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);
897 write_unlock_bh(&bond->curr_slave_lock);
898 read_unlock(&bond->lock);
899 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
900 read_lock(&bond->lock);
901 write_lock_bh(&bond->curr_slave_lock);
904 case BOND_FOM_FOLLOW:
906 * if new_active && old_active, swap them
907 * if just old_active, do nothing (going to no active slave)
908 * if just new_active, set new_active to bond's MAC
913 write_unlock_bh(&bond->curr_slave_lock);
914 read_unlock(&bond->lock);
917 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
918 memcpy(saddr.sa_data, old_active->dev->dev_addr,
920 saddr.sa_family = new_active->dev->type;
922 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
923 saddr.sa_family = bond->dev->type;
926 rv = dev_set_mac_address(new_active->dev, &saddr);
928 pr_err("%s: Error %d setting MAC of slave %s\n",
929 bond->dev->name, -rv, new_active->dev->name);
936 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
937 saddr.sa_family = old_active->dev->type;
939 rv = dev_set_mac_address(old_active->dev, &saddr);
941 pr_err("%s: Error %d setting MAC of slave %s\n",
942 bond->dev->name, -rv, new_active->dev->name);
944 read_lock(&bond->lock);
945 write_lock_bh(&bond->curr_slave_lock);
948 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
949 bond->dev->name, bond->params.fail_over_mac);
955 static bool bond_should_change_active(struct bonding *bond)
957 struct slave *prim = bond->primary_slave;
958 struct slave *curr = bond->curr_active_slave;
960 if (!prim || !curr || curr->link != BOND_LINK_UP)
962 if (bond->force_primary) {
963 bond->force_primary = false;
966 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
967 (prim->speed < curr->speed ||
968 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
970 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
976 * find_best_interface - select the best available slave to be the active one
977 * @bond: our bonding struct
979 * Warning: Caller must hold curr_slave_lock for writing.
981 static struct slave *bond_find_best_slave(struct bonding *bond)
983 struct slave *new_active, *old_active;
984 struct slave *bestslave = NULL;
985 int mintime = bond->params.updelay;
988 new_active = bond->curr_active_slave;
990 if (!new_active) { /* there were no active slaves left */
991 if (bond->slave_cnt > 0) /* found one slave */
992 new_active = bond->first_slave;
994 return NULL; /* still no slave, return NULL */
997 if ((bond->primary_slave) &&
998 bond->primary_slave->link == BOND_LINK_UP &&
999 bond_should_change_active(bond)) {
1000 new_active = bond->primary_slave;
1003 /* remember where to stop iterating over the slaves */
1004 old_active = new_active;
1006 bond_for_each_slave_from(bond, new_active, i, old_active) {
1007 if (new_active->link == BOND_LINK_UP) {
1009 } else if (new_active->link == BOND_LINK_BACK &&
1010 IS_UP(new_active->dev)) {
1011 /* link up, but waiting for stabilization */
1012 if (new_active->delay < mintime) {
1013 mintime = new_active->delay;
1014 bestslave = new_active;
1022 static bool bond_should_notify_peers(struct bonding *bond)
1024 struct slave *slave = bond->curr_active_slave;
1026 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1027 bond->dev->name, slave ? slave->dev->name : "NULL");
1029 if (!slave || !bond->send_peer_notif ||
1030 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1033 bond->send_peer_notif--;
1038 * change_active_interface - change the active slave into the specified one
1039 * @bond: our bonding struct
1040 * @new: the new slave to make the active one
1042 * Set the new slave to the bond's settings and unset them on the old
1043 * curr_active_slave.
1044 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1046 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1047 * because it is apparently the best available slave we have, even though its
1048 * updelay hasn't timed out yet.
1050 * If new_active is not NULL, caller must hold bond->lock for read and
1051 * curr_slave_lock for write_bh.
1053 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1055 struct slave *old_active = bond->curr_active_slave;
1057 if (old_active == new_active)
1061 new_active->jiffies = jiffies;
1063 if (new_active->link == BOND_LINK_BACK) {
1064 if (USES_PRIMARY(bond->params.mode)) {
1065 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1066 bond->dev->name, new_active->dev->name,
1067 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1070 new_active->delay = 0;
1071 new_active->link = BOND_LINK_UP;
1073 if (bond->params.mode == BOND_MODE_8023AD)
1074 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1076 if (bond_is_lb(bond))
1077 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1079 if (USES_PRIMARY(bond->params.mode)) {
1080 pr_info("%s: making interface %s the new active one.\n",
1081 bond->dev->name, new_active->dev->name);
1086 if (USES_PRIMARY(bond->params.mode))
1087 bond_mc_swap(bond, new_active, old_active);
1089 if (bond_is_lb(bond)) {
1090 bond_alb_handle_active_change(bond, new_active);
1092 bond_set_slave_inactive_flags(old_active);
1094 bond_set_slave_active_flags(new_active);
1096 bond->curr_active_slave = new_active;
1099 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1101 bond_set_slave_inactive_flags(old_active);
1104 bool should_notify_peers = false;
1106 bond_set_slave_active_flags(new_active);
1108 if (bond->params.fail_over_mac)
1109 bond_do_fail_over_mac(bond, new_active,
1112 if (netif_running(bond->dev)) {
1113 bond->send_peer_notif =
1114 bond->params.num_peer_notif;
1115 should_notify_peers =
1116 bond_should_notify_peers(bond);
1119 write_unlock_bh(&bond->curr_slave_lock);
1120 read_unlock(&bond->lock);
1122 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1123 if (should_notify_peers)
1124 netdev_bonding_change(bond->dev,
1125 NETDEV_NOTIFY_PEERS);
1127 read_lock(&bond->lock);
1128 write_lock_bh(&bond->curr_slave_lock);
1132 /* resend IGMP joins since active slave has changed or
1133 * all were sent on curr_active_slave.
1134 * resend only if bond is brought up with the affected
1135 * bonding modes and the retransmission is enabled */
1136 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1137 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1138 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1139 bond->igmp_retrans = bond->params.resend_igmp;
1140 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1145 * bond_select_active_slave - select a new active slave, if needed
1146 * @bond: our bonding struct
1148 * This functions should be called when one of the following occurs:
1149 * - The old curr_active_slave has been released or lost its link.
1150 * - The primary_slave has got its link back.
1151 * - A slave has got its link back and there's no old curr_active_slave.
1153 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1155 void bond_select_active_slave(struct bonding *bond)
1157 struct slave *best_slave;
1160 best_slave = bond_find_best_slave(bond);
1161 if (best_slave != bond->curr_active_slave) {
1162 bond_change_active_slave(bond, best_slave);
1163 rv = bond_set_carrier(bond);
1167 if (netif_carrier_ok(bond->dev)) {
1168 pr_info("%s: first active interface up!\n",
1171 pr_info("%s: now running without any active interface !\n",
1177 /*--------------------------- slave list handling ---------------------------*/
1180 * This function attaches the slave to the end of list.
1182 * bond->lock held for writing by caller.
1184 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1186 if (bond->first_slave == NULL) { /* attaching the first slave */
1187 new_slave->next = new_slave;
1188 new_slave->prev = new_slave;
1189 bond->first_slave = new_slave;
1191 new_slave->next = bond->first_slave;
1192 new_slave->prev = bond->first_slave->prev;
1193 new_slave->next->prev = new_slave;
1194 new_slave->prev->next = new_slave;
1201 * This function detaches the slave from the list.
1202 * WARNING: no check is made to verify if the slave effectively
1203 * belongs to <bond>.
1204 * Nothing is freed on return, structures are just unchained.
1205 * If any slave pointer in bond was pointing to <slave>,
1206 * it should be changed by the calling function.
1208 * bond->lock held for writing by caller.
1210 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1213 slave->next->prev = slave->prev;
1216 slave->prev->next = slave->next;
1218 if (bond->first_slave == slave) { /* slave is the first slave */
1219 if (bond->slave_cnt > 1) { /* there are more slave */
1220 bond->first_slave = slave->next;
1222 bond->first_slave = NULL; /* slave was the last one */
1231 #ifdef CONFIG_NET_POLL_CONTROLLER
1232 static inline int slave_enable_netpoll(struct slave *slave)
1237 np = kzalloc(sizeof(*np), GFP_KERNEL);
1242 np->dev = slave->dev;
1243 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1244 err = __netpoll_setup(np);
1253 static inline void slave_disable_netpoll(struct slave *slave)
1255 struct netpoll *np = slave->np;
1261 synchronize_rcu_bh();
1262 __netpoll_cleanup(np);
1265 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1267 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1269 if (!slave_dev->netdev_ops->ndo_poll_controller)
1274 static void bond_poll_controller(struct net_device *bond_dev)
1278 static void __bond_netpoll_cleanup(struct bonding *bond)
1280 struct slave *slave;
1283 bond_for_each_slave(bond, slave, i)
1284 if (IS_UP(slave->dev))
1285 slave_disable_netpoll(slave);
1287 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1289 struct bonding *bond = netdev_priv(bond_dev);
1291 read_lock(&bond->lock);
1292 __bond_netpoll_cleanup(bond);
1293 read_unlock(&bond->lock);
1296 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1298 struct bonding *bond = netdev_priv(dev);
1299 struct slave *slave;
1302 read_lock(&bond->lock);
1303 bond_for_each_slave(bond, slave, i) {
1304 err = slave_enable_netpoll(slave);
1306 __bond_netpoll_cleanup(bond);
1310 read_unlock(&bond->lock);
1314 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1316 return bond->dev->npinfo;
1320 static inline int slave_enable_netpoll(struct slave *slave)
1324 static inline void slave_disable_netpoll(struct slave *slave)
1327 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1332 /*---------------------------------- IOCTL ----------------------------------*/
1334 static int bond_sethwaddr(struct net_device *bond_dev,
1335 struct net_device *slave_dev)
1337 pr_debug("bond_dev=%p\n", bond_dev);
1338 pr_debug("slave_dev=%p\n", slave_dev);
1339 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1340 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1344 static netdev_features_t bond_fix_features(struct net_device *dev,
1345 netdev_features_t features)
1347 struct slave *slave;
1348 struct bonding *bond = netdev_priv(dev);
1349 netdev_features_t mask;
1352 read_lock(&bond->lock);
1354 if (!bond->first_slave) {
1355 /* Disable adding VLANs to empty bond. But why? --mq */
1356 features |= NETIF_F_VLAN_CHALLENGED;
1361 features &= ~NETIF_F_ONE_FOR_ALL;
1362 features |= NETIF_F_ALL_FOR_ALL;
1364 bond_for_each_slave(bond, slave, i) {
1365 features = netdev_increment_features(features,
1366 slave->dev->features,
1371 read_unlock(&bond->lock);
1375 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1376 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1377 NETIF_F_HIGHDMA | NETIF_F_LRO)
1379 static void bond_compute_features(struct bonding *bond)
1381 struct slave *slave;
1382 struct net_device *bond_dev = bond->dev;
1383 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1384 unsigned short max_hard_header_len = ETH_HLEN;
1387 read_lock(&bond->lock);
1389 if (!bond->first_slave)
1392 bond_for_each_slave(bond, slave, i) {
1393 vlan_features = netdev_increment_features(vlan_features,
1394 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1396 if (slave->dev->hard_header_len > max_hard_header_len)
1397 max_hard_header_len = slave->dev->hard_header_len;
1401 bond_dev->vlan_features = vlan_features;
1402 bond_dev->hard_header_len = max_hard_header_len;
1404 read_unlock(&bond->lock);
1406 netdev_change_features(bond_dev);
1409 static void bond_setup_by_slave(struct net_device *bond_dev,
1410 struct net_device *slave_dev)
1412 struct bonding *bond = netdev_priv(bond_dev);
1414 bond_dev->header_ops = slave_dev->header_ops;
1416 bond_dev->type = slave_dev->type;
1417 bond_dev->hard_header_len = slave_dev->hard_header_len;
1418 bond_dev->addr_len = slave_dev->addr_len;
1420 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1421 slave_dev->addr_len);
1422 bond->setup_by_slave = 1;
1425 /* On bonding slaves other than the currently active slave, suppress
1426 * duplicates except for alb non-mcast/bcast.
1428 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1429 struct slave *slave,
1430 struct bonding *bond)
1432 if (bond_is_slave_inactive(slave)) {
1433 if (bond->params.mode == BOND_MODE_ALB &&
1434 skb->pkt_type != PACKET_BROADCAST &&
1435 skb->pkt_type != PACKET_MULTICAST)
1442 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1444 struct sk_buff *skb = *pskb;
1445 struct slave *slave;
1446 struct bonding *bond;
1447 int (*recv_probe)(struct sk_buff *, struct bonding *,
1449 int ret = RX_HANDLER_ANOTHER;
1451 skb = skb_share_check(skb, GFP_ATOMIC);
1453 return RX_HANDLER_CONSUMED;
1457 slave = bond_slave_get_rcu(skb->dev);
1460 if (bond->params.arp_interval)
1461 slave->dev->last_rx = jiffies;
1463 recv_probe = ACCESS_ONCE(bond->recv_probe);
1465 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1468 ret = recv_probe(nskb, bond, slave);
1469 dev_kfree_skb(nskb);
1470 if (ret == RX_HANDLER_CONSUMED) {
1477 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1478 return RX_HANDLER_EXACT;
1481 skb->dev = bond->dev;
1483 if (bond->params.mode == BOND_MODE_ALB &&
1484 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1485 skb->pkt_type == PACKET_HOST) {
1487 if (unlikely(skb_cow_head(skb,
1488 skb->data - skb_mac_header(skb)))) {
1490 return RX_HANDLER_CONSUMED;
1492 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1498 /* enslave device <slave> to bond device <master> */
1499 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1501 struct bonding *bond = netdev_priv(bond_dev);
1502 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1503 struct slave *new_slave = NULL;
1504 struct netdev_hw_addr *ha;
1505 struct sockaddr addr;
1509 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1510 slave_ops->ndo_do_ioctl == NULL) {
1511 pr_warning("%s: Warning: no link monitoring support for %s\n",
1512 bond_dev->name, slave_dev->name);
1515 /* already enslaved */
1516 if (slave_dev->flags & IFF_SLAVE) {
1517 pr_debug("Error, Device was already enslaved\n");
1521 /* vlan challenged mutual exclusion */
1522 /* no need to lock since we're protected by rtnl_lock */
1523 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1524 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1525 if (bond_vlan_used(bond)) {
1526 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1527 bond_dev->name, slave_dev->name, bond_dev->name);
1530 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1531 bond_dev->name, slave_dev->name,
1532 slave_dev->name, bond_dev->name);
1535 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1539 * Old ifenslave binaries are no longer supported. These can
1540 * be identified with moderate accuracy by the state of the slave:
1541 * the current ifenslave will set the interface down prior to
1542 * enslaving it; the old ifenslave will not.
1544 if ((slave_dev->flags & IFF_UP)) {
1545 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1548 goto err_undo_flags;
1551 /* set bonding device ether type by slave - bonding netdevices are
1552 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1553 * there is a need to override some of the type dependent attribs/funcs.
1555 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1556 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1558 if (bond->slave_cnt == 0) {
1559 if (bond_dev->type != slave_dev->type) {
1560 pr_debug("%s: change device type from %d to %d\n",
1562 bond_dev->type, slave_dev->type);
1564 res = netdev_bonding_change(bond_dev,
1565 NETDEV_PRE_TYPE_CHANGE);
1566 res = notifier_to_errno(res);
1568 pr_err("%s: refused to change device type\n",
1571 goto err_undo_flags;
1574 /* Flush unicast and multicast addresses */
1575 dev_uc_flush(bond_dev);
1576 dev_mc_flush(bond_dev);
1578 if (slave_dev->type != ARPHRD_ETHER)
1579 bond_setup_by_slave(bond_dev, slave_dev);
1581 ether_setup(bond_dev);
1582 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1585 netdev_bonding_change(bond_dev,
1586 NETDEV_POST_TYPE_CHANGE);
1588 } else if (bond_dev->type != slave_dev->type) {
1589 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1591 slave_dev->type, bond_dev->type);
1593 goto err_undo_flags;
1596 if (slave_ops->ndo_set_mac_address == NULL) {
1597 if (bond->slave_cnt == 0) {
1598 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1600 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1601 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1602 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",
1605 goto err_undo_flags;
1609 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1611 /* If this is the first slave, then we need to set the master's hardware
1612 * address to be the same as the slave's. */
1613 if (is_zero_ether_addr(bond->dev->dev_addr))
1614 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1615 slave_dev->addr_len);
1618 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1621 goto err_undo_flags;
1625 * Set the new_slave's queue_id to be zero. Queue ID mapping
1626 * is set via sysfs or module option if desired.
1628 new_slave->queue_id = 0;
1630 /* Save slave's original mtu and then set it to match the bond */
1631 new_slave->original_mtu = slave_dev->mtu;
1632 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1634 pr_debug("Error %d calling dev_set_mtu\n", res);
1639 * Save slave's original ("permanent") mac address for modes
1640 * that need it, and for restoring it upon release, and then
1641 * set it to the master's address
1643 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1645 if (!bond->params.fail_over_mac) {
1647 * Set slave to master's mac address. The application already
1648 * set the master's mac address to that of the first slave
1650 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1651 addr.sa_family = slave_dev->type;
1652 res = dev_set_mac_address(slave_dev, &addr);
1654 pr_debug("Error %d calling set_mac_address\n", res);
1655 goto err_restore_mtu;
1659 res = netdev_set_bond_master(slave_dev, bond_dev);
1661 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1662 goto err_restore_mac;
1665 /* open the slave since the application closed it */
1666 res = dev_open(slave_dev);
1668 pr_debug("Opening slave %s failed\n", slave_dev->name);
1669 goto err_unset_master;
1672 new_slave->bond = bond;
1673 new_slave->dev = slave_dev;
1674 slave_dev->priv_flags |= IFF_BONDING;
1676 if (bond_is_lb(bond)) {
1677 /* bond_alb_init_slave() must be called before all other stages since
1678 * it might fail and we do not want to have to undo everything
1680 res = bond_alb_init_slave(bond, new_slave);
1685 /* If the mode USES_PRIMARY, then the new slave gets the
1686 * master's promisc (and mc) settings only if it becomes the
1687 * curr_active_slave, and that is taken care of later when calling
1688 * bond_change_active()
1690 if (!USES_PRIMARY(bond->params.mode)) {
1691 /* set promiscuity level to new slave */
1692 if (bond_dev->flags & IFF_PROMISC) {
1693 res = dev_set_promiscuity(slave_dev, 1);
1698 /* set allmulti level to new slave */
1699 if (bond_dev->flags & IFF_ALLMULTI) {
1700 res = dev_set_allmulti(slave_dev, 1);
1705 netif_addr_lock_bh(bond_dev);
1706 /* upload master's mc_list to new slave */
1707 netdev_for_each_mc_addr(ha, bond_dev)
1708 dev_mc_add(slave_dev, ha->addr);
1709 netif_addr_unlock_bh(bond_dev);
1712 if (bond->params.mode == BOND_MODE_8023AD) {
1713 /* add lacpdu mc addr to mc list */
1714 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1716 dev_mc_add(slave_dev, lacpdu_multicast);
1719 bond_add_vlans_on_slave(bond, slave_dev);
1721 write_lock_bh(&bond->lock);
1723 bond_attach_slave(bond, new_slave);
1725 new_slave->delay = 0;
1726 new_slave->link_failure_count = 0;
1728 write_unlock_bh(&bond->lock);
1730 bond_compute_features(bond);
1732 read_lock(&bond->lock);
1734 new_slave->last_arp_rx = jiffies;
1736 if (bond->params.miimon && !bond->params.use_carrier) {
1737 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1739 if ((link_reporting == -1) && !bond->params.arp_interval) {
1741 * miimon is set but a bonded network driver
1742 * does not support ETHTOOL/MII and
1743 * arp_interval is not set. Note: if
1744 * use_carrier is enabled, we will never go
1745 * here (because netif_carrier is always
1746 * supported); thus, we don't need to change
1747 * the messages for netif_carrier.
1749 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",
1750 bond_dev->name, slave_dev->name);
1751 } else if (link_reporting == -1) {
1752 /* unable get link status using mii/ethtool */
1753 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",
1754 bond_dev->name, slave_dev->name);
1758 /* check for initial state */
1759 if (!bond->params.miimon ||
1760 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1761 if (bond->params.updelay) {
1762 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1763 new_slave->link = BOND_LINK_BACK;
1764 new_slave->delay = bond->params.updelay;
1766 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1767 new_slave->link = BOND_LINK_UP;
1769 new_slave->jiffies = jiffies;
1771 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1772 new_slave->link = BOND_LINK_DOWN;
1775 bond_update_speed_duplex(new_slave);
1777 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1778 /* if there is a primary slave, remember it */
1779 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1780 bond->primary_slave = new_slave;
1781 bond->force_primary = true;
1785 write_lock_bh(&bond->curr_slave_lock);
1787 switch (bond->params.mode) {
1788 case BOND_MODE_ACTIVEBACKUP:
1789 bond_set_slave_inactive_flags(new_slave);
1790 bond_select_active_slave(bond);
1792 case BOND_MODE_8023AD:
1793 /* in 802.3ad mode, the internal mechanism
1794 * will activate the slaves in the selected
1797 bond_set_slave_inactive_flags(new_slave);
1798 /* if this is the first slave */
1799 if (bond->slave_cnt == 1) {
1800 SLAVE_AD_INFO(new_slave).id = 1;
1801 /* Initialize AD with the number of times that the AD timer is called in 1 second
1802 * can be called only after the mac address of the bond is set
1804 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1806 SLAVE_AD_INFO(new_slave).id =
1807 SLAVE_AD_INFO(new_slave->prev).id + 1;
1810 bond_3ad_bind_slave(new_slave);
1814 bond_set_active_slave(new_slave);
1815 bond_set_slave_inactive_flags(new_slave);
1816 bond_select_active_slave(bond);
1819 pr_debug("This slave is always active in trunk mode\n");
1821 /* always active in trunk mode */
1822 bond_set_active_slave(new_slave);
1824 /* In trunking mode there is little meaning to curr_active_slave
1825 * anyway (it holds no special properties of the bond device),
1826 * so we can change it without calling change_active_interface()
1828 if (!bond->curr_active_slave)
1829 bond->curr_active_slave = new_slave;
1832 } /* switch(bond_mode) */
1834 write_unlock_bh(&bond->curr_slave_lock);
1836 bond_set_carrier(bond);
1838 #ifdef CONFIG_NET_POLL_CONTROLLER
1839 slave_dev->npinfo = bond_netpoll_info(bond);
1840 if (slave_dev->npinfo) {
1841 if (slave_enable_netpoll(new_slave)) {
1842 read_unlock(&bond->lock);
1843 pr_info("Error, %s: master_dev is using netpoll, "
1844 "but new slave device does not support netpoll.\n",
1852 read_unlock(&bond->lock);
1854 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1858 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1861 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1862 goto err_dest_symlinks;
1865 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1866 bond_dev->name, slave_dev->name,
1867 bond_is_active_slave(new_slave) ? "n active" : " backup",
1868 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1870 /* enslave is successful */
1873 /* Undo stages on error */
1875 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1878 write_lock_bh(&bond->lock);
1879 bond_detach_slave(bond, new_slave);
1880 write_unlock_bh(&bond->lock);
1883 dev_close(slave_dev);
1886 netdev_set_bond_master(slave_dev, NULL);
1889 if (!bond->params.fail_over_mac) {
1890 /* XXX TODO - fom follow mode needs to change master's
1891 * MAC if this slave's MAC is in use by the bond, or at
1892 * least print a warning.
1894 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1895 addr.sa_family = slave_dev->type;
1896 dev_set_mac_address(slave_dev, &addr);
1900 dev_set_mtu(slave_dev, new_slave->original_mtu);
1906 bond_compute_features(bond);
1912 * Try to release the slave device <slave> from the bond device <master>
1913 * It is legal to access curr_active_slave without a lock because all the function
1916 * The rules for slave state should be:
1917 * for Active/Backup:
1918 * Active stays on all backups go down
1919 * for Bonded connections:
1920 * The first up interface should be left on and all others downed.
1922 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1924 struct bonding *bond = netdev_priv(bond_dev);
1925 struct slave *slave, *oldcurrent;
1926 struct sockaddr addr;
1927 netdev_features_t old_features = bond_dev->features;
1929 /* slave is not a slave or master is not master of this slave */
1930 if (!(slave_dev->flags & IFF_SLAVE) ||
1931 (slave_dev->master != bond_dev)) {
1932 pr_err("%s: Error: cannot release %s.\n",
1933 bond_dev->name, slave_dev->name);
1938 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1939 write_lock_bh(&bond->lock);
1941 slave = bond_get_slave_by_dev(bond, slave_dev);
1943 /* not a slave of this bond */
1944 pr_info("%s: %s not enslaved\n",
1945 bond_dev->name, slave_dev->name);
1946 write_unlock_bh(&bond->lock);
1947 unblock_netpoll_tx();
1951 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1952 * for this slave anymore.
1954 netdev_rx_handler_unregister(slave_dev);
1955 write_unlock_bh(&bond->lock);
1957 write_lock_bh(&bond->lock);
1959 if (!bond->params.fail_over_mac) {
1960 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1961 bond->slave_cnt > 1)
1962 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",
1963 bond_dev->name, slave_dev->name,
1965 bond_dev->name, slave_dev->name);
1968 /* Inform AD package of unbinding of slave. */
1969 if (bond->params.mode == BOND_MODE_8023AD) {
1970 /* must be called before the slave is
1971 * detached from the list
1973 bond_3ad_unbind_slave(slave);
1976 pr_info("%s: releasing %s interface %s\n",
1978 bond_is_active_slave(slave) ? "active" : "backup",
1981 oldcurrent = bond->curr_active_slave;
1983 bond->current_arp_slave = NULL;
1985 /* release the slave from its bond */
1986 bond_detach_slave(bond, slave);
1988 if (bond->primary_slave == slave)
1989 bond->primary_slave = NULL;
1991 if (oldcurrent == slave)
1992 bond_change_active_slave(bond, NULL);
1994 if (bond_is_lb(bond)) {
1995 /* Must be called only after the slave has been
1996 * detached from the list and the curr_active_slave
1997 * has been cleared (if our_slave == old_current),
1998 * but before a new active slave is selected.
2000 write_unlock_bh(&bond->lock);
2001 bond_alb_deinit_slave(bond, slave);
2002 write_lock_bh(&bond->lock);
2005 if (oldcurrent == slave) {
2007 * Note that we hold RTNL over this sequence, so there
2008 * is no concern that another slave add/remove event
2011 write_unlock_bh(&bond->lock);
2012 read_lock(&bond->lock);
2013 write_lock_bh(&bond->curr_slave_lock);
2015 bond_select_active_slave(bond);
2017 write_unlock_bh(&bond->curr_slave_lock);
2018 read_unlock(&bond->lock);
2019 write_lock_bh(&bond->lock);
2022 if (bond->slave_cnt == 0) {
2023 bond_set_carrier(bond);
2025 /* if the last slave was removed, zero the mac address
2026 * of the master so it will be set by the application
2027 * to the mac address of the first slave
2029 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2031 if (bond_vlan_used(bond)) {
2032 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2033 bond_dev->name, bond_dev->name);
2034 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2039 write_unlock_bh(&bond->lock);
2040 unblock_netpoll_tx();
2042 if (bond->slave_cnt == 0)
2043 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2045 bond_compute_features(bond);
2046 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2047 (old_features & NETIF_F_VLAN_CHALLENGED))
2048 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2049 bond_dev->name, slave_dev->name, bond_dev->name);
2051 /* must do this from outside any spinlocks */
2052 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2054 bond_del_vlans_from_slave(bond, slave_dev);
2056 /* If the mode USES_PRIMARY, then we should only remove its
2057 * promisc and mc settings if it was the curr_active_slave, but that was
2058 * already taken care of above when we detached the slave
2060 if (!USES_PRIMARY(bond->params.mode)) {
2061 /* unset promiscuity level from slave */
2062 if (bond_dev->flags & IFF_PROMISC)
2063 dev_set_promiscuity(slave_dev, -1);
2065 /* unset allmulti level from slave */
2066 if (bond_dev->flags & IFF_ALLMULTI)
2067 dev_set_allmulti(slave_dev, -1);
2069 /* flush master's mc_list from slave */
2070 netif_addr_lock_bh(bond_dev);
2071 bond_mc_list_flush(bond_dev, slave_dev);
2072 netif_addr_unlock_bh(bond_dev);
2075 netdev_set_bond_master(slave_dev, NULL);
2077 slave_disable_netpoll(slave);
2079 /* close slave before restoring its mac address */
2080 dev_close(slave_dev);
2082 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2083 /* restore original ("permanent") mac address */
2084 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2085 addr.sa_family = slave_dev->type;
2086 dev_set_mac_address(slave_dev, &addr);
2089 dev_set_mtu(slave_dev, slave->original_mtu);
2091 slave_dev->priv_flags &= ~IFF_BONDING;
2095 return 0; /* deletion OK */
2099 * First release a slave and then destroy the bond if no more slaves are left.
2100 * Must be under rtnl_lock when this function is called.
2102 static int bond_release_and_destroy(struct net_device *bond_dev,
2103 struct net_device *slave_dev)
2105 struct bonding *bond = netdev_priv(bond_dev);
2108 ret = bond_release(bond_dev, slave_dev);
2109 if ((ret == 0) && (bond->slave_cnt == 0)) {
2110 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2111 pr_info("%s: destroying bond %s.\n",
2112 bond_dev->name, bond_dev->name);
2113 unregister_netdevice(bond_dev);
2119 * This function releases all slaves.
2121 static int bond_release_all(struct net_device *bond_dev)
2123 struct bonding *bond = netdev_priv(bond_dev);
2124 struct slave *slave;
2125 struct net_device *slave_dev;
2126 struct sockaddr addr;
2128 write_lock_bh(&bond->lock);
2130 netif_carrier_off(bond_dev);
2132 if (bond->slave_cnt == 0)
2135 bond->current_arp_slave = NULL;
2136 bond->primary_slave = NULL;
2137 bond_change_active_slave(bond, NULL);
2139 while ((slave = bond->first_slave) != NULL) {
2140 /* Inform AD package of unbinding of slave
2141 * before slave is detached from the list.
2143 if (bond->params.mode == BOND_MODE_8023AD)
2144 bond_3ad_unbind_slave(slave);
2146 slave_dev = slave->dev;
2147 bond_detach_slave(bond, slave);
2149 /* now that the slave is detached, unlock and perform
2150 * all the undo steps that should not be called from
2153 write_unlock_bh(&bond->lock);
2155 /* unregister rx_handler early so bond_handle_frame wouldn't
2156 * be called for this slave anymore.
2158 netdev_rx_handler_unregister(slave_dev);
2161 if (bond_is_lb(bond)) {
2162 /* must be called only after the slave
2163 * has been detached from the list
2165 bond_alb_deinit_slave(bond, slave);
2168 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2169 bond_del_vlans_from_slave(bond, slave_dev);
2171 /* If the mode USES_PRIMARY, then we should only remove its
2172 * promisc and mc settings if it was the curr_active_slave, but that was
2173 * already taken care of above when we detached the slave
2175 if (!USES_PRIMARY(bond->params.mode)) {
2176 /* unset promiscuity level from slave */
2177 if (bond_dev->flags & IFF_PROMISC)
2178 dev_set_promiscuity(slave_dev, -1);
2180 /* unset allmulti level from slave */
2181 if (bond_dev->flags & IFF_ALLMULTI)
2182 dev_set_allmulti(slave_dev, -1);
2184 /* flush master's mc_list from slave */
2185 netif_addr_lock_bh(bond_dev);
2186 bond_mc_list_flush(bond_dev, slave_dev);
2187 netif_addr_unlock_bh(bond_dev);
2190 netdev_set_bond_master(slave_dev, NULL);
2192 slave_disable_netpoll(slave);
2194 /* close slave before restoring its mac address */
2195 dev_close(slave_dev);
2197 if (!bond->params.fail_over_mac) {
2198 /* restore original ("permanent") mac address*/
2199 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2200 addr.sa_family = slave_dev->type;
2201 dev_set_mac_address(slave_dev, &addr);
2206 /* re-acquire the lock before getting the next slave */
2207 write_lock_bh(&bond->lock);
2210 /* zero the mac address of the master so it will be
2211 * set by the application to the mac address of the
2214 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2216 if (bond_vlan_used(bond)) {
2217 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2218 bond_dev->name, bond_dev->name);
2219 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2223 pr_info("%s: released all slaves\n", bond_dev->name);
2226 write_unlock_bh(&bond->lock);
2228 bond_compute_features(bond);
2234 * This function changes the active slave to slave <slave_dev>.
2235 * It returns -EINVAL in the following cases.
2236 * - <slave_dev> is not found in the list.
2237 * - There is not active slave now.
2238 * - <slave_dev> is already active.
2239 * - The link state of <slave_dev> is not BOND_LINK_UP.
2240 * - <slave_dev> is not running.
2241 * In these cases, this function does nothing.
2242 * In the other cases, current_slave pointer is changed and 0 is returned.
2244 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2246 struct bonding *bond = netdev_priv(bond_dev);
2247 struct slave *old_active = NULL;
2248 struct slave *new_active = NULL;
2251 if (!USES_PRIMARY(bond->params.mode))
2254 /* Verify that master_dev is indeed the master of slave_dev */
2255 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2258 read_lock(&bond->lock);
2260 read_lock(&bond->curr_slave_lock);
2261 old_active = bond->curr_active_slave;
2262 read_unlock(&bond->curr_slave_lock);
2264 new_active = bond_get_slave_by_dev(bond, slave_dev);
2267 * Changing to the current active: do nothing; return success.
2269 if (new_active && (new_active == old_active)) {
2270 read_unlock(&bond->lock);
2276 (new_active->link == BOND_LINK_UP) &&
2277 IS_UP(new_active->dev)) {
2279 write_lock_bh(&bond->curr_slave_lock);
2280 bond_change_active_slave(bond, new_active);
2281 write_unlock_bh(&bond->curr_slave_lock);
2282 unblock_netpoll_tx();
2286 read_unlock(&bond->lock);
2291 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2293 struct bonding *bond = netdev_priv(bond_dev);
2295 info->bond_mode = bond->params.mode;
2296 info->miimon = bond->params.miimon;
2298 read_lock(&bond->lock);
2299 info->num_slaves = bond->slave_cnt;
2300 read_unlock(&bond->lock);
2305 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2307 struct bonding *bond = netdev_priv(bond_dev);
2308 struct slave *slave;
2309 int i, res = -ENODEV;
2311 read_lock(&bond->lock);
2313 bond_for_each_slave(bond, slave, i) {
2314 if (i == (int)info->slave_id) {
2316 strcpy(info->slave_name, slave->dev->name);
2317 info->link = slave->link;
2318 info->state = bond_slave_state(slave);
2319 info->link_failure_count = slave->link_failure_count;
2324 read_unlock(&bond->lock);
2329 /*-------------------------------- Monitoring -------------------------------*/
2332 static int bond_miimon_inspect(struct bonding *bond)
2334 struct slave *slave;
2335 int i, link_state, commit = 0;
2336 bool ignore_updelay;
2338 ignore_updelay = !bond->curr_active_slave ? true : false;
2340 bond_for_each_slave(bond, slave, i) {
2341 slave->new_link = BOND_LINK_NOCHANGE;
2343 link_state = bond_check_dev_link(bond, slave->dev, 0);
2345 switch (slave->link) {
2350 slave->link = BOND_LINK_FAIL;
2351 slave->delay = bond->params.downdelay;
2353 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2355 (bond->params.mode ==
2356 BOND_MODE_ACTIVEBACKUP) ?
2357 (bond_is_active_slave(slave) ?
2358 "active " : "backup ") : "",
2360 bond->params.downdelay * bond->params.miimon);
2363 case BOND_LINK_FAIL:
2366 * recovered before downdelay expired
2368 slave->link = BOND_LINK_UP;
2369 slave->jiffies = jiffies;
2370 pr_info("%s: link status up again after %d ms for interface %s.\n",
2372 (bond->params.downdelay - slave->delay) *
2373 bond->params.miimon,
2378 if (slave->delay <= 0) {
2379 slave->new_link = BOND_LINK_DOWN;
2387 case BOND_LINK_DOWN:
2391 slave->link = BOND_LINK_BACK;
2392 slave->delay = bond->params.updelay;
2395 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2396 bond->dev->name, slave->dev->name,
2397 ignore_updelay ? 0 :
2398 bond->params.updelay *
2399 bond->params.miimon);
2402 case BOND_LINK_BACK:
2404 slave->link = BOND_LINK_DOWN;
2405 pr_info("%s: link status down again after %d ms for interface %s.\n",
2407 (bond->params.updelay - slave->delay) *
2408 bond->params.miimon,
2417 if (slave->delay <= 0) {
2418 slave->new_link = BOND_LINK_UP;
2420 ignore_updelay = false;
2432 static void bond_miimon_commit(struct bonding *bond)
2434 struct slave *slave;
2437 bond_for_each_slave(bond, slave, i) {
2438 switch (slave->new_link) {
2439 case BOND_LINK_NOCHANGE:
2443 slave->link = BOND_LINK_UP;
2444 slave->jiffies = jiffies;
2446 if (bond->params.mode == BOND_MODE_8023AD) {
2447 /* prevent it from being the active one */
2448 bond_set_backup_slave(slave);
2449 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2450 /* make it immediately active */
2451 bond_set_active_slave(slave);
2452 } else if (slave != bond->primary_slave) {
2453 /* prevent it from being the active one */
2454 bond_set_backup_slave(slave);
2457 bond_update_speed_duplex(slave);
2459 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2460 bond->dev->name, slave->dev->name,
2461 slave->speed, slave->duplex ? "full" : "half");
2463 /* notify ad that the link status has changed */
2464 if (bond->params.mode == BOND_MODE_8023AD)
2465 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2467 if (bond_is_lb(bond))
2468 bond_alb_handle_link_change(bond, slave,
2471 if (!bond->curr_active_slave ||
2472 (slave == bond->primary_slave))
2477 case BOND_LINK_DOWN:
2478 if (slave->link_failure_count < UINT_MAX)
2479 slave->link_failure_count++;
2481 slave->link = BOND_LINK_DOWN;
2483 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2484 bond->params.mode == BOND_MODE_8023AD)
2485 bond_set_slave_inactive_flags(slave);
2487 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2488 bond->dev->name, slave->dev->name);
2490 if (bond->params.mode == BOND_MODE_8023AD)
2491 bond_3ad_handle_link_change(slave,
2494 if (bond_is_lb(bond))
2495 bond_alb_handle_link_change(bond, slave,
2498 if (slave == bond->curr_active_slave)
2504 pr_err("%s: invalid new link %d on slave %s\n",
2505 bond->dev->name, slave->new_link,
2507 slave->new_link = BOND_LINK_NOCHANGE;
2515 write_lock_bh(&bond->curr_slave_lock);
2516 bond_select_active_slave(bond);
2517 write_unlock_bh(&bond->curr_slave_lock);
2518 unblock_netpoll_tx();
2521 bond_set_carrier(bond);
2527 * Really a wrapper that splits the mii monitor into two phases: an
2528 * inspection, then (if inspection indicates something needs to be done)
2529 * an acquisition of appropriate locks followed by a commit phase to
2530 * implement whatever link state changes are indicated.
2532 void bond_mii_monitor(struct work_struct *work)
2534 struct bonding *bond = container_of(work, struct bonding,
2536 bool should_notify_peers = false;
2537 unsigned long delay;
2539 read_lock(&bond->lock);
2541 delay = msecs_to_jiffies(bond->params.miimon);
2543 if (bond->slave_cnt == 0)
2546 should_notify_peers = bond_should_notify_peers(bond);
2548 if (bond_miimon_inspect(bond)) {
2549 read_unlock(&bond->lock);
2551 /* Race avoidance with bond_close cancel of workqueue */
2552 if (!rtnl_trylock()) {
2553 read_lock(&bond->lock);
2555 should_notify_peers = false;
2559 read_lock(&bond->lock);
2561 bond_miimon_commit(bond);
2563 read_unlock(&bond->lock);
2564 rtnl_unlock(); /* might sleep, hold no other locks */
2565 read_lock(&bond->lock);
2569 if (bond->params.miimon)
2570 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2572 read_unlock(&bond->lock);
2574 if (should_notify_peers) {
2575 if (!rtnl_trylock()) {
2576 read_lock(&bond->lock);
2577 bond->send_peer_notif++;
2578 read_unlock(&bond->lock);
2581 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2586 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2588 struct vlan_entry *vlan;
2589 struct net_device *vlan_dev;
2591 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2594 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2596 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2598 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2606 * We go to the (large) trouble of VLAN tagging ARP frames because
2607 * switches in VLAN mode (especially if ports are configured as
2608 * "native" to a VLAN) might not pass non-tagged frames.
2610 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2612 struct sk_buff *skb;
2614 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2615 slave_dev->name, dest_ip, src_ip, vlan_id);
2617 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2618 NULL, slave_dev->dev_addr, NULL);
2621 pr_err("ARP packet allocation failed\n");
2625 skb = vlan_put_tag(skb, vlan_id);
2627 pr_err("failed to insert VLAN tag\n");
2635 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2638 __be32 *targets = bond->params.arp_targets;
2639 struct vlan_entry *vlan;
2640 struct net_device *vlan_dev = NULL;
2643 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2647 pr_debug("basa: target %x\n", targets[i]);
2648 if (!bond_vlan_used(bond)) {
2649 pr_debug("basa: empty vlan: arp_send\n");
2650 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2651 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2657 * If VLANs are configured, we do a route lookup to
2658 * determine which VLAN interface would be used, so we
2659 * can tag the ARP with the proper VLAN tag.
2661 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2664 if (net_ratelimit()) {
2665 pr_warning("%s: no route to arp_ip_target %pI4\n",
2666 bond->dev->name, &targets[i]);
2672 * This target is not on a VLAN
2674 if (rt->dst.dev == bond->dev) {
2676 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2677 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2678 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2684 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2686 vlan_dev = __vlan_find_dev_deep(bond->dev,
2689 if (vlan_dev == rt->dst.dev) {
2690 vlan_id = vlan->vlan_id;
2691 pr_debug("basa: vlan match on %s %d\n",
2692 vlan_dev->name, vlan_id);
2697 if (vlan_id && vlan_dev) {
2699 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2700 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2705 if (net_ratelimit()) {
2706 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2707 bond->dev->name, &targets[i],
2708 rt->dst.dev ? rt->dst.dev->name : "NULL");
2714 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2717 __be32 *targets = bond->params.arp_targets;
2719 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2720 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2721 &sip, &tip, i, &targets[i],
2722 bond_has_this_ip(bond, tip));
2723 if (sip == targets[i]) {
2724 if (bond_has_this_ip(bond, tip))
2725 slave->last_arp_rx = jiffies;
2731 static int bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2732 struct slave *slave)
2735 unsigned char *arp_ptr;
2738 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2739 return RX_HANDLER_ANOTHER;
2741 read_lock(&bond->lock);
2743 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2744 bond->dev->name, skb->dev->name);
2746 if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2750 if (arp->ar_hln != bond->dev->addr_len ||
2751 skb->pkt_type == PACKET_OTHERHOST ||
2752 skb->pkt_type == PACKET_LOOPBACK ||
2753 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2754 arp->ar_pro != htons(ETH_P_IP) ||
2758 arp_ptr = (unsigned char *)(arp + 1);
2759 arp_ptr += bond->dev->addr_len;
2760 memcpy(&sip, arp_ptr, 4);
2761 arp_ptr += 4 + bond->dev->addr_len;
2762 memcpy(&tip, arp_ptr, 4);
2764 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2765 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2766 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2770 * Backup slaves won't see the ARP reply, but do come through
2771 * here for each ARP probe (so we swap the sip/tip to validate
2772 * the probe). In a "redundant switch, common router" type of
2773 * configuration, the ARP probe will (hopefully) travel from
2774 * the active, through one switch, the router, then the other
2775 * switch before reaching the backup.
2777 if (bond_is_active_slave(slave))
2778 bond_validate_arp(bond, slave, sip, tip);
2780 bond_validate_arp(bond, slave, tip, sip);
2783 read_unlock(&bond->lock);
2784 return RX_HANDLER_ANOTHER;
2788 * this function is called regularly to monitor each slave's link
2789 * ensuring that traffic is being sent and received when arp monitoring
2790 * is used in load-balancing mode. if the adapter has been dormant, then an
2791 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2792 * arp monitoring in active backup mode.
2794 void bond_loadbalance_arp_mon(struct work_struct *work)
2796 struct bonding *bond = container_of(work, struct bonding,
2798 struct slave *slave, *oldcurrent;
2799 int do_failover = 0;
2803 read_lock(&bond->lock);
2805 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2807 if (bond->slave_cnt == 0)
2810 read_lock(&bond->curr_slave_lock);
2811 oldcurrent = bond->curr_active_slave;
2812 read_unlock(&bond->curr_slave_lock);
2814 /* see if any of the previous devices are up now (i.e. they have
2815 * xmt and rcv traffic). the curr_active_slave does not come into
2816 * the picture unless it is null. also, slave->jiffies is not needed
2817 * here because we send an arp on each slave and give a slave as
2818 * long as it needs to get the tx/rx within the delta.
2819 * TODO: what about up/down delay in arp mode? it wasn't here before
2822 bond_for_each_slave(bond, slave, i) {
2823 unsigned long trans_start = dev_trans_start(slave->dev);
2825 if (slave->link != BOND_LINK_UP) {
2826 if (time_in_range(jiffies,
2827 trans_start - delta_in_ticks,
2828 trans_start + delta_in_ticks) &&
2829 time_in_range(jiffies,
2830 slave->dev->last_rx - delta_in_ticks,
2831 slave->dev->last_rx + delta_in_ticks)) {
2833 slave->link = BOND_LINK_UP;
2834 bond_set_active_slave(slave);
2836 /* primary_slave has no meaning in round-robin
2837 * mode. the window of a slave being up and
2838 * curr_active_slave being null after enslaving
2842 pr_info("%s: link status definitely up for interface %s, ",
2847 pr_info("%s: interface %s is now up\n",
2853 /* slave->link == BOND_LINK_UP */
2855 /* not all switches will respond to an arp request
2856 * when the source ip is 0, so don't take the link down
2857 * if we don't know our ip yet
2859 if (!time_in_range(jiffies,
2860 trans_start - delta_in_ticks,
2861 trans_start + 2 * delta_in_ticks) ||
2862 !time_in_range(jiffies,
2863 slave->dev->last_rx - delta_in_ticks,
2864 slave->dev->last_rx + 2 * delta_in_ticks)) {
2866 slave->link = BOND_LINK_DOWN;
2867 bond_set_backup_slave(slave);
2869 if (slave->link_failure_count < UINT_MAX)
2870 slave->link_failure_count++;
2872 pr_info("%s: interface %s is now down.\n",
2876 if (slave == oldcurrent)
2881 /* note: if switch is in round-robin mode, all links
2882 * must tx arp to ensure all links rx an arp - otherwise
2883 * links may oscillate or not come up at all; if switch is
2884 * in something like xor mode, there is nothing we can
2885 * do - all replies will be rx'ed on same link causing slaves
2886 * to be unstable during low/no traffic periods
2888 if (IS_UP(slave->dev))
2889 bond_arp_send_all(bond, slave);
2894 write_lock_bh(&bond->curr_slave_lock);
2896 bond_select_active_slave(bond);
2898 write_unlock_bh(&bond->curr_slave_lock);
2899 unblock_netpoll_tx();
2903 if (bond->params.arp_interval)
2904 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2906 read_unlock(&bond->lock);
2910 * Called to inspect slaves for active-backup mode ARP monitor link state
2911 * changes. Sets new_link in slaves to specify what action should take
2912 * place for the slave. Returns 0 if no changes are found, >0 if changes
2913 * to link states must be committed.
2915 * Called with bond->lock held for read.
2917 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2919 struct slave *slave;
2921 unsigned long trans_start;
2923 bond_for_each_slave(bond, slave, i) {
2924 slave->new_link = BOND_LINK_NOCHANGE;
2926 if (slave->link != BOND_LINK_UP) {
2927 if (time_in_range(jiffies,
2928 slave_last_rx(bond, slave) - delta_in_ticks,
2929 slave_last_rx(bond, slave) + delta_in_ticks)) {
2931 slave->new_link = BOND_LINK_UP;
2939 * Give slaves 2*delta after being enslaved or made
2940 * active. This avoids bouncing, as the last receive
2941 * times need a full ARP monitor cycle to be updated.
2943 if (time_in_range(jiffies,
2944 slave->jiffies - delta_in_ticks,
2945 slave->jiffies + 2 * delta_in_ticks))
2949 * Backup slave is down if:
2950 * - No current_arp_slave AND
2951 * - more than 3*delta since last receive AND
2952 * - the bond has an IP address
2954 * Note: a non-null current_arp_slave indicates
2955 * the curr_active_slave went down and we are
2956 * searching for a new one; under this condition
2957 * we only take the curr_active_slave down - this
2958 * gives each slave a chance to tx/rx traffic
2959 * before being taken out
2961 if (!bond_is_active_slave(slave) &&
2962 !bond->current_arp_slave &&
2963 !time_in_range(jiffies,
2964 slave_last_rx(bond, slave) - delta_in_ticks,
2965 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2967 slave->new_link = BOND_LINK_DOWN;
2972 * Active slave is down if:
2973 * - more than 2*delta since transmitting OR
2974 * - (more than 2*delta since receive AND
2975 * the bond has an IP address)
2977 trans_start = dev_trans_start(slave->dev);
2978 if (bond_is_active_slave(slave) &&
2979 (!time_in_range(jiffies,
2980 trans_start - delta_in_ticks,
2981 trans_start + 2 * delta_in_ticks) ||
2982 !time_in_range(jiffies,
2983 slave_last_rx(bond, slave) - delta_in_ticks,
2984 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2986 slave->new_link = BOND_LINK_DOWN;
2995 * Called to commit link state changes noted by inspection step of
2996 * active-backup mode ARP monitor.
2998 * Called with RTNL and bond->lock for read.
3000 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3002 struct slave *slave;
3004 unsigned long trans_start;
3006 bond_for_each_slave(bond, slave, i) {
3007 switch (slave->new_link) {
3008 case BOND_LINK_NOCHANGE:
3012 trans_start = dev_trans_start(slave->dev);
3013 if ((!bond->curr_active_slave &&
3014 time_in_range(jiffies,
3015 trans_start - delta_in_ticks,
3016 trans_start + delta_in_ticks)) ||
3017 bond->curr_active_slave != slave) {
3018 slave->link = BOND_LINK_UP;
3019 if (bond->current_arp_slave) {
3020 bond_set_slave_inactive_flags(
3021 bond->current_arp_slave);
3022 bond->current_arp_slave = NULL;
3025 pr_info("%s: link status definitely up for interface %s.\n",
3026 bond->dev->name, slave->dev->name);
3028 if (!bond->curr_active_slave ||
3029 (slave == bond->primary_slave))
3036 case BOND_LINK_DOWN:
3037 if (slave->link_failure_count < UINT_MAX)
3038 slave->link_failure_count++;
3040 slave->link = BOND_LINK_DOWN;
3041 bond_set_slave_inactive_flags(slave);
3043 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3044 bond->dev->name, slave->dev->name);
3046 if (slave == bond->curr_active_slave) {
3047 bond->current_arp_slave = NULL;
3054 pr_err("%s: impossible: new_link %d on slave %s\n",
3055 bond->dev->name, slave->new_link,
3063 write_lock_bh(&bond->curr_slave_lock);
3064 bond_select_active_slave(bond);
3065 write_unlock_bh(&bond->curr_slave_lock);
3066 unblock_netpoll_tx();
3069 bond_set_carrier(bond);
3073 * Send ARP probes for active-backup mode ARP monitor.
3075 * Called with bond->lock held for read.
3077 static void bond_ab_arp_probe(struct bonding *bond)
3079 struct slave *slave;
3082 read_lock(&bond->curr_slave_lock);
3084 if (bond->current_arp_slave && bond->curr_active_slave)
3085 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3086 bond->current_arp_slave->dev->name,
3087 bond->curr_active_slave->dev->name);
3089 if (bond->curr_active_slave) {
3090 bond_arp_send_all(bond, bond->curr_active_slave);
3091 read_unlock(&bond->curr_slave_lock);
3095 read_unlock(&bond->curr_slave_lock);
3097 /* if we don't have a curr_active_slave, search for the next available
3098 * backup slave from the current_arp_slave and make it the candidate
3099 * for becoming the curr_active_slave
3102 if (!bond->current_arp_slave) {
3103 bond->current_arp_slave = bond->first_slave;
3104 if (!bond->current_arp_slave)
3108 bond_set_slave_inactive_flags(bond->current_arp_slave);
3110 /* search for next candidate */
3111 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3112 if (IS_UP(slave->dev)) {
3113 slave->link = BOND_LINK_BACK;
3114 bond_set_slave_active_flags(slave);
3115 bond_arp_send_all(bond, slave);
3116 slave->jiffies = jiffies;
3117 bond->current_arp_slave = slave;
3121 /* if the link state is up at this point, we
3122 * mark it down - this can happen if we have
3123 * simultaneous link failures and
3124 * reselect_active_interface doesn't make this
3125 * one the current slave so it is still marked
3126 * up when it is actually down
3128 if (slave->link == BOND_LINK_UP) {
3129 slave->link = BOND_LINK_DOWN;
3130 if (slave->link_failure_count < UINT_MAX)
3131 slave->link_failure_count++;
3133 bond_set_slave_inactive_flags(slave);
3135 pr_info("%s: backup interface %s is now down.\n",
3136 bond->dev->name, slave->dev->name);
3141 void bond_activebackup_arp_mon(struct work_struct *work)
3143 struct bonding *bond = container_of(work, struct bonding,
3145 bool should_notify_peers = false;
3148 read_lock(&bond->lock);
3150 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3152 if (bond->slave_cnt == 0)
3155 should_notify_peers = bond_should_notify_peers(bond);
3157 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3158 read_unlock(&bond->lock);
3160 /* Race avoidance with bond_close flush of workqueue */
3161 if (!rtnl_trylock()) {
3162 read_lock(&bond->lock);
3164 should_notify_peers = false;
3168 read_lock(&bond->lock);
3170 bond_ab_arp_commit(bond, delta_in_ticks);
3172 read_unlock(&bond->lock);
3174 read_lock(&bond->lock);
3177 bond_ab_arp_probe(bond);
3180 if (bond->params.arp_interval)
3181 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3183 read_unlock(&bond->lock);
3185 if (should_notify_peers) {
3186 if (!rtnl_trylock()) {
3187 read_lock(&bond->lock);
3188 bond->send_peer_notif++;
3189 read_unlock(&bond->lock);
3192 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3197 /*-------------------------- netdev event handling --------------------------*/
3200 * Change device name
3202 static int bond_event_changename(struct bonding *bond)
3204 bond_remove_proc_entry(bond);
3205 bond_create_proc_entry(bond);
3207 bond_debug_reregister(bond);
3212 static int bond_master_netdev_event(unsigned long event,
3213 struct net_device *bond_dev)
3215 struct bonding *event_bond = netdev_priv(bond_dev);
3218 case NETDEV_CHANGENAME:
3219 return bond_event_changename(event_bond);
3227 static int bond_slave_netdev_event(unsigned long event,
3228 struct net_device *slave_dev)
3230 struct net_device *bond_dev = slave_dev->master;
3231 struct bonding *bond = netdev_priv(bond_dev);
3232 struct slave *slave = NULL;
3235 case NETDEV_UNREGISTER:
3237 if (bond->setup_by_slave)
3238 bond_release_and_destroy(bond_dev, slave_dev);
3240 bond_release(bond_dev, slave_dev);
3245 slave = bond_get_slave_by_dev(bond, slave_dev);
3247 u32 old_speed = slave->speed;
3248 u8 old_duplex = slave->duplex;
3250 bond_update_speed_duplex(slave);
3252 if (bond->params.mode == BOND_MODE_8023AD) {
3253 if (old_speed != slave->speed)
3254 bond_3ad_adapter_speed_changed(slave);
3255 if (old_duplex != slave->duplex)
3256 bond_3ad_adapter_duplex_changed(slave);
3263 * ... Or is it this?
3266 case NETDEV_CHANGEMTU:
3268 * TODO: Should slaves be allowed to
3269 * independently alter their MTU? For
3270 * an active-backup bond, slaves need
3271 * not be the same type of device, so
3272 * MTUs may vary. For other modes,
3273 * slaves arguably should have the
3274 * same MTUs. To do this, we'd need to
3275 * take over the slave's change_mtu
3276 * function for the duration of their
3280 case NETDEV_CHANGENAME:
3282 * TODO: handle changing the primary's name
3285 case NETDEV_FEAT_CHANGE:
3286 bond_compute_features(bond);
3296 * bond_netdev_event: handle netdev notifier chain events.
3298 * This function receives events for the netdev chain. The caller (an
3299 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3300 * locks for us to safely manipulate the slave devices (RTNL lock,
3303 static int bond_netdev_event(struct notifier_block *this,
3304 unsigned long event, void *ptr)
3306 struct net_device *event_dev = (struct net_device *)ptr;
3308 pr_debug("event_dev: %s, event: %lx\n",
3309 event_dev ? event_dev->name : "None",
3312 if (!(event_dev->priv_flags & IFF_BONDING))
3315 if (event_dev->flags & IFF_MASTER) {
3316 pr_debug("IFF_MASTER\n");
3317 return bond_master_netdev_event(event, event_dev);
3320 if (event_dev->flags & IFF_SLAVE) {
3321 pr_debug("IFF_SLAVE\n");
3322 return bond_slave_netdev_event(event, event_dev);
3328 static struct notifier_block bond_netdev_notifier = {
3329 .notifier_call = bond_netdev_event,
3332 /*---------------------------- Hashing Policies -----------------------------*/
3335 * Hash for the output device based upon layer 2 and layer 3 data. If
3336 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3338 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3340 struct ethhdr *data = (struct ethhdr *)skb->data;
3341 struct iphdr *iph = ip_hdr(skb);
3343 if (skb->protocol == htons(ETH_P_IP)) {
3344 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3345 (data->h_dest[5] ^ data->h_source[5])) % count;
3348 return (data->h_dest[5] ^ data->h_source[5]) % count;
3352 * Hash for the output device based upon layer 3 and layer 4 data. If
3353 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3354 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3356 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3358 struct ethhdr *data = (struct ethhdr *)skb->data;
3359 struct iphdr *iph = ip_hdr(skb);
3360 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3363 if (skb->protocol == htons(ETH_P_IP)) {
3364 if (!ip_is_fragment(iph) &&
3365 (iph->protocol == IPPROTO_TCP ||
3366 iph->protocol == IPPROTO_UDP)) {
3367 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3369 return (layer4_xor ^
3370 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3374 return (data->h_dest[5] ^ data->h_source[5]) % count;
3378 * Hash for the output device based upon layer 2 data
3380 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3382 struct ethhdr *data = (struct ethhdr *)skb->data;
3384 return (data->h_dest[5] ^ data->h_source[5]) % count;
3387 /*-------------------------- Device entry points ----------------------------*/
3389 static int bond_open(struct net_device *bond_dev)
3391 struct bonding *bond = netdev_priv(bond_dev);
3392 struct slave *slave;
3395 /* reset slave->backup and slave->inactive */
3396 read_lock(&bond->lock);
3397 if (bond->slave_cnt > 0) {
3398 read_lock(&bond->curr_slave_lock);
3399 bond_for_each_slave(bond, slave, i) {
3400 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3401 && (slave != bond->curr_active_slave)) {
3402 bond_set_slave_inactive_flags(slave);
3404 bond_set_slave_active_flags(slave);
3407 read_unlock(&bond->curr_slave_lock);
3409 read_unlock(&bond->lock);
3411 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3413 if (bond_is_lb(bond)) {
3414 /* bond_alb_initialize must be called before the timer
3417 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3418 /* something went wrong - fail the open operation */
3422 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3423 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3426 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3427 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3428 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3431 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3432 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3433 INIT_DELAYED_WORK(&bond->arp_work,
3434 bond_activebackup_arp_mon);
3436 INIT_DELAYED_WORK(&bond->arp_work,
3437 bond_loadbalance_arp_mon);
3439 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3440 if (bond->params.arp_validate)
3441 bond->recv_probe = bond_arp_rcv;
3444 if (bond->params.mode == BOND_MODE_8023AD) {
3445 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3446 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3447 /* register to receive LACPDUs */
3448 bond->recv_probe = bond_3ad_lacpdu_recv;
3449 bond_3ad_initiate_agg_selection(bond, 1);
3455 static int bond_close(struct net_device *bond_dev)
3457 struct bonding *bond = netdev_priv(bond_dev);
3459 write_lock_bh(&bond->lock);
3461 bond->send_peer_notif = 0;
3463 write_unlock_bh(&bond->lock);
3465 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3466 cancel_delayed_work_sync(&bond->mii_work);
3469 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3470 cancel_delayed_work_sync(&bond->arp_work);
3473 switch (bond->params.mode) {
3474 case BOND_MODE_8023AD:
3475 cancel_delayed_work_sync(&bond->ad_work);
3479 cancel_delayed_work_sync(&bond->alb_work);
3485 if (delayed_work_pending(&bond->mcast_work))
3486 cancel_delayed_work_sync(&bond->mcast_work);
3488 if (bond_is_lb(bond)) {
3489 /* Must be called only after all
3490 * slaves have been released
3492 bond_alb_deinitialize(bond);
3494 bond->recv_probe = NULL;
3499 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3500 struct rtnl_link_stats64 *stats)
3502 struct bonding *bond = netdev_priv(bond_dev);
3503 struct rtnl_link_stats64 temp;
3504 struct slave *slave;
3507 memset(stats, 0, sizeof(*stats));
3509 read_lock_bh(&bond->lock);
3511 bond_for_each_slave(bond, slave, i) {
3512 const struct rtnl_link_stats64 *sstats =
3513 dev_get_stats(slave->dev, &temp);
3515 stats->rx_packets += sstats->rx_packets;
3516 stats->rx_bytes += sstats->rx_bytes;
3517 stats->rx_errors += sstats->rx_errors;
3518 stats->rx_dropped += sstats->rx_dropped;
3520 stats->tx_packets += sstats->tx_packets;
3521 stats->tx_bytes += sstats->tx_bytes;
3522 stats->tx_errors += sstats->tx_errors;
3523 stats->tx_dropped += sstats->tx_dropped;
3525 stats->multicast += sstats->multicast;
3526 stats->collisions += sstats->collisions;
3528 stats->rx_length_errors += sstats->rx_length_errors;
3529 stats->rx_over_errors += sstats->rx_over_errors;
3530 stats->rx_crc_errors += sstats->rx_crc_errors;
3531 stats->rx_frame_errors += sstats->rx_frame_errors;
3532 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3533 stats->rx_missed_errors += sstats->rx_missed_errors;
3535 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3536 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3537 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3538 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3539 stats->tx_window_errors += sstats->tx_window_errors;
3542 read_unlock_bh(&bond->lock);
3547 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3549 struct net_device *slave_dev = NULL;
3550 struct ifbond k_binfo;
3551 struct ifbond __user *u_binfo = NULL;
3552 struct ifslave k_sinfo;
3553 struct ifslave __user *u_sinfo = NULL;
3554 struct mii_ioctl_data *mii = NULL;
3557 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3569 * We do this again just in case we were called by SIOCGMIIREG
3570 * instead of SIOCGMIIPHY.
3577 if (mii->reg_num == 1) {
3578 struct bonding *bond = netdev_priv(bond_dev);
3580 read_lock(&bond->lock);
3581 read_lock(&bond->curr_slave_lock);
3582 if (netif_carrier_ok(bond->dev))
3583 mii->val_out = BMSR_LSTATUS;
3585 read_unlock(&bond->curr_slave_lock);
3586 read_unlock(&bond->lock);
3590 case BOND_INFO_QUERY_OLD:
3591 case SIOCBONDINFOQUERY:
3592 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3594 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3597 res = bond_info_query(bond_dev, &k_binfo);
3599 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3603 case BOND_SLAVE_INFO_QUERY_OLD:
3604 case SIOCBONDSLAVEINFOQUERY:
3605 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3607 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3610 res = bond_slave_info_query(bond_dev, &k_sinfo);
3612 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3621 if (!capable(CAP_NET_ADMIN))
3624 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3626 pr_debug("slave_dev=%p:\n", slave_dev);
3631 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3633 case BOND_ENSLAVE_OLD:
3634 case SIOCBONDENSLAVE:
3635 res = bond_enslave(bond_dev, slave_dev);
3637 case BOND_RELEASE_OLD:
3638 case SIOCBONDRELEASE:
3639 res = bond_release(bond_dev, slave_dev);
3641 case BOND_SETHWADDR_OLD:
3642 case SIOCBONDSETHWADDR:
3643 res = bond_sethwaddr(bond_dev, slave_dev);
3645 case BOND_CHANGE_ACTIVE_OLD:
3646 case SIOCBONDCHANGEACTIVE:
3647 res = bond_ioctl_change_active(bond_dev, slave_dev);
3659 static bool bond_addr_in_mc_list(unsigned char *addr,
3660 struct netdev_hw_addr_list *list,
3663 struct netdev_hw_addr *ha;
3665 netdev_hw_addr_list_for_each(ha, list)
3666 if (!memcmp(ha->addr, addr, addrlen))
3672 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3674 struct bonding *bond = netdev_priv(bond_dev);
3676 if (change & IFF_PROMISC)
3677 bond_set_promiscuity(bond,
3678 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3680 if (change & IFF_ALLMULTI)
3681 bond_set_allmulti(bond,
3682 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3685 static void bond_set_multicast_list(struct net_device *bond_dev)
3687 struct bonding *bond = netdev_priv(bond_dev);
3688 struct netdev_hw_addr *ha;
3691 read_lock(&bond->lock);
3693 /* looking for addresses to add to slaves' mc list */
3694 netdev_for_each_mc_addr(ha, bond_dev) {
3695 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3696 bond_dev->addr_len);
3698 bond_mc_add(bond, ha->addr);
3701 /* looking for addresses to delete from slaves' list */
3702 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3703 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3704 bond_dev->addr_len);
3706 bond_mc_del(bond, ha->addr);
3709 /* save master's multicast list */
3710 __hw_addr_flush(&bond->mc_list);
3711 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3712 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3714 read_unlock(&bond->lock);
3717 static int bond_neigh_init(struct neighbour *n)
3719 struct bonding *bond = netdev_priv(n->dev);
3720 struct slave *slave = bond->first_slave;
3721 const struct net_device_ops *slave_ops;
3722 struct neigh_parms parms;
3728 slave_ops = slave->dev->netdev_ops;
3730 if (!slave_ops->ndo_neigh_setup)
3733 parms.neigh_setup = NULL;
3734 parms.neigh_cleanup = NULL;
3735 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3740 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3741 * after the last slave has been detached. Assumes that all slaves
3742 * utilize the same neigh_cleanup (true at this writing as only user
3745 n->parms->neigh_cleanup = parms.neigh_cleanup;
3747 if (!parms.neigh_setup)
3750 return parms.neigh_setup(n);
3754 * The bonding ndo_neigh_setup is called at init time beofre any
3755 * slave exists. So we must declare proxy setup function which will
3756 * be used at run time to resolve the actual slave neigh param setup.
3758 static int bond_neigh_setup(struct net_device *dev,
3759 struct neigh_parms *parms)
3761 parms->neigh_setup = bond_neigh_init;
3767 * Change the MTU of all of a master's slaves to match the master
3769 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3771 struct bonding *bond = netdev_priv(bond_dev);
3772 struct slave *slave, *stop_at;
3776 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3777 (bond_dev ? bond_dev->name : "None"), new_mtu);
3779 /* Can't hold bond->lock with bh disabled here since
3780 * some base drivers panic. On the other hand we can't
3781 * hold bond->lock without bh disabled because we'll
3782 * deadlock. The only solution is to rely on the fact
3783 * that we're under rtnl_lock here, and the slaves
3784 * list won't change. This doesn't solve the problem
3785 * of setting the slave's MTU while it is
3786 * transmitting, but the assumption is that the base
3787 * driver can handle that.
3789 * TODO: figure out a way to safely iterate the slaves
3790 * list, but without holding a lock around the actual
3791 * call to the base driver.
3794 bond_for_each_slave(bond, slave, i) {
3795 pr_debug("s %p s->p %p c_m %p\n",
3798 slave->dev->netdev_ops->ndo_change_mtu);
3800 res = dev_set_mtu(slave->dev, new_mtu);
3803 /* If we failed to set the slave's mtu to the new value
3804 * we must abort the operation even in ACTIVE_BACKUP
3805 * mode, because if we allow the backup slaves to have
3806 * different mtu values than the active slave we'll
3807 * need to change their mtu when doing a failover. That
3808 * means changing their mtu from timer context, which
3809 * is probably not a good idea.
3811 pr_debug("err %d %s\n", res, slave->dev->name);
3816 bond_dev->mtu = new_mtu;
3821 /* unwind from head to the slave that failed */
3823 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3826 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3828 pr_debug("unwind err %d dev %s\n",
3829 tmp_res, slave->dev->name);
3839 * Note that many devices must be down to change the HW address, and
3840 * downing the master releases all slaves. We can make bonds full of
3841 * bonding devices to test this, however.
3843 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3845 struct bonding *bond = netdev_priv(bond_dev);
3846 struct sockaddr *sa = addr, tmp_sa;
3847 struct slave *slave, *stop_at;
3851 if (bond->params.mode == BOND_MODE_ALB)
3852 return bond_alb_set_mac_address(bond_dev, addr);
3855 pr_debug("bond=%p, name=%s\n",
3856 bond, bond_dev ? bond_dev->name : "None");
3859 * If fail_over_mac is set to active, do nothing and return
3860 * success. Returning an error causes ifenslave to fail.
3862 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3865 if (!is_valid_ether_addr(sa->sa_data))
3866 return -EADDRNOTAVAIL;
3868 /* Can't hold bond->lock with bh disabled here since
3869 * some base drivers panic. On the other hand we can't
3870 * hold bond->lock without bh disabled because we'll
3871 * deadlock. The only solution is to rely on the fact
3872 * that we're under rtnl_lock here, and the slaves
3873 * list won't change. This doesn't solve the problem
3874 * of setting the slave's hw address while it is
3875 * transmitting, but the assumption is that the base
3876 * driver can handle that.
3878 * TODO: figure out a way to safely iterate the slaves
3879 * list, but without holding a lock around the actual
3880 * call to the base driver.
3883 bond_for_each_slave(bond, slave, i) {
3884 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3885 pr_debug("slave %p %s\n", slave, slave->dev->name);
3887 if (slave_ops->ndo_set_mac_address == NULL) {
3889 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3893 res = dev_set_mac_address(slave->dev, addr);
3895 /* TODO: consider downing the slave
3897 * User should expect communications
3898 * breakage anyway until ARP finish
3901 pr_debug("err %d %s\n", res, slave->dev->name);
3907 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3911 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3912 tmp_sa.sa_family = bond_dev->type;
3914 /* unwind from head to the slave that failed */
3916 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3919 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3921 pr_debug("unwind err %d dev %s\n",
3922 tmp_res, slave->dev->name);
3929 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3931 struct bonding *bond = netdev_priv(bond_dev);
3932 struct slave *slave, *start_at;
3933 int i, slave_no, res = 1;
3934 struct iphdr *iph = ip_hdr(skb);
3937 * Start with the curr_active_slave that joined the bond as the
3938 * default for sending IGMP traffic. For failover purposes one
3939 * needs to maintain some consistency for the interface that will
3940 * send the join/membership reports. The curr_active_slave found
3941 * will send all of this type of traffic.
3943 if ((iph->protocol == IPPROTO_IGMP) &&
3944 (skb->protocol == htons(ETH_P_IP))) {
3946 read_lock(&bond->curr_slave_lock);
3947 slave = bond->curr_active_slave;
3948 read_unlock(&bond->curr_slave_lock);
3954 * Concurrent TX may collide on rr_tx_counter; we accept
3955 * that as being rare enough not to justify using an
3958 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3960 bond_for_each_slave(bond, slave, i) {
3968 bond_for_each_slave_from(bond, slave, i, start_at) {
3969 if (IS_UP(slave->dev) &&
3970 (slave->link == BOND_LINK_UP) &&
3971 bond_is_active_slave(slave)) {
3972 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3979 /* no suitable interface, frame not sent */
3983 return NETDEV_TX_OK;
3988 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3989 * the bond has a usable interface.
3991 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3993 struct bonding *bond = netdev_priv(bond_dev);
3996 read_lock(&bond->curr_slave_lock);
3998 if (bond->curr_active_slave)
3999 res = bond_dev_queue_xmit(bond, skb,
4000 bond->curr_active_slave->dev);
4003 /* no suitable interface, frame not sent */
4006 read_unlock(&bond->curr_slave_lock);
4008 return NETDEV_TX_OK;
4012 * In bond_xmit_xor() , we determine the output device by using a pre-
4013 * determined xmit_hash_policy(), If the selected device is not enabled,
4014 * find the next active slave.
4016 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4018 struct bonding *bond = netdev_priv(bond_dev);
4019 struct slave *slave, *start_at;
4024 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4026 bond_for_each_slave(bond, slave, i) {
4034 bond_for_each_slave_from(bond, slave, i, start_at) {
4035 if (IS_UP(slave->dev) &&
4036 (slave->link == BOND_LINK_UP) &&
4037 bond_is_active_slave(slave)) {
4038 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4044 /* no suitable interface, frame not sent */
4048 return NETDEV_TX_OK;
4052 * in broadcast mode, we send everything to all usable interfaces.
4054 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4056 struct bonding *bond = netdev_priv(bond_dev);
4057 struct slave *slave, *start_at;
4058 struct net_device *tx_dev = NULL;
4062 read_lock(&bond->curr_slave_lock);
4063 start_at = bond->curr_active_slave;
4064 read_unlock(&bond->curr_slave_lock);
4069 bond_for_each_slave_from(bond, slave, i, start_at) {
4070 if (IS_UP(slave->dev) &&
4071 (slave->link == BOND_LINK_UP) &&
4072 bond_is_active_slave(slave)) {
4074 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4076 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4081 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4083 dev_kfree_skb(skb2);
4087 tx_dev = slave->dev;
4092 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4096 /* no suitable interface, frame not sent */
4099 /* frame sent to all suitable interfaces */
4100 return NETDEV_TX_OK;
4103 /*------------------------- Device initialization ---------------------------*/
4105 static void bond_set_xmit_hash_policy(struct bonding *bond)
4107 switch (bond->params.xmit_policy) {
4108 case BOND_XMIT_POLICY_LAYER23:
4109 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4111 case BOND_XMIT_POLICY_LAYER34:
4112 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4114 case BOND_XMIT_POLICY_LAYER2:
4116 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4122 * Lookup the slave that corresponds to a qid
4124 static inline int bond_slave_override(struct bonding *bond,
4125 struct sk_buff *skb)
4128 struct slave *slave = NULL;
4129 struct slave *check_slave;
4131 if (!skb->queue_mapping)
4134 /* Find out if any slaves have the same mapping as this skb. */
4135 bond_for_each_slave(bond, check_slave, i) {
4136 if (check_slave->queue_id == skb->queue_mapping) {
4137 slave = check_slave;
4142 /* If the slave isn't UP, use default transmit policy. */
4143 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4144 (slave->link == BOND_LINK_UP)) {
4145 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4152 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4155 * This helper function exists to help dev_pick_tx get the correct
4156 * destination queue. Using a helper function skips a call to
4157 * skb_tx_hash and will put the skbs in the queue we expect on their
4158 * way down to the bonding driver.
4160 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4163 * Save the original txq to restore before passing to the driver
4165 bond_queue_mapping(skb) = skb->queue_mapping;
4167 if (unlikely(txq >= dev->real_num_tx_queues)) {
4169 txq -= dev->real_num_tx_queues;
4170 } while (txq >= dev->real_num_tx_queues);
4175 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4177 struct bonding *bond = netdev_priv(dev);
4179 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4180 if (!bond_slave_override(bond, skb))
4181 return NETDEV_TX_OK;
4184 switch (bond->params.mode) {
4185 case BOND_MODE_ROUNDROBIN:
4186 return bond_xmit_roundrobin(skb, dev);
4187 case BOND_MODE_ACTIVEBACKUP:
4188 return bond_xmit_activebackup(skb, dev);
4190 return bond_xmit_xor(skb, dev);
4191 case BOND_MODE_BROADCAST:
4192 return bond_xmit_broadcast(skb, dev);
4193 case BOND_MODE_8023AD:
4194 return bond_3ad_xmit_xor(skb, dev);
4197 return bond_alb_xmit(skb, dev);
4199 /* Should never happen, mode already checked */
4200 pr_err("%s: Error: Unknown bonding mode %d\n",
4201 dev->name, bond->params.mode);
4204 return NETDEV_TX_OK;
4208 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4210 struct bonding *bond = netdev_priv(dev);
4211 netdev_tx_t ret = NETDEV_TX_OK;
4214 * If we risk deadlock from transmitting this in the
4215 * netpoll path, tell netpoll to queue the frame for later tx
4217 if (is_netpoll_tx_blocked(dev))
4218 return NETDEV_TX_BUSY;
4220 read_lock(&bond->lock);
4222 if (bond->slave_cnt)
4223 ret = __bond_start_xmit(skb, dev);
4227 read_unlock(&bond->lock);
4233 * set bond mode specific net device operations
4235 void bond_set_mode_ops(struct bonding *bond, int mode)
4237 struct net_device *bond_dev = bond->dev;
4240 case BOND_MODE_ROUNDROBIN:
4242 case BOND_MODE_ACTIVEBACKUP:
4245 bond_set_xmit_hash_policy(bond);
4247 case BOND_MODE_BROADCAST:
4249 case BOND_MODE_8023AD:
4250 bond_set_xmit_hash_policy(bond);
4257 /* Should never happen, mode already checked */
4258 pr_err("%s: Error: Unknown bonding mode %d\n",
4259 bond_dev->name, mode);
4264 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4265 struct ethtool_drvinfo *drvinfo)
4267 strncpy(drvinfo->driver, DRV_NAME, 32);
4268 strncpy(drvinfo->version, DRV_VERSION, 32);
4269 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4272 static const struct ethtool_ops bond_ethtool_ops = {
4273 .get_drvinfo = bond_ethtool_get_drvinfo,
4274 .get_link = ethtool_op_get_link,
4277 static const struct net_device_ops bond_netdev_ops = {
4278 .ndo_init = bond_init,
4279 .ndo_uninit = bond_uninit,
4280 .ndo_open = bond_open,
4281 .ndo_stop = bond_close,
4282 .ndo_start_xmit = bond_start_xmit,
4283 .ndo_select_queue = bond_select_queue,
4284 .ndo_get_stats64 = bond_get_stats,
4285 .ndo_do_ioctl = bond_do_ioctl,
4286 .ndo_change_rx_flags = bond_change_rx_flags,
4287 .ndo_set_rx_mode = bond_set_multicast_list,
4288 .ndo_change_mtu = bond_change_mtu,
4289 .ndo_set_mac_address = bond_set_mac_address,
4290 .ndo_neigh_setup = bond_neigh_setup,
4291 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4292 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4293 #ifdef CONFIG_NET_POLL_CONTROLLER
4294 .ndo_netpoll_setup = bond_netpoll_setup,
4295 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4296 .ndo_poll_controller = bond_poll_controller,
4298 .ndo_add_slave = bond_enslave,
4299 .ndo_del_slave = bond_release,
4300 .ndo_fix_features = bond_fix_features,
4303 static void bond_destructor(struct net_device *bond_dev)
4305 struct bonding *bond = netdev_priv(bond_dev);
4307 destroy_workqueue(bond->wq);
4308 free_netdev(bond_dev);
4311 static void bond_setup(struct net_device *bond_dev)
4313 struct bonding *bond = netdev_priv(bond_dev);
4315 /* initialize rwlocks */
4316 rwlock_init(&bond->lock);
4317 rwlock_init(&bond->curr_slave_lock);
4319 bond->params = bonding_defaults;
4321 /* Initialize pointers */
4322 bond->dev = bond_dev;
4323 INIT_LIST_HEAD(&bond->vlan_list);
4325 /* Initialize the device entry points */
4326 ether_setup(bond_dev);
4327 bond_dev->netdev_ops = &bond_netdev_ops;
4328 bond_dev->ethtool_ops = &bond_ethtool_ops;
4329 bond_set_mode_ops(bond, bond->params.mode);
4331 bond_dev->destructor = bond_destructor;
4333 /* Initialize the device options */
4334 bond_dev->tx_queue_len = 0;
4335 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4336 bond_dev->priv_flags |= IFF_BONDING;
4337 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4339 /* At first, we block adding VLANs. That's the only way to
4340 * prevent problems that occur when adding VLANs over an
4341 * empty bond. The block will be removed once non-challenged
4342 * slaves are enslaved.
4344 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4346 /* don't acquire bond device's netif_tx_lock when
4348 bond_dev->features |= NETIF_F_LLTX;
4350 /* By default, we declare the bond to be fully
4351 * VLAN hardware accelerated capable. Special
4352 * care is taken in the various xmit functions
4353 * when there are slaves that are not hw accel
4357 bond_dev->hw_features = BOND_VLAN_FEATURES |
4358 NETIF_F_HW_VLAN_TX |
4359 NETIF_F_HW_VLAN_RX |
4360 NETIF_F_HW_VLAN_FILTER;
4362 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4363 bond_dev->features |= bond_dev->hw_features;
4366 static void bond_work_cancel_all(struct bonding *bond)
4368 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4369 cancel_delayed_work_sync(&bond->mii_work);
4371 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4372 cancel_delayed_work_sync(&bond->arp_work);
4374 if (bond->params.mode == BOND_MODE_ALB &&
4375 delayed_work_pending(&bond->alb_work))
4376 cancel_delayed_work_sync(&bond->alb_work);
4378 if (bond->params.mode == BOND_MODE_8023AD &&
4379 delayed_work_pending(&bond->ad_work))
4380 cancel_delayed_work_sync(&bond->ad_work);
4382 if (delayed_work_pending(&bond->mcast_work))
4383 cancel_delayed_work_sync(&bond->mcast_work);
4387 * Destroy a bonding device.
4388 * Must be under rtnl_lock when this function is called.
4390 static void bond_uninit(struct net_device *bond_dev)
4392 struct bonding *bond = netdev_priv(bond_dev);
4393 struct vlan_entry *vlan, *tmp;
4395 bond_netpoll_cleanup(bond_dev);
4397 /* Release the bonded slaves */
4398 bond_release_all(bond_dev);
4400 list_del(&bond->bond_list);
4402 bond_work_cancel_all(bond);
4404 bond_remove_proc_entry(bond);
4406 bond_debug_unregister(bond);
4408 __hw_addr_flush(&bond->mc_list);
4410 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4411 list_del(&vlan->vlan_list);
4416 /*------------------------- Module initialization ---------------------------*/
4419 * Convert string input module parms. Accept either the
4420 * number of the mode or its string name. A bit complicated because
4421 * some mode names are substrings of other names, and calls from sysfs
4422 * may have whitespace in the name (trailing newlines, for example).
4424 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4426 int modeint = -1, i, rv;
4427 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4429 for (p = (char *)buf; *p; p++)
4430 if (!(isdigit(*p) || isspace(*p)))
4434 rv = sscanf(buf, "%20s", modestr);
4436 rv = sscanf(buf, "%d", &modeint);
4441 for (i = 0; tbl[i].modename; i++) {
4442 if (modeint == tbl[i].mode)
4444 if (strcmp(modestr, tbl[i].modename) == 0)
4451 static int bond_check_params(struct bond_params *params)
4453 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4456 * Convert string parameters.
4459 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4460 if (bond_mode == -1) {
4461 pr_err("Error: Invalid bonding mode \"%s\"\n",
4462 mode == NULL ? "NULL" : mode);
4467 if (xmit_hash_policy) {
4468 if ((bond_mode != BOND_MODE_XOR) &&
4469 (bond_mode != BOND_MODE_8023AD)) {
4470 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4471 bond_mode_name(bond_mode));
4473 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4475 if (xmit_hashtype == -1) {
4476 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4477 xmit_hash_policy == NULL ? "NULL" :
4485 if (bond_mode != BOND_MODE_8023AD) {
4486 pr_info("lacp_rate param is irrelevant in mode %s\n",
4487 bond_mode_name(bond_mode));
4489 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4490 if (lacp_fast == -1) {
4491 pr_err("Error: Invalid lacp rate \"%s\"\n",
4492 lacp_rate == NULL ? "NULL" : lacp_rate);
4499 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4500 if (params->ad_select == -1) {
4501 pr_err("Error: Invalid ad_select \"%s\"\n",
4502 ad_select == NULL ? "NULL" : ad_select);
4506 if (bond_mode != BOND_MODE_8023AD) {
4507 pr_warning("ad_select param only affects 802.3ad mode\n");
4510 params->ad_select = BOND_AD_STABLE;
4513 if (max_bonds < 0) {
4514 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4515 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4516 max_bonds = BOND_DEFAULT_MAX_BONDS;
4520 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4521 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4522 miimon = BOND_LINK_MON_INTERV;
4526 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4531 if (downdelay < 0) {
4532 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4533 downdelay, INT_MAX);
4537 if ((use_carrier != 0) && (use_carrier != 1)) {
4538 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4543 if (num_peer_notif < 0 || num_peer_notif > 255) {
4544 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4549 /* reset values for 802.3ad */
4550 if (bond_mode == BOND_MODE_8023AD) {
4552 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");
4553 pr_warning("Forcing miimon to 100msec\n");
4558 if (tx_queues < 1 || tx_queues > 255) {
4559 pr_warning("Warning: tx_queues (%d) should be between "
4560 "1 and 255, resetting to %d\n",
4561 tx_queues, BOND_DEFAULT_TX_QUEUES);
4562 tx_queues = BOND_DEFAULT_TX_QUEUES;
4565 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4566 pr_warning("Warning: all_slaves_active module parameter (%d), "
4567 "not of valid value (0/1), so it was set to "
4568 "0\n", all_slaves_active);
4569 all_slaves_active = 0;
4572 if (resend_igmp < 0 || resend_igmp > 255) {
4573 pr_warning("Warning: resend_igmp (%d) should be between "
4574 "0 and 255, resetting to %d\n",
4575 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4576 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4579 /* reset values for TLB/ALB */
4580 if ((bond_mode == BOND_MODE_TLB) ||
4581 (bond_mode == BOND_MODE_ALB)) {
4583 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");
4584 pr_warning("Forcing miimon to 100msec\n");
4589 if (bond_mode == BOND_MODE_ALB) {
4590 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",
4595 if (updelay || downdelay) {
4596 /* just warn the user the up/down delay will have
4597 * no effect since miimon is zero...
4599 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",
4600 updelay, downdelay);
4603 /* don't allow arp monitoring */
4605 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4606 miimon, arp_interval);
4610 if ((updelay % miimon) != 0) {
4611 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4613 (updelay / miimon) * miimon);
4618 if ((downdelay % miimon) != 0) {
4619 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4621 (downdelay / miimon) * miimon);
4624 downdelay /= miimon;
4627 if (arp_interval < 0) {
4628 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4629 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4630 arp_interval = BOND_LINK_ARP_INTERV;
4633 for (arp_ip_count = 0;
4634 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4636 /* not complete check, but should be good enough to
4638 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4639 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4640 arp_ip_target[arp_ip_count]);
4643 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4644 arp_target[arp_ip_count] = ip;
4648 if (arp_interval && !arp_ip_count) {
4649 /* don't allow arping if no arp_ip_target given... */
4650 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4656 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4657 pr_err("arp_validate only supported in active-backup mode\n");
4660 if (!arp_interval) {
4661 pr_err("arp_validate requires arp_interval\n");
4665 arp_validate_value = bond_parse_parm(arp_validate,
4667 if (arp_validate_value == -1) {
4668 pr_err("Error: invalid arp_validate \"%s\"\n",
4669 arp_validate == NULL ? "NULL" : arp_validate);
4673 arp_validate_value = 0;
4676 pr_info("MII link monitoring set to %d ms\n", miimon);
4677 } else if (arp_interval) {
4680 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4682 arp_validate_tbl[arp_validate_value].modename,
4685 for (i = 0; i < arp_ip_count; i++)
4686 pr_info(" %s", arp_ip_target[i]);
4690 } else if (max_bonds) {
4691 /* miimon and arp_interval not set, we need one so things
4692 * work as expected, see bonding.txt for details
4694 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");
4697 if (primary && !USES_PRIMARY(bond_mode)) {
4698 /* currently, using a primary only makes sense
4699 * in active backup, TLB or ALB modes
4701 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4702 primary, bond_mode_name(bond_mode));
4706 if (primary && primary_reselect) {
4707 primary_reselect_value = bond_parse_parm(primary_reselect,
4709 if (primary_reselect_value == -1) {
4710 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4712 NULL ? "NULL" : primary_reselect);
4716 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4719 if (fail_over_mac) {
4720 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4722 if (fail_over_mac_value == -1) {
4723 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4724 arp_validate == NULL ? "NULL" : arp_validate);
4728 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4729 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4731 fail_over_mac_value = BOND_FOM_NONE;
4734 /* fill params struct with the proper values */
4735 params->mode = bond_mode;
4736 params->xmit_policy = xmit_hashtype;
4737 params->miimon = miimon;
4738 params->num_peer_notif = num_peer_notif;
4739 params->arp_interval = arp_interval;
4740 params->arp_validate = arp_validate_value;
4741 params->updelay = updelay;
4742 params->downdelay = downdelay;
4743 params->use_carrier = use_carrier;
4744 params->lacp_fast = lacp_fast;
4745 params->primary[0] = 0;
4746 params->primary_reselect = primary_reselect_value;
4747 params->fail_over_mac = fail_over_mac_value;
4748 params->tx_queues = tx_queues;
4749 params->all_slaves_active = all_slaves_active;
4750 params->resend_igmp = resend_igmp;
4751 params->min_links = min_links;
4754 strncpy(params->primary, primary, IFNAMSIZ);
4755 params->primary[IFNAMSIZ - 1] = 0;
4758 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4763 static struct lock_class_key bonding_netdev_xmit_lock_key;
4764 static struct lock_class_key bonding_netdev_addr_lock_key;
4766 static void bond_set_lockdep_class_one(struct net_device *dev,
4767 struct netdev_queue *txq,
4770 lockdep_set_class(&txq->_xmit_lock,
4771 &bonding_netdev_xmit_lock_key);
4774 static void bond_set_lockdep_class(struct net_device *dev)
4776 lockdep_set_class(&dev->addr_list_lock,
4777 &bonding_netdev_addr_lock_key);
4778 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4782 * Called from registration process
4784 static int bond_init(struct net_device *bond_dev)
4786 struct bonding *bond = netdev_priv(bond_dev);
4787 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4788 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4790 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4793 * Initialize locks that may be required during
4794 * en/deslave operations. All of the bond_open work
4795 * (of which this is part) should really be moved to
4796 * a phase prior to dev_open
4798 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4799 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4801 bond->wq = create_singlethread_workqueue(bond_dev->name);
4805 bond_set_lockdep_class(bond_dev);
4807 bond_create_proc_entry(bond);
4808 list_add_tail(&bond->bond_list, &bn->dev_list);
4810 bond_prepare_sysfs_group(bond);
4812 bond_debug_register(bond);
4814 __hw_addr_init(&bond->mc_list);
4818 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4820 if (tb[IFLA_ADDRESS]) {
4821 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4823 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4824 return -EADDRNOTAVAIL;
4829 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[],
4830 unsigned int *num_queues,
4831 unsigned int *real_num_queues)
4833 *num_queues = tx_queues;
4837 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4839 .priv_size = sizeof(struct bonding),
4840 .setup = bond_setup,
4841 .validate = bond_validate,
4842 .get_tx_queues = bond_get_tx_queues,
4845 /* Create a new bond based on the specified name and bonding parameters.
4846 * If name is NULL, obtain a suitable "bond%d" name for us.
4847 * Caller must NOT hold rtnl_lock; we need to release it here before we
4848 * set up our sysfs entries.
4850 int bond_create(struct net *net, const char *name)
4852 struct net_device *bond_dev;
4857 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4858 name ? name : "bond%d",
4859 bond_setup, tx_queues);
4861 pr_err("%s: eek! can't alloc netdev!\n", name);
4866 dev_net_set(bond_dev, net);
4867 bond_dev->rtnl_link_ops = &bond_link_ops;
4869 res = register_netdevice(bond_dev);
4871 netif_carrier_off(bond_dev);
4875 bond_destructor(bond_dev);
4879 static int __net_init bond_net_init(struct net *net)
4881 struct bond_net *bn = net_generic(net, bond_net_id);
4884 INIT_LIST_HEAD(&bn->dev_list);
4886 bond_create_proc_dir(bn);
4887 bond_create_sysfs(bn);
4892 static void __net_exit bond_net_exit(struct net *net)
4894 struct bond_net *bn = net_generic(net, bond_net_id);
4896 bond_destroy_sysfs(bn);
4897 bond_destroy_proc_dir(bn);
4900 static struct pernet_operations bond_net_ops = {
4901 .init = bond_net_init,
4902 .exit = bond_net_exit,
4904 .size = sizeof(struct bond_net),
4907 static int __init bonding_init(void)
4912 pr_info("%s", bond_version);
4914 res = bond_check_params(&bonding_defaults);
4918 res = register_pernet_subsys(&bond_net_ops);
4922 res = rtnl_link_register(&bond_link_ops);
4926 bond_create_debugfs();
4928 for (i = 0; i < max_bonds; i++) {
4929 res = bond_create(&init_net, NULL);
4934 register_netdevice_notifier(&bond_netdev_notifier);
4938 rtnl_link_unregister(&bond_link_ops);
4940 unregister_pernet_subsys(&bond_net_ops);
4945 static void __exit bonding_exit(void)
4947 unregister_netdevice_notifier(&bond_netdev_notifier);
4949 bond_destroy_debugfs();
4951 rtnl_link_unregister(&bond_link_ops);
4952 unregister_pernet_subsys(&bond_net_ops);
4954 #ifdef CONFIG_NET_POLL_CONTROLLER
4956 * Make sure we don't have an imbalance on our netpoll blocking
4958 WARN_ON(atomic_read(&netpoll_block_tx));
4962 module_init(bonding_init);
4963 module_exit(bonding_exit);
4964 MODULE_LICENSE("GPL");
4965 MODULE_VERSION(DRV_VERSION);
4966 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4967 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4968 MODULE_ALIAS_RTNL_LINK("bond");