2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
83 /*---------------------------- Module parameters ----------------------------*/
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV 0
87 #define BOND_LINK_ARP_INTERV 0
89 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
90 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
91 static int num_peer_notif = 1;
92 static int miimon = BOND_LINK_MON_INTERV;
95 static int use_carrier = 1;
98 static char *primary_reselect;
99 static char *lacp_rate;
100 static int min_links;
101 static char *ad_select;
102 static char *xmit_hash_policy;
103 static int arp_interval = BOND_LINK_ARP_INTERV;
104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
105 static char *arp_validate;
106 static char *fail_over_mac;
107 static int all_slaves_active = 0;
108 static struct bond_params bonding_defaults;
109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111 module_param(max_bonds, int, 0);
112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
113 module_param(tx_queues, int, 0);
114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
117 "failover event (alias of num_unsol_na)");
118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
120 "failover event (alias of num_grat_arp)");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130 "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
133 "1 for active-backup, 2 for balance-xor, "
134 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135 "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "0 for always (default), "
142 "1 for only if speed of primary is "
144 "2 for only on active slave "
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
148 "0 for slow, 1 for fast");
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
151 "0 for stable (default), 1 for bandwidth, "
153 module_param(min_links, int, 0);
154 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156 module_param(xmit_hash_policy, charp, 0);
157 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
158 "0 for layer 2 (default), 1 for layer 3+4, "
160 module_param(arp_interval, int, 0);
161 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
162 module_param_array(arp_ip_target, charp, NULL, 0);
163 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
164 module_param(arp_validate, charp, 0);
165 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
166 "0 for none (default), 1 for active, "
167 "2 for backup, 3 for all");
168 module_param(fail_over_mac, charp, 0);
169 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
170 "the same MAC; 0 for none (default), "
171 "1 for active, 2 for follow");
172 module_param(all_slaves_active, int, 0);
173 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
174 "by setting active flag for all slaves; "
175 "0 for never (default), 1 for always.");
176 module_param(resend_igmp, int, 0);
177 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
180 /*----------------------------- Global variables ----------------------------*/
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
186 int bond_net_id __read_mostly;
188 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
189 static int arp_ip_count;
190 static int bond_mode = BOND_MODE_ROUNDROBIN;
191 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
192 static int lacp_fast;
194 const struct bond_parm_tbl bond_lacp_tbl[] = {
195 { "slow", AD_LACP_SLOW},
196 { "fast", AD_LACP_FAST},
200 const struct bond_parm_tbl bond_mode_tbl[] = {
201 { "balance-rr", BOND_MODE_ROUNDROBIN},
202 { "active-backup", BOND_MODE_ACTIVEBACKUP},
203 { "balance-xor", BOND_MODE_XOR},
204 { "broadcast", BOND_MODE_BROADCAST},
205 { "802.3ad", BOND_MODE_8023AD},
206 { "balance-tlb", BOND_MODE_TLB},
207 { "balance-alb", BOND_MODE_ALB},
211 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
212 { "layer2", BOND_XMIT_POLICY_LAYER2},
213 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
214 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
218 const struct bond_parm_tbl arp_validate_tbl[] = {
219 { "none", BOND_ARP_VALIDATE_NONE},
220 { "active", BOND_ARP_VALIDATE_ACTIVE},
221 { "backup", BOND_ARP_VALIDATE_BACKUP},
222 { "all", BOND_ARP_VALIDATE_ALL},
226 const struct bond_parm_tbl fail_over_mac_tbl[] = {
227 { "none", BOND_FOM_NONE},
228 { "active", BOND_FOM_ACTIVE},
229 { "follow", BOND_FOM_FOLLOW},
233 const struct bond_parm_tbl pri_reselect_tbl[] = {
234 { "always", BOND_PRI_RESELECT_ALWAYS},
235 { "better", BOND_PRI_RESELECT_BETTER},
236 { "failure", BOND_PRI_RESELECT_FAILURE},
240 struct bond_parm_tbl ad_select_tbl[] = {
241 { "stable", BOND_AD_STABLE},
242 { "bandwidth", BOND_AD_BANDWIDTH},
243 { "count", BOND_AD_COUNT},
247 /*-------------------------- Forward declarations ---------------------------*/
249 static int bond_init(struct net_device *bond_dev);
250 static void bond_uninit(struct net_device *bond_dev);
252 /*---------------------------- General routines -----------------------------*/
254 const char *bond_mode_name(int mode)
256 static const char *names[] = {
257 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
258 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
259 [BOND_MODE_XOR] = "load balancing (xor)",
260 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
261 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
262 [BOND_MODE_TLB] = "transmit load balancing",
263 [BOND_MODE_ALB] = "adaptive load balancing",
266 if (mode < 0 || mode > BOND_MODE_ALB)
272 /*---------------------------------- VLAN -----------------------------------*/
275 * bond_add_vlan - add a new vlan id on bond
276 * @bond: bond that got the notification
277 * @vlan_id: the vlan id to add
279 * Returns -ENOMEM if allocation failed.
281 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283 struct vlan_entry *vlan;
285 pr_debug("bond: %s, vlan id %d\n",
286 (bond ? bond->dev->name : "None"), vlan_id);
288 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
292 INIT_LIST_HEAD(&vlan->vlan_list);
293 vlan->vlan_id = vlan_id;
295 write_lock_bh(&bond->lock);
297 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299 write_unlock_bh(&bond->lock);
301 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
307 * bond_del_vlan - delete a vlan id from bond
308 * @bond: bond that got the notification
309 * @vlan_id: the vlan id to delete
311 * returns -ENODEV if @vlan_id was not found in @bond.
313 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315 struct vlan_entry *vlan;
318 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
321 write_lock_bh(&bond->lock);
323 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
324 if (vlan->vlan_id == vlan_id) {
325 list_del(&vlan->vlan_list);
327 if (bond_is_lb(bond))
328 bond_alb_clear_vlan(bond, vlan_id);
330 pr_debug("removed VLAN ID %d from bond %s\n",
331 vlan_id, bond->dev->name);
340 pr_debug("couldn't find VLAN ID %d in bond %s\n",
341 vlan_id, bond->dev->name);
344 write_unlock_bh(&bond->lock);
345 unblock_netpoll_tx();
350 * bond_next_vlan - safely skip to the next item in the vlans list.
351 * @bond: the bond we're working on
352 * @curr: item we're advancing from
354 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
355 * or @curr->next otherwise (even if it is @curr itself again).
357 * Caller must hold bond->lock
359 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361 struct vlan_entry *next, *last;
363 if (list_empty(&bond->vlan_list))
367 next = list_entry(bond->vlan_list.next,
368 struct vlan_entry, vlan_list);
370 last = list_entry(bond->vlan_list.prev,
371 struct vlan_entry, vlan_list);
373 next = list_entry(bond->vlan_list.next,
374 struct vlan_entry, vlan_list);
376 next = list_entry(curr->vlan_list.next,
377 struct vlan_entry, vlan_list);
384 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
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 skb->queue_mapping = bond_queue_mapping(skb);
400 if (unlikely(netpoll_tx_running(slave_dev)))
401 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
409 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
410 * We don't protect the slave list iteration with a lock because:
411 * a. This operation is performed in IOCTL context,
412 * b. The operation is protected by the RTNL semaphore in the 8021q code,
413 * c. Holding a lock with BH disabled while directly calling a base driver
414 * entry point is generally a BAD idea.
416 * The design of synchronization/protection for this operation in the 8021q
417 * module is good for one or more VLAN devices over a single physical device
418 * and cannot be extended for a teaming solution like bonding, so there is a
419 * potential race condition here where a net device from the vlan group might
420 * be referenced (either by a base driver or the 8021q code) while it is being
421 * removed from the system. However, it turns out we're not making matters
422 * worse, and if it works for regular VLAN usage it will work here too.
426 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
427 * @bond_dev: bonding net device that got called
428 * @vid: vlan id being added
430 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
432 struct bonding *bond = netdev_priv(bond_dev);
433 struct slave *slave, *stop_at;
436 bond_for_each_slave(bond, slave, i) {
437 res = vlan_vid_add(slave->dev, vid);
442 res = bond_add_vlan(bond, vid);
444 pr_err("%s: Error: Failed to add vlan id %d\n",
445 bond_dev->name, vid);
452 /* unwind from head to the slave that failed */
454 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
455 vlan_vid_del(slave->dev, vid);
461 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
462 * @bond_dev: bonding net device that got called
463 * @vid: vlan id being removed
465 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
467 struct bonding *bond = netdev_priv(bond_dev);
471 bond_for_each_slave(bond, slave, i)
472 vlan_vid_del(slave->dev, vid);
474 res = bond_del_vlan(bond, vid);
476 pr_err("%s: Error: Failed to remove vlan id %d\n",
477 bond_dev->name, vid);
484 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
486 struct vlan_entry *vlan;
489 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
490 res = vlan_vid_add(slave_dev, vlan->vlan_id);
492 pr_warning("%s: Failed to add vlan id %d to device %s\n",
493 bond->dev->name, vlan->vlan_id,
498 static void bond_del_vlans_from_slave(struct bonding *bond,
499 struct net_device *slave_dev)
501 struct vlan_entry *vlan;
503 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
506 vlan_vid_del(slave_dev, vlan->vlan_id);
510 /*------------------------------- Link status -------------------------------*/
513 * Set the carrier state for the master according to the state of its
514 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
515 * do special 802.3ad magic.
517 * Returns zero if carrier state does not change, nonzero if it does.
519 static int bond_set_carrier(struct bonding *bond)
524 if (bond->slave_cnt == 0)
527 if (bond->params.mode == BOND_MODE_8023AD)
528 return bond_3ad_set_carrier(bond);
530 bond_for_each_slave(bond, slave, i) {
531 if (slave->link == BOND_LINK_UP) {
532 if (!netif_carrier_ok(bond->dev)) {
533 netif_carrier_on(bond->dev);
541 if (netif_carrier_ok(bond->dev)) {
542 netif_carrier_off(bond->dev);
549 * Get link speed and duplex from the slave's base driver
550 * using ethtool. If for some reason the call fails or the
551 * values are invalid, set speed and duplex to -1,
554 static int bond_update_speed_duplex(struct slave *slave)
556 struct net_device *slave_dev = slave->dev;
557 struct ethtool_cmd ecmd;
561 slave->speed = SPEED_UNKNOWN;
562 slave->duplex = DUPLEX_UNKNOWN;
564 res = __ethtool_get_settings(slave_dev, &ecmd);
568 slave_speed = ethtool_cmd_speed(&ecmd);
569 if (slave_speed == 0 || slave_speed == ((__u32) -1))
572 switch (ecmd.duplex) {
580 slave->speed = slave_speed;
581 slave->duplex = ecmd.duplex;
587 * if <dev> supports MII link status reporting, check its link status.
589 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
590 * depending upon the setting of the use_carrier parameter.
592 * Return either BMSR_LSTATUS, meaning that the link is up (or we
593 * can't tell and just pretend it is), or 0, meaning that the link is
596 * If reporting is non-zero, instead of faking link up, return -1 if
597 * both ETHTOOL and MII ioctls fail (meaning the device does not
598 * support them). If use_carrier is set, return whatever it says.
599 * It'd be nice if there was a good way to tell if a driver supports
600 * netif_carrier, but there really isn't.
602 static int bond_check_dev_link(struct bonding *bond,
603 struct net_device *slave_dev, int reporting)
605 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
606 int (*ioctl)(struct net_device *, struct ifreq *, int);
608 struct mii_ioctl_data *mii;
610 if (!reporting && !netif_running(slave_dev))
613 if (bond->params.use_carrier)
614 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
616 /* Try to get link status using Ethtool first. */
617 if (slave_dev->ethtool_ops) {
618 if (slave_dev->ethtool_ops->get_link) {
621 link = slave_dev->ethtool_ops->get_link(slave_dev);
623 return link ? BMSR_LSTATUS : 0;
627 /* Ethtool can't be used, fallback to MII ioctls. */
628 ioctl = slave_ops->ndo_do_ioctl;
630 /* TODO: set pointer to correct ioctl on a per team member */
631 /* bases to make this more efficient. that is, once */
632 /* we determine the correct ioctl, we will always */
633 /* call it and not the others for that team */
637 * We cannot assume that SIOCGMIIPHY will also read a
638 * register; not all network drivers (e.g., e100)
642 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
643 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
645 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
646 mii->reg_num = MII_BMSR;
647 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
648 return mii->val_out & BMSR_LSTATUS;
653 * If reporting, report that either there's no dev->do_ioctl,
654 * or both SIOCGMIIREG and get_link failed (meaning that we
655 * cannot report link status). If not reporting, pretend
658 return reporting ? -1 : BMSR_LSTATUS;
661 /*----------------------------- Multicast list ------------------------------*/
664 * Push the promiscuity flag down to appropriate slaves
666 static int bond_set_promiscuity(struct bonding *bond, int inc)
669 if (USES_PRIMARY(bond->params.mode)) {
670 /* write lock already acquired */
671 if (bond->curr_active_slave) {
672 err = dev_set_promiscuity(bond->curr_active_slave->dev,
678 bond_for_each_slave(bond, slave, i) {
679 err = dev_set_promiscuity(slave->dev, inc);
688 * Push the allmulti flag down to all slaves
690 static int bond_set_allmulti(struct bonding *bond, int inc)
693 if (USES_PRIMARY(bond->params.mode)) {
694 /* write lock already acquired */
695 if (bond->curr_active_slave) {
696 err = dev_set_allmulti(bond->curr_active_slave->dev,
702 bond_for_each_slave(bond, slave, i) {
703 err = dev_set_allmulti(slave->dev, inc);
712 * Add a Multicast address to slaves
715 static void bond_mc_add(struct bonding *bond, void *addr)
717 if (USES_PRIMARY(bond->params.mode)) {
718 /* write lock already acquired */
719 if (bond->curr_active_slave)
720 dev_mc_add(bond->curr_active_slave->dev, addr);
725 bond_for_each_slave(bond, slave, i)
726 dev_mc_add(slave->dev, addr);
731 * Remove a multicast address from slave
734 static void bond_mc_del(struct bonding *bond, void *addr)
736 if (USES_PRIMARY(bond->params.mode)) {
737 /* write lock already acquired */
738 if (bond->curr_active_slave)
739 dev_mc_del(bond->curr_active_slave->dev, addr);
743 bond_for_each_slave(bond, slave, i) {
744 dev_mc_del(slave->dev, addr);
750 static void __bond_resend_igmp_join_requests(struct net_device *dev)
752 struct in_device *in_dev;
755 in_dev = __in_dev_get_rcu(dev);
757 ip_mc_rejoin_groups(in_dev);
762 * Retrieve the list of registered multicast addresses for the bonding
763 * device and retransmit an IGMP JOIN request to the current active
766 static void bond_resend_igmp_join_requests(struct bonding *bond)
768 struct net_device *bond_dev, *vlan_dev, *master_dev;
769 struct vlan_entry *vlan;
771 read_lock(&bond->lock);
773 bond_dev = bond->dev;
775 /* rejoin all groups on bond device */
776 __bond_resend_igmp_join_requests(bond_dev);
779 * if bond is enslaved to a bridge,
780 * then rejoin all groups on its master
782 master_dev = bond_dev->master;
784 if ((master_dev->priv_flags & IFF_EBRIDGE)
785 && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
786 __bond_resend_igmp_join_requests(master_dev);
788 /* rejoin all groups on vlan devices */
789 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
791 vlan_dev = __vlan_find_dev_deep(bond_dev,
795 __bond_resend_igmp_join_requests(vlan_dev);
798 if (--bond->igmp_retrans > 0)
799 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
801 read_unlock(&bond->lock);
804 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
806 struct bonding *bond = container_of(work, struct bonding,
808 bond_resend_igmp_join_requests(bond);
812 * flush all members of flush->mc_list from device dev->mc_list
814 static void bond_mc_list_flush(struct net_device *bond_dev,
815 struct net_device *slave_dev)
817 struct bonding *bond = netdev_priv(bond_dev);
818 struct netdev_hw_addr *ha;
820 netdev_for_each_mc_addr(ha, bond_dev)
821 dev_mc_del(slave_dev, ha->addr);
823 if (bond->params.mode == BOND_MODE_8023AD) {
824 /* del lacpdu mc addr from mc list */
825 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
827 dev_mc_del(slave_dev, lacpdu_multicast);
831 /*--------------------------- Active slave change ---------------------------*/
834 * Update the mc list and multicast-related flags for the new and
835 * old active slaves (if any) according to the multicast mode, and
836 * promiscuous flags unconditionally.
838 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
839 struct slave *old_active)
841 struct netdev_hw_addr *ha;
843 if (!USES_PRIMARY(bond->params.mode))
844 /* nothing to do - mc list is already up-to-date on
850 if (bond->dev->flags & IFF_PROMISC)
851 dev_set_promiscuity(old_active->dev, -1);
853 if (bond->dev->flags & IFF_ALLMULTI)
854 dev_set_allmulti(old_active->dev, -1);
856 netdev_for_each_mc_addr(ha, bond->dev)
857 dev_mc_del(old_active->dev, ha->addr);
861 /* FIXME: Signal errors upstream. */
862 if (bond->dev->flags & IFF_PROMISC)
863 dev_set_promiscuity(new_active->dev, 1);
865 if (bond->dev->flags & IFF_ALLMULTI)
866 dev_set_allmulti(new_active->dev, 1);
868 netdev_for_each_mc_addr(ha, bond->dev)
869 dev_mc_add(new_active->dev, ha->addr);
874 * bond_do_fail_over_mac
876 * Perform special MAC address swapping for fail_over_mac settings
878 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
880 static void bond_do_fail_over_mac(struct bonding *bond,
881 struct slave *new_active,
882 struct slave *old_active)
883 __releases(&bond->curr_slave_lock)
884 __releases(&bond->lock)
885 __acquires(&bond->lock)
886 __acquires(&bond->curr_slave_lock)
888 u8 tmp_mac[ETH_ALEN];
889 struct sockaddr saddr;
892 switch (bond->params.fail_over_mac) {
893 case BOND_FOM_ACTIVE:
895 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
896 new_active->dev->addr_len);
898 case BOND_FOM_FOLLOW:
900 * if new_active && old_active, swap them
901 * if just old_active, do nothing (going to no active slave)
902 * if just new_active, set new_active to bond's MAC
907 write_unlock_bh(&bond->curr_slave_lock);
908 read_unlock(&bond->lock);
911 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
912 memcpy(saddr.sa_data, old_active->dev->dev_addr,
914 saddr.sa_family = new_active->dev->type;
916 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
917 saddr.sa_family = bond->dev->type;
920 rv = dev_set_mac_address(new_active->dev, &saddr);
922 pr_err("%s: Error %d setting MAC of slave %s\n",
923 bond->dev->name, -rv, new_active->dev->name);
930 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
931 saddr.sa_family = old_active->dev->type;
933 rv = dev_set_mac_address(old_active->dev, &saddr);
935 pr_err("%s: Error %d setting MAC of slave %s\n",
936 bond->dev->name, -rv, new_active->dev->name);
938 read_lock(&bond->lock);
939 write_lock_bh(&bond->curr_slave_lock);
942 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
943 bond->dev->name, bond->params.fail_over_mac);
949 static bool bond_should_change_active(struct bonding *bond)
951 struct slave *prim = bond->primary_slave;
952 struct slave *curr = bond->curr_active_slave;
954 if (!prim || !curr || curr->link != BOND_LINK_UP)
956 if (bond->force_primary) {
957 bond->force_primary = false;
960 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
961 (prim->speed < curr->speed ||
962 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
964 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
970 * find_best_interface - select the best available slave to be the active one
971 * @bond: our bonding struct
973 * Warning: Caller must hold curr_slave_lock for writing.
975 static struct slave *bond_find_best_slave(struct bonding *bond)
977 struct slave *new_active, *old_active;
978 struct slave *bestslave = NULL;
979 int mintime = bond->params.updelay;
982 new_active = bond->curr_active_slave;
984 if (!new_active) { /* there were no active slaves left */
985 if (bond->slave_cnt > 0) /* found one slave */
986 new_active = bond->first_slave;
988 return NULL; /* still no slave, return NULL */
991 if ((bond->primary_slave) &&
992 bond->primary_slave->link == BOND_LINK_UP &&
993 bond_should_change_active(bond)) {
994 new_active = bond->primary_slave;
997 /* remember where to stop iterating over the slaves */
998 old_active = new_active;
1000 bond_for_each_slave_from(bond, new_active, i, old_active) {
1001 if (new_active->link == BOND_LINK_UP) {
1003 } else if (new_active->link == BOND_LINK_BACK &&
1004 IS_UP(new_active->dev)) {
1005 /* link up, but waiting for stabilization */
1006 if (new_active->delay < mintime) {
1007 mintime = new_active->delay;
1008 bestslave = new_active;
1016 static bool bond_should_notify_peers(struct bonding *bond)
1018 struct slave *slave = bond->curr_active_slave;
1020 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1021 bond->dev->name, slave ? slave->dev->name : "NULL");
1023 if (!slave || !bond->send_peer_notif ||
1024 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1027 bond->send_peer_notif--;
1032 * change_active_interface - change the active slave into the specified one
1033 * @bond: our bonding struct
1034 * @new: the new slave to make the active one
1036 * Set the new slave to the bond's settings and unset them on the old
1037 * curr_active_slave.
1038 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1040 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1041 * because it is apparently the best available slave we have, even though its
1042 * updelay hasn't timed out yet.
1044 * If new_active is not NULL, caller must hold bond->lock for read and
1045 * curr_slave_lock for write_bh.
1047 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1049 struct slave *old_active = bond->curr_active_slave;
1051 if (old_active == new_active)
1055 new_active->jiffies = jiffies;
1057 if (new_active->link == BOND_LINK_BACK) {
1058 if (USES_PRIMARY(bond->params.mode)) {
1059 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1060 bond->dev->name, new_active->dev->name,
1061 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1064 new_active->delay = 0;
1065 new_active->link = BOND_LINK_UP;
1067 if (bond->params.mode == BOND_MODE_8023AD)
1068 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1070 if (bond_is_lb(bond))
1071 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1073 if (USES_PRIMARY(bond->params.mode)) {
1074 pr_info("%s: making interface %s the new active one.\n",
1075 bond->dev->name, new_active->dev->name);
1080 if (USES_PRIMARY(bond->params.mode))
1081 bond_mc_swap(bond, new_active, old_active);
1083 if (bond_is_lb(bond)) {
1084 bond_alb_handle_active_change(bond, new_active);
1086 bond_set_slave_inactive_flags(old_active);
1088 bond_set_slave_active_flags(new_active);
1090 bond->curr_active_slave = new_active;
1093 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1095 bond_set_slave_inactive_flags(old_active);
1098 bool should_notify_peers = false;
1100 bond_set_slave_active_flags(new_active);
1102 if (bond->params.fail_over_mac)
1103 bond_do_fail_over_mac(bond, new_active,
1106 if (netif_running(bond->dev)) {
1107 bond->send_peer_notif =
1108 bond->params.num_peer_notif;
1109 should_notify_peers =
1110 bond_should_notify_peers(bond);
1113 write_unlock_bh(&bond->curr_slave_lock);
1114 read_unlock(&bond->lock);
1116 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1117 if (should_notify_peers)
1118 netdev_bonding_change(bond->dev,
1119 NETDEV_NOTIFY_PEERS);
1121 read_lock(&bond->lock);
1122 write_lock_bh(&bond->curr_slave_lock);
1126 /* resend IGMP joins since active slave has changed or
1127 * all were sent on curr_active_slave.
1128 * resend only if bond is brought up with the affected
1129 * bonding modes and the retransmission is enabled */
1130 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1131 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1132 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1133 bond->igmp_retrans = bond->params.resend_igmp;
1134 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1139 * bond_select_active_slave - select a new active slave, if needed
1140 * @bond: our bonding struct
1142 * This functions should be called when one of the following occurs:
1143 * - The old curr_active_slave has been released or lost its link.
1144 * - The primary_slave has got its link back.
1145 * - A slave has got its link back and there's no old curr_active_slave.
1147 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1149 void bond_select_active_slave(struct bonding *bond)
1151 struct slave *best_slave;
1154 best_slave = bond_find_best_slave(bond);
1155 if (best_slave != bond->curr_active_slave) {
1156 bond_change_active_slave(bond, best_slave);
1157 rv = bond_set_carrier(bond);
1161 if (netif_carrier_ok(bond->dev)) {
1162 pr_info("%s: first active interface up!\n",
1165 pr_info("%s: now running without any active interface !\n",
1171 /*--------------------------- slave list handling ---------------------------*/
1174 * This function attaches the slave to the end of list.
1176 * bond->lock held for writing by caller.
1178 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1180 if (bond->first_slave == NULL) { /* attaching the first slave */
1181 new_slave->next = new_slave;
1182 new_slave->prev = new_slave;
1183 bond->first_slave = new_slave;
1185 new_slave->next = bond->first_slave;
1186 new_slave->prev = bond->first_slave->prev;
1187 new_slave->next->prev = new_slave;
1188 new_slave->prev->next = new_slave;
1195 * This function detaches the slave from the list.
1196 * WARNING: no check is made to verify if the slave effectively
1197 * belongs to <bond>.
1198 * Nothing is freed on return, structures are just unchained.
1199 * If any slave pointer in bond was pointing to <slave>,
1200 * it should be changed by the calling function.
1202 * bond->lock held for writing by caller.
1204 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1207 slave->next->prev = slave->prev;
1210 slave->prev->next = slave->next;
1212 if (bond->first_slave == slave) { /* slave is the first slave */
1213 if (bond->slave_cnt > 1) { /* there are more slave */
1214 bond->first_slave = slave->next;
1216 bond->first_slave = NULL; /* slave was the last one */
1225 #ifdef CONFIG_NET_POLL_CONTROLLER
1226 static inline int slave_enable_netpoll(struct slave *slave)
1231 np = kzalloc(sizeof(*np), GFP_KERNEL);
1236 np->dev = slave->dev;
1237 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1238 err = __netpoll_setup(np);
1247 static inline void slave_disable_netpoll(struct slave *slave)
1249 struct netpoll *np = slave->np;
1255 synchronize_rcu_bh();
1256 __netpoll_cleanup(np);
1259 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1261 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1263 if (!slave_dev->netdev_ops->ndo_poll_controller)
1268 static void bond_poll_controller(struct net_device *bond_dev)
1272 static void __bond_netpoll_cleanup(struct bonding *bond)
1274 struct slave *slave;
1277 bond_for_each_slave(bond, slave, i)
1278 if (IS_UP(slave->dev))
1279 slave_disable_netpoll(slave);
1281 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1283 struct bonding *bond = netdev_priv(bond_dev);
1285 read_lock(&bond->lock);
1286 __bond_netpoll_cleanup(bond);
1287 read_unlock(&bond->lock);
1290 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1292 struct bonding *bond = netdev_priv(dev);
1293 struct slave *slave;
1296 read_lock(&bond->lock);
1297 bond_for_each_slave(bond, slave, i) {
1298 err = slave_enable_netpoll(slave);
1300 __bond_netpoll_cleanup(bond);
1304 read_unlock(&bond->lock);
1308 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1310 return bond->dev->npinfo;
1314 static inline int slave_enable_netpoll(struct slave *slave)
1318 static inline void slave_disable_netpoll(struct slave *slave)
1321 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1326 /*---------------------------------- IOCTL ----------------------------------*/
1328 static int bond_sethwaddr(struct net_device *bond_dev,
1329 struct net_device *slave_dev)
1331 pr_debug("bond_dev=%p\n", bond_dev);
1332 pr_debug("slave_dev=%p\n", slave_dev);
1333 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1334 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1338 static netdev_features_t bond_fix_features(struct net_device *dev,
1339 netdev_features_t features)
1341 struct slave *slave;
1342 struct bonding *bond = netdev_priv(dev);
1343 netdev_features_t mask;
1346 read_lock(&bond->lock);
1348 if (!bond->first_slave) {
1349 /* Disable adding VLANs to empty bond. But why? --mq */
1350 features |= NETIF_F_VLAN_CHALLENGED;
1355 features &= ~NETIF_F_ONE_FOR_ALL;
1356 features |= NETIF_F_ALL_FOR_ALL;
1358 bond_for_each_slave(bond, slave, i) {
1359 features = netdev_increment_features(features,
1360 slave->dev->features,
1365 read_unlock(&bond->lock);
1369 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1370 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1371 NETIF_F_HIGHDMA | NETIF_F_LRO)
1373 static void bond_compute_features(struct bonding *bond)
1375 struct slave *slave;
1376 struct net_device *bond_dev = bond->dev;
1377 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1378 unsigned short max_hard_header_len = ETH_HLEN;
1381 read_lock(&bond->lock);
1383 if (!bond->first_slave)
1386 bond_for_each_slave(bond, slave, i) {
1387 vlan_features = netdev_increment_features(vlan_features,
1388 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1390 if (slave->dev->hard_header_len > max_hard_header_len)
1391 max_hard_header_len = slave->dev->hard_header_len;
1395 bond_dev->vlan_features = vlan_features;
1396 bond_dev->hard_header_len = max_hard_header_len;
1398 read_unlock(&bond->lock);
1400 netdev_change_features(bond_dev);
1403 static void bond_setup_by_slave(struct net_device *bond_dev,
1404 struct net_device *slave_dev)
1406 struct bonding *bond = netdev_priv(bond_dev);
1408 bond_dev->header_ops = slave_dev->header_ops;
1410 bond_dev->type = slave_dev->type;
1411 bond_dev->hard_header_len = slave_dev->hard_header_len;
1412 bond_dev->addr_len = slave_dev->addr_len;
1414 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1415 slave_dev->addr_len);
1416 bond->setup_by_slave = 1;
1419 /* On bonding slaves other than the currently active slave, suppress
1420 * duplicates except for alb non-mcast/bcast.
1422 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1423 struct slave *slave,
1424 struct bonding *bond)
1426 if (bond_is_slave_inactive(slave)) {
1427 if (bond->params.mode == BOND_MODE_ALB &&
1428 skb->pkt_type != PACKET_BROADCAST &&
1429 skb->pkt_type != PACKET_MULTICAST)
1436 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1438 struct sk_buff *skb = *pskb;
1439 struct slave *slave;
1440 struct bonding *bond;
1441 void (*recv_probe)(struct sk_buff *, struct bonding *,
1444 skb = skb_share_check(skb, GFP_ATOMIC);
1446 return RX_HANDLER_CONSUMED;
1450 slave = bond_slave_get_rcu(skb->dev);
1453 if (bond->params.arp_interval)
1454 slave->dev->last_rx = jiffies;
1456 recv_probe = ACCESS_ONCE(bond->recv_probe);
1458 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1461 recv_probe(nskb, bond, slave);
1462 dev_kfree_skb(nskb);
1466 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1467 return RX_HANDLER_EXACT;
1470 skb->dev = bond->dev;
1472 if (bond->params.mode == BOND_MODE_ALB &&
1473 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1474 skb->pkt_type == PACKET_HOST) {
1476 if (unlikely(skb_cow_head(skb,
1477 skb->data - skb_mac_header(skb)))) {
1479 return RX_HANDLER_CONSUMED;
1481 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1484 return RX_HANDLER_ANOTHER;
1487 /* enslave device <slave> to bond device <master> */
1488 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1490 struct bonding *bond = netdev_priv(bond_dev);
1491 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1492 struct slave *new_slave = NULL;
1493 struct netdev_hw_addr *ha;
1494 struct sockaddr addr;
1498 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1499 slave_ops->ndo_do_ioctl == NULL) {
1500 pr_warning("%s: Warning: no link monitoring support for %s\n",
1501 bond_dev->name, slave_dev->name);
1504 /* already enslaved */
1505 if (slave_dev->flags & IFF_SLAVE) {
1506 pr_debug("Error, Device was already enslaved\n");
1510 /* vlan challenged mutual exclusion */
1511 /* no need to lock since we're protected by rtnl_lock */
1512 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1513 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1514 if (bond_vlan_used(bond)) {
1515 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1516 bond_dev->name, slave_dev->name, bond_dev->name);
1519 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1520 bond_dev->name, slave_dev->name,
1521 slave_dev->name, bond_dev->name);
1524 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1528 * Old ifenslave binaries are no longer supported. These can
1529 * be identified with moderate accuracy by the state of the slave:
1530 * the current ifenslave will set the interface down prior to
1531 * enslaving it; the old ifenslave will not.
1533 if ((slave_dev->flags & IFF_UP)) {
1534 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1537 goto err_undo_flags;
1540 /* set bonding device ether type by slave - bonding netdevices are
1541 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1542 * there is a need to override some of the type dependent attribs/funcs.
1544 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1545 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1547 if (bond->slave_cnt == 0) {
1548 if (bond_dev->type != slave_dev->type) {
1549 pr_debug("%s: change device type from %d to %d\n",
1551 bond_dev->type, slave_dev->type);
1553 res = netdev_bonding_change(bond_dev,
1554 NETDEV_PRE_TYPE_CHANGE);
1555 res = notifier_to_errno(res);
1557 pr_err("%s: refused to change device type\n",
1560 goto err_undo_flags;
1563 /* Flush unicast and multicast addresses */
1564 dev_uc_flush(bond_dev);
1565 dev_mc_flush(bond_dev);
1567 if (slave_dev->type != ARPHRD_ETHER)
1568 bond_setup_by_slave(bond_dev, slave_dev);
1570 ether_setup(bond_dev);
1571 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1574 netdev_bonding_change(bond_dev,
1575 NETDEV_POST_TYPE_CHANGE);
1577 } else if (bond_dev->type != slave_dev->type) {
1578 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1580 slave_dev->type, bond_dev->type);
1582 goto err_undo_flags;
1585 if (slave_ops->ndo_set_mac_address == NULL) {
1586 if (bond->slave_cnt == 0) {
1587 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1589 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1590 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1591 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",
1594 goto err_undo_flags;
1598 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1600 /* If this is the first slave, then we need to set the master's hardware
1601 * address to be the same as the slave's. */
1602 if (is_zero_ether_addr(bond->dev->dev_addr))
1603 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1604 slave_dev->addr_len);
1607 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1610 goto err_undo_flags;
1614 * Set the new_slave's queue_id to be zero. Queue ID mapping
1615 * is set via sysfs or module option if desired.
1617 new_slave->queue_id = 0;
1619 /* Save slave's original mtu and then set it to match the bond */
1620 new_slave->original_mtu = slave_dev->mtu;
1621 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1623 pr_debug("Error %d calling dev_set_mtu\n", res);
1628 * Save slave's original ("permanent") mac address for modes
1629 * that need it, and for restoring it upon release, and then
1630 * set it to the master's address
1632 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1634 if (!bond->params.fail_over_mac) {
1636 * Set slave to master's mac address. The application already
1637 * set the master's mac address to that of the first slave
1639 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1640 addr.sa_family = slave_dev->type;
1641 res = dev_set_mac_address(slave_dev, &addr);
1643 pr_debug("Error %d calling set_mac_address\n", res);
1644 goto err_restore_mtu;
1648 res = netdev_set_bond_master(slave_dev, bond_dev);
1650 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1651 goto err_restore_mac;
1654 /* open the slave since the application closed it */
1655 res = dev_open(slave_dev);
1657 pr_debug("Opening slave %s failed\n", slave_dev->name);
1658 goto err_unset_master;
1661 new_slave->bond = bond;
1662 new_slave->dev = slave_dev;
1663 slave_dev->priv_flags |= IFF_BONDING;
1665 if (bond_is_lb(bond)) {
1666 /* bond_alb_init_slave() must be called before all other stages since
1667 * it might fail and we do not want to have to undo everything
1669 res = bond_alb_init_slave(bond, new_slave);
1674 /* If the mode USES_PRIMARY, then the new slave gets the
1675 * master's promisc (and mc) settings only if it becomes the
1676 * curr_active_slave, and that is taken care of later when calling
1677 * bond_change_active()
1679 if (!USES_PRIMARY(bond->params.mode)) {
1680 /* set promiscuity level to new slave */
1681 if (bond_dev->flags & IFF_PROMISC) {
1682 res = dev_set_promiscuity(slave_dev, 1);
1687 /* set allmulti level to new slave */
1688 if (bond_dev->flags & IFF_ALLMULTI) {
1689 res = dev_set_allmulti(slave_dev, 1);
1694 netif_addr_lock_bh(bond_dev);
1695 /* upload master's mc_list to new slave */
1696 netdev_for_each_mc_addr(ha, bond_dev)
1697 dev_mc_add(slave_dev, ha->addr);
1698 netif_addr_unlock_bh(bond_dev);
1701 if (bond->params.mode == BOND_MODE_8023AD) {
1702 /* add lacpdu mc addr to mc list */
1703 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1705 dev_mc_add(slave_dev, lacpdu_multicast);
1708 bond_add_vlans_on_slave(bond, slave_dev);
1710 write_lock_bh(&bond->lock);
1712 bond_attach_slave(bond, new_slave);
1714 new_slave->delay = 0;
1715 new_slave->link_failure_count = 0;
1717 write_unlock_bh(&bond->lock);
1719 bond_compute_features(bond);
1721 read_lock(&bond->lock);
1723 new_slave->last_arp_rx = jiffies;
1725 if (bond->params.miimon && !bond->params.use_carrier) {
1726 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1728 if ((link_reporting == -1) && !bond->params.arp_interval) {
1730 * miimon is set but a bonded network driver
1731 * does not support ETHTOOL/MII and
1732 * arp_interval is not set. Note: if
1733 * use_carrier is enabled, we will never go
1734 * here (because netif_carrier is always
1735 * supported); thus, we don't need to change
1736 * the messages for netif_carrier.
1738 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",
1739 bond_dev->name, slave_dev->name);
1740 } else if (link_reporting == -1) {
1741 /* unable get link status using mii/ethtool */
1742 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",
1743 bond_dev->name, slave_dev->name);
1747 /* check for initial state */
1748 if (!bond->params.miimon ||
1749 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1750 if (bond->params.updelay) {
1751 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1752 new_slave->link = BOND_LINK_BACK;
1753 new_slave->delay = bond->params.updelay;
1755 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1756 new_slave->link = BOND_LINK_UP;
1758 new_slave->jiffies = jiffies;
1760 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1761 new_slave->link = BOND_LINK_DOWN;
1764 bond_update_speed_duplex(new_slave);
1766 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1767 /* if there is a primary slave, remember it */
1768 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1769 bond->primary_slave = new_slave;
1770 bond->force_primary = true;
1774 write_lock_bh(&bond->curr_slave_lock);
1776 switch (bond->params.mode) {
1777 case BOND_MODE_ACTIVEBACKUP:
1778 bond_set_slave_inactive_flags(new_slave);
1779 bond_select_active_slave(bond);
1781 case BOND_MODE_8023AD:
1782 /* in 802.3ad mode, the internal mechanism
1783 * will activate the slaves in the selected
1786 bond_set_slave_inactive_flags(new_slave);
1787 /* if this is the first slave */
1788 if (bond->slave_cnt == 1) {
1789 SLAVE_AD_INFO(new_slave).id = 1;
1790 /* Initialize AD with the number of times that the AD timer is called in 1 second
1791 * can be called only after the mac address of the bond is set
1793 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1795 SLAVE_AD_INFO(new_slave).id =
1796 SLAVE_AD_INFO(new_slave->prev).id + 1;
1799 bond_3ad_bind_slave(new_slave);
1803 bond_set_active_slave(new_slave);
1804 bond_set_slave_inactive_flags(new_slave);
1805 bond_select_active_slave(bond);
1808 pr_debug("This slave is always active in trunk mode\n");
1810 /* always active in trunk mode */
1811 bond_set_active_slave(new_slave);
1813 /* In trunking mode there is little meaning to curr_active_slave
1814 * anyway (it holds no special properties of the bond device),
1815 * so we can change it without calling change_active_interface()
1817 if (!bond->curr_active_slave)
1818 bond->curr_active_slave = new_slave;
1821 } /* switch(bond_mode) */
1823 write_unlock_bh(&bond->curr_slave_lock);
1825 bond_set_carrier(bond);
1827 #ifdef CONFIG_NET_POLL_CONTROLLER
1828 slave_dev->npinfo = bond_netpoll_info(bond);
1829 if (slave_dev->npinfo) {
1830 if (slave_enable_netpoll(new_slave)) {
1831 read_unlock(&bond->lock);
1832 pr_info("Error, %s: master_dev is using netpoll, "
1833 "but new slave device does not support netpoll.\n",
1841 read_unlock(&bond->lock);
1843 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1847 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1850 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1851 goto err_dest_symlinks;
1854 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1855 bond_dev->name, slave_dev->name,
1856 bond_is_active_slave(new_slave) ? "n active" : " backup",
1857 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1859 /* enslave is successful */
1862 /* Undo stages on error */
1864 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1867 write_lock_bh(&bond->lock);
1868 bond_detach_slave(bond, new_slave);
1869 write_unlock_bh(&bond->lock);
1872 dev_close(slave_dev);
1875 netdev_set_bond_master(slave_dev, NULL);
1878 if (!bond->params.fail_over_mac) {
1879 /* XXX TODO - fom follow mode needs to change master's
1880 * MAC if this slave's MAC is in use by the bond, or at
1881 * least print a warning.
1883 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1884 addr.sa_family = slave_dev->type;
1885 dev_set_mac_address(slave_dev, &addr);
1889 dev_set_mtu(slave_dev, new_slave->original_mtu);
1895 bond_compute_features(bond);
1901 * Try to release the slave device <slave> from the bond device <master>
1902 * It is legal to access curr_active_slave without a lock because all the function
1905 * The rules for slave state should be:
1906 * for Active/Backup:
1907 * Active stays on all backups go down
1908 * for Bonded connections:
1909 * The first up interface should be left on and all others downed.
1911 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1913 struct bonding *bond = netdev_priv(bond_dev);
1914 struct slave *slave, *oldcurrent;
1915 struct sockaddr addr;
1916 netdev_features_t old_features = bond_dev->features;
1918 /* slave is not a slave or master is not master of this slave */
1919 if (!(slave_dev->flags & IFF_SLAVE) ||
1920 (slave_dev->master != bond_dev)) {
1921 pr_err("%s: Error: cannot release %s.\n",
1922 bond_dev->name, slave_dev->name);
1927 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1928 write_lock_bh(&bond->lock);
1930 slave = bond_get_slave_by_dev(bond, slave_dev);
1932 /* not a slave of this bond */
1933 pr_info("%s: %s not enslaved\n",
1934 bond_dev->name, slave_dev->name);
1935 write_unlock_bh(&bond->lock);
1936 unblock_netpoll_tx();
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_unlock_bh(&bond->lock);
1946 write_lock_bh(&bond->lock);
1948 if (!bond->params.fail_over_mac) {
1949 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1950 bond->slave_cnt > 1)
1951 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",
1952 bond_dev->name, slave_dev->name,
1954 bond_dev->name, slave_dev->name);
1957 /* Inform AD package of unbinding of slave. */
1958 if (bond->params.mode == BOND_MODE_8023AD) {
1959 /* must be called before the slave is
1960 * detached from the list
1962 bond_3ad_unbind_slave(slave);
1965 pr_info("%s: releasing %s interface %s\n",
1967 bond_is_active_slave(slave) ? "active" : "backup",
1970 oldcurrent = bond->curr_active_slave;
1972 bond->current_arp_slave = NULL;
1974 /* release the slave from its bond */
1975 bond_detach_slave(bond, slave);
1977 if (bond->primary_slave == slave)
1978 bond->primary_slave = NULL;
1980 if (oldcurrent == slave)
1981 bond_change_active_slave(bond, NULL);
1983 if (bond_is_lb(bond)) {
1984 /* Must be called only after the slave has been
1985 * detached from the list and the curr_active_slave
1986 * has been cleared (if our_slave == old_current),
1987 * but before a new active slave is selected.
1989 write_unlock_bh(&bond->lock);
1990 bond_alb_deinit_slave(bond, slave);
1991 write_lock_bh(&bond->lock);
1994 if (oldcurrent == slave) {
1996 * Note that we hold RTNL over this sequence, so there
1997 * is no concern that another slave add/remove event
2000 write_unlock_bh(&bond->lock);
2001 read_lock(&bond->lock);
2002 write_lock_bh(&bond->curr_slave_lock);
2004 bond_select_active_slave(bond);
2006 write_unlock_bh(&bond->curr_slave_lock);
2007 read_unlock(&bond->lock);
2008 write_lock_bh(&bond->lock);
2011 if (bond->slave_cnt == 0) {
2012 bond_set_carrier(bond);
2014 /* if the last slave was removed, zero the mac address
2015 * of the master so it will be set by the application
2016 * to the mac address of the first slave
2018 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2020 if (bond_vlan_used(bond)) {
2021 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2022 bond_dev->name, bond_dev->name);
2023 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2028 write_unlock_bh(&bond->lock);
2029 unblock_netpoll_tx();
2031 bond_compute_features(bond);
2032 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2033 (old_features & NETIF_F_VLAN_CHALLENGED))
2034 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2035 bond_dev->name, slave_dev->name, bond_dev->name);
2037 /* must do this from outside any spinlocks */
2038 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2040 bond_del_vlans_from_slave(bond, slave_dev);
2042 /* If the mode USES_PRIMARY, then we should only remove its
2043 * promisc and mc settings if it was the curr_active_slave, but that was
2044 * already taken care of above when we detached the slave
2046 if (!USES_PRIMARY(bond->params.mode)) {
2047 /* unset promiscuity level from slave */
2048 if (bond_dev->flags & IFF_PROMISC)
2049 dev_set_promiscuity(slave_dev, -1);
2051 /* unset allmulti level from slave */
2052 if (bond_dev->flags & IFF_ALLMULTI)
2053 dev_set_allmulti(slave_dev, -1);
2055 /* flush master's mc_list from slave */
2056 netif_addr_lock_bh(bond_dev);
2057 bond_mc_list_flush(bond_dev, slave_dev);
2058 netif_addr_unlock_bh(bond_dev);
2061 netdev_set_bond_master(slave_dev, NULL);
2063 slave_disable_netpoll(slave);
2065 /* close slave before restoring its mac address */
2066 dev_close(slave_dev);
2068 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2069 /* restore original ("permanent") mac address */
2070 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2071 addr.sa_family = slave_dev->type;
2072 dev_set_mac_address(slave_dev, &addr);
2075 dev_set_mtu(slave_dev, slave->original_mtu);
2077 slave_dev->priv_flags &= ~IFF_BONDING;
2081 return 0; /* deletion OK */
2085 * First release a slave and then destroy the bond if no more slaves are left.
2086 * Must be under rtnl_lock when this function is called.
2088 static int bond_release_and_destroy(struct net_device *bond_dev,
2089 struct net_device *slave_dev)
2091 struct bonding *bond = netdev_priv(bond_dev);
2094 ret = bond_release(bond_dev, slave_dev);
2095 if ((ret == 0) && (bond->slave_cnt == 0)) {
2096 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2097 pr_info("%s: destroying bond %s.\n",
2098 bond_dev->name, bond_dev->name);
2099 unregister_netdevice(bond_dev);
2105 * This function releases all slaves.
2107 static int bond_release_all(struct net_device *bond_dev)
2109 struct bonding *bond = netdev_priv(bond_dev);
2110 struct slave *slave;
2111 struct net_device *slave_dev;
2112 struct sockaddr addr;
2114 write_lock_bh(&bond->lock);
2116 netif_carrier_off(bond_dev);
2118 if (bond->slave_cnt == 0)
2121 bond->current_arp_slave = NULL;
2122 bond->primary_slave = NULL;
2123 bond_change_active_slave(bond, NULL);
2125 while ((slave = bond->first_slave) != NULL) {
2126 /* Inform AD package of unbinding of slave
2127 * before slave is detached from the list.
2129 if (bond->params.mode == BOND_MODE_8023AD)
2130 bond_3ad_unbind_slave(slave);
2132 slave_dev = slave->dev;
2133 bond_detach_slave(bond, slave);
2135 /* now that the slave is detached, unlock and perform
2136 * all the undo steps that should not be called from
2139 write_unlock_bh(&bond->lock);
2141 /* unregister rx_handler early so bond_handle_frame wouldn't
2142 * be called for this slave anymore.
2144 netdev_rx_handler_unregister(slave_dev);
2147 if (bond_is_lb(bond)) {
2148 /* must be called only after the slave
2149 * has been detached from the list
2151 bond_alb_deinit_slave(bond, slave);
2154 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2155 bond_del_vlans_from_slave(bond, slave_dev);
2157 /* If the mode USES_PRIMARY, then we should only remove its
2158 * promisc and mc settings if it was the curr_active_slave, but that was
2159 * already taken care of above when we detached the slave
2161 if (!USES_PRIMARY(bond->params.mode)) {
2162 /* unset promiscuity level from slave */
2163 if (bond_dev->flags & IFF_PROMISC)
2164 dev_set_promiscuity(slave_dev, -1);
2166 /* unset allmulti level from slave */
2167 if (bond_dev->flags & IFF_ALLMULTI)
2168 dev_set_allmulti(slave_dev, -1);
2170 /* flush master's mc_list from slave */
2171 netif_addr_lock_bh(bond_dev);
2172 bond_mc_list_flush(bond_dev, slave_dev);
2173 netif_addr_unlock_bh(bond_dev);
2176 netdev_set_bond_master(slave_dev, NULL);
2178 slave_disable_netpoll(slave);
2180 /* close slave before restoring its mac address */
2181 dev_close(slave_dev);
2183 if (!bond->params.fail_over_mac) {
2184 /* restore original ("permanent") mac address*/
2185 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2186 addr.sa_family = slave_dev->type;
2187 dev_set_mac_address(slave_dev, &addr);
2192 /* re-acquire the lock before getting the next slave */
2193 write_lock_bh(&bond->lock);
2196 /* zero the mac address of the master so it will be
2197 * set by the application to the mac address of the
2200 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2202 if (bond_vlan_used(bond)) {
2203 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2204 bond_dev->name, bond_dev->name);
2205 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2209 pr_info("%s: released all slaves\n", bond_dev->name);
2212 write_unlock_bh(&bond->lock);
2214 bond_compute_features(bond);
2220 * This function changes the active slave to slave <slave_dev>.
2221 * It returns -EINVAL in the following cases.
2222 * - <slave_dev> is not found in the list.
2223 * - There is not active slave now.
2224 * - <slave_dev> is already active.
2225 * - The link state of <slave_dev> is not BOND_LINK_UP.
2226 * - <slave_dev> is not running.
2227 * In these cases, this function does nothing.
2228 * In the other cases, current_slave pointer is changed and 0 is returned.
2230 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2232 struct bonding *bond = netdev_priv(bond_dev);
2233 struct slave *old_active = NULL;
2234 struct slave *new_active = NULL;
2237 if (!USES_PRIMARY(bond->params.mode))
2240 /* Verify that master_dev is indeed the master of slave_dev */
2241 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2244 read_lock(&bond->lock);
2246 read_lock(&bond->curr_slave_lock);
2247 old_active = bond->curr_active_slave;
2248 read_unlock(&bond->curr_slave_lock);
2250 new_active = bond_get_slave_by_dev(bond, slave_dev);
2253 * Changing to the current active: do nothing; return success.
2255 if (new_active && (new_active == old_active)) {
2256 read_unlock(&bond->lock);
2262 (new_active->link == BOND_LINK_UP) &&
2263 IS_UP(new_active->dev)) {
2265 write_lock_bh(&bond->curr_slave_lock);
2266 bond_change_active_slave(bond, new_active);
2267 write_unlock_bh(&bond->curr_slave_lock);
2268 unblock_netpoll_tx();
2272 read_unlock(&bond->lock);
2277 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2279 struct bonding *bond = netdev_priv(bond_dev);
2281 info->bond_mode = bond->params.mode;
2282 info->miimon = bond->params.miimon;
2284 read_lock(&bond->lock);
2285 info->num_slaves = bond->slave_cnt;
2286 read_unlock(&bond->lock);
2291 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2293 struct bonding *bond = netdev_priv(bond_dev);
2294 struct slave *slave;
2295 int i, res = -ENODEV;
2297 read_lock(&bond->lock);
2299 bond_for_each_slave(bond, slave, i) {
2300 if (i == (int)info->slave_id) {
2302 strcpy(info->slave_name, slave->dev->name);
2303 info->link = slave->link;
2304 info->state = bond_slave_state(slave);
2305 info->link_failure_count = slave->link_failure_count;
2310 read_unlock(&bond->lock);
2315 /*-------------------------------- Monitoring -------------------------------*/
2318 static int bond_miimon_inspect(struct bonding *bond)
2320 struct slave *slave;
2321 int i, link_state, commit = 0;
2322 bool ignore_updelay;
2324 ignore_updelay = !bond->curr_active_slave ? true : false;
2326 bond_for_each_slave(bond, slave, i) {
2327 slave->new_link = BOND_LINK_NOCHANGE;
2329 link_state = bond_check_dev_link(bond, slave->dev, 0);
2331 switch (slave->link) {
2336 slave->link = BOND_LINK_FAIL;
2337 slave->delay = bond->params.downdelay;
2339 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2341 (bond->params.mode ==
2342 BOND_MODE_ACTIVEBACKUP) ?
2343 (bond_is_active_slave(slave) ?
2344 "active " : "backup ") : "",
2346 bond->params.downdelay * bond->params.miimon);
2349 case BOND_LINK_FAIL:
2352 * recovered before downdelay expired
2354 slave->link = BOND_LINK_UP;
2355 slave->jiffies = jiffies;
2356 pr_info("%s: link status up again after %d ms for interface %s.\n",
2358 (bond->params.downdelay - slave->delay) *
2359 bond->params.miimon,
2364 if (slave->delay <= 0) {
2365 slave->new_link = BOND_LINK_DOWN;
2373 case BOND_LINK_DOWN:
2377 slave->link = BOND_LINK_BACK;
2378 slave->delay = bond->params.updelay;
2381 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2382 bond->dev->name, slave->dev->name,
2383 ignore_updelay ? 0 :
2384 bond->params.updelay *
2385 bond->params.miimon);
2388 case BOND_LINK_BACK:
2390 slave->link = BOND_LINK_DOWN;
2391 pr_info("%s: link status down again after %d ms for interface %s.\n",
2393 (bond->params.updelay - slave->delay) *
2394 bond->params.miimon,
2403 if (slave->delay <= 0) {
2404 slave->new_link = BOND_LINK_UP;
2406 ignore_updelay = false;
2418 static void bond_miimon_commit(struct bonding *bond)
2420 struct slave *slave;
2423 bond_for_each_slave(bond, slave, i) {
2424 switch (slave->new_link) {
2425 case BOND_LINK_NOCHANGE:
2429 slave->link = BOND_LINK_UP;
2430 slave->jiffies = jiffies;
2432 if (bond->params.mode == BOND_MODE_8023AD) {
2433 /* prevent it from being the active one */
2434 bond_set_backup_slave(slave);
2435 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2436 /* make it immediately active */
2437 bond_set_active_slave(slave);
2438 } else if (slave != bond->primary_slave) {
2439 /* prevent it from being the active one */
2440 bond_set_backup_slave(slave);
2443 bond_update_speed_duplex(slave);
2445 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2446 bond->dev->name, slave->dev->name,
2447 slave->speed, slave->duplex ? "full" : "half");
2449 /* notify ad that the link status has changed */
2450 if (bond->params.mode == BOND_MODE_8023AD)
2451 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2453 if (bond_is_lb(bond))
2454 bond_alb_handle_link_change(bond, slave,
2457 if (!bond->curr_active_slave ||
2458 (slave == bond->primary_slave))
2463 case BOND_LINK_DOWN:
2464 if (slave->link_failure_count < UINT_MAX)
2465 slave->link_failure_count++;
2467 slave->link = BOND_LINK_DOWN;
2469 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2470 bond->params.mode == BOND_MODE_8023AD)
2471 bond_set_slave_inactive_flags(slave);
2473 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2474 bond->dev->name, slave->dev->name);
2476 if (bond->params.mode == BOND_MODE_8023AD)
2477 bond_3ad_handle_link_change(slave,
2480 if (bond_is_lb(bond))
2481 bond_alb_handle_link_change(bond, slave,
2484 if (slave == bond->curr_active_slave)
2490 pr_err("%s: invalid new link %d on slave %s\n",
2491 bond->dev->name, slave->new_link,
2493 slave->new_link = BOND_LINK_NOCHANGE;
2501 write_lock_bh(&bond->curr_slave_lock);
2502 bond_select_active_slave(bond);
2503 write_unlock_bh(&bond->curr_slave_lock);
2504 unblock_netpoll_tx();
2507 bond_set_carrier(bond);
2513 * Really a wrapper that splits the mii monitor into two phases: an
2514 * inspection, then (if inspection indicates something needs to be done)
2515 * an acquisition of appropriate locks followed by a commit phase to
2516 * implement whatever link state changes are indicated.
2518 void bond_mii_monitor(struct work_struct *work)
2520 struct bonding *bond = container_of(work, struct bonding,
2522 bool should_notify_peers = false;
2523 unsigned long delay;
2525 read_lock(&bond->lock);
2527 delay = msecs_to_jiffies(bond->params.miimon);
2529 if (bond->slave_cnt == 0)
2532 should_notify_peers = bond_should_notify_peers(bond);
2534 if (bond_miimon_inspect(bond)) {
2535 read_unlock(&bond->lock);
2537 /* Race avoidance with bond_close cancel of workqueue */
2538 if (!rtnl_trylock()) {
2539 read_lock(&bond->lock);
2541 should_notify_peers = false;
2545 read_lock(&bond->lock);
2547 bond_miimon_commit(bond);
2549 read_unlock(&bond->lock);
2550 rtnl_unlock(); /* might sleep, hold no other locks */
2551 read_lock(&bond->lock);
2555 if (bond->params.miimon)
2556 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2558 read_unlock(&bond->lock);
2560 if (should_notify_peers) {
2561 if (!rtnl_trylock()) {
2562 read_lock(&bond->lock);
2563 bond->send_peer_notif++;
2564 read_unlock(&bond->lock);
2567 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2572 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2574 struct vlan_entry *vlan;
2575 struct net_device *vlan_dev;
2577 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2580 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2582 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2584 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2592 * We go to the (large) trouble of VLAN tagging ARP frames because
2593 * switches in VLAN mode (especially if ports are configured as
2594 * "native" to a VLAN) might not pass non-tagged frames.
2596 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2598 struct sk_buff *skb;
2600 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2601 slave_dev->name, dest_ip, src_ip, vlan_id);
2603 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2604 NULL, slave_dev->dev_addr, NULL);
2607 pr_err("ARP packet allocation failed\n");
2611 skb = vlan_put_tag(skb, vlan_id);
2613 pr_err("failed to insert VLAN tag\n");
2621 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2624 __be32 *targets = bond->params.arp_targets;
2625 struct vlan_entry *vlan;
2626 struct net_device *vlan_dev = NULL;
2629 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2633 pr_debug("basa: target %x\n", targets[i]);
2634 if (!bond_vlan_used(bond)) {
2635 pr_debug("basa: empty vlan: arp_send\n");
2636 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2637 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2643 * If VLANs are configured, we do a route lookup to
2644 * determine which VLAN interface would be used, so we
2645 * can tag the ARP with the proper VLAN tag.
2647 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2650 if (net_ratelimit()) {
2651 pr_warning("%s: no route to arp_ip_target %pI4\n",
2652 bond->dev->name, &targets[i]);
2658 * This target is not on a VLAN
2660 if (rt->dst.dev == bond->dev) {
2662 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2663 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2664 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2670 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2672 vlan_dev = __vlan_find_dev_deep(bond->dev,
2675 if (vlan_dev == rt->dst.dev) {
2676 vlan_id = vlan->vlan_id;
2677 pr_debug("basa: vlan match on %s %d\n",
2678 vlan_dev->name, vlan_id);
2683 if (vlan_id && vlan_dev) {
2685 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2686 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2691 if (net_ratelimit()) {
2692 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2693 bond->dev->name, &targets[i],
2694 rt->dst.dev ? rt->dst.dev->name : "NULL");
2700 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2703 __be32 *targets = bond->params.arp_targets;
2705 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2706 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2707 &sip, &tip, i, &targets[i],
2708 bond_has_this_ip(bond, tip));
2709 if (sip == targets[i]) {
2710 if (bond_has_this_ip(bond, tip))
2711 slave->last_arp_rx = jiffies;
2717 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2718 struct slave *slave)
2721 unsigned char *arp_ptr;
2724 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2727 read_lock(&bond->lock);
2729 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2730 bond->dev->name, skb->dev->name);
2732 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2736 if (arp->ar_hln != bond->dev->addr_len ||
2737 skb->pkt_type == PACKET_OTHERHOST ||
2738 skb->pkt_type == PACKET_LOOPBACK ||
2739 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2740 arp->ar_pro != htons(ETH_P_IP) ||
2744 arp_ptr = (unsigned char *)(arp + 1);
2745 arp_ptr += bond->dev->addr_len;
2746 memcpy(&sip, arp_ptr, 4);
2747 arp_ptr += 4 + bond->dev->addr_len;
2748 memcpy(&tip, arp_ptr, 4);
2750 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2751 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2752 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2756 * Backup slaves won't see the ARP reply, but do come through
2757 * here for each ARP probe (so we swap the sip/tip to validate
2758 * the probe). In a "redundant switch, common router" type of
2759 * configuration, the ARP probe will (hopefully) travel from
2760 * the active, through one switch, the router, then the other
2761 * switch before reaching the backup.
2763 if (bond_is_active_slave(slave))
2764 bond_validate_arp(bond, slave, sip, tip);
2766 bond_validate_arp(bond, slave, tip, sip);
2769 read_unlock(&bond->lock);
2773 * this function is called regularly to monitor each slave's link
2774 * ensuring that traffic is being sent and received when arp monitoring
2775 * is used in load-balancing mode. if the adapter has been dormant, then an
2776 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2777 * arp monitoring in active backup mode.
2779 void bond_loadbalance_arp_mon(struct work_struct *work)
2781 struct bonding *bond = container_of(work, struct bonding,
2783 struct slave *slave, *oldcurrent;
2784 int do_failover = 0;
2788 read_lock(&bond->lock);
2790 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2792 if (bond->slave_cnt == 0)
2795 read_lock(&bond->curr_slave_lock);
2796 oldcurrent = bond->curr_active_slave;
2797 read_unlock(&bond->curr_slave_lock);
2799 /* see if any of the previous devices are up now (i.e. they have
2800 * xmt and rcv traffic). the curr_active_slave does not come into
2801 * the picture unless it is null. also, slave->jiffies is not needed
2802 * here because we send an arp on each slave and give a slave as
2803 * long as it needs to get the tx/rx within the delta.
2804 * TODO: what about up/down delay in arp mode? it wasn't here before
2807 bond_for_each_slave(bond, slave, i) {
2808 unsigned long trans_start = dev_trans_start(slave->dev);
2810 if (slave->link != BOND_LINK_UP) {
2811 if (time_in_range(jiffies,
2812 trans_start - delta_in_ticks,
2813 trans_start + delta_in_ticks) &&
2814 time_in_range(jiffies,
2815 slave->dev->last_rx - delta_in_ticks,
2816 slave->dev->last_rx + delta_in_ticks)) {
2818 slave->link = BOND_LINK_UP;
2819 bond_set_active_slave(slave);
2821 /* primary_slave has no meaning in round-robin
2822 * mode. the window of a slave being up and
2823 * curr_active_slave being null after enslaving
2827 pr_info("%s: link status definitely up for interface %s, ",
2832 pr_info("%s: interface %s is now up\n",
2838 /* slave->link == BOND_LINK_UP */
2840 /* not all switches will respond to an arp request
2841 * when the source ip is 0, so don't take the link down
2842 * if we don't know our ip yet
2844 if (!time_in_range(jiffies,
2845 trans_start - delta_in_ticks,
2846 trans_start + 2 * delta_in_ticks) ||
2847 !time_in_range(jiffies,
2848 slave->dev->last_rx - delta_in_ticks,
2849 slave->dev->last_rx + 2 * delta_in_ticks)) {
2851 slave->link = BOND_LINK_DOWN;
2852 bond_set_backup_slave(slave);
2854 if (slave->link_failure_count < UINT_MAX)
2855 slave->link_failure_count++;
2857 pr_info("%s: interface %s is now down.\n",
2861 if (slave == oldcurrent)
2866 /* note: if switch is in round-robin mode, all links
2867 * must tx arp to ensure all links rx an arp - otherwise
2868 * links may oscillate or not come up at all; if switch is
2869 * in something like xor mode, there is nothing we can
2870 * do - all replies will be rx'ed on same link causing slaves
2871 * to be unstable during low/no traffic periods
2873 if (IS_UP(slave->dev))
2874 bond_arp_send_all(bond, slave);
2879 write_lock_bh(&bond->curr_slave_lock);
2881 bond_select_active_slave(bond);
2883 write_unlock_bh(&bond->curr_slave_lock);
2884 unblock_netpoll_tx();
2888 if (bond->params.arp_interval)
2889 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2891 read_unlock(&bond->lock);
2895 * Called to inspect slaves for active-backup mode ARP monitor link state
2896 * changes. Sets new_link in slaves to specify what action should take
2897 * place for the slave. Returns 0 if no changes are found, >0 if changes
2898 * to link states must be committed.
2900 * Called with bond->lock held for read.
2902 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2904 struct slave *slave;
2906 unsigned long trans_start;
2908 bond_for_each_slave(bond, slave, i) {
2909 slave->new_link = BOND_LINK_NOCHANGE;
2911 if (slave->link != BOND_LINK_UP) {
2912 if (time_in_range(jiffies,
2913 slave_last_rx(bond, slave) - delta_in_ticks,
2914 slave_last_rx(bond, slave) + delta_in_ticks)) {
2916 slave->new_link = BOND_LINK_UP;
2924 * Give slaves 2*delta after being enslaved or made
2925 * active. This avoids bouncing, as the last receive
2926 * times need a full ARP monitor cycle to be updated.
2928 if (time_in_range(jiffies,
2929 slave->jiffies - delta_in_ticks,
2930 slave->jiffies + 2 * delta_in_ticks))
2934 * Backup slave is down if:
2935 * - No current_arp_slave AND
2936 * - more than 3*delta since last receive AND
2937 * - the bond has an IP address
2939 * Note: a non-null current_arp_slave indicates
2940 * the curr_active_slave went down and we are
2941 * searching for a new one; under this condition
2942 * we only take the curr_active_slave down - this
2943 * gives each slave a chance to tx/rx traffic
2944 * before being taken out
2946 if (!bond_is_active_slave(slave) &&
2947 !bond->current_arp_slave &&
2948 !time_in_range(jiffies,
2949 slave_last_rx(bond, slave) - delta_in_ticks,
2950 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2952 slave->new_link = BOND_LINK_DOWN;
2957 * Active slave is down if:
2958 * - more than 2*delta since transmitting OR
2959 * - (more than 2*delta since receive AND
2960 * the bond has an IP address)
2962 trans_start = dev_trans_start(slave->dev);
2963 if (bond_is_active_slave(slave) &&
2964 (!time_in_range(jiffies,
2965 trans_start - delta_in_ticks,
2966 trans_start + 2 * delta_in_ticks) ||
2967 !time_in_range(jiffies,
2968 slave_last_rx(bond, slave) - delta_in_ticks,
2969 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2971 slave->new_link = BOND_LINK_DOWN;
2980 * Called to commit link state changes noted by inspection step of
2981 * active-backup mode ARP monitor.
2983 * Called with RTNL and bond->lock for read.
2985 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2987 struct slave *slave;
2989 unsigned long trans_start;
2991 bond_for_each_slave(bond, slave, i) {
2992 switch (slave->new_link) {
2993 case BOND_LINK_NOCHANGE:
2997 trans_start = dev_trans_start(slave->dev);
2998 if ((!bond->curr_active_slave &&
2999 time_in_range(jiffies,
3000 trans_start - delta_in_ticks,
3001 trans_start + delta_in_ticks)) ||
3002 bond->curr_active_slave != slave) {
3003 slave->link = BOND_LINK_UP;
3004 bond->current_arp_slave = NULL;
3006 pr_info("%s: link status definitely up for interface %s.\n",
3007 bond->dev->name, slave->dev->name);
3009 if (!bond->curr_active_slave ||
3010 (slave == bond->primary_slave))
3017 case BOND_LINK_DOWN:
3018 if (slave->link_failure_count < UINT_MAX)
3019 slave->link_failure_count++;
3021 slave->link = BOND_LINK_DOWN;
3022 bond_set_slave_inactive_flags(slave);
3024 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3025 bond->dev->name, slave->dev->name);
3027 if (slave == bond->curr_active_slave) {
3028 bond->current_arp_slave = NULL;
3035 pr_err("%s: impossible: new_link %d on slave %s\n",
3036 bond->dev->name, slave->new_link,
3044 write_lock_bh(&bond->curr_slave_lock);
3045 bond_select_active_slave(bond);
3046 write_unlock_bh(&bond->curr_slave_lock);
3047 unblock_netpoll_tx();
3050 bond_set_carrier(bond);
3054 * Send ARP probes for active-backup mode ARP monitor.
3056 * Called with bond->lock held for read.
3058 static void bond_ab_arp_probe(struct bonding *bond)
3060 struct slave *slave;
3063 read_lock(&bond->curr_slave_lock);
3065 if (bond->current_arp_slave && bond->curr_active_slave)
3066 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3067 bond->current_arp_slave->dev->name,
3068 bond->curr_active_slave->dev->name);
3070 if (bond->curr_active_slave) {
3071 bond_arp_send_all(bond, bond->curr_active_slave);
3072 read_unlock(&bond->curr_slave_lock);
3076 read_unlock(&bond->curr_slave_lock);
3078 /* if we don't have a curr_active_slave, search for the next available
3079 * backup slave from the current_arp_slave and make it the candidate
3080 * for becoming the curr_active_slave
3083 if (!bond->current_arp_slave) {
3084 bond->current_arp_slave = bond->first_slave;
3085 if (!bond->current_arp_slave)
3089 bond_set_slave_inactive_flags(bond->current_arp_slave);
3091 /* search for next candidate */
3092 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3093 if (IS_UP(slave->dev)) {
3094 slave->link = BOND_LINK_BACK;
3095 bond_set_slave_active_flags(slave);
3096 bond_arp_send_all(bond, slave);
3097 slave->jiffies = jiffies;
3098 bond->current_arp_slave = slave;
3102 /* if the link state is up at this point, we
3103 * mark it down - this can happen if we have
3104 * simultaneous link failures and
3105 * reselect_active_interface doesn't make this
3106 * one the current slave so it is still marked
3107 * up when it is actually down
3109 if (slave->link == BOND_LINK_UP) {
3110 slave->link = BOND_LINK_DOWN;
3111 if (slave->link_failure_count < UINT_MAX)
3112 slave->link_failure_count++;
3114 bond_set_slave_inactive_flags(slave);
3116 pr_info("%s: backup interface %s is now down.\n",
3117 bond->dev->name, slave->dev->name);
3122 void bond_activebackup_arp_mon(struct work_struct *work)
3124 struct bonding *bond = container_of(work, struct bonding,
3126 bool should_notify_peers = false;
3129 read_lock(&bond->lock);
3131 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3133 if (bond->slave_cnt == 0)
3136 should_notify_peers = bond_should_notify_peers(bond);
3138 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3139 read_unlock(&bond->lock);
3141 /* Race avoidance with bond_close flush of workqueue */
3142 if (!rtnl_trylock()) {
3143 read_lock(&bond->lock);
3145 should_notify_peers = false;
3149 read_lock(&bond->lock);
3151 bond_ab_arp_commit(bond, delta_in_ticks);
3153 read_unlock(&bond->lock);
3155 read_lock(&bond->lock);
3158 bond_ab_arp_probe(bond);
3161 if (bond->params.arp_interval)
3162 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3164 read_unlock(&bond->lock);
3166 if (should_notify_peers) {
3167 if (!rtnl_trylock()) {
3168 read_lock(&bond->lock);
3169 bond->send_peer_notif++;
3170 read_unlock(&bond->lock);
3173 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3178 /*-------------------------- netdev event handling --------------------------*/
3181 * Change device name
3183 static int bond_event_changename(struct bonding *bond)
3185 bond_remove_proc_entry(bond);
3186 bond_create_proc_entry(bond);
3188 bond_debug_reregister(bond);
3193 static int bond_master_netdev_event(unsigned long event,
3194 struct net_device *bond_dev)
3196 struct bonding *event_bond = netdev_priv(bond_dev);
3199 case NETDEV_CHANGENAME:
3200 return bond_event_changename(event_bond);
3208 static int bond_slave_netdev_event(unsigned long event,
3209 struct net_device *slave_dev)
3211 struct net_device *bond_dev = slave_dev->master;
3212 struct bonding *bond = netdev_priv(bond_dev);
3213 struct slave *slave = NULL;
3216 case NETDEV_UNREGISTER:
3218 if (bond->setup_by_slave)
3219 bond_release_and_destroy(bond_dev, slave_dev);
3221 bond_release(bond_dev, slave_dev);
3226 slave = bond_get_slave_by_dev(bond, slave_dev);
3228 u32 old_speed = slave->speed;
3229 u8 old_duplex = slave->duplex;
3231 bond_update_speed_duplex(slave);
3233 if (bond->params.mode == BOND_MODE_8023AD) {
3234 if (old_speed != slave->speed)
3235 bond_3ad_adapter_speed_changed(slave);
3236 if (old_duplex != slave->duplex)
3237 bond_3ad_adapter_duplex_changed(slave);
3244 * ... Or is it this?
3247 case NETDEV_CHANGEMTU:
3249 * TODO: Should slaves be allowed to
3250 * independently alter their MTU? For
3251 * an active-backup bond, slaves need
3252 * not be the same type of device, so
3253 * MTUs may vary. For other modes,
3254 * slaves arguably should have the
3255 * same MTUs. To do this, we'd need to
3256 * take over the slave's change_mtu
3257 * function for the duration of their
3261 case NETDEV_CHANGENAME:
3263 * TODO: handle changing the primary's name
3266 case NETDEV_FEAT_CHANGE:
3267 bond_compute_features(bond);
3277 * bond_netdev_event: handle netdev notifier chain events.
3279 * This function receives events for the netdev chain. The caller (an
3280 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3281 * locks for us to safely manipulate the slave devices (RTNL lock,
3284 static int bond_netdev_event(struct notifier_block *this,
3285 unsigned long event, void *ptr)
3287 struct net_device *event_dev = (struct net_device *)ptr;
3289 pr_debug("event_dev: %s, event: %lx\n",
3290 event_dev ? event_dev->name : "None",
3293 if (!(event_dev->priv_flags & IFF_BONDING))
3296 if (event_dev->flags & IFF_MASTER) {
3297 pr_debug("IFF_MASTER\n");
3298 return bond_master_netdev_event(event, event_dev);
3301 if (event_dev->flags & IFF_SLAVE) {
3302 pr_debug("IFF_SLAVE\n");
3303 return bond_slave_netdev_event(event, event_dev);
3309 static struct notifier_block bond_netdev_notifier = {
3310 .notifier_call = bond_netdev_event,
3313 /*---------------------------- Hashing Policies -----------------------------*/
3316 * Hash for the output device based upon layer 2 and layer 3 data. If
3317 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3319 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3321 struct ethhdr *data = (struct ethhdr *)skb->data;
3322 struct iphdr *iph = ip_hdr(skb);
3324 if (skb->protocol == htons(ETH_P_IP)) {
3325 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3326 (data->h_dest[5] ^ data->h_source[5])) % count;
3329 return (data->h_dest[5] ^ data->h_source[5]) % count;
3333 * Hash for the output device based upon layer 3 and layer 4 data. If
3334 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3335 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3337 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3339 struct ethhdr *data = (struct ethhdr *)skb->data;
3340 struct iphdr *iph = ip_hdr(skb);
3341 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3344 if (skb->protocol == htons(ETH_P_IP)) {
3345 if (!ip_is_fragment(iph) &&
3346 (iph->protocol == IPPROTO_TCP ||
3347 iph->protocol == IPPROTO_UDP)) {
3348 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3350 return (layer4_xor ^
3351 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3355 return (data->h_dest[5] ^ data->h_source[5]) % count;
3359 * Hash for the output device based upon layer 2 data
3361 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3363 struct ethhdr *data = (struct ethhdr *)skb->data;
3365 return (data->h_dest[5] ^ data->h_source[5]) % count;
3368 /*-------------------------- Device entry points ----------------------------*/
3370 static int bond_open(struct net_device *bond_dev)
3372 struct bonding *bond = netdev_priv(bond_dev);
3373 struct slave *slave;
3376 /* reset slave->backup and slave->inactive */
3377 read_lock(&bond->lock);
3378 if (bond->slave_cnt > 0) {
3379 read_lock(&bond->curr_slave_lock);
3380 bond_for_each_slave(bond, slave, i) {
3381 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3382 && (slave != bond->curr_active_slave)) {
3383 bond_set_slave_inactive_flags(slave);
3385 bond_set_slave_active_flags(slave);
3388 read_unlock(&bond->curr_slave_lock);
3390 read_unlock(&bond->lock);
3392 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3394 if (bond_is_lb(bond)) {
3395 /* bond_alb_initialize must be called before the timer
3398 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3399 /* something went wrong - fail the open operation */
3403 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3404 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3407 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3408 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3409 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3412 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3413 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3414 INIT_DELAYED_WORK(&bond->arp_work,
3415 bond_activebackup_arp_mon);
3417 INIT_DELAYED_WORK(&bond->arp_work,
3418 bond_loadbalance_arp_mon);
3420 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3421 if (bond->params.arp_validate)
3422 bond->recv_probe = bond_arp_rcv;
3425 if (bond->params.mode == BOND_MODE_8023AD) {
3426 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3427 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3428 /* register to receive LACPDUs */
3429 bond->recv_probe = bond_3ad_lacpdu_recv;
3430 bond_3ad_initiate_agg_selection(bond, 1);
3436 static int bond_close(struct net_device *bond_dev)
3438 struct bonding *bond = netdev_priv(bond_dev);
3440 write_lock_bh(&bond->lock);
3442 bond->send_peer_notif = 0;
3444 write_unlock_bh(&bond->lock);
3446 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3447 cancel_delayed_work_sync(&bond->mii_work);
3450 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3451 cancel_delayed_work_sync(&bond->arp_work);
3454 switch (bond->params.mode) {
3455 case BOND_MODE_8023AD:
3456 cancel_delayed_work_sync(&bond->ad_work);
3460 cancel_delayed_work_sync(&bond->alb_work);
3466 if (delayed_work_pending(&bond->mcast_work))
3467 cancel_delayed_work_sync(&bond->mcast_work);
3469 if (bond_is_lb(bond)) {
3470 /* Must be called only after all
3471 * slaves have been released
3473 bond_alb_deinitialize(bond);
3475 bond->recv_probe = NULL;
3480 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3481 struct rtnl_link_stats64 *stats)
3483 struct bonding *bond = netdev_priv(bond_dev);
3484 struct rtnl_link_stats64 temp;
3485 struct slave *slave;
3488 memset(stats, 0, sizeof(*stats));
3490 read_lock_bh(&bond->lock);
3492 bond_for_each_slave(bond, slave, i) {
3493 const struct rtnl_link_stats64 *sstats =
3494 dev_get_stats(slave->dev, &temp);
3496 stats->rx_packets += sstats->rx_packets;
3497 stats->rx_bytes += sstats->rx_bytes;
3498 stats->rx_errors += sstats->rx_errors;
3499 stats->rx_dropped += sstats->rx_dropped;
3501 stats->tx_packets += sstats->tx_packets;
3502 stats->tx_bytes += sstats->tx_bytes;
3503 stats->tx_errors += sstats->tx_errors;
3504 stats->tx_dropped += sstats->tx_dropped;
3506 stats->multicast += sstats->multicast;
3507 stats->collisions += sstats->collisions;
3509 stats->rx_length_errors += sstats->rx_length_errors;
3510 stats->rx_over_errors += sstats->rx_over_errors;
3511 stats->rx_crc_errors += sstats->rx_crc_errors;
3512 stats->rx_frame_errors += sstats->rx_frame_errors;
3513 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3514 stats->rx_missed_errors += sstats->rx_missed_errors;
3516 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3517 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3518 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3519 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3520 stats->tx_window_errors += sstats->tx_window_errors;
3523 read_unlock_bh(&bond->lock);
3528 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3530 struct net_device *slave_dev = NULL;
3531 struct ifbond k_binfo;
3532 struct ifbond __user *u_binfo = NULL;
3533 struct ifslave k_sinfo;
3534 struct ifslave __user *u_sinfo = NULL;
3535 struct mii_ioctl_data *mii = NULL;
3538 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3550 * We do this again just in case we were called by SIOCGMIIREG
3551 * instead of SIOCGMIIPHY.
3558 if (mii->reg_num == 1) {
3559 struct bonding *bond = netdev_priv(bond_dev);
3561 read_lock(&bond->lock);
3562 read_lock(&bond->curr_slave_lock);
3563 if (netif_carrier_ok(bond->dev))
3564 mii->val_out = BMSR_LSTATUS;
3566 read_unlock(&bond->curr_slave_lock);
3567 read_unlock(&bond->lock);
3571 case BOND_INFO_QUERY_OLD:
3572 case SIOCBONDINFOQUERY:
3573 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3575 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3578 res = bond_info_query(bond_dev, &k_binfo);
3580 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3584 case BOND_SLAVE_INFO_QUERY_OLD:
3585 case SIOCBONDSLAVEINFOQUERY:
3586 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3588 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3591 res = bond_slave_info_query(bond_dev, &k_sinfo);
3593 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3602 if (!capable(CAP_NET_ADMIN))
3605 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3607 pr_debug("slave_dev=%p:\n", slave_dev);
3612 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3614 case BOND_ENSLAVE_OLD:
3615 case SIOCBONDENSLAVE:
3616 res = bond_enslave(bond_dev, slave_dev);
3618 case BOND_RELEASE_OLD:
3619 case SIOCBONDRELEASE:
3620 res = bond_release(bond_dev, slave_dev);
3622 case BOND_SETHWADDR_OLD:
3623 case SIOCBONDSETHWADDR:
3624 res = bond_sethwaddr(bond_dev, slave_dev);
3626 case BOND_CHANGE_ACTIVE_OLD:
3627 case SIOCBONDCHANGEACTIVE:
3628 res = bond_ioctl_change_active(bond_dev, slave_dev);
3640 static bool bond_addr_in_mc_list(unsigned char *addr,
3641 struct netdev_hw_addr_list *list,
3644 struct netdev_hw_addr *ha;
3646 netdev_hw_addr_list_for_each(ha, list)
3647 if (!memcmp(ha->addr, addr, addrlen))
3653 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3655 struct bonding *bond = netdev_priv(bond_dev);
3657 if (change & IFF_PROMISC)
3658 bond_set_promiscuity(bond,
3659 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3661 if (change & IFF_ALLMULTI)
3662 bond_set_allmulti(bond,
3663 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3666 static void bond_set_multicast_list(struct net_device *bond_dev)
3668 struct bonding *bond = netdev_priv(bond_dev);
3669 struct netdev_hw_addr *ha;
3672 read_lock(&bond->lock);
3674 /* looking for addresses to add to slaves' mc list */
3675 netdev_for_each_mc_addr(ha, bond_dev) {
3676 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3677 bond_dev->addr_len);
3679 bond_mc_add(bond, ha->addr);
3682 /* looking for addresses to delete from slaves' list */
3683 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3684 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3685 bond_dev->addr_len);
3687 bond_mc_del(bond, ha->addr);
3690 /* save master's multicast list */
3691 __hw_addr_flush(&bond->mc_list);
3692 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3693 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3695 read_unlock(&bond->lock);
3698 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3700 struct bonding *bond = netdev_priv(dev);
3701 struct slave *slave = bond->first_slave;
3704 const struct net_device_ops *slave_ops
3705 = slave->dev->netdev_ops;
3706 if (slave_ops->ndo_neigh_setup)
3707 return slave_ops->ndo_neigh_setup(slave->dev, parms);
3713 * Change the MTU of all of a master's slaves to match the master
3715 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3717 struct bonding *bond = netdev_priv(bond_dev);
3718 struct slave *slave, *stop_at;
3722 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3723 (bond_dev ? bond_dev->name : "None"), new_mtu);
3725 /* Can't hold bond->lock with bh disabled here since
3726 * some base drivers panic. On the other hand we can't
3727 * hold bond->lock without bh disabled because we'll
3728 * deadlock. The only solution is to rely on the fact
3729 * that we're under rtnl_lock here, and the slaves
3730 * list won't change. This doesn't solve the problem
3731 * of setting the slave's MTU while it is
3732 * transmitting, but the assumption is that the base
3733 * driver can handle that.
3735 * TODO: figure out a way to safely iterate the slaves
3736 * list, but without holding a lock around the actual
3737 * call to the base driver.
3740 bond_for_each_slave(bond, slave, i) {
3741 pr_debug("s %p s->p %p c_m %p\n",
3744 slave->dev->netdev_ops->ndo_change_mtu);
3746 res = dev_set_mtu(slave->dev, new_mtu);
3749 /* If we failed to set the slave's mtu to the new value
3750 * we must abort the operation even in ACTIVE_BACKUP
3751 * mode, because if we allow the backup slaves to have
3752 * different mtu values than the active slave we'll
3753 * need to change their mtu when doing a failover. That
3754 * means changing their mtu from timer context, which
3755 * is probably not a good idea.
3757 pr_debug("err %d %s\n", res, slave->dev->name);
3762 bond_dev->mtu = new_mtu;
3767 /* unwind from head to the slave that failed */
3769 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3772 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3774 pr_debug("unwind err %d dev %s\n",
3775 tmp_res, slave->dev->name);
3785 * Note that many devices must be down to change the HW address, and
3786 * downing the master releases all slaves. We can make bonds full of
3787 * bonding devices to test this, however.
3789 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3791 struct bonding *bond = netdev_priv(bond_dev);
3792 struct sockaddr *sa = addr, tmp_sa;
3793 struct slave *slave, *stop_at;
3797 if (bond->params.mode == BOND_MODE_ALB)
3798 return bond_alb_set_mac_address(bond_dev, addr);
3801 pr_debug("bond=%p, name=%s\n",
3802 bond, bond_dev ? bond_dev->name : "None");
3805 * If fail_over_mac is set to active, do nothing and return
3806 * success. Returning an error causes ifenslave to fail.
3808 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3811 if (!is_valid_ether_addr(sa->sa_data))
3812 return -EADDRNOTAVAIL;
3814 /* Can't hold bond->lock with bh disabled here since
3815 * some base drivers panic. On the other hand we can't
3816 * hold bond->lock without bh disabled because we'll
3817 * deadlock. The only solution is to rely on the fact
3818 * that we're under rtnl_lock here, and the slaves
3819 * list won't change. This doesn't solve the problem
3820 * of setting the slave's hw address while it is
3821 * transmitting, but the assumption is that the base
3822 * driver can handle that.
3824 * TODO: figure out a way to safely iterate the slaves
3825 * list, but without holding a lock around the actual
3826 * call to the base driver.
3829 bond_for_each_slave(bond, slave, i) {
3830 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3831 pr_debug("slave %p %s\n", slave, slave->dev->name);
3833 if (slave_ops->ndo_set_mac_address == NULL) {
3835 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3839 res = dev_set_mac_address(slave->dev, addr);
3841 /* TODO: consider downing the slave
3843 * User should expect communications
3844 * breakage anyway until ARP finish
3847 pr_debug("err %d %s\n", res, slave->dev->name);
3853 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3857 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3858 tmp_sa.sa_family = bond_dev->type;
3860 /* unwind from head to the slave that failed */
3862 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3865 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3867 pr_debug("unwind err %d dev %s\n",
3868 tmp_res, slave->dev->name);
3875 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3877 struct bonding *bond = netdev_priv(bond_dev);
3878 struct slave *slave, *start_at;
3879 int i, slave_no, res = 1;
3880 struct iphdr *iph = ip_hdr(skb);
3883 * Start with the curr_active_slave that joined the bond as the
3884 * default for sending IGMP traffic. For failover purposes one
3885 * needs to maintain some consistency for the interface that will
3886 * send the join/membership reports. The curr_active_slave found
3887 * will send all of this type of traffic.
3889 if ((iph->protocol == IPPROTO_IGMP) &&
3890 (skb->protocol == htons(ETH_P_IP))) {
3892 read_lock(&bond->curr_slave_lock);
3893 slave = bond->curr_active_slave;
3894 read_unlock(&bond->curr_slave_lock);
3900 * Concurrent TX may collide on rr_tx_counter; we accept
3901 * that as being rare enough not to justify using an
3904 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3906 bond_for_each_slave(bond, slave, i) {
3914 bond_for_each_slave_from(bond, slave, i, start_at) {
3915 if (IS_UP(slave->dev) &&
3916 (slave->link == BOND_LINK_UP) &&
3917 bond_is_active_slave(slave)) {
3918 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3925 /* no suitable interface, frame not sent */
3929 return NETDEV_TX_OK;
3934 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3935 * the bond has a usable interface.
3937 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3939 struct bonding *bond = netdev_priv(bond_dev);
3942 read_lock(&bond->curr_slave_lock);
3944 if (bond->curr_active_slave)
3945 res = bond_dev_queue_xmit(bond, skb,
3946 bond->curr_active_slave->dev);
3949 /* no suitable interface, frame not sent */
3952 read_unlock(&bond->curr_slave_lock);
3954 return NETDEV_TX_OK;
3958 * In bond_xmit_xor() , we determine the output device by using a pre-
3959 * determined xmit_hash_policy(), If the selected device is not enabled,
3960 * find the next active slave.
3962 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3964 struct bonding *bond = netdev_priv(bond_dev);
3965 struct slave *slave, *start_at;
3970 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3972 bond_for_each_slave(bond, slave, i) {
3980 bond_for_each_slave_from(bond, slave, i, start_at) {
3981 if (IS_UP(slave->dev) &&
3982 (slave->link == BOND_LINK_UP) &&
3983 bond_is_active_slave(slave)) {
3984 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3990 /* no suitable interface, frame not sent */
3994 return NETDEV_TX_OK;
3998 * in broadcast mode, we send everything to all usable interfaces.
4000 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4002 struct bonding *bond = netdev_priv(bond_dev);
4003 struct slave *slave, *start_at;
4004 struct net_device *tx_dev = NULL;
4008 read_lock(&bond->curr_slave_lock);
4009 start_at = bond->curr_active_slave;
4010 read_unlock(&bond->curr_slave_lock);
4015 bond_for_each_slave_from(bond, slave, i, start_at) {
4016 if (IS_UP(slave->dev) &&
4017 (slave->link == BOND_LINK_UP) &&
4018 bond_is_active_slave(slave)) {
4020 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4022 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4027 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4029 dev_kfree_skb(skb2);
4033 tx_dev = slave->dev;
4038 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4042 /* no suitable interface, frame not sent */
4045 /* frame sent to all suitable interfaces */
4046 return NETDEV_TX_OK;
4049 /*------------------------- Device initialization ---------------------------*/
4051 static void bond_set_xmit_hash_policy(struct bonding *bond)
4053 switch (bond->params.xmit_policy) {
4054 case BOND_XMIT_POLICY_LAYER23:
4055 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4057 case BOND_XMIT_POLICY_LAYER34:
4058 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4060 case BOND_XMIT_POLICY_LAYER2:
4062 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4068 * Lookup the slave that corresponds to a qid
4070 static inline int bond_slave_override(struct bonding *bond,
4071 struct sk_buff *skb)
4074 struct slave *slave = NULL;
4075 struct slave *check_slave;
4077 if (!skb->queue_mapping)
4080 /* Find out if any slaves have the same mapping as this skb. */
4081 bond_for_each_slave(bond, check_slave, i) {
4082 if (check_slave->queue_id == skb->queue_mapping) {
4083 slave = check_slave;
4088 /* If the slave isn't UP, use default transmit policy. */
4089 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4090 (slave->link == BOND_LINK_UP)) {
4091 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4098 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4101 * This helper function exists to help dev_pick_tx get the correct
4102 * destination queue. Using a helper function skips a call to
4103 * skb_tx_hash and will put the skbs in the queue we expect on their
4104 * way down to the bonding driver.
4106 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4109 * Save the original txq to restore before passing to the driver
4111 bond_queue_mapping(skb) = skb->queue_mapping;
4113 if (unlikely(txq >= dev->real_num_tx_queues)) {
4115 txq -= dev->real_num_tx_queues;
4116 } while (txq >= dev->real_num_tx_queues);
4121 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4123 struct bonding *bond = netdev_priv(dev);
4125 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4126 if (!bond_slave_override(bond, skb))
4127 return NETDEV_TX_OK;
4130 switch (bond->params.mode) {
4131 case BOND_MODE_ROUNDROBIN:
4132 return bond_xmit_roundrobin(skb, dev);
4133 case BOND_MODE_ACTIVEBACKUP:
4134 return bond_xmit_activebackup(skb, dev);
4136 return bond_xmit_xor(skb, dev);
4137 case BOND_MODE_BROADCAST:
4138 return bond_xmit_broadcast(skb, dev);
4139 case BOND_MODE_8023AD:
4140 return bond_3ad_xmit_xor(skb, dev);
4143 return bond_alb_xmit(skb, dev);
4145 /* Should never happen, mode already checked */
4146 pr_err("%s: Error: Unknown bonding mode %d\n",
4147 dev->name, bond->params.mode);
4150 return NETDEV_TX_OK;
4154 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4156 struct bonding *bond = netdev_priv(dev);
4157 netdev_tx_t ret = NETDEV_TX_OK;
4160 * If we risk deadlock from transmitting this in the
4161 * netpoll path, tell netpoll to queue the frame for later tx
4163 if (is_netpoll_tx_blocked(dev))
4164 return NETDEV_TX_BUSY;
4166 read_lock(&bond->lock);
4168 if (bond->slave_cnt)
4169 ret = __bond_start_xmit(skb, dev);
4173 read_unlock(&bond->lock);
4179 * set bond mode specific net device operations
4181 void bond_set_mode_ops(struct bonding *bond, int mode)
4183 struct net_device *bond_dev = bond->dev;
4186 case BOND_MODE_ROUNDROBIN:
4188 case BOND_MODE_ACTIVEBACKUP:
4191 bond_set_xmit_hash_policy(bond);
4193 case BOND_MODE_BROADCAST:
4195 case BOND_MODE_8023AD:
4196 bond_set_xmit_hash_policy(bond);
4203 /* Should never happen, mode already checked */
4204 pr_err("%s: Error: Unknown bonding mode %d\n",
4205 bond_dev->name, mode);
4210 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4211 struct ethtool_drvinfo *drvinfo)
4213 strncpy(drvinfo->driver, DRV_NAME, 32);
4214 strncpy(drvinfo->version, DRV_VERSION, 32);
4215 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4218 static const struct ethtool_ops bond_ethtool_ops = {
4219 .get_drvinfo = bond_ethtool_get_drvinfo,
4220 .get_link = ethtool_op_get_link,
4223 static const struct net_device_ops bond_netdev_ops = {
4224 .ndo_init = bond_init,
4225 .ndo_uninit = bond_uninit,
4226 .ndo_open = bond_open,
4227 .ndo_stop = bond_close,
4228 .ndo_start_xmit = bond_start_xmit,
4229 .ndo_select_queue = bond_select_queue,
4230 .ndo_get_stats64 = bond_get_stats,
4231 .ndo_do_ioctl = bond_do_ioctl,
4232 .ndo_change_rx_flags = bond_change_rx_flags,
4233 .ndo_set_rx_mode = bond_set_multicast_list,
4234 .ndo_change_mtu = bond_change_mtu,
4235 .ndo_set_mac_address = bond_set_mac_address,
4236 .ndo_neigh_setup = bond_neigh_setup,
4237 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4238 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4239 #ifdef CONFIG_NET_POLL_CONTROLLER
4240 .ndo_netpoll_setup = bond_netpoll_setup,
4241 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4242 .ndo_poll_controller = bond_poll_controller,
4244 .ndo_add_slave = bond_enslave,
4245 .ndo_del_slave = bond_release,
4246 .ndo_fix_features = bond_fix_features,
4249 static void bond_destructor(struct net_device *bond_dev)
4251 struct bonding *bond = netdev_priv(bond_dev);
4253 destroy_workqueue(bond->wq);
4254 free_netdev(bond_dev);
4257 static void bond_setup(struct net_device *bond_dev)
4259 struct bonding *bond = netdev_priv(bond_dev);
4261 /* initialize rwlocks */
4262 rwlock_init(&bond->lock);
4263 rwlock_init(&bond->curr_slave_lock);
4265 bond->params = bonding_defaults;
4267 /* Initialize pointers */
4268 bond->dev = bond_dev;
4269 INIT_LIST_HEAD(&bond->vlan_list);
4271 /* Initialize the device entry points */
4272 ether_setup(bond_dev);
4273 bond_dev->netdev_ops = &bond_netdev_ops;
4274 bond_dev->ethtool_ops = &bond_ethtool_ops;
4275 bond_set_mode_ops(bond, bond->params.mode);
4277 bond_dev->destructor = bond_destructor;
4279 /* Initialize the device options */
4280 bond_dev->tx_queue_len = 0;
4281 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4282 bond_dev->priv_flags |= IFF_BONDING;
4283 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4285 /* At first, we block adding VLANs. That's the only way to
4286 * prevent problems that occur when adding VLANs over an
4287 * empty bond. The block will be removed once non-challenged
4288 * slaves are enslaved.
4290 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4292 /* don't acquire bond device's netif_tx_lock when
4294 bond_dev->features |= NETIF_F_LLTX;
4296 /* By default, we declare the bond to be fully
4297 * VLAN hardware accelerated capable. Special
4298 * care is taken in the various xmit functions
4299 * when there are slaves that are not hw accel
4303 bond_dev->hw_features = BOND_VLAN_FEATURES |
4304 NETIF_F_HW_VLAN_TX |
4305 NETIF_F_HW_VLAN_RX |
4306 NETIF_F_HW_VLAN_FILTER;
4308 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4309 bond_dev->features |= bond_dev->hw_features;
4312 static void bond_work_cancel_all(struct bonding *bond)
4314 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4315 cancel_delayed_work_sync(&bond->mii_work);
4317 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4318 cancel_delayed_work_sync(&bond->arp_work);
4320 if (bond->params.mode == BOND_MODE_ALB &&
4321 delayed_work_pending(&bond->alb_work))
4322 cancel_delayed_work_sync(&bond->alb_work);
4324 if (bond->params.mode == BOND_MODE_8023AD &&
4325 delayed_work_pending(&bond->ad_work))
4326 cancel_delayed_work_sync(&bond->ad_work);
4328 if (delayed_work_pending(&bond->mcast_work))
4329 cancel_delayed_work_sync(&bond->mcast_work);
4333 * Destroy a bonding device.
4334 * Must be under rtnl_lock when this function is called.
4336 static void bond_uninit(struct net_device *bond_dev)
4338 struct bonding *bond = netdev_priv(bond_dev);
4339 struct vlan_entry *vlan, *tmp;
4341 bond_netpoll_cleanup(bond_dev);
4343 /* Release the bonded slaves */
4344 bond_release_all(bond_dev);
4346 list_del(&bond->bond_list);
4348 bond_work_cancel_all(bond);
4350 bond_remove_proc_entry(bond);
4352 bond_debug_unregister(bond);
4354 __hw_addr_flush(&bond->mc_list);
4356 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4357 list_del(&vlan->vlan_list);
4362 /*------------------------- Module initialization ---------------------------*/
4365 * Convert string input module parms. Accept either the
4366 * number of the mode or its string name. A bit complicated because
4367 * some mode names are substrings of other names, and calls from sysfs
4368 * may have whitespace in the name (trailing newlines, for example).
4370 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4372 int modeint = -1, i, rv;
4373 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4375 for (p = (char *)buf; *p; p++)
4376 if (!(isdigit(*p) || isspace(*p)))
4380 rv = sscanf(buf, "%20s", modestr);
4382 rv = sscanf(buf, "%d", &modeint);
4387 for (i = 0; tbl[i].modename; i++) {
4388 if (modeint == tbl[i].mode)
4390 if (strcmp(modestr, tbl[i].modename) == 0)
4397 static int bond_check_params(struct bond_params *params)
4399 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4402 * Convert string parameters.
4405 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4406 if (bond_mode == -1) {
4407 pr_err("Error: Invalid bonding mode \"%s\"\n",
4408 mode == NULL ? "NULL" : mode);
4413 if (xmit_hash_policy) {
4414 if ((bond_mode != BOND_MODE_XOR) &&
4415 (bond_mode != BOND_MODE_8023AD)) {
4416 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4417 bond_mode_name(bond_mode));
4419 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4421 if (xmit_hashtype == -1) {
4422 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4423 xmit_hash_policy == NULL ? "NULL" :
4431 if (bond_mode != BOND_MODE_8023AD) {
4432 pr_info("lacp_rate param is irrelevant in mode %s\n",
4433 bond_mode_name(bond_mode));
4435 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4436 if (lacp_fast == -1) {
4437 pr_err("Error: Invalid lacp rate \"%s\"\n",
4438 lacp_rate == NULL ? "NULL" : lacp_rate);
4445 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4446 if (params->ad_select == -1) {
4447 pr_err("Error: Invalid ad_select \"%s\"\n",
4448 ad_select == NULL ? "NULL" : ad_select);
4452 if (bond_mode != BOND_MODE_8023AD) {
4453 pr_warning("ad_select param only affects 802.3ad mode\n");
4456 params->ad_select = BOND_AD_STABLE;
4459 if (max_bonds < 0) {
4460 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4461 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4462 max_bonds = BOND_DEFAULT_MAX_BONDS;
4466 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4467 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4468 miimon = BOND_LINK_MON_INTERV;
4472 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4477 if (downdelay < 0) {
4478 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4479 downdelay, INT_MAX);
4483 if ((use_carrier != 0) && (use_carrier != 1)) {
4484 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4489 if (num_peer_notif < 0 || num_peer_notif > 255) {
4490 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4495 /* reset values for 802.3ad */
4496 if (bond_mode == BOND_MODE_8023AD) {
4498 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");
4499 pr_warning("Forcing miimon to 100msec\n");
4504 if (tx_queues < 1 || tx_queues > 255) {
4505 pr_warning("Warning: tx_queues (%d) should be between "
4506 "1 and 255, resetting to %d\n",
4507 tx_queues, BOND_DEFAULT_TX_QUEUES);
4508 tx_queues = BOND_DEFAULT_TX_QUEUES;
4511 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4512 pr_warning("Warning: all_slaves_active module parameter (%d), "
4513 "not of valid value (0/1), so it was set to "
4514 "0\n", all_slaves_active);
4515 all_slaves_active = 0;
4518 if (resend_igmp < 0 || resend_igmp > 255) {
4519 pr_warning("Warning: resend_igmp (%d) should be between "
4520 "0 and 255, resetting to %d\n",
4521 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4522 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4525 /* reset values for TLB/ALB */
4526 if ((bond_mode == BOND_MODE_TLB) ||
4527 (bond_mode == BOND_MODE_ALB)) {
4529 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");
4530 pr_warning("Forcing miimon to 100msec\n");
4535 if (bond_mode == BOND_MODE_ALB) {
4536 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",
4541 if (updelay || downdelay) {
4542 /* just warn the user the up/down delay will have
4543 * no effect since miimon is zero...
4545 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",
4546 updelay, downdelay);
4549 /* don't allow arp monitoring */
4551 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4552 miimon, arp_interval);
4556 if ((updelay % miimon) != 0) {
4557 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4559 (updelay / miimon) * miimon);
4564 if ((downdelay % miimon) != 0) {
4565 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4567 (downdelay / miimon) * miimon);
4570 downdelay /= miimon;
4573 if (arp_interval < 0) {
4574 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4575 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4576 arp_interval = BOND_LINK_ARP_INTERV;
4579 for (arp_ip_count = 0;
4580 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4582 /* not complete check, but should be good enough to
4584 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4585 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4586 arp_ip_target[arp_ip_count]);
4589 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4590 arp_target[arp_ip_count] = ip;
4594 if (arp_interval && !arp_ip_count) {
4595 /* don't allow arping if no arp_ip_target given... */
4596 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4602 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4603 pr_err("arp_validate only supported in active-backup mode\n");
4606 if (!arp_interval) {
4607 pr_err("arp_validate requires arp_interval\n");
4611 arp_validate_value = bond_parse_parm(arp_validate,
4613 if (arp_validate_value == -1) {
4614 pr_err("Error: invalid arp_validate \"%s\"\n",
4615 arp_validate == NULL ? "NULL" : arp_validate);
4619 arp_validate_value = 0;
4622 pr_info("MII link monitoring set to %d ms\n", miimon);
4623 } else if (arp_interval) {
4626 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4628 arp_validate_tbl[arp_validate_value].modename,
4631 for (i = 0; i < arp_ip_count; i++)
4632 pr_info(" %s", arp_ip_target[i]);
4636 } else if (max_bonds) {
4637 /* miimon and arp_interval not set, we need one so things
4638 * work as expected, see bonding.txt for details
4640 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4643 if (primary && !USES_PRIMARY(bond_mode)) {
4644 /* currently, using a primary only makes sense
4645 * in active backup, TLB or ALB modes
4647 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4648 primary, bond_mode_name(bond_mode));
4652 if (primary && primary_reselect) {
4653 primary_reselect_value = bond_parse_parm(primary_reselect,
4655 if (primary_reselect_value == -1) {
4656 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4658 NULL ? "NULL" : primary_reselect);
4662 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4665 if (fail_over_mac) {
4666 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4668 if (fail_over_mac_value == -1) {
4669 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4670 arp_validate == NULL ? "NULL" : arp_validate);
4674 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4675 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4677 fail_over_mac_value = BOND_FOM_NONE;
4680 /* fill params struct with the proper values */
4681 params->mode = bond_mode;
4682 params->xmit_policy = xmit_hashtype;
4683 params->miimon = miimon;
4684 params->num_peer_notif = num_peer_notif;
4685 params->arp_interval = arp_interval;
4686 params->arp_validate = arp_validate_value;
4687 params->updelay = updelay;
4688 params->downdelay = downdelay;
4689 params->use_carrier = use_carrier;
4690 params->lacp_fast = lacp_fast;
4691 params->primary[0] = 0;
4692 params->primary_reselect = primary_reselect_value;
4693 params->fail_over_mac = fail_over_mac_value;
4694 params->tx_queues = tx_queues;
4695 params->all_slaves_active = all_slaves_active;
4696 params->resend_igmp = resend_igmp;
4697 params->min_links = min_links;
4700 strncpy(params->primary, primary, IFNAMSIZ);
4701 params->primary[IFNAMSIZ - 1] = 0;
4704 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4709 static struct lock_class_key bonding_netdev_xmit_lock_key;
4710 static struct lock_class_key bonding_netdev_addr_lock_key;
4712 static void bond_set_lockdep_class_one(struct net_device *dev,
4713 struct netdev_queue *txq,
4716 lockdep_set_class(&txq->_xmit_lock,
4717 &bonding_netdev_xmit_lock_key);
4720 static void bond_set_lockdep_class(struct net_device *dev)
4722 lockdep_set_class(&dev->addr_list_lock,
4723 &bonding_netdev_addr_lock_key);
4724 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4728 * Called from registration process
4730 static int bond_init(struct net_device *bond_dev)
4732 struct bonding *bond = netdev_priv(bond_dev);
4733 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4734 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4736 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4739 * Initialize locks that may be required during
4740 * en/deslave operations. All of the bond_open work
4741 * (of which this is part) should really be moved to
4742 * a phase prior to dev_open
4744 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4745 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4747 bond->wq = create_singlethread_workqueue(bond_dev->name);
4751 bond_set_lockdep_class(bond_dev);
4753 bond_create_proc_entry(bond);
4754 list_add_tail(&bond->bond_list, &bn->dev_list);
4756 bond_prepare_sysfs_group(bond);
4758 bond_debug_register(bond);
4760 __hw_addr_init(&bond->mc_list);
4764 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4766 if (tb[IFLA_ADDRESS]) {
4767 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4769 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4770 return -EADDRNOTAVAIL;
4775 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4776 unsigned int *num_queues,
4777 unsigned int *real_num_queues)
4779 *num_queues = tx_queues;
4783 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4785 .priv_size = sizeof(struct bonding),
4786 .setup = bond_setup,
4787 .validate = bond_validate,
4788 .get_tx_queues = bond_get_tx_queues,
4791 /* Create a new bond based on the specified name and bonding parameters.
4792 * If name is NULL, obtain a suitable "bond%d" name for us.
4793 * Caller must NOT hold rtnl_lock; we need to release it here before we
4794 * set up our sysfs entries.
4796 int bond_create(struct net *net, const char *name)
4798 struct net_device *bond_dev;
4803 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4804 name ? name : "bond%d",
4805 bond_setup, tx_queues);
4807 pr_err("%s: eek! can't alloc netdev!\n", name);
4812 dev_net_set(bond_dev, net);
4813 bond_dev->rtnl_link_ops = &bond_link_ops;
4815 res = register_netdevice(bond_dev);
4817 netif_carrier_off(bond_dev);
4821 bond_destructor(bond_dev);
4825 static int __net_init bond_net_init(struct net *net)
4827 struct bond_net *bn = net_generic(net, bond_net_id);
4830 INIT_LIST_HEAD(&bn->dev_list);
4832 bond_create_proc_dir(bn);
4833 bond_create_sysfs(bn);
4838 static void __net_exit bond_net_exit(struct net *net)
4840 struct bond_net *bn = net_generic(net, bond_net_id);
4842 bond_destroy_sysfs(bn);
4843 bond_destroy_proc_dir(bn);
4846 static struct pernet_operations bond_net_ops = {
4847 .init = bond_net_init,
4848 .exit = bond_net_exit,
4850 .size = sizeof(struct bond_net),
4853 static int __init bonding_init(void)
4858 pr_info("%s", bond_version);
4860 res = bond_check_params(&bonding_defaults);
4864 res = register_pernet_subsys(&bond_net_ops);
4868 res = rtnl_link_register(&bond_link_ops);
4872 bond_create_debugfs();
4874 for (i = 0; i < max_bonds; i++) {
4875 res = bond_create(&init_net, NULL);
4880 register_netdevice_notifier(&bond_netdev_notifier);
4884 rtnl_link_unregister(&bond_link_ops);
4886 unregister_pernet_subsys(&bond_net_ops);
4891 static void __exit bonding_exit(void)
4893 unregister_netdevice_notifier(&bond_netdev_notifier);
4895 bond_destroy_debugfs();
4897 rtnl_link_unregister(&bond_link_ops);
4898 unregister_pernet_subsys(&bond_net_ops);
4900 #ifdef CONFIG_NET_POLL_CONTROLLER
4902 * Make sure we don't have an imbalance on our netpoll blocking
4904 WARN_ON(atomic_read(&netpoll_block_tx));
4908 module_init(bonding_init);
4909 module_exit(bonding_exit);
4910 MODULE_LICENSE("GPL");
4911 MODULE_VERSION(DRV_VERSION);
4912 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4913 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4914 MODULE_ALIAS_RTNL_LINK("bond");