_dev = __vlan_find_dev_deep(bond->dev,
786 __bond_resend_igmp_join_requests(vlan_dev);
789 if (--bond->igmp_retrans > 0)
790 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
792 read_unlock(&bond->lock);
795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
797 struct bonding *bond = container_of(work, struct bonding,
799 bond_resend_igmp_join_requests(bond);
803 * flush all members of flush->mc_list from device dev->mc_list
805 static void bond_mc_list_flush(struct net_device *bond_dev,
806 struct net_device *slave_dev)
808 struct bonding *bond = netdev_priv(bond_dev);
809 struct netdev_hw_addr *ha;
811 netdev_for_each_mc_addr(ha, bond_dev)
812 dev_mc_del(slave_dev, ha->addr);
814 if (bond->params.mode == BOND_MODE_8023AD) {
815 /* del lacpdu mc addr from mc list */
816 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
818 dev_mc_del(slave_dev, lacpdu_multicast);
822 /*--------------------------- Active slave change ---------------------------*/
825 * Update the mc list and multicast-related flags for the new and
826 * old active slaves (if any) according to the multicast mode, and
827 * promiscuous flags unconditionally.
829 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
830 struct slave *old_active)
832 struct netdev_hw_addr *ha;
834 if (!USES_PRIMARY(bond->params.mode))
835 /* nothing to do - mc list is already up-to-date on
841 if (bond->dev->flags & IFF_PROMISC)
842 dev_set_promiscuity(old_active->dev, -1);
844 if (bond->dev->flags & IFF_ALLMULTI)
845 dev_set_allmulti(old_active->dev, -1);
847 netdev_for_each_mc_addr(ha, bond->dev)
848 dev_mc_del(old_active->dev, ha->addr);
852 /* FIXME: Signal errors upstream. */
853 if (bond->dev->flags & IFF_PROMISC)
854 dev_set_promiscuity(new_active->dev, 1);
856 if (bond->dev->flags & IFF_ALLMULTI)
857 dev_set_allmulti(new_active->dev, 1);
859 netdev_for_each_mc_addr(ha, bond->dev)
860 dev_mc_add(new_active->dev, ha->addr);
865 * bond_do_fail_over_mac
867 * Perform special MAC address swapping for fail_over_mac settings
869 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
871 static void bond_do_fail_over_mac(struct bonding *bond,
872 struct slave *new_active,
873 struct slave *old_active)
874 __releases(&bond->curr_slave_lock)
875 __releases(&bond->lock)
876 __acquires(&bond->lock)
877 __acquires(&bond->curr_slave_lock)
879 u8 tmp_mac[ETH_ALEN];
880 struct sockaddr saddr;
883 switch (bond->params.fail_over_mac) {
884 case BOND_FOM_ACTIVE:
886 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
887 new_active->dev->addr_len);
889 case BOND_FOM_FOLLOW:
891 * if new_active && old_active, swap them
892 * if just old_active, do nothing (going to no active slave)
893 * if just new_active, set new_active to bond's MAC
898 write_unlock_bh(&bond->curr_slave_lock);
899 read_unlock(&bond->lock);
902 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
903 memcpy(saddr.sa_data, old_active->dev->dev_addr,
905 saddr.sa_family = new_active->dev->type;
907 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
908 saddr.sa_family = bond->dev->type;
911 rv = dev_set_mac_address(new_active->dev, &saddr);
913 pr_err("%s: Error %d setting MAC of slave %s\n",
914 bond->dev->name, -rv, new_active->dev->name);
921 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
922 saddr.sa_family = old_active->dev->type;
924 rv = dev_set_mac_address(old_active->dev, &saddr);
926 pr_err("%s: Error %d setting MAC of slave %s\n",
927 bond->dev->name, -rv, new_active->dev->name);
929 read_lock(&bond->lock);
930 write_lock_bh(&bond->curr_slave_lock);
933 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
934 bond->dev->name, bond->params.fail_over_mac);
940 static bool bond_should_change_active(struct bonding *bond)
942 struct slave *prim = bond->primary_slave;
943 struct slave *curr = bond->curr_active_slave;
945 if (!prim || !curr || curr->link != BOND_LINK_UP)
947 if (bond->force_primary) {
948 bond->force_primary = false;
951 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
952 (prim->speed < curr->speed ||
953 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
955 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
961 * find_best_interface - select the best available slave to be the active one
962 * @bond: our bonding struct
964 * Warning: Caller must hold curr_slave_lock for writing.
966 static struct slave *bond_find_best_slave(struct bonding *bond)
968 struct slave *new_active, *old_active;
969 struct slave *bestslave = NULL;
970 int mintime = bond->params.updelay;
973 new_active = bond->curr_active_slave;
975 if (!new_active) { /* there were no active slaves left */
976 if (bond->slave_cnt > 0) /* found one slave */
977 new_active = bond->first_slave;
979 return NULL; /* still no slave, return NULL */
982 if ((bond->primary_slave) &&
983 bond->primary_slave->link == BOND_LINK_UP &&
984 bond_should_change_active(bond)) {
985 new_active = bond->primary_slave;
988 /* remember where to stop iterating over the slaves */
989 old_active = new_active;
991 bond_for_each_slave_from(bond, new_active, i, old_active) {
992 if (new_active->link == BOND_LINK_UP) {
994 } else if (new_active->link == BOND_LINK_BACK &&
995 IS_UP(new_active->dev)) {
996 /* link up, but waiting for stabilization */
997 if (new_active->delay < mintime) {
998 mintime = new_active->delay;
999 bestslave = new_active;
1007 static bool bond_should_notify_peers(struct bonding *bond)
1009 struct slave *slave = bond->curr_active_slave;
1011 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1012 bond->dev->name, slave ? slave->dev->name : "NULL");
1014 if (!slave || !bond->send_peer_notif ||
1015 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1018 bond->send_peer_notif--;
1023 * change_active_interface - change the active slave into the specified one
1024 * @bond: our bonding struct
1025 * @new: the new slave to make the active one
1027 * Set the new slave to the bond's settings and unset them on the old
1028 * curr_active_slave.
1029 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1031 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1032 * because it is apparently the best available slave we have, even though its
1033 * updelay hasn't timed out yet.
1035 * If new_active is not NULL, caller must hold bond->lock for read and
1036 * curr_slave_lock for write_bh.
1038 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1040 struct slave *old_active = bond->curr_active_slave;
1042 if (old_active == new_active)
1046 new_active->jiffies = jiffies;
1048 if (new_active->link == BOND_LINK_BACK) {
1049 if (USES_PRIMARY(bond->params.mode)) {
1050 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1051 bond->dev->name, new_active->dev->name,
1052 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1055 new_active->delay = 0;
1056 new_active->link = BOND_LINK_UP;
1058 if (bond->params.mode == BOND_MODE_8023AD)
1059 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1061 if (bond_is_lb(bond))
1062 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1064 if (USES_PRIMARY(bond->params.mode)) {
1065 pr_info("%s: making interface %s the new active one.\n",
1066 bond->dev->name, new_active->dev->name);
1071 if (USES_PRIMARY(bond->params.mode))
1072 bond_mc_swap(bond, new_active, old_active);
1074 if (bond_is_lb(bond)) {
1075 bond_alb_handle_active_change(bond, new_active);
1077 bond_set_slave_inactive_flags(old_active);
1079 bond_set_slave_active_flags(new_active);
1081 bond->curr_active_slave = new_active;
1084 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1086 bond_set_slave_inactive_flags(old_active);
1089 bool should_notify_peers = false;
1091 bond_set_slave_active_flags(new_active);
1093 if (bond->params.fail_over_mac)
1094 bond_do_fail_over_mac(bond, new_active,
1097 if (netif_running(bond->dev)) {
1098 bond->send_peer_notif =
1099 bond->params.num_peer_notif;
1100 should_notify_peers =
1101 bond_should_notify_peers(bond);
1104 write_unlock_bh(&bond->curr_slave_lock);
1105 read_unlock(&bond->lock);
1107 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1108 if (should_notify_peers)
1109 netdev_bonding_change(bond->dev,
1110 NETDEV_NOTIFY_PEERS);
1112 read_lock(&bond->lock);
1113 write_lock_bh(&bond->curr_slave_lock);
1117 /* resend IGMP joins since active slave has changed or
1118 * all were sent on curr_active_slave.
1119 * resend only if bond is brought up with the affected
1120 * bonding modes and the retransmission is enabled */
1121 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1122 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1123 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1124 bond->igmp_retrans = bond->params.resend_igmp;
1125 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1130 * bond_select_active_slave - select a new active slave, if needed
1131 * @bond: our bonding struct
1133 * This functions should be called when one of the following occurs:
1134 * - The old curr_active_slave has been released or lost its link.
1135 * - The primary_slave has got its link back.
1136 * - A slave has got its link back and there's no old curr_active_slave.
1138 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1140 void bond_select_active_slave(struct bonding *bond)
1142 struct slave *best_slave;
1145 best_slave = bond_find_best_slave(bond);
1146 if (best_slave != bond->curr_active_slave) {
1147 bond_change_active_slave(bond, best_slave);
1148 rv = bond_set_carrier(bond);
1152 if (netif_carrier_ok(bond->dev)) {
1153 pr_info("%s: first active interface up!\n",
1156 pr_info("%s: now running without any active interface !\n",
1162 /*--------------------------- slave list handling ---------------------------*/
1165 * This function attaches the slave to the end of list.
1167 * bond->lock held for writing by caller.
1169 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1171 if (bond->first_slave == NULL) { /* attaching the first slave */
1172 new_slave->next = new_slave;
1173 new_slave->prev = new_slave;
1174 bond->first_slave = new_slave;
1176 new_slave->next = bond->first_slave;
1177 new_slave->prev = bond->first_slave->prev;
1178 new_slave->next->prev = new_slave;
1179 new_slave->prev->next = new_slave;
1186 * This function detaches the slave from the list.
1187 * WARNING: no check is made to verify if the slave effectively
1188 * belongs to <bond>.
1189 * Nothing is freed on return, structures are just unchained.
1190 * If any slave pointer in bond was pointing to <slave>,
1191 * it should be changed by the calling function.
1193 * bond->lock held for writing by caller.
1195 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1198 slave->next->prev = slave->prev;
1201 slave->prev->next = slave->next;
1203 if (bond->first_slave == slave) { /* slave is the first slave */
1204 if (bond->slave_cnt > 1) { /* there are more slave */
1205 bond->first_slave = slave->next;
1207 bond->first_slave = NULL; /* slave was the last one */
1216 #ifdef CONFIG_NET_POLL_CONTROLLER
1217 static inline int slave_enable_netpoll(struct slave *slave)
1222 np = kzalloc(sizeof(*np), GFP_KERNEL);
1227 np->dev = slave->dev;
1228 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1229 err = __netpoll_setup(np);
1238 static inline void slave_disable_netpoll(struct slave *slave)
1240 struct netpoll *np = slave->np;
1246 synchronize_rcu_bh();
1247 __netpoll_cleanup(np);
1250 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1252 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1254 if (!slave_dev->netdev_ops->ndo_poll_controller)
1259 static void bond_poll_controller(struct net_device *bond_dev)
1263 static void __bond_netpoll_cleanup(struct bonding *bond)
1265 struct slave *slave;
1268 bond_for_each_slave(bond, slave, i)
1269 if (IS_UP(slave->dev))
1270 slave_disable_netpoll(slave);
1272 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1274 struct bonding *bond = netdev_priv(bond_dev);
1276 read_lock(&bond->lock);
1277 __bond_netpoll_cleanup(bond);
1278 read_unlock(&bond->lock);
1281 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1283 struct bonding *bond = netdev_priv(dev);
1284 struct slave *slave;
1287 read_lock(&bond->lock);
1288 bond_for_each_slave(bond, slave, i) {
1289 err = slave_enable_netpoll(slave);
1291 __bond_netpoll_cleanup(bond);
1295 read_unlock(&bond->lock);
1299 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1301 return bond->dev->npinfo;
1305 static inline int slave_enable_netpoll(struct slave *slave)
1309 static inline void slave_disable_netpoll(struct slave *slave)
1312 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 /*---------------------------------- IOCTL ----------------------------------*/
1319 static int bond_sethwaddr(struct net_device *bond_dev,
1320 struct net_device *slave_dev)
1322 pr_debug("bond_dev=%p\n", bond_dev);
1323 pr_debug("slave_dev=%p\n", slave_dev);
1324 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1325 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1329 static u32 bond_fix_features(struct net_device *dev, u32 features)
1331 struct slave *slave;
1332 struct bonding *bond = netdev_priv(dev);
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 u32 vlan_features = BOND_VLAN_FEATURES;
1368 unsigned short max_hard_header_len = ETH_HLEN;
1369 unsigned int gso_max_size = GSO_MAX_SIZE;
1370 u16 gso_max_segs = GSO_MAX_SEGS;
1373 read_lock(&bond->lock);
1375 if (!bond->first_slave)
1378 bond_for_each_slave(bond, slave, i) {
1379 vlan_features = netdev_increment_features(vlan_features,
1380 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1382 if (slave->dev->hard_header_len > max_hard_header_len)
1383 max_hard_header_len = slave->dev->hard_header_len;
1385 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1386 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1390 bond_dev->vlan_features = vlan_features;
1391 bond_dev->hard_header_len = max_hard_header_len;
1392 bond_dev->gso_max_segs = gso_max_segs;
1393 netif_set_gso_max_size(bond_dev, gso_max_size);
1395 read_unlock(&bond->lock);
1397 netdev_change_features(bond_dev);
1400 static void bond_setup_by_slave(struct net_device *bond_dev,
1401 struct net_device *slave_dev)
1403 struct bonding *bond = netdev_priv(bond_dev);
1405 bond_dev->header_ops = slave_dev->header_ops;
1407 bond_dev->type = slave_dev->type;
1408 bond_dev->hard_header_len = slave_dev->hard_header_len;
1409 bond_dev->addr_len = slave_dev->addr_len;
1411 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1412 slave_dev->addr_len);
1413 bond->setup_by_slave = 1;
1416 /* On bonding slaves other than the currently active slave, suppress
1417 * duplicates except for alb non-mcast/bcast.
1419 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1420 struct slave *slave,
1421 struct bonding *bond)
1423 if (bond_is_slave_inactive(slave)) {
1424 if (bond->params.mode == BOND_MODE_ALB &&
1425 skb->pkt_type != PACKET_BROADCAST &&
1426 skb->pkt_type != PACKET_MULTICAST)
1433 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1435 struct sk_buff *skb = *pskb;
1436 struct slave *slave;
1437 struct bonding *bond;
1438 void (*recv_probe)(struct sk_buff *, struct bonding *,
1441 skb = skb_share_check(skb, GFP_ATOMIC);
1443 return RX_HANDLER_CONSUMED;
1447 slave = bond_slave_get_rcu(skb->dev);
1450 if (bond->params.arp_interval)
1451 slave->dev->last_rx = jiffies;
1453 recv_probe = ACCESS_ONCE(bond->recv_probe);
1455 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1458 recv_probe(nskb, bond, slave);
1459 dev_kfree_skb(nskb);
1463 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1464 return RX_HANDLER_EXACT;
1467 skb->dev = bond->dev;
1469 if (bond->params.mode == BOND_MODE_ALB &&
1470 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1471 skb->pkt_type == PACKET_HOST) {
1473 if (unlikely(skb_cow_head(skb,
1474 skb->data - skb_mac_header(skb)))) {
1476 return RX_HANDLER_CONSUMED;
1478 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1481 return RX_HANDLER_ANOTHER;
1484 /* enslave device <slave> to bond device <master> */
1485 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1487 struct bonding *bond = netdev_priv(bond_dev);
1488 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1489 struct slave *new_slave = NULL;
1490 struct netdev_hw_addr *ha;
1491 struct sockaddr addr;
1495 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1496 slave_ops->ndo_do_ioctl == NULL) {
1497 pr_warning("%s: Warning: no link monitoring support for %s\n",
1498 bond_dev->name, slave_dev->name);
1501 /* already enslaved */
1502 if (slave_dev->flags & IFF_SLAVE) {
1503 pr_debug("Error, Device was already enslaved\n");
1507 /* vlan challenged mutual exclusion */
1508 /* no need to lock since we're protected by rtnl_lock */
1509 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1510 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1511 if (bond_vlan_used(bond)) {
1512 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1513 bond_dev->name, slave_dev->name, bond_dev->name);
1516 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1517 bond_dev->name, slave_dev->name,
1518 slave_dev->name, bond_dev->name);
1521 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1525 * Old ifenslave binaries are no longer supported. These can
1526 * be identified with moderate accuracy by the state of the slave:
1527 * the current ifenslave will set the interface down prior to
1528 * enslaving it; the old ifenslave will not.
1530 if ((slave_dev->flags & IFF_UP)) {
1531 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1534 goto err_undo_flags;
1537 /* set bonding device ether type by slave - bonding netdevices are
1538 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1539 * there is a need to override some of the type dependent attribs/funcs.
1541 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1542 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1544 if (bond->slave_cnt == 0) {
1545 if (bond_dev->type != slave_dev->type) {
1546 pr_debug("%s: change device type from %d to %d\n",
1548 bond_dev->type, slave_dev->type);
1550 res = netdev_bonding_change(bond_dev,
1551 NETDEV_PRE_TYPE_CHANGE);
1552 res = notifier_to_errno(res);
1554 pr_err("%s: refused to change device type\n",
1557 goto err_undo_flags;
1560 /* Flush unicast and multicast addresses */
1561 dev_uc_flush(bond_dev);
1562 dev_mc_flush(bond_dev);
1564 if (slave_dev->type != ARPHRD_ETHER)
1565 bond_setup_by_slave(bond_dev, slave_dev);
1567 ether_setup(bond_dev);
1568 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1571 netdev_bonding_change(bond_dev,
1572 NETDEV_POST_TYPE_CHANGE);
1574 } else if (bond_dev->type != slave_dev->type) {
1575 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1577 slave_dev->type, bond_dev->type);
1579 goto err_undo_flags;
1582 if (slave_ops->ndo_set_mac_address == NULL) {
1583 if (bond->slave_cnt == 0) {
1584 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1586 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1587 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1588 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1591 goto err_undo_flags;
1595 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1597 /* If this is the first slave, then we need to set the master's hardware
1598 * address to be the same as the slave's. */
1599 if (is_zero_ether_addr(bond->dev->dev_addr))
1600 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1601 slave_dev->addr_len);
1604 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1607 goto err_undo_flags;
1611 * Set the new_slave's queue_id to be zero. Queue ID mapping
1612 * is set via sysfs or module option if desired.
1614 new_slave->queue_id = 0;
1616 /* Save slave's original mtu and then set it to match the bond */
1617 new_slave->original_mtu = slave_dev->mtu;
1618 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1620 pr_debug("Error %d calling dev_set_mtu\n", res);
1625 * Save slave's original ("permanent") mac address for modes
1626 * that need it, and for restoring it upon release, and then
1627 * set it to the master's address
1629 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1631 if (!bond->params.fail_over_mac) {
1633 * Set slave to master's mac address. The application already
1634 * set the master's mac address to that of the first slave
1636 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1637 addr.sa_family = slave_dev->type;
1638 res = dev_set_mac_address(slave_dev, &addr);
1640 pr_debug("Error %d calling set_mac_address\n", res);
1641 goto err_restore_mtu;
1645 res = netdev_set_bond_master(slave_dev, bond_dev);
1647 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1648 goto err_restore_mac;
1651 /* open the slave since the application closed it */
1652 res = dev_open(slave_dev);
1654 pr_debug("Opening slave %s failed\n", slave_dev->name);
1655 goto err_unset_master;
1658 new_slave->bond = bond;
1659 new_slave->dev = slave_dev;
1660 slave_dev->priv_flags |= IFF_BONDING;
1662 if (bond_is_lb(bond)) {
1663 /* bond_alb_init_slave() must be called before all other stages since
1664 * it might fail and we do not want to have to undo everything
1666 res = bond_alb_init_slave(bond, new_slave);
1671 /* If the mode USES_PRIMARY, then the new slave gets the
1672 * master's promisc (and mc) settings only if it becomes the
1673 * curr_active_slave, and that is taken care of later when calling
1674 * bond_change_active()
1676 if (!USES_PRIMARY(bond->params.mode)) {
1677 /* set promiscuity level to new slave */
1678 if (bond_dev->flags & IFF_PROMISC) {
1679 res = dev_set_promiscuity(slave_dev, 1);
1684 /* set allmulti level to new slave */
1685 if (bond_dev->flags & IFF_ALLMULTI) {
1686 res = dev_set_allmulti(slave_dev, 1);
1691 netif_addr_lock_bh(bond_dev);
1692 /* upload master's mc_list to new slave */
1693 netdev_for_each_mc_addr(ha, bond_dev)
1694 dev_mc_add(slave_dev, ha->addr);
1695 netif_addr_unlock_bh(bond_dev);
1698 if (bond->params.mode == BOND_MODE_8023AD) {
1699 /* add lacpdu mc addr to mc list */
1700 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1702 dev_mc_add(slave_dev, lacpdu_multicast);
1705 bond_add_vlans_on_slave(bond, slave_dev);
1707 write_lock_bh(&bond->lock);
1709 bond_attach_slave(bond, new_slave);
1711 new_slave->delay = 0;
1712 new_slave->link_failure_count = 0;
1714 write_unlock_bh(&bond->lock);
1716 bond_compute_features(bond);
1718 bond_update_speed_duplex(new_slave);
1720 read_lock(&bond->lock);
1722 new_slave->last_arp_rx = jiffies;
1724 if (bond->params.miimon && !bond->params.use_carrier) {
1725 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1727 if ((link_reporting == -1) && !bond->params.arp_interval) {
1729 * miimon is set but a bonded network driver
1730 * does not support ETHTOOL/MII and
1731 * arp_interval is not set. Note: if
1732 * use_carrier is enabled, we will never go
1733 * here (because netif_carrier is always
1734 * supported); thus, we don't need to change
1735 * the messages for netif_carrier.
1737 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",
1738 bond_dev->name, slave_dev->name);
1739 } else if (link_reporting == -1) {
1740 /* unable get link status using mii/ethtool */
1741 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",
1742 bond_dev->name, slave_dev->name);
1746 /* check for initial state */
1747 if (!bond->params.miimon ||
1748 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1749 if (bond->params.updelay) {
1750 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1751 new_slave->link = BOND_LINK_BACK;
1752 new_slave->delay = bond->params.updelay;
1754 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1755 new_slave->link = BOND_LINK_UP;
1757 new_slave->jiffies = jiffies;
1759 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1760 new_slave->link = BOND_LINK_DOWN;
1763 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1764 /* if there is a primary slave, remember it */
1765 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1766 bond->primary_slave = new_slave;
1767 bond->force_primary = true;
1771 write_lock_bh(&bond->curr_slave_lock);
1773 switch (bond->params.mode) {
1774 case BOND_MODE_ACTIVEBACKUP:
1775 bond_set_slave_inactive_flags(new_slave);
1776 bond_select_active_slave(bond);
1778 case BOND_MODE_8023AD:
1779 /* in 802.3ad mode, the internal mechanism
1780 * will activate the slaves in the selected
1783 bond_set_slave_inactive_flags(new_slave);
1784 /* if this is the first slave */
1785 if (bond->slave_cnt == 1) {
1786 SLAVE_AD_INFO(new_slave).id = 1;
1787 /* Initialize AD with the number of times that the AD timer is called in 1 second
1788 * can be called only after the mac address of the bond is set
1790 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1792 SLAVE_AD_INFO(new_slave).id =
1793 SLAVE_AD_INFO(new_slave->prev).id + 1;
1796 bond_3ad_bind_slave(new_slave);
1800 bond_set_active_slave(new_slave);
1801 bond_set_slave_inactive_flags(new_slave);
1802 bond_select_active_slave(bond);
1805 pr_debug("This slave is always active in trunk mode\n");
1807 /* always active in trunk mode */
1808 bond_set_active_slave(new_slave);
1810 /* In trunking mode there is little meaning to curr_active_slave
1811 * anyway (it holds no special properties of the bond device),
1812 * so we can change it without calling change_active_interface()
1814 if (!bond->curr_active_slave)
1815 bond->curr_active_slave = new_slave;
1818 } /* switch(bond_mode) */
1820 write_unlock_bh(&bond->curr_slave_lock);
1822 bond_set_carrier(bond);
1824 #ifdef CONFIG_NET_POLL_CONTROLLER
1825 slave_dev->npinfo = bond_netpoll_info(bond);
1826 if (slave_dev->npinfo) {
1827 if (slave_enable_netpoll(new_slave)) {
1828 read_unlock(&bond->lock);
1829 pr_info("Error, %s: master_dev is using netpoll, "
1830 "but new slave device does not support netpoll.\n",
1838 read_unlock(&bond->lock);
1840 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1844 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1847 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1848 goto err_dest_symlinks;
1851 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1852 bond_dev->name, slave_dev->name,
1853 bond_is_active_slave(new_slave) ? "n active" : " backup",
1854 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1856 /* enslave is successful */
1859 /* Undo stages on error */
1861 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1864 write_lock_bh(&bond->lock);
1865 bond_detach_slave(bond, new_slave);
1866 write_unlock_bh(&bond->lock);
1869 slave_dev->priv_flags &= ~IFF_BONDING;
1870 dev_close(slave_dev);
1873 netdev_set_bond_master(slave_dev, NULL);
1876 if (!bond->params.fail_over_mac) {
1877 /* XXX TODO - fom follow mode needs to change master's
1878 * MAC if this slave's MAC is in use by the bond, or at
1879 * least print a warning.
1881 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1882 addr.sa_family = slave_dev->type;
1883 dev_set_mac_address(slave_dev, &addr);
1887 dev_set_mtu(slave_dev, new_slave->original_mtu);
1893 bond_compute_features(bond);
1899 * Try to release the slave device <slave> from the bond device <master>
1900 * It is legal to access curr_active_slave without a lock because all the function
1903 * The rules for slave state should be:
1904 * for Active/Backup:
1905 * Active stays on all backups go down
1906 * for Bonded connections:
1907 * The first up interface should be left on and all others downed.
1909 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1911 struct bonding *bond = netdev_priv(bond_dev);
1912 struct slave *slave, *oldcurrent;
1913 struct sockaddr addr;
1914 int old_flags = bond_dev->flags;
1915 u32 old_features = bond_dev->features;
1917 /* slave is not a slave or master is not master of this slave */
1918 if (!(slave_dev->flags & IFF_SLAVE) ||
1919 (slave_dev->master != bond_dev)) {
1920 pr_err("%s: Error: cannot release %s.\n",
1921 bond_dev->name, slave_dev->name);
1926 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1927 write_lock_bh(&bond->lock);
1929 slave = bond_get_slave_by_dev(bond, slave_dev);
1931 /* not a slave of this bond */
1932 pr_info("%s: %s not enslaved\n",
1933 bond_dev->name, slave_dev->name);
1934 write_unlock_bh(&bond->lock);
1935 unblock_netpoll_tx();
1939 write_unlock_bh(&bond->lock);
1940 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1941 * for this slave anymore.
1943 netdev_rx_handler_unregister(slave_dev);
1944 write_lock_bh(&bond->lock);
1946 if (!bond->params.fail_over_mac) {
1947 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1948 bond->slave_cnt > 1)
1949 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1950 bond_dev->name, slave_dev->name,
1952 bond_dev->name, slave_dev->name);
1955 /* Inform AD package of unbinding of slave. */
1956 if (bond->params.mode == BOND_MODE_8023AD) {
1957 /* must be called before the slave is
1958 * detached from the list
1960 bond_3ad_unbind_slave(slave);
1963 pr_info("%s: releasing %s interface %s\n",
1965 bond_is_active_slave(slave) ? "active" : "backup",
1968 oldcurrent = bond->curr_active_slave;
1970 bond->current_arp_slave = NULL;
1972 /* release the slave from its bond */
1973 bond_detach_slave(bond, slave);
1975 if (bond->primary_slave == slave)
1976 bond->primary_slave = NULL;
1978 if (oldcurrent == slave)
1979 bond_change_active_slave(bond, NULL);
1981 if (bond_is_lb(bond)) {
1982 /* Must be called only after the slave has been
1983 * detached from the list and the curr_active_slave
1984 * has been cleared (if our_slave == old_current),
1985 * but before a new active slave is selected.
1987 write_unlock_bh(&bond->lock);
1988 bond_alb_deinit_slave(bond, slave);
1989 write_lock_bh(&bond->lock);
1992 if (oldcurrent == slave) {
1994 * Note that we hold RTNL over this sequence, so there
1995 * is no concern that another slave add/remove event
1998 write_unlock_bh(&bond->lock);
1999 read_lock(&bond->lock);
2000 write_lock_bh(&bond->curr_slave_lock);
2002 bond_select_active_slave(bond);
2004 write_unlock_bh(&bond->curr_slave_lock);
2005 read_unlock(&bond->lock);
2006 write_lock_bh(&bond->lock);
2009 if (bond->slave_cnt == 0) {
2010 bond_set_carrier(bond);
2012 /* if the last slave was removed, zero the mac address
2013 * of the master so it will be set by the application
2014 * to the mac address of the first slave
2016 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2018 if (bond_vlan_used(bond)) {
2019 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2020 bond_dev->name, bond_dev->name);
2021 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2026 write_unlock_bh(&bond->lock);
2027 unblock_netpoll_tx();
2029 bond_compute_features(bond);
2030 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2031 (old_features & NETIF_F_VLAN_CHALLENGED))
2032 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2033 bond_dev->name, slave_dev->name, bond_dev->name);
2035 /* must do this from outside any spinlocks */
2036 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2038 bond_del_vlans_from_slave(bond, slave_dev);
2040 /* If the mode USES_PRIMARY, then we should only remove its
2041 * promisc and mc settings if it was the curr_active_slave, but that was
2042 * already taken care of above when we detached the slave
2044 if (!USES_PRIMARY(bond->params.mode)) {
2045 /* unset promiscuity level from slave
2046 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2047 * of the IFF_PROMISC flag in the bond_dev, but we need the
2048 * value of that flag before that change, as that was the value
2049 * when this slave was attached, so we cache at the start of the
2050 * function and use it here. Same goes for ALLMULTI below
2052 if (old_flags & IFF_PROMISC)
2053 dev_set_promiscuity(slave_dev, -1);
2055 /* unset allmulti level from slave */
2056 if (old_flags & IFF_ALLMULTI)
2057 dev_set_allmulti(slave_dev, -1);
2059 /* flush master's mc_list from slave */
2060 netif_addr_lock_bh(bond_dev);
2061 bond_mc_list_flush(bond_dev, slave_dev);
2062 netif_addr_unlock_bh(bond_dev);
2065 netdev_set_bond_master(slave_dev, NULL);
2067 slave_disable_netpoll(slave);
2069 /* close slave before restoring its mac address */
2070 dev_close(slave_dev);
2072 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2073 /* restore original ("permanent") mac address */
2074 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2075 addr.sa_family = slave_dev->type;
2076 dev_set_mac_address(slave_dev, &addr);
2079 dev_set_mtu(slave_dev, slave->original_mtu);
2081 slave_dev->priv_flags &= ~IFF_BONDING;
2085 return 0; /* deletion OK */
2089 * First release a slave and then destroy the bond if no more slaves are left.
2090 * Must be under rtnl_lock when this function is called.
2092 static int bond_release_and_destroy(struct net_device *bond_dev,
2093 struct net_device *slave_dev)
2095 struct bonding *bond = netdev_priv(bond_dev);
2098 ret = bond_release(bond_dev, slave_dev);
2099 if ((ret == 0) && (bond->slave_cnt == 0)) {
2100 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2101 pr_info("%s: destroying bond %s.\n",
2102 bond_dev->name, bond_dev->name);
2103 unregister_netdevice(bond_dev);
2109 * This function releases all slaves.
2111 static int bond_release_all(struct net_device *bond_dev)
2113 struct bonding *bond = netdev_priv(bond_dev);
2114 struct slave *slave;
2115 struct net_device *slave_dev;
2116 struct sockaddr addr;
2118 write_lock_bh(&bond->lock);
2120 netif_carrier_off(bond_dev);
2122 if (bond->slave_cnt == 0)
2125 bond->current_arp_slave = NULL;
2126 bond->primary_slave = NULL;
2127 bond_change_active_slave(bond, NULL);
2129 while ((slave = bond->first_slave) != NULL) {
2130 /* Inform AD package of unbinding of slave
2131 * before slave is detached from the list.
2133 if (bond->params.mode == BOND_MODE_8023AD)
2134 bond_3ad_unbind_slave(slave);
2136 slave_dev = slave->dev;
2137 bond_detach_slave(bond, slave);
2139 /* now that the slave is detached, unlock and perform
2140 * all the undo steps that should not be called from
2143 write_unlock_bh(&bond->lock);
2145 /* unregister rx_handler early so bond_handle_frame wouldn't
2146 * be called for this slave anymore.
2148 netdev_rx_handler_unregister(slave_dev);
2151 if (bond_is_lb(bond)) {
2152 /* must be called only after the slave
2153 * has been detached from the list
2155 bond_alb_deinit_slave(bond, slave);
2158 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2159 bond_del_vlans_from_slave(bond, slave_dev);
2161 /* If the mode USES_PRIMARY, then we should only remove its
2162 * promisc and mc settings if it was the curr_active_slave, but that was
2163 * already taken care of above when we detached the slave
2165 if (!USES_PRIMARY(bond->params.mode)) {
2166 /* unset promiscuity level from slave */
2167 if (bond_dev->flags & IFF_PROMISC)
2168 dev_set_promiscuity(slave_dev, -1);
2170 /* unset allmulti level from slave */
2171 if (bond_dev->flags & IFF_ALLMULTI)
2172 dev_set_allmulti(slave_dev, -1);
2174 /* flush master's mc_list from slave */
2175 netif_addr_lock_bh(bond_dev);
2176 bond_mc_list_flush(bond_dev, slave_dev);
2177 netif_addr_unlock_bh(bond_dev);
2180 netdev_set_bond_master(slave_dev, NULL);
2182 slave_disable_netpoll(slave);
2184 /* close slave before restoring its mac address */
2185 dev_close(slave_dev);
2187 if (!bond->params.fail_over_mac) {
2188 /* restore original ("permanent") mac address*/
2189 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2190 addr.sa_family = slave_dev->type;
2191 dev_set_mac_address(slave_dev, &addr);
2196 /* re-acquire the lock before getting the next slave */
2197 write_lock_bh(&bond->lock);
2200 /* zero the mac address of the master so it will be
2201 * set by the application to the mac address of the
2204 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2206 if (bond_vlan_used(bond)) {
2207 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2208 bond_dev->name, bond_dev->name);
2209 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2213 pr_info("%s: released all slaves\n", bond_dev->name);
2216 write_unlock_bh(&bond->lock);
2218 bond_compute_features(bond);
2224 * This function changes the active slave to slave <slave_dev>.
2225 * It returns -EINVAL in the following cases.
2226 * - <slave_dev> is not found in the list.
2227 * - There is not active slave now.
2228 * - <slave_dev> is already active.
2229 * - The link state of <slave_dev> is not BOND_LINK_UP.
2230 * - <slave_dev> is not running.
2231 * In these cases, this function does nothing.
2232 * In the other cases, current_slave pointer is changed and 0 is returned.
2234 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2236 struct bonding *bond = netdev_priv(bond_dev);
2237 struct slave *old_active = NULL;
2238 struct slave *new_active = NULL;
2241 if (!USES_PRIMARY(bond->params.mode))
2244 /* Verify that master_dev is indeed the master of slave_dev */
2245 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2248 read_lock(&bond->lock);
2250 read_lock(&bond->curr_slave_lock);
2251 old_active = bond->curr_active_slave;
2252 read_unlock(&bond->curr_slave_lock);
2254 new_active = bond_get_slave_by_dev(bond, slave_dev);
2257 * Changing to the current active: do nothing; return success.
2259 if (new_active && (new_active == old_active)) {
2260 read_unlock(&bond->lock);
2266 (new_active->link == BOND_LINK_UP) &&
2267 IS_UP(new_active->dev)) {
2269 write_lock_bh(&bond->curr_slave_lock);
2270 bond_change_active_slave(bond, new_active);
2271 write_unlock_bh(&bond->curr_slave_lock);
2272 unblock_netpoll_tx();
2276 read_unlock(&bond->lock);
2281 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2283 struct bonding *bond = netdev_priv(bond_dev);
2285 info->bond_mode = bond->params.mode;
2286 info->miimon = bond->params.miimon;
2288 read_lock(&bond->lock);
2289 info->num_slaves = bond->slave_cnt;
2290 read_unlock(&bond->lock);
2295 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2297 struct bonding *bond = netdev_priv(bond_dev);
2298 struct slave *slave;
2299 int i, res = -ENODEV;
2301 read_lock(&bond->lock);
2303 bond_for_each_slave(bond, slave, i) {
2304 if (i == (int)info->slave_id) {
2306 strcpy(info->slave_name, slave->dev->name);
2307 info->link = slave->link;
2308 info->state = bond_slave_state(slave);
2309 info->link_failure_count = slave->link_failure_count;
2314 read_unlock(&bond->lock);
2319 /*-------------------------------- Monitoring -------------------------------*/
2322 static int bond_miimon_inspect(struct bonding *bond)
2324 struct slave *slave;
2325 int i, link_state, commit = 0;
2326 bool ignore_updelay;
2328 ignore_updelay = !bond->curr_active_slave ? true : false;
2330 bond_for_each_slave(bond, slave, i) {
2331 slave->new_link = BOND_LINK_NOCHANGE;
2333 link_state = bond_check_dev_link(bond, slave->dev, 0);
2335 switch (slave->link) {
2340 slave->link = BOND_LINK_FAIL;
2341 slave->delay = bond->params.downdelay;
2343 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2345 (bond->params.mode ==
2346 BOND_MODE_ACTIVEBACKUP) ?
2347 (bond_is_active_slave(slave) ?
2348 "active " : "backup ") : "",
2350 bond->params.downdelay * bond->params.miimon);
2353 case BOND_LINK_FAIL:
2356 * recovered before downdelay expired
2358 slave->link = BOND_LINK_UP;
2359 slave->jiffies = jiffies;
2360 pr_info("%s: link status up again after %d ms for interface %s.\n",
2362 (bond->params.downdelay - slave->delay) *
2363 bond->params.miimon,
2368 if (slave->delay <= 0) {
2369 slave->new_link = BOND_LINK_DOWN;
2377 case BOND_LINK_DOWN:
2381 slave->link = BOND_LINK_BACK;
2382 slave->delay = bond->params.updelay;
2385 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2386 bond->dev->name, slave->dev->name,
2387 ignore_updelay ? 0 :
2388 bond->params.updelay *
2389 bond->params.miimon);
2392 case BOND_LINK_BACK:
2394 slave->link = BOND_LINK_DOWN;
2395 pr_info("%s: link status down again after %d ms for interface %s.\n",
2397 (bond->params.updelay - slave->delay) *
2398 bond->params.miimon,
2407 if (slave->delay <= 0) {
2408 slave->new_link = BOND_LINK_UP;
2410 ignore_updelay = false;
2422 static void bond_miimon_commit(struct bonding *bond)
2424 struct slave *slave;
2427 bond_for_each_slave(bond, slave, i) {
2428 switch (slave->new_link) {
2429 case BOND_LINK_NOCHANGE:
2433 slave->link = BOND_LINK_UP;
2434 slave->jiffies = jiffies;
2436 if (bond->params.mode == BOND_MODE_8023AD) {
2437 /* prevent it from being the active one */
2438 bond_set_backup_slave(slave);
2439 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2440 /* make it immediately active */
2441 bond_set_active_slave(slave);
2442 } else if (slave != bond->primary_slave) {
2443 /* prevent it from being the active one */
2444 bond_set_backup_slave(slave);
2447 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2448 bond->dev->name, slave->dev->name,
2449 slave->speed, slave->duplex ? "full" : "half");
2451 /* notify ad that the link status has changed */
2452 if (bond->params.mode == BOND_MODE_8023AD)
2453 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2455 if (bond_is_lb(bond))
2456 bond_alb_handle_link_change(bond, slave,
2459 if (!bond->curr_active_slave ||
2460 (slave == bond->primary_slave))
2465 case BOND_LINK_DOWN:
2466 if (slave->link_failure_count < UINT_MAX)
2467 slave->link_failure_count++;
2469 slave->link = BOND_LINK_DOWN;
2471 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2472 bond->params.mode == BOND_MODE_8023AD)
2473 bond_set_slave_inactive_flags(slave);
2475 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2476 bond->dev->name, slave->dev->name);
2478 if (bond->params.mode == BOND_MODE_8023AD)
2479 bond_3ad_handle_link_change(slave,
2482 if (bond_is_lb(bond))
2483 bond_alb_handle_link_change(bond, slave,
2486 if (slave == bond->curr_active_slave)
2492 pr_err("%s: invalid new link %d on slave %s\n",
2493 bond->dev->name, slave->new_link,
2495 slave->new_link = BOND_LINK_NOCHANGE;
2503 write_lock_bh(&bond->curr_slave_lock);
2504 bond_select_active_slave(bond);
2505 write_unlock_bh(&bond->curr_slave_lock);
2506 unblock_netpoll_tx();
2509 bond_set_carrier(bond);
2515 * Really a wrapper that splits the mii monitor into two phases: an
2516 * inspection, then (if inspection indicates something needs to be done)
2517 * an acquisition of appropriate locks followed by a commit phase to
2518 * implement whatever link state changes are indicated.
2520 void bond_mii_monitor(struct work_struct *work)
2522 struct bonding *bond = container_of(work, struct bonding,
2524 bool should_notify_peers = false;
2525 unsigned long delay;
2527 read_lock(&bond->lock);
2529 delay = msecs_to_jiffies(bond->params.miimon);
2531 if (bond->slave_cnt == 0)
2534 should_notify_peers = bond_should_notify_peers(bond);
2536 if (bond_miimon_inspect(bond)) {
2537 read_unlock(&bond->lock);
2539 /* Race avoidance with bond_close cancel of workqueue */
2540 if (!rtnl_trylock()) {
2541 read_lock(&bond->lock);
2543 should_notify_peers = false;
2547 read_lock(&bond->lock);
2549 bond_miimon_commit(bond);
2551 read_unlock(&bond->lock);
2552 rtnl_unlock(); /* might sleep, hold no other locks */
2553 read_lock(&bond->lock);
2557 if (bond->params.miimon)
2558 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2560 read_unlock(&bond->lock);
2562 if (should_notify_peers) {
2563 if (!rtnl_trylock()) {
2564 read_lock(&bond->lock);
2565 bond->send_peer_notif++;
2566 read_unlock(&bond->lock);
2569 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2574 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2576 struct vlan_entry *vlan;
2578 if (ip == bond->master_ip)
2581 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2582 if (ip == vlan->vlan_ip)
2590 * We go to the (large) trouble of VLAN tagging ARP frames because
2591 * switches in VLAN mode (especially if ports are configured as
2592 * "native" to a VLAN) might not pass non-tagged frames.
2594 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2596 struct sk_buff *skb;
2598 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2599 slave_dev->name, dest_ip, src_ip, vlan_id);
2601 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2602 NULL, slave_dev->dev_addr, NULL);
2605 pr_err("ARP packet allocation failed\n");
2609 skb = vlan_put_tag(skb, vlan_id);
2611 pr_err("failed to insert VLAN tag\n");
2619 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2622 __be32 *targets = bond->params.arp_targets;
2623 struct vlan_entry *vlan;
2624 struct net_device *vlan_dev;
2627 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2630 pr_debug("basa: target %x\n", targets[i]);
2631 if (!bond_vlan_used(bond)) {
2632 pr_debug("basa: empty vlan: arp_send\n");
2633 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2634 bond->master_ip, 0);
2639 * If VLANs are configured, we do a route lookup to
2640 * determine which VLAN interface would be used, so we
2641 * can tag the ARP with the proper VLAN tag.
2643 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2646 if (net_ratelimit()) {
2647 pr_warning("%s: no route to arp_ip_target %pI4\n",
2648 bond->dev->name, &targets[i]);
2654 * This target is not on a VLAN
2656 if (rt->dst.dev == bond->dev) {
2658 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2659 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2660 bond->master_ip, 0);
2665 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2667 vlan_dev = __vlan_find_dev_deep(bond->dev,
2670 if (vlan_dev == rt->dst.dev) {
2671 vlan_id = vlan->vlan_id;
2672 pr_debug("basa: vlan match on %s %d\n",
2673 vlan_dev->name, vlan_id);
2680 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2681 vlan->vlan_ip, vlan_id);
2685 if (net_ratelimit()) {
2686 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2687 bond->dev->name, &targets[i],
2688 rt->dst.dev ? rt->dst.dev->name : "NULL");
2694 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2697 __be32 *targets = bond->params.arp_targets;
2699 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2700 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2701 &sip, &tip, i, &targets[i],
2702 bond_has_this_ip(bond, tip));
2703 if (sip == targets[i]) {
2704 if (bond_has_this_ip(bond, tip))
2705 slave->last_arp_rx = jiffies;
2711 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2712 struct slave *slave)
2715 unsigned char *arp_ptr;
2718 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2721 read_lock(&bond->lock);
2723 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2724 bond->dev->name, skb->dev->name);
2726 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2730 if (arp->ar_hln != bond->dev->addr_len ||
2731 skb->pkt_type == PACKET_OTHERHOST ||
2732 skb->pkt_type == PACKET_LOOPBACK ||
2733 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2734 arp->ar_pro != htons(ETH_P_IP) ||
2738 arp_ptr = (unsigned char *)(arp + 1);
2739 arp_ptr += bond->dev->addr_len;
2740 memcpy(&sip, arp_ptr, 4);
2741 arp_ptr += 4 + bond->dev->addr_len;
2742 memcpy(&tip, arp_ptr, 4);
2744 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2745 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2746 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2750 * Backup slaves won't see the ARP reply, but do come through
2751 * here for each ARP probe (so we swap the sip/tip to validate
2752 * the probe). In a "redundant switch, common router" type of
2753 * configuration, the ARP probe will (hopefully) travel from
2754 * the active, through one switch, the router, then the other
2755 * switch before reaching the backup.
2757 if (bond_is_active_slave(slave))
2758 bond_validate_arp(bond, slave, sip, tip);
2760 bond_validate_arp(bond, slave, tip, sip);
2763 read_unlock(&bond->lock);
2767 * this function is called regularly to monitor each slave's link
2768 * ensuring that traffic is being sent and received when arp monitoring
2769 * is used in load-balancing mode. if the adapter has been dormant, then an
2770 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2771 * arp monitoring in active backup mode.
2773 void bond_loadbalance_arp_mon(struct work_struct *work)
2775 struct bonding *bond = container_of(work, struct bonding,
2777 struct slave *slave, *oldcurrent;
2778 int do_failover = 0;
2782 read_lock(&bond->lock);
2784 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2786 if (bond->slave_cnt == 0)
2789 read_lock(&bond->curr_slave_lock);
2790 oldcurrent = bond->curr_active_slave;
2791 read_unlock(&bond->curr_slave_lock);
2793 /* see if any of the previous devices are up now (i.e. they have
2794 * xmt and rcv traffic). the curr_active_slave does not come into
2795 * the picture unless it is null. also, slave->jiffies is not needed
2796 * here because we send an arp on each slave and give a slave as
2797 * long as it needs to get the tx/rx within the delta.
2798 * TODO: what about up/down delay in arp mode? it wasn't here before
2801 bond_for_each_slave(bond, slave, i) {
2802 unsigned long trans_start = dev_trans_start(slave->dev);
2804 if (slave->link != BOND_LINK_UP) {
2805 if (time_in_range(jiffies,
2806 trans_start - delta_in_ticks,
2807 trans_start + delta_in_ticks) &&
2808 time_in_range(jiffies,
2809 slave->dev->last_rx - delta_in_ticks,
2810 slave->dev->last_rx + delta_in_ticks)) {
2812 slave->link = BOND_LINK_UP;
2813 bond_set_active_slave(slave);
2815 /* primary_slave has no meaning in round-robin
2816 * mode. the window of a slave being up and
2817 * curr_active_slave being null after enslaving
2821 pr_info("%s: link status definitely up for interface %s, ",
2826 pr_info("%s: interface %s is now up\n",
2832 /* slave->link == BOND_LINK_UP */
2834 /* not all switches will respond to an arp request
2835 * when the source ip is 0, so don't take the link down
2836 * if we don't know our ip yet
2838 if (!time_in_range(jiffies,
2839 trans_start - delta_in_ticks,
2840 trans_start + 2 * delta_in_ticks) ||
2841 !time_in_range(jiffies,
2842 slave->dev->last_rx - delta_in_ticks,
2843 slave->dev->last_rx + 2 * delta_in_ticks)) {
2845 slave->link = BOND_LINK_DOWN;
2846 bond_set_backup_slave(slave);
2848 if (slave->link_failure_count < UINT_MAX)
2849 slave->link_failure_count++;
2851 pr_info("%s: interface %s is now down.\n",
2855 if (slave == oldcurrent)
2860 /* note: if switch is in round-robin mode, all links
2861 * must tx arp to ensure all links rx an arp - otherwise
2862 * links may oscillate or not come up at all; if switch is
2863 * in something like xor mode, there is nothing we can
2864 * do - all replies will be rx'ed on same link causing slaves
2865 * to be unstable during low/no traffic periods
2867 if (IS_UP(slave->dev))
2868 bond_arp_send_all(bond, slave);
2873 write_lock_bh(&bond->curr_slave_lock);
2875 bond_select_active_slave(bond);
2877 write_unlock_bh(&bond->curr_slave_lock);
2878 unblock_netpoll_tx();
2882 if (bond->params.arp_interval)
2883 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2885 read_unlock(&bond->lock);
2889 * Called to inspect slaves for active-backup mode ARP monitor link state
2890 * changes. Sets new_link in slaves to specify what action should take
2891 * place for the slave. Returns 0 if no changes are found, >0 if changes
2892 * to link states must be committed.
2894 * Called with bond->lock held for read.
2896 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2898 struct slave *slave;
2900 unsigned long trans_start;
2902 bond_for_each_slave(bond, slave, i) {
2903 slave->new_link = BOND_LINK_NOCHANGE;
2905 if (slave->link != BOND_LINK_UP) {
2906 if (time_in_range(jiffies,
2907 slave_last_rx(bond, slave) - delta_in_ticks,
2908 slave_last_rx(bond, slave) + delta_in_ticks)) {
2910 slave->new_link = BOND_LINK_UP;
2918 * Give slaves 2*delta after being enslaved or made
2919 * active. This avoids bouncing, as the last receive
2920 * times need a full ARP monitor cycle to be updated.
2922 if (time_in_range(jiffies,
2923 slave->jiffies - delta_in_ticks,
2924 slave->jiffies + 2 * delta_in_ticks))
2928 * Backup slave is down if:
2929 * - No current_arp_slave AND
2930 * - more than 3*delta since last receive AND
2931 * - the bond has an IP address
2933 * Note: a non-null current_arp_slave indicates
2934 * the curr_active_slave went down and we are
2935 * searching for a new one; under this condition
2936 * we only take the curr_active_slave down - this
2937 * gives each slave a chance to tx/rx traffic
2938 * before being taken out
2940 if (!bond_is_active_slave(slave) &&
2941 !bond->current_arp_slave &&
2942 !time_in_range(jiffies,
2943 slave_last_rx(bond, slave) - delta_in_ticks,
2944 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2946 slave->new_link = BOND_LINK_DOWN;
2951 * Active slave is down if:
2952 * - more than 2*delta since transmitting OR
2953 * - (more than 2*delta since receive AND
2954 * the bond has an IP address)
2956 trans_start = dev_trans_start(slave->dev);
2957 if (bond_is_active_slave(slave) &&
2958 (!time_in_range(jiffies,
2959 trans_start - delta_in_ticks,
2960 trans_start + 2 * delta_in_ticks) ||
2961 !time_in_range(jiffies,
2962 slave_last_rx(bond, slave) - delta_in_ticks,
2963 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2965 slave->new_link = BOND_LINK_DOWN;
2974 * Called to commit link state changes noted by inspection step of
2975 * active-backup mode ARP monitor.
2977 * Called with RTNL and bond->lock for read.
2979 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2981 struct slave *slave;
2983 unsigned long trans_start;
2985 bond_for_each_slave(bond, slave, i) {
2986 switch (slave->new_link) {
2987 case BOND_LINK_NOCHANGE:
2991 trans_start = dev_trans_start(slave->dev);
2992 if ((!bond->curr_active_slave &&
2993 time_in_range(jiffies,
2994 trans_start - delta_in_ticks,
2995 trans_start + delta_in_ticks)) ||
2996 bond->curr_active_slave != slave) {
2997 slave->link = BOND_LINK_UP;
2998 if (bond->current_arp_slave) {
2999 bond_set_slave_inactive_flags(
3000 bond->current_arp_slave);
3001 bond->current_arp_slave = NULL;
3004 pr_info("%s: link status definitely up for interface %s.\n",
3005 bond->dev->name, slave->dev->name);
3007 if (!bond->curr_active_slave ||
3008 (slave == bond->primary_slave))
3015 case BOND_LINK_DOWN:
3016 if (slave->link_failure_count < UINT_MAX)
3017 slave->link_failure_count++;
3019 slave->link = BOND_LINK_DOWN;
3020 bond_set_slave_inactive_flags(slave);
3022 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3023 bond->dev->name, slave->dev->name);
3025 if (slave == bond->curr_active_slave) {
3026 bond->current_arp_slave = NULL;
3033 pr_err("%s: impossible: new_link %d on slave %s\n",
3034 bond->dev->name, slave->new_link,
3042 write_lock_bh(&bond->curr_slave_lock);
3043 bond_select_active_slave(bond);
3044 write_unlock_bh(&bond->curr_slave_lock);
3045 unblock_netpoll_tx();
3048 bond_set_carrier(bond);
3052 * Send ARP probes for active-backup mode ARP monitor.
3054 * Called with bond->lock held for read.
3056 static void bond_ab_arp_probe(struct bonding *bond)
3058 struct slave *slave;
3061 read_lock(&bond->curr_slave_lock);
3063 if (bond->current_arp_slave && bond->curr_active_slave)
3064 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3065 bond->current_arp_slave->dev->name,
3066 bond->curr_active_slave->dev->name);
3068 if (bond->curr_active_slave) {
3069 bond_arp_send_all(bond, bond->curr_active_slave);
3070 read_unlock(&bond->curr_slave_lock);
3074 read_unlock(&bond->curr_slave_lock);
3076 /* if we don't have a curr_active_slave, search for the next available
3077 * backup slave from the current_arp_slave and make it the candidate
3078 * for becoming the curr_active_slave
3081 if (!bond->current_arp_slave) {
3082 bond->current_arp_slave = bond->first_slave;
3083 if (!bond->current_arp_slave)
3087 bond_set_slave_inactive_flags(bond->current_arp_slave);
3089 /* search for next candidate */
3090 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3091 if (IS_UP(slave->dev)) {
3092 slave->link = BOND_LINK_BACK;
3093 bond_set_slave_active_flags(slave);
3094 bond_arp_send_all(bond, slave);
3095 slave->jiffies = jiffies;
3096 bond->current_arp_slave = slave;
3100 /* if the link state is up at this point, we
3101 * mark it down - this can happen if we have
3102 * simultaneous link failures and
3103 * reselect_active_interface doesn't make this
3104 * one the current slave so it is still marked
3105 * up when it is actually down
3107 if (slave->link == BOND_LINK_UP) {
3108 slave->link = BOND_LINK_DOWN;
3109 if (slave->link_failure_count < UINT_MAX)
3110 slave->link_failure_count++;
3112 bond_set_slave_inactive_flags(slave);
3114 pr_info("%s: backup interface %s is now down.\n",
3115 bond->dev->name, slave->dev->name);
3120 void bond_activebackup_arp_mon(struct work_struct *work)
3122 struct bonding *bond = container_of(work, struct bonding,
3124 bool should_notify_peers = false;
3127 read_lock(&bond->lock);
3129 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3131 if (bond->slave_cnt == 0)
3134 should_notify_peers = bond_should_notify_peers(bond);
3136 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3137 read_unlock(&bond->lock);
3139 /* Race avoidance with bond_close flush of workqueue */
3140 if (!rtnl_trylock()) {
3141 read_lock(&bond->lock);
3143 should_notify_peers = false;
3147 read_lock(&bond->lock);
3149 bond_ab_arp_commit(bond, delta_in_ticks);
3151 read_unlock(&bond->lock);
3153 read_lock(&bond->lock);
3156 bond_ab_arp_probe(bond);
3159 if (bond->params.arp_interval)
3160 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3162 read_unlock(&bond->lock);
3164 if (should_notify_peers) {
3165 if (!rtnl_trylock()) {
3166 read_lock(&bond->lock);
3167 bond->send_peer_notif++;
3168 read_unlock(&bond->lock);
3171 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3176 /*-------------------------- netdev event handling --------------------------*/
3179 * Change device name
3181 static int bond_event_changename(struct bonding *bond)
3183 bond_remove_proc_entry(bond);
3184 bond_create_proc_entry(bond);
3186 bond_debug_reregister(bond);
3191 static int bond_master_netdev_event(unsigned long event,
3192 struct net_device *bond_dev)
3194 struct bonding *event_bond = netdev_priv(bond_dev);
3197 case NETDEV_CHANGENAME:
3198 return bond_event_changename(event_bond);
3199 case NETDEV_UNREGISTER:
3200 bond_remove_proc_entry(event_bond);
3202 case NETDEV_REGISTER:
3203 bond_create_proc_entry(event_bond);
3212 static int bond_slave_netdev_event(unsigned long event,
3213 struct net_device *slave_dev)
3215 struct net_device *bond_dev = slave_dev->master;
3216 struct bonding *bond = netdev_priv(bond_dev);
3217 struct slave *slave = NULL;
3220 case NETDEV_UNREGISTER:
3222 if (bond->setup_by_slave)
3223 bond_release_and_destroy(bond_dev, slave_dev);
3225 bond_release(bond_dev, slave_dev);
3230 slave = bond_get_slave_by_dev(bond, slave_dev);
3232 u32 old_speed = slave->speed;
3233 u8 old_duplex = slave->duplex;
3235 bond_update_speed_duplex(slave);
3237 if (bond->params.mode == BOND_MODE_8023AD) {
3238 if (old_speed != slave->speed)
3239 bond_3ad_adapter_speed_changed(slave);
3240 if (old_duplex != slave->duplex)
3241 bond_3ad_adapter_duplex_changed(slave);
3248 * ... Or is it this?
3251 case NETDEV_CHANGEMTU:
3253 * TODO: Should slaves be allowed to
3254 * independently alter their MTU? For
3255 * an active-backup bond, slaves need
3256 * not be the same type of device, so
3257 * MTUs may vary. For other modes,
3258 * slaves arguably should have the
3259 * same MTUs. To do this, we'd need to
3260 * take over the slave's change_mtu
3261 * function for the duration of their
3265 case NETDEV_CHANGENAME:
3267 * TODO: handle changing the primary's name
3270 case NETDEV_FEAT_CHANGE:
3271 bond_compute_features(bond);
3281 * bond_netdev_event: handle netdev notifier chain events.
3283 * This function receives events for the netdev chain. The caller (an
3284 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3285 * locks for us to safely manipulate the slave devices (RTNL lock,
3288 static int bond_netdev_event(struct notifier_block *this,
3289 unsigned long event, void *ptr)
3291 struct net_device *event_dev = (struct net_device *)ptr;
3293 pr_debug("event_dev: %s, event: %lx\n",
3294 event_dev ? event_dev->name : "None",
3297 if (!(event_dev->priv_flags & IFF_BONDING))
3300 if (event_dev->flags & IFF_MASTER) {
3301 pr_debug("IFF_MASTER\n");
3302 return bond_master_netdev_event(event, event_dev);
3305 if (event_dev->flags & IFF_SLAVE) {
3306 pr_debug("IFF_SLAVE\n");
3307 return bond_slave_netdev_event(event, event_dev);
3314 * bond_inetaddr_event: handle inetaddr notifier chain events.
3316 * We keep track of device IPs primarily to use as source addresses in
3317 * ARP monitor probes (rather than spewing out broadcasts all the time).
3319 * We track one IP for the main device (if it has one), plus one per VLAN.
3321 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3323 struct in_ifaddr *ifa = ptr;
3324 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3325 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3326 struct bonding *bond;
3327 struct vlan_entry *vlan;
3329 /* we only care about primary address */
3330 if(ifa->ifa_flags & IFA_F_SECONDARY)
3333 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3334 if (bond->dev == event_dev) {
3337 bond->master_ip = ifa->ifa_local;
3340 bond->master_ip = 0;
3347 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3348 vlan_dev = __vlan_find_dev_deep(bond->dev,
3350 if (vlan_dev == event_dev) {
3353 vlan->vlan_ip = ifa->ifa_local;
3367 static struct notifier_block bond_netdev_notifier = {
3368 .notifier_call = bond_netdev_event,
3371 static struct notifier_block bond_inetaddr_notifier = {
3372 .notifier_call = bond_inetaddr_event,
3375 /*---------------------------- Hashing Policies -----------------------------*/
3378 * Hash for the output device based upon layer 2 and layer 3 data. If
3379 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3381 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3383 struct ethhdr *data = (struct ethhdr *)skb->data;
3384 struct iphdr *iph = ip_hdr(skb);
3386 if (skb->protocol == htons(ETH_P_IP)) {
3387 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3388 (data->h_dest[5] ^ data->h_source[5])) % count;
3391 return (data->h_dest[5] ^ data->h_source[5]) % count;
3395 * Hash for the output device based upon layer 3 and layer 4 data. If
3396 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3397 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3399 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3401 struct ethhdr *data = (struct ethhdr *)skb->data;
3402 struct iphdr *iph = ip_hdr(skb);
3403 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3406 if (skb->protocol == htons(ETH_P_IP)) {
3407 if (!ip_is_fragment(iph) &&
3408 (iph->protocol == IPPROTO_TCP ||
3409 iph->protocol == IPPROTO_UDP)) {
3410 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3412 return (layer4_xor ^
3413 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3417 return (data->h_dest[5] ^ data->h_source[5]) % count;
3421 * Hash for the output device based upon layer 2 data
3423 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3425 struct ethhdr *data = (struct ethhdr *)skb->data;
3427 return (data->h_dest[5] ^ data->h_source[5]) % count;
3430 /*-------------------------- Device entry points ----------------------------*/
3432 static void bond_work_init_all(struct bonding *bond)
3434 INIT_DELAYED_WORK(&bond->mcast_work,
3435 bond_resend_igmp_join_requests_delayed);
3436 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3437 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3438 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3439 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3441 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3442 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3445 static void bond_work_cancel_all(struct bonding *bond)
3447 cancel_delayed_work_sync(&bond->mii_work);
3448 cancel_delayed_work_sync(&bond->arp_work);
3449 cancel_delayed_work_sync(&bond->alb_work);
3450 cancel_delayed_work_sync(&bond->ad_work);
3451 cancel_delayed_work_sync(&bond->mcast_work);
3454 static int bond_open(struct net_device *bond_dev)
3456 struct bonding *bond = netdev_priv(bond_dev);
3457 struct slave *slave;
3460 /* reset slave->backup and slave->inactive */
3461 read_lock(&bond->lock);
3462 if (bond->slave_cnt > 0) {
3463 read_lock(&bond->curr_slave_lock);
3464 bond_for_each_slave(bond, slave, i) {
3465 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3466 && (slave != bond->curr_active_slave)) {
3467 bond_set_slave_inactive_flags(slave);
3469 bond_set_slave_active_flags(slave);
3472 read_unlock(&bond->curr_slave_lock);
3474 read_unlock(&bond->lock);
3476 bond_work_init_all(bond);
3478 if (bond_is_lb(bond)) {
3479 /* bond_alb_initialize must be called before the timer
3482 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3484 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3487 if (bond->params.miimon) /* link check interval, in milliseconds. */
3488 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3490 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3491 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3492 if (bond->params.arp_validate)
3493 bond->recv_probe = bond_arp_rcv;
3496 if (bond->params.mode == BOND_MODE_8023AD) {
3497 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3498 /* register to receive LACPDUs */
3499 bond->recv_probe = bond_3ad_lacpdu_recv;
3500 bond_3ad_initiate_agg_selection(bond, 1);
3506 static int bond_close(struct net_device *bond_dev)
3508 struct bonding *bond = netdev_priv(bond_dev);
3510 write_lock_bh(&bond->lock);
3511 bond->send_peer_notif = 0;
3512 write_unlock_bh(&bond->lock);
3514 bond_work_cancel_all(bond);
3515 if (bond_is_lb(bond)) {
3516 /* Must be called only after all
3517 * slaves have been released
3519 bond_alb_deinitialize(bond);
3521 bond->recv_probe = NULL;
3526 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3527 struct rtnl_link_stats64 *stats)
3529 struct bonding *bond = netdev_priv(bond_dev);
3530 struct rtnl_link_stats64 temp;
3531 struct slave *slave;
3534 memset(stats, 0, sizeof(*stats));
3536 read_lock_bh(&bond->lock);
3538 bond_for_each_slave(bond, slave, i) {
3539 const struct rtnl_link_stats64 *sstats =
3540 dev_get_stats(slave->dev, &temp);
3542 stats->rx_packets += sstats->rx_packets;
3543 stats->rx_bytes += sstats->rx_bytes;
3544 stats->rx_errors += sstats->rx_errors;
3545 stats->rx_dropped += sstats->rx_dropped;
3547 stats->tx_packets += sstats->tx_packets;
3548 stats->tx_bytes += sstats->tx_bytes;
3549 stats->tx_errors += sstats->tx_errors;
3550 stats->tx_dropped += sstats->tx_dropped;
3552 stats->multicast += sstats->multicast;
3553 stats->collisions += sstats->collisions;
3555 stats->rx_length_errors += sstats->rx_length_errors;
3556 stats->rx_over_errors += sstats->rx_over_errors;
3557 stats->rx_crc_errors += sstats->rx_crc_errors;
3558 stats->rx_frame_errors += sstats->rx_frame_errors;
3559 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3560 stats->rx_missed_errors += sstats->rx_missed_errors;
3562 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3563 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3564 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3565 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3566 stats->tx_window_errors += sstats->tx_window_errors;
3569 read_unlock_bh(&bond->lock);
3574 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3576 struct net_device *slave_dev = NULL;
3577 struct ifbond k_binfo;
3578 struct ifbond __user *u_binfo = NULL;
3579 struct ifslave k_sinfo;
3580 struct ifslave __user *u_sinfo = NULL;
3581 struct mii_ioctl_data *mii = NULL;
3584 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3596 * We do this again just in case we were called by SIOCGMIIREG
3597 * instead of SIOCGMIIPHY.
3604 if (mii->reg_num == 1) {
3605 struct bonding *bond = netdev_priv(bond_dev);
3607 read_lock(&bond->lock);
3608 read_lock(&bond->curr_slave_lock);
3609 if (netif_carrier_ok(bond->dev))
3610 mii->val_out = BMSR_LSTATUS;
3612 read_unlock(&bond->curr_slave_lock);
3613 read_unlock(&bond->lock);
3617 case BOND_INFO_QUERY_OLD:
3618 case SIOCBONDINFOQUERY:
3619 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3621 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3624 res = bond_info_query(bond_dev, &k_binfo);
3626 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3630 case BOND_SLAVE_INFO_QUERY_OLD:
3631 case SIOCBONDSLAVEINFOQUERY:
3632 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3634 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3637 res = bond_slave_info_query(bond_dev, &k_sinfo);
3639 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3648 if (!capable(CAP_NET_ADMIN))
3651 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3653 pr_debug("slave_dev=%p:\n", slave_dev);
3658 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3660 case BOND_ENSLAVE_OLD:
3661 case SIOCBONDENSLAVE:
3662 res = bond_enslave(bond_dev, slave_dev);
3664 case BOND_RELEASE_OLD:
3665 case SIOCBONDRELEASE:
3666 res = bond_release(bond_dev, slave_dev);
3668 case BOND_SETHWADDR_OLD:
3669 case SIOCBONDSETHWADDR:
3670 res = bond_sethwaddr(bond_dev, slave_dev);
3672 case BOND_CHANGE_ACTIVE_OLD:
3673 case SIOCBONDCHANGEACTIVE:
3674 res = bond_ioctl_change_active(bond_dev, slave_dev);
3686 static bool bond_addr_in_mc_list(unsigned char *addr,
3687 struct netdev_hw_addr_list *list,
3690 struct netdev_hw_addr *ha;
3692 netdev_hw_addr_list_for_each(ha, list)
3693 if (!memcmp(ha->addr, addr, addrlen))
3699 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3701 struct bonding *bond = netdev_priv(bond_dev);
3703 if (change & IFF_PROMISC)
3704 bond_set_promiscuity(bond,
3705 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3707 if (change & IFF_ALLMULTI)
3708 bond_set_allmulti(bond,
3709 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3712 static void bond_set_multicast_list(struct net_device *bond_dev)
3714 struct bonding *bond = netdev_priv(bond_dev);
3715 struct netdev_hw_addr *ha;
3718 read_lock(&bond->lock);
3720 /* looking for addresses to add to slaves' mc list */
3721 netdev_for_each_mc_addr(ha, bond_dev) {
3722 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3723 bond_dev->addr_len);
3725 bond_mc_add(bond, ha->addr);
3728 /* looking for addresses to delete from slaves' list */
3729 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3730 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3731 bond_dev->addr_len);
3733 bond_mc_del(bond, ha->addr);
3736 /* save master's multicast list */
3737 __hw_addr_flush(&bond->mc_list);
3738 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3739 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3741 read_unlock(&bond->lock);
3744 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3746 struct bonding *bond = netdev_priv(dev);
3747 struct slave *slave = bond->first_slave;
3750 const struct net_device_ops *slave_ops
3751 = slave->dev->netdev_ops;
3752 if (slave_ops->ndo_neigh_setup)
3753 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3759 * Change the MTU of all of a master's slaves to match the master
3761 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3763 struct bonding *bond = netdev_priv(bond_dev);
3764 struct slave *slave, *stop_at;
3768 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3769 (bond_dev ? bond_dev->name : "None"), new_mtu);
3771 /* Can't hold bond->lock with bh disabled here since
3772 * some base drivers panic. On the other hand we can't
3773 * hold bond->lock without bh disabled because we'll
3774 * deadlock. The only solution is to rely on the fact
3775 * that we're under rtnl_lock here, and the slaves
3776 * list won't change. This doesn't solve the problem
3777 * of setting the slave's MTU while it is
3778 * transmitting, but the assumption is that the base
3779 * driver can handle that.
3781 * TODO: figure out a way to safely iterate the slaves
3782 * list, but without holding a lock around the actual
3783 * call to the base driver.
3786 bond_for_each_slave(bond, slave, i) {
3787 pr_debug("s %p s->p %p c_m %p\n",
3790 slave->dev->netdev_ops->ndo_change_mtu);
3792 res = dev_set_mtu(slave->dev, new_mtu);
3795 /* If we failed to set the slave's mtu to the new value
3796 * we must abort the operation even in ACTIVE_BACKUP
3797 * mode, because if we allow the backup slaves to have
3798 * different mtu values than the active slave we'll
3799 * need to change their mtu when doing a failover. That
3800 * means changing their mtu from timer context, which
3801 * is probably not a good idea.
3803 pr_debug("err %d %s\n", res, slave->dev->name);
3808 bond_dev->mtu = new_mtu;
3813 /* unwind from head to the slave that failed */
3815 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3818 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3820 pr_debug("unwind err %d dev %s\n",
3821 tmp_res, slave->dev->name);
3831 * Note that many devices must be down to change the HW address, and
3832 * downing the master releases all slaves. We can make bonds full of
3833 * bonding devices to test this, however.
3835 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3837 struct bonding *bond = netdev_priv(bond_dev);
3838 struct sockaddr *sa = addr, tmp_sa;
3839 struct slave *slave, *stop_at;
3843 if (bond->params.mode == BOND_MODE_ALB)
3844 return bond_alb_set_mac_address(bond_dev, addr);
3847 pr_debug("bond=%p, name=%s\n",
3848 bond, bond_dev ? bond_dev->name : "None");
3851 * If fail_over_mac is set to active, do nothing and return
3852 * success. Returning an error causes ifenslave to fail.
3854 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3857 if (!is_valid_ether_addr(sa->sa_data))
3858 return -EADDRNOTAVAIL;
3860 /* Can't hold bond->lock with bh disabled here since
3861 * some base drivers panic. On the other hand we can't
3862 * hold bond->lock without bh disabled because we'll
3863 * deadlock. The only solution is to rely on the fact
3864 * that we're under rtnl_lock here, and the slaves
3865 * list won't change. This doesn't solve the problem
3866 * of setting the slave's hw address while it is
3867 * transmitting, but the assumption is that the base
3868 * driver can handle that.
3870 * TODO: figure out a way to safely iterate the slaves
3871 * list, but without holding a lock around the actual
3872 * call to the base driver.
3875 bond_for_each_slave(bond, slave, i) {
3876 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3877 pr_debug("slave %p %s\n", slave, slave->dev->name);
3879 if (slave_ops->ndo_set_mac_address == NULL) {
3881 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3885 res = dev_set_mac_address(slave->dev, addr);
3887 /* TODO: consider downing the slave
3889 * User should expect communications
3890 * breakage anyway until ARP finish
3893 pr_debug("err %d %s\n", res, slave->dev->name);
3899 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3903 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3904 tmp_sa.sa_family = bond_dev->type;
3906 /* unwind from head to the slave that failed */
3908 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3911 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3913 pr_debug("unwind err %d dev %s\n",
3914 tmp_res, slave->dev->name);
3921 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3923 struct bonding *bond = netdev_priv(bond_dev);
3924 struct slave *slave, *start_at;
3925 int i, slave_no, res = 1;
3926 struct iphdr *iph = ip_hdr(skb);
3929 * Start with the curr_active_slave that joined the bond as the
3930 * default for sending IGMP traffic. For failover purposes one
3931 * needs to maintain some consistency for the interface that will
3932 * send the join/membership reports. The curr_active_slave found
3933 * will send all of this type of traffic.
3935 if ((iph->protocol == IPPROTO_IGMP) &&
3936 (skb->protocol == htons(ETH_P_IP))) {
3938 read_lock(&bond->curr_slave_lock);
3939 slave = bond->curr_active_slave;
3940 read_unlock(&bond->curr_slave_lock);
3946 * Concurrent TX may collide on rr_tx_counter; we accept
3947 * that as being rare enough not to justify using an
3950 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3952 bond_for_each_slave(bond, slave, i) {
3960 bond_for_each_slave_from(bond, slave, i, start_at) {
3961 if (IS_UP(slave->dev) &&
3962 (slave->link == BOND_LINK_UP) &&
3963 bond_is_active_slave(slave)) {
3964 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3971 /* no suitable interface, frame not sent */
3975 return NETDEV_TX_OK;
3980 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3981 * the bond has a usable interface.
3983 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3985 struct bonding *bond = netdev_priv(bond_dev);
3988 read_lock(&bond->curr_slave_lock);
3990 if (bond->curr_active_slave)
3991 res = bond_dev_queue_xmit(bond, skb,
3992 bond->curr_active_slave->dev);
3995 /* no suitable interface, frame not sent */
3998 read_unlock(&bond->curr_slave_lock);
4000 return NETDEV_TX_OK;
4004 * In bond_xmit_xor() , we determine the output device by using a pre-
4005 * determined xmit_hash_policy(), If the selected device is not enabled,
4006 * find the next active slave.
4008 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4010 struct bonding *bond = netdev_priv(bond_dev);
4011 struct slave *slave, *start_at;
4016 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4018 bond_for_each_slave(bond, slave, i) {
4026 bond_for_each_slave_from(bond, slave, i, start_at) {
4027 if (IS_UP(slave->dev) &&
4028 (slave->link == BOND_LINK_UP) &&
4029 bond_is_active_slave(slave)) {
4030 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4036 /* no suitable interface, frame not sent */
4040 return NETDEV_TX_OK;
4044 * in broadcast mode, we send everything to all usable interfaces.
4046 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4048 struct bonding *bond = netdev_priv(bond_dev);
4049 struct slave *slave, *start_at;
4050 struct net_device *tx_dev = NULL;
4054 read_lock(&bond->curr_slave_lock);
4055 start_at = bond->curr_active_slave;
4056 read_unlock(&bond->curr_slave_lock);
4061 bond_for_each_slave_from(bond, slave, i, start_at) {
4062 if (IS_UP(slave->dev) &&
4063 (slave->link == BOND_LINK_UP) &&
4064 bond_is_active_slave(slave)) {
4066 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4068 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4073 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4075 dev_kfree_skb(skb2);
4079 tx_dev = slave->dev;
4084 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4088 /* no suitable interface, frame not sent */
4091 /* frame sent to all suitable interfaces */
4092 return NETDEV_TX_OK;
4095 /*------------------------- Device initialization ---------------------------*/
4097 static void bond_set_xmit_hash_policy(struct bonding *bond)
4099 switch (bond->params.xmit_policy) {
4100 case BOND_XMIT_POLICY_LAYER23:
4101 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4103 case BOND_XMIT_POLICY_LAYER34:
4104 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4106 case BOND_XMIT_POLICY_LAYER2:
4108 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4114 * Lookup the slave that corresponds to a qid
4116 static inline int bond_slave_override(struct bonding *bond,
4117 struct sk_buff *skb)
4120 struct slave *slave = NULL;
4121 struct slave *check_slave;
4123 if (!skb->queue_mapping)
4126 /* Find out if any slaves have the same mapping as this skb. */
4127 bond_for_each_slave(bond, check_slave, i) {
4128 if (check_slave->queue_id == skb->queue_mapping) {
4129 slave = check_slave;
4134 /* If the slave isn't UP, use default transmit policy. */
4135 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4136 (slave->link == BOND_LINK_UP)) {
4137 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4144 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4147 * This helper function exists to help dev_pick_tx get the correct
4148 * destination queue. Using a helper function skips a call to
4149 * skb_tx_hash and will put the skbs in the queue we expect on their
4150 * way down to the bonding driver.
4152 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4155 * Save the original txq to restore before passing to the driver
4157 qdisc_skb_cb(skb)->bond_queue_mapping = skb->queue_mapping;
4159 if (unlikely(txq >= dev->real_num_tx_queues)) {
4161 txq -= dev->real_num_tx_queues;
4162 } while (txq >= dev->real_num_tx_queues);
4167 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4169 struct bonding *bond = netdev_priv(dev);
4171 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4172 if (!bond_slave_override(bond, skb))
4173 return NETDEV_TX_OK;
4176 switch (bond->params.mode) {
4177 case BOND_MODE_ROUNDROBIN:
4178 return bond_xmit_roundrobin(skb, dev);
4179 case BOND_MODE_ACTIVEBACKUP:
4180 return bond_xmit_activebackup(skb, dev);
4182 return bond_xmit_xor(skb, dev);
4183 case BOND_MODE_BROADCAST:
4184 return bond_xmit_broadcast(skb, dev);
4185 case BOND_MODE_8023AD:
4186 return bond_3ad_xmit_xor(skb, dev);
4189 return bond_alb_xmit(skb, dev);
4191 /* Should never happen, mode already checked */
4192 pr_err("%s: Error: Unknown bonding mode %d\n",
4193 dev->name, bond->params.mode);
4196 return NETDEV_TX_OK;
4200 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4202 struct bonding *bond = netdev_priv(dev);
4203 netdev_tx_t ret = NETDEV_TX_OK;
4206 * If we risk deadlock from transmitting this in the
4207 * netpoll path, tell netpoll to queue the frame for later tx
4209 if (is_netpoll_tx_blocked(dev))
4210 return NETDEV_TX_BUSY;
4212 read_lock(&bond->lock);
4214 if (bond->slave_cnt)
4215 ret = __bond_start_xmit(skb, dev);
4219 read_unlock(&bond->lock);
4225 * set bond mode specific net device operations
4227 void bond_set_mode_ops(struct bonding *bond, int mode)
4229 struct net_device *bond_dev = bond->dev;
4232 case BOND_MODE_ROUNDROBIN:
4234 case BOND_MODE_ACTIVEBACKUP:
4237 bond_set_xmit_hash_policy(bond);
4239 case BOND_MODE_BROADCAST:
4241 case BOND_MODE_8023AD:
4242 bond_set_xmit_hash_policy(bond);
4249 /* Should never happen, mode already checked */
4250 pr_err("%s: Error: Unknown bonding mode %d\n",
4251 bond_dev->name, mode);
4256 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4257 struct ethtool_drvinfo *drvinfo)
4259 strncpy(drvinfo->driver, DRV_NAME, 32);
4260 strncpy(drvinfo->version, DRV_VERSION, 32);
4261 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4264 static const struct ethtool_ops bond_ethtool_ops = {
4265 .get_drvinfo = bond_ethtool_get_drvinfo,
4266 .get_link = ethtool_op_get_link,
4269 static const struct net_device_ops bond_netdev_ops = {
4270 .ndo_init = bond_init,
4271 .ndo_uninit = bond_uninit,
4272 .ndo_open = bond_open,
4273 .ndo_stop = bond_close,
4274 .ndo_start_xmit = bond_start_xmit,
4275 .ndo_select_queue = bond_select_queue,
4276 .ndo_get_stats64 = bond_get_stats,
4277 .ndo_do_ioctl = bond_do_ioctl,
4278 .ndo_change_rx_flags = bond_change_rx_flags,
4279 .ndo_set_rx_mode = bond_set_multicast_list,
4280 .ndo_change_mtu = bond_change_mtu,
4281 .ndo_set_mac_address = bond_set_mac_address,
4282 .ndo_neigh_setup = bond_neigh_setup,
4283 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4284 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4285 #ifdef CONFIG_NET_POLL_CONTROLLER
4286 .ndo_netpoll_setup = bond_netpoll_setup,
4287 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4288 .ndo_poll_controller = bond_poll_controller,
4290 .ndo_add_slave = bond_enslave,
4291 .ndo_del_slave = bond_release,
4292 .ndo_fix_features = bond_fix_features,
4295 static void bond_destructor(struct net_device *bond_dev)
4297 struct bonding *bond = netdev_priv(bond_dev);
4299 destroy_workqueue(bond->wq);
4300 free_netdev(bond_dev);
4303 static void bond_setup(struct net_device *bond_dev)
4305 struct bonding *bond = netdev_priv(bond_dev);
4307 /* initialize rwlocks */
4308 rwlock_init(&bond->lock);
4309 rwlock_init(&bond->curr_slave_lock);
4311 bond->params = bonding_defaults;
4313 /* Initialize pointers */
4314 bond->dev = bond_dev;
4315 INIT_LIST_HEAD(&bond->vlan_list);
4317 /* Initialize the device entry points */
4318 ether_setup(bond_dev);
4319 bond_dev->netdev_ops = &bond_netdev_ops;
4320 bond_dev->ethtool_ops = &bond_ethtool_ops;
4321 bond_set_mode_ops(bond, bond->params.mode);
4323 bond_dev->destructor = bond_destructor;
4325 /* Initialize the device options */
4326 bond_dev->tx_queue_len = 0;
4327 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4328 bond_dev->priv_flags |= IFF_BONDING;
4329 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4331 /* At first, we block adding VLANs. That's the only way to
4332 * prevent problems that occur when adding VLANs over an
4333 * empty bond. The block will be removed once non-challenged
4334 * slaves are enslaved.
4336 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4338 /* don't acquire bond device's netif_tx_lock when
4340 bond_dev->features |= NETIF_F_LLTX;
4342 /* By default, we declare the bond to be fully
4343 * VLAN hardware accelerated capable. Special
4344 * care is taken in the various xmit functions
4345 * when there are slaves that are not hw accel
4349 bond_dev->hw_features = BOND_VLAN_FEATURES |
4350 NETIF_F_HW_VLAN_TX |
4351 NETIF_F_HW_VLAN_RX |
4352 NETIF_F_HW_VLAN_FILTER;
4354 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4355 bond_dev->features |= bond_dev->hw_features;
4359 * Destroy a bonding device.
4360 * Must be under rtnl_lock when this function is called.
4362 static void bond_uninit(struct net_device *bond_dev)
4364 struct bonding *bond = netdev_priv(bond_dev);
4365 struct vlan_entry *vlan, *tmp;
4367 bond_netpoll_cleanup(bond_dev);
4369 /* Release the bonded slaves */
4370 bond_release_all(bond_dev);
4372 list_del(&bond->bond_list);
4374 bond_work_cancel_all(bond);
4376 bond_debug_unregister(bond);
4378 __hw_addr_flush(&bond->mc_list);
4380 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4381 list_del(&vlan->vlan_list);
4386 /*------------------------- Module initialization ---------------------------*/
4389 * Convert string input module parms. Accept either the
4390 * number of the mode or its string name. A bit complicated because
4391 * some mode names are substrings of other names, and calls from sysfs
4392 * may have whitespace in the name (trailing newlines, for example).
4394 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4396 int modeint = -1, i, rv;
4397 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4399 for (p = (char *)buf; *p; p++)
4400 if (!(isdigit(*p) || isspace(*p)))
4404 rv = sscanf(buf, "%20s", modestr);
4406 rv = sscanf(buf, "%d", &modeint);
4411 for (i = 0; tbl[i].modename; i++) {
4412 if (modeint == tbl[i].mode)
4414 if (strcmp(modestr, tbl[i].modename) == 0)
4421 static int bond_check_params(struct bond_params *params)
4423 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4426 * Convert string parameters.
4429 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4430 if (bond_mode == -1) {
4431 pr_err("Error: Invalid bonding mode \"%s\"\n",
4432 mode == NULL ? "NULL" : mode);
4437 if (xmit_hash_policy) {
4438 if ((bond_mode != BOND_MODE_XOR) &&
4439 (bond_mode != BOND_MODE_8023AD)) {
4440 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4441 bond_mode_name(bond_mode));
4443 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4445 if (xmit_hashtype == -1) {
4446 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4447 xmit_hash_policy == NULL ? "NULL" :
4455 if (bond_mode != BOND_MODE_8023AD) {
4456 pr_info("lacp_rate param is irrelevant in mode %s\n",
4457 bond_mode_name(bond_mode));
4459 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4460 if (lacp_fast == -1) {
4461 pr_err("Error: Invalid lacp rate \"%s\"\n",
4462 lacp_rate == NULL ? "NULL" : lacp_rate);
4469 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4470 if (params->ad_select == -1) {
4471 pr_err("Error: Invalid ad_select \"%s\"\n",
4472 ad_select == NULL ? "NULL" : ad_select);
4476 if (bond_mode != BOND_MODE_8023AD) {
4477 pr_warning("ad_select param only affects 802.3ad mode\n");
4480 params->ad_select = BOND_AD_STABLE;
4483 if (max_bonds < 0) {
4484 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4485 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4486 max_bonds = BOND_DEFAULT_MAX_BONDS;
4490 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4491 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4492 miimon = BOND_LINK_MON_INTERV;
4496 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4501 if (downdelay < 0) {
4502 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4503 downdelay, INT_MAX);
4507 if ((use_carrier != 0) && (use_carrier != 1)) {
4508 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4513 if (num_peer_notif < 0 || num_peer_notif > 255) {
4514 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4519 /* reset values for 802.3ad */
4520 if (bond_mode == BOND_MODE_8023AD) {
4522 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");
4523 pr_warning("Forcing miimon to 100msec\n");
4528 if (tx_queues < 1 || tx_queues > 255) {
4529 pr_warning("Warning: tx_queues (%d) should be between "
4530 "1 and 255, resetting to %d\n",
4531 tx_queues, BOND_DEFAULT_TX_QUEUES);
4532 tx_queues = BOND_DEFAULT_TX_QUEUES;
4535 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4536 pr_warning("Warning: all_slaves_active module parameter (%d), "
4537 "not of valid value (0/1), so it was set to "
4538 "0\n", all_slaves_active);
4539 all_slaves_active = 0;
4542 if (resend_igmp < 0 || resend_igmp > 255) {
4543 pr_warning("Warning: resend_igmp (%d) should be between "
4544 "0 and 255, resetting to %d\n",
4545 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4546 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4549 /* reset values for TLB/ALB */
4550 if ((bond_mode == BOND_MODE_TLB) ||
4551 (bond_mode == BOND_MODE_ALB)) {
4553 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");
4554 pr_warning("Forcing miimon to 100msec\n");
4559 if (bond_mode == BOND_MODE_ALB) {
4560 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",
4565 if (updelay || downdelay) {
4566 /* just warn the user the up/down delay will have
4567 * no effect since miimon is zero...
4569 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",
4570 updelay, downdelay);
4573 /* don't allow arp monitoring */
4575 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4576 miimon, arp_interval);
4580 if ((updelay % miimon) != 0) {
4581 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4583 (updelay / miimon) * miimon);
4588 if ((downdelay % miimon) != 0) {
4589 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4591 (downdelay / miimon) * miimon);
4594 downdelay /= miimon;
4597 if (arp_interval < 0) {
4598 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4599 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4600 arp_interval = BOND_LINK_ARP_INTERV;
4603 for (arp_ip_count = 0;
4604 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4606 /* not complete check, but should be good enough to
4608 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4609 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4610 arp_ip_target[arp_ip_count]);
4613 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4614 arp_target[arp_ip_count] = ip;
4618 if (arp_interval && !arp_ip_count) {
4619 /* don't allow arping if no arp_ip_target given... */
4620 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4626 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4627 pr_err("arp_validate only supported in active-backup mode\n");
4630 if (!arp_interval) {
4631 pr_err("arp_validate requires arp_interval\n");
4635 arp_validate_value = bond_parse_parm(arp_validate,
4637 if (arp_validate_value == -1) {
4638 pr_err("Error: invalid arp_validate \"%s\"\n",
4639 arp_validate == NULL ? "NULL" : arp_validate);
4643 arp_validate_value = 0;
4646 pr_info("MII link monitoring set to %d ms\n", miimon);
4647 } else if (arp_interval) {
4650 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4652 arp_validate_tbl[arp_validate_value].modename,
4655 for (i = 0; i < arp_ip_count; i++)
4656 pr_info(" %s", arp_ip_target[i]);
4660 } else if (max_bonds) {
4661 /* miimon and arp_interval not set, we need one so things
4662 * work as expected, see bonding.txt for details
4664 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");
4667 if (primary && !USES_PRIMARY(bond_mode)) {
4668 /* currently, using a primary only makes sense
4669 * in active backup, TLB or ALB modes
4671 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4672 primary, bond_mode_name(bond_mode));
4676 if (primary && primary_reselect) {
4677 primary_reselect_value = bond_parse_parm(primary_reselect,
4679 if (primary_reselect_value == -1) {
4680 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4682 NULL ? "NULL" : primary_reselect);
4686 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4689 if (fail_over_mac) {
4690 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4692 if (fail_over_mac_value == -1) {
4693 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4694 arp_validate == NULL ? "NULL" : arp_validate);
4698 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4699 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4701 fail_over_mac_value = BOND_FOM_NONE;
4704 /* fill params struct with the proper values */
4705 params->mode = bond_mode;
4706 params->xmit_policy = xmit_hashtype;
4707 params->miimon = miimon;
4708 params->num_peer_notif = num_peer_notif;
4709 params->arp_interval = arp_interval;
4710 params->arp_validate = arp_validate_value;
4711 params->updelay = updelay;
4712 params->downdelay = downdelay;
4713 params->use_carrier = use_carrier;
4714 params->lacp_fast = lacp_fast;
4715 params->primary[0] = 0;
4716 params->primary_reselect = primary_reselect_value;
4717 params->fail_over_mac = fail_over_mac_value;
4718 params->tx_queues = tx_queues;
4719 params->all_slaves_active = all_slaves_active;
4720 params->resend_igmp = resend_igmp;
4721 params->min_links = min_links;
4724 strncpy(params->primary, primary, IFNAMSIZ);
4725 params->primary[IFNAMSIZ - 1] = 0;
4728 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4733 static struct lock_class_key bonding_netdev_xmit_lock_key;
4734 static struct lock_class_key bonding_netdev_addr_lock_key;
4736 static void bond_set_lockdep_class_one(struct net_device *dev,
4737 struct netdev_queue *txq,
4740 lockdep_set_class(&txq->_xmit_lock,
4741 &bonding_netdev_xmit_lock_key);
4744 static void bond_set_lockdep_class(struct net_device *dev)
4746 lockdep_set_class(&dev->addr_list_lock,
4747 &bonding_netdev_addr_lock_key);
4748 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4752 * Called from registration process
4754 static int bond_init(struct net_device *bond_dev)
4756 struct bonding *bond = netdev_priv(bond_dev);
4757 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4758 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4760 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4763 * Initialize locks that may be required during
4764 * en/deslave operations. All of the bond_open work
4765 * (of which this is part) should really be moved to
4766 * a phase prior to dev_open
4768 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4769 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4771 bond->wq = create_singlethread_workqueue(bond_dev->name);
4775 bond_set_lockdep_class(bond_dev);
4777 list_add_tail(&bond->bond_list, &bn->dev_list);
4779 bond_prepare_sysfs_group(bond);
4781 bond_debug_register(bond);
4783 __hw_addr_init(&bond->mc_list);
4787 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4789 if (tb[IFLA_ADDRESS]) {
4790 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4792 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4793 return -EADDRNOTAVAIL;
4798 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4799 unsigned int *num_queues,
4800 unsigned int *real_num_queues)
4802 *num_queues = tx_queues;
4806 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4808 .priv_size = sizeof(struct bonding),
4809 .setup = bond_setup,
4810 .validate = bond_validate,
4811 .get_tx_queues = bond_get_tx_queues,
4814 /* Create a new bond based on the specified name and bonding parameters.
4815 * If name is NULL, obtain a suitable "bond%d" name for us.
4816 * Caller must NOT hold rtnl_lock; we need to release it here before we
4817 * set up our sysfs entries.
4819 int bond_create(struct net *net, const char *name)
4821 struct net_device *bond_dev;
4826 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4827 name ? name : "bond%d",
4828 bond_setup, tx_queues);
4830 pr_err("%s: eek! can't alloc netdev!\n", name);
4835 dev_net_set(bond_dev, net);
4836 bond_dev->rtnl_link_ops = &bond_link_ops;
4838 res = register_netdevice(bond_dev);
4840 netif_carrier_off(bond_dev);
4844 bond_destructor(bond_dev);
4848 static int __net_init bond_net_init(struct net *net)
4850 struct bond_net *bn = net_generic(net, bond_net_id);
4853 INIT_LIST_HEAD(&bn->dev_list);
4855 bond_create_proc_dir(bn);
4856 bond_create_sysfs(bn);
4861 static void __net_exit bond_net_exit(struct net *net)
4863 struct bond_net *bn = net_generic(net, bond_net_id);
4864 struct bonding *bond, *tmp_bond;
4867 bond_destroy_sysfs(bn);
4868 bond_destroy_proc_dir(bn);
4870 /* Kill off any bonds created after unregistering bond rtnl ops */
4872 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4873 unregister_netdevice_queue(bond->dev, &list);
4874 unregister_netdevice_many(&list);
4878 static struct pernet_operations bond_net_ops = {
4879 .init = bond_net_init,
4880 .exit = bond_net_exit,
4882 .size = sizeof(struct bond_net),
4885 static int __init bonding_init(void)
4890 pr_info("%s", bond_version);
4892 res = bond_check_params(&bonding_defaults);
4896 res = register_pernet_subsys(&bond_net_ops);
4900 res = rtnl_link_register(&bond_link_ops);
4904 bond_create_debugfs();
4906 for (i = 0; i < max_bonds; i++) {
4907 res = bond_create(&init_net, NULL);
4912 register_netdevice_notifier(&bond_netdev_notifier);
4913 register_inetaddr_notifier(&bond_inetaddr_notifier);
4917 rtnl_link_unregister(&bond_link_ops);
4919 unregister_pernet_subsys(&bond_net_ops);
4924 static void __exit bonding_exit(void)
4926 unregister_netdevice_notifier(&bond_netdev_notifier);
4927 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4929 bond_destroy_debugfs();
4931 rtnl_link_unregister(&bond_link_ops);
4932 unregister_pernet_subsys(&bond_net_ops);
4934 #ifdef CONFIG_NET_POLL_CONTROLLER
4936 * Make sure we don't have an imbalance on our netpoll blocking
4938 WARN_ON(atomic_read(&netpoll_block_tx));
4942 module_init(bonding_init);
4943 module_exit(bonding_exit);
4944 MODULE_LICENSE("GPL");
4945 MODULE_VERSION(DRV_VERSION);
4946 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4947 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4948 MODULE_ALIAS_RTNL_LINK("bond");