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>
79 #include <net/pkt_sched.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV 0
88 #define BOND_LINK_ARP_INTERV 0
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon = BOND_LINK_MON_INTERV;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118 "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121 "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149 "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152 "0 for stable (default), 1 for bandwidth, "
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159 "0 for layer 2 (default), 1 for layer 3+4, "
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167 "0 for none (default), 1 for active, "
168 "2 for backup, 3 for all");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171 "the same MAC; 0 for none (default), "
172 "1 for active, 2 for follow");
173 module_param(all_slaves_active, int, 0);
174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175 "by setting active flag for all slaves; "
176 "0 for never (default), 1 for always.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
181 /*----------------------------- Global variables ----------------------------*/
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
187 int bond_net_id __read_mostly;
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode = BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 { "slow", AD_LACP_SLOW},
197 { "fast", AD_LACP_FAST},
201 const struct bond_parm_tbl bond_mode_tbl[] = {
202 { "balance-rr", BOND_MODE_ROUNDROBIN},
203 { "active-backup", BOND_MODE_ACTIVEBACKUP},
204 { "balance-xor", BOND_MODE_XOR},
205 { "broadcast", BOND_MODE_BROADCAST},
206 { "802.3ad", BOND_MODE_8023AD},
207 { "balance-tlb", BOND_MODE_TLB},
208 { "balance-alb", BOND_MODE_ALB},
212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213 { "layer2", BOND_XMIT_POLICY_LAYER2},
214 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
215 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
219 const struct bond_parm_tbl arp_validate_tbl[] = {
220 { "none", BOND_ARP_VALIDATE_NONE},
221 { "active", BOND_ARP_VALIDATE_ACTIVE},
222 { "backup", BOND_ARP_VALIDATE_BACKUP},
223 { "all", BOND_ARP_VALIDATE_ALL},
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 { "none", BOND_FOM_NONE},
229 { "active", BOND_FOM_ACTIVE},
230 { "follow", BOND_FOM_FOLLOW},
234 const struct bond_parm_tbl pri_reselect_tbl[] = {
235 { "always", BOND_PRI_RESELECT_ALWAYS},
236 { "better", BOND_PRI_RESELECT_BETTER},
237 { "failure", BOND_PRI_RESELECT_FAILURE},
241 struct bond_parm_tbl ad_select_tbl[] = {
242 { "stable", BOND_AD_STABLE},
243 { "bandwidth", BOND_AD_BANDWIDTH},
244 { "count", BOND_AD_COUNT},
248 /*-------------------------- Forward declarations ---------------------------*/
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
253 /*---------------------------- General routines -----------------------------*/
255 const char *bond_mode_name(int mode)
257 static const char *names[] = {
258 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260 [BOND_MODE_XOR] = "load balancing (xor)",
261 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263 [BOND_MODE_TLB] = "transmit load balancing",
264 [BOND_MODE_ALB] = "adaptive load balancing",
267 if (mode < 0 || mode > BOND_MODE_ALB)
273 /*---------------------------------- VLAN -----------------------------------*/
276 * bond_add_vlan - add a new vlan id on bond
277 * @bond: bond that got the notification
278 * @vlan_id: the vlan id to add
280 * Returns -ENOMEM if allocation failed.
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
284 struct vlan_entry *vlan;
286 pr_debug("bond: %s, vlan id %d\n",
287 (bond ? bond->dev->name : "None"), vlan_id);
289 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
293 INIT_LIST_HEAD(&vlan->vlan_list);
294 vlan->vlan_id = vlan_id;
296 write_lock_bh(&bond->lock);
298 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
300 write_unlock_bh(&bond->lock);
302 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
308 * bond_del_vlan - delete a vlan id from bond
309 * @bond: bond that got the notification
310 * @vlan_id: the vlan id to delete
312 * returns -ENODEV if @vlan_id was not found in @bond.
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
316 struct vlan_entry *vlan;
319 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
322 write_lock_bh(&bond->lock);
324 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325 if (vlan->vlan_id == vlan_id) {
326 list_del(&vlan->vlan_list);
328 if (bond_is_lb(bond))
329 bond_alb_clear_vlan(bond, vlan_id);
331 pr_debug("removed VLAN ID %d from bond %s\n",
332 vlan_id, bond->dev->name);
341 pr_debug("couldn't find VLAN ID %d in bond %s\n",
342 vlan_id, bond->dev->name);
345 write_unlock_bh(&bond->lock);
346 unblock_netpoll_tx();
351 * bond_next_vlan - safely skip to the next item in the vlans list.
352 * @bond: the bond we're working on
353 * @curr: item we're advancing from
355 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356 * or @curr->next otherwise (even if it is @curr itself again).
358 * Caller must hold bond->lock
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
362 struct vlan_entry *next, *last;
364 if (list_empty(&bond->vlan_list))
368 next = list_entry(bond->vlan_list.next,
369 struct vlan_entry, vlan_list);
371 last = list_entry(bond->vlan_list.prev,
372 struct vlan_entry, vlan_list);
374 next = list_entry(bond->vlan_list.next,
375 struct vlan_entry, vlan_list);
377 next = list_entry(curr->vlan_list.next,
378 struct vlan_entry, vlan_list);
386 * bond_dev_queue_xmit - Prepare skb for xmit.
388 * @bond: bond device that got this skb for tx.
389 * @skb: hw accel VLAN tagged skb to transmit
390 * @slave_dev: slave that is supposed to xmit this skbuff
392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
393 struct net_device *slave_dev)
395 skb->dev = slave_dev;
397 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
398 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
399 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
401 if (unlikely(netpoll_tx_running(bond->dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
432 __be16 proto, u16 vid)
434 struct bonding *bond = netdev_priv(bond_dev);
435 struct slave *slave, *stop_at;
438 bond_for_each_slave(bond, slave, i) {
439 res = vlan_vid_add(slave->dev, proto, vid);
444 res = bond_add_vlan(bond, vid);
446 pr_err("%s: Error: Failed to add vlan id %d\n",
447 bond_dev->name, vid);
454 /* unwind from head to the slave that failed */
456 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
457 vlan_vid_del(slave->dev, proto, vid);
463 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464 * @bond_dev: bonding net device that got called
465 * @vid: vlan id being removed
467 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
468 __be16 proto, u16 vid)
470 struct bonding *bond = netdev_priv(bond_dev);
474 bond_for_each_slave(bond, slave, i)
475 vlan_vid_del(slave->dev, proto, vid);
477 res = bond_del_vlan(bond, vid);
479 pr_err("%s: Error: Failed to remove vlan id %d\n",
480 bond_dev->name, vid);
487 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
489 struct vlan_entry *vlan;
492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
493 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
496 pr_warning("%s: Failed to add vlan id %d to device %s\n",
497 bond->dev->name, vlan->vlan_id,
502 static void bond_del_vlans_from_slave(struct bonding *bond,
503 struct net_device *slave_dev)
505 struct vlan_entry *vlan;
507 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
510 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
514 /*------------------------------- Link status -------------------------------*/
517 * Set the carrier state for the master according to the state of its
518 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
519 * do special 802.3ad magic.
521 * Returns zero if carrier state does not change, nonzero if it does.
523 static int bond_set_carrier(struct bonding *bond)
528 if (bond->slave_cnt == 0)
531 if (bond->params.mode == BOND_MODE_8023AD)
532 return bond_3ad_set_carrier(bond);
534 bond_for_each_slave(bond, slave, i) {
535 if (slave->link == BOND_LINK_UP) {
536 if (!netif_carrier_ok(bond->dev)) {
537 netif_carrier_on(bond->dev);
545 if (netif_carrier_ok(bond->dev)) {
546 netif_carrier_off(bond->dev);
553 * Get link speed and duplex from the slave's base driver
554 * using ethtool. If for some reason the call fails or the
555 * values are invalid, set speed and duplex to -1,
558 static void bond_update_speed_duplex(struct slave *slave)
560 struct net_device *slave_dev = slave->dev;
561 struct ethtool_cmd ecmd;
565 slave->speed = SPEED_UNKNOWN;
566 slave->duplex = DUPLEX_UNKNOWN;
568 res = __ethtool_get_settings(slave_dev, &ecmd);
572 slave_speed = ethtool_cmd_speed(&ecmd);
573 if (slave_speed == 0 || slave_speed == ((__u32) -1))
576 switch (ecmd.duplex) {
584 slave->speed = slave_speed;
585 slave->duplex = ecmd.duplex;
591 * if <dev> supports MII link status reporting, check its link status.
593 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
594 * depending upon the setting of the use_carrier parameter.
596 * Return either BMSR_LSTATUS, meaning that the link is up (or we
597 * can't tell and just pretend it is), or 0, meaning that the link is
600 * If reporting is non-zero, instead of faking link up, return -1 if
601 * both ETHTOOL and MII ioctls fail (meaning the device does not
602 * support them). If use_carrier is set, return whatever it says.
603 * It'd be nice if there was a good way to tell if a driver supports
604 * netif_carrier, but there really isn't.
606 static int bond_check_dev_link(struct bonding *bond,
607 struct net_device *slave_dev, int reporting)
609 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
610 int (*ioctl)(struct net_device *, struct ifreq *, int);
612 struct mii_ioctl_data *mii;
614 if (!reporting && !netif_running(slave_dev))
617 if (bond->params.use_carrier)
618 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
620 /* Try to get link status using Ethtool first. */
621 if (slave_dev->ethtool_ops->get_link)
622 return slave_dev->ethtool_ops->get_link(slave_dev) ?
625 /* Ethtool can't be used, fallback to MII ioctls. */
626 ioctl = slave_ops->ndo_do_ioctl;
628 /* TODO: set pointer to correct ioctl on a per team member */
629 /* bases to make this more efficient. that is, once */
630 /* we determine the correct ioctl, we will always */
631 /* call it and not the others for that team */
635 * We cannot assume that SIOCGMIIPHY will also read a
636 * register; not all network drivers (e.g., e100)
640 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
641 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
643 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
644 mii->reg_num = MII_BMSR;
645 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
646 return mii->val_out & BMSR_LSTATUS;
651 * If reporting, report that either there's no dev->do_ioctl,
652 * or both SIOCGMIIREG and get_link failed (meaning that we
653 * cannot report link status). If not reporting, pretend
656 return reporting ? -1 : BMSR_LSTATUS;
659 /*----------------------------- Multicast list ------------------------------*/
662 * Push the promiscuity flag down to appropriate slaves
664 static int bond_set_promiscuity(struct bonding *bond, int inc)
667 if (USES_PRIMARY(bond->params.mode)) {
668 /* write lock already acquired */
669 if (bond->curr_active_slave) {
670 err = dev_set_promiscuity(bond->curr_active_slave->dev,
676 bond_for_each_slave(bond, slave, i) {
677 err = dev_set_promiscuity(slave->dev, inc);
686 * Push the allmulti flag down to all slaves
688 static int bond_set_allmulti(struct bonding *bond, int inc)
691 if (USES_PRIMARY(bond->params.mode)) {
692 /* write lock already acquired */
693 if (bond->curr_active_slave) {
694 err = dev_set_allmulti(bond->curr_active_slave->dev,
700 bond_for_each_slave(bond, slave, i) {
701 err = dev_set_allmulti(slave->dev, inc);
710 * Add a Multicast address to slaves
713 static void bond_mc_add(struct bonding *bond, void *addr)
715 if (USES_PRIMARY(bond->params.mode)) {
716 /* write lock already acquired */
717 if (bond->curr_active_slave)
718 dev_mc_add(bond->curr_active_slave->dev, addr);
723 bond_for_each_slave(bond, slave, i)
724 dev_mc_add(slave->dev, addr);
729 * Remove a multicast address from slave
732 static void bond_mc_del(struct bonding *bond, void *addr)
734 if (USES_PRIMARY(bond->params.mode)) {
735 /* write lock already acquired */
736 if (bond->curr_active_slave)
737 dev_mc_del(bond->curr_active_slave->dev, addr);
741 bond_for_each_slave(bond, slave, i) {
742 dev_mc_del(slave->dev, addr);
748 static void __bond_resend_igmp_join_requests(struct net_device *dev)
750 struct in_device *in_dev;
752 in_dev = __in_dev_get_rcu(dev);
754 ip_mc_rejoin_groups(in_dev);
758 * Retrieve the list of registered multicast addresses for the bonding
759 * device and retransmit an IGMP JOIN request to the current active
762 static void bond_resend_igmp_join_requests(struct bonding *bond)
764 struct net_device *bond_dev, *vlan_dev, *upper_dev;
765 struct vlan_entry *vlan;
768 read_lock(&bond->lock);
770 bond_dev = bond->dev;
772 /* rejoin all groups on bond device */
773 __bond_resend_igmp_join_requests(bond_dev);
776 * if bond is enslaved to a bridge,
777 * then rejoin all groups on its master
779 upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
780 if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
781 __bond_resend_igmp_join_requests(upper_dev);
783 /* rejoin all groups on vlan devices */
784 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
785 vlan_dev = __vlan_find_dev_deep(bond_dev, htons(ETH_P_8021Q),
788 __bond_resend_igmp_join_requests(vlan_dev);
791 if (--bond->igmp_retrans > 0)
792 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
794 read_unlock(&bond->lock);
798 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
800 struct bonding *bond = container_of(work, struct bonding,
803 bond_resend_igmp_join_requests(bond);
807 * flush all members of flush->mc_list from device dev->mc_list
809 static void bond_mc_list_flush(struct net_device *bond_dev,
810 struct net_device *slave_dev)
812 struct bonding *bond = netdev_priv(bond_dev);
813 struct netdev_hw_addr *ha;
815 netdev_for_each_mc_addr(ha, bond_dev)
816 dev_mc_del(slave_dev, ha->addr);
818 if (bond->params.mode == BOND_MODE_8023AD) {
819 /* del lacpdu mc addr from mc list */
820 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
822 dev_mc_del(slave_dev, lacpdu_multicast);
826 /*--------------------------- Active slave change ---------------------------*/
829 * Update the mc list and multicast-related flags for the new and
830 * old active slaves (if any) according to the multicast mode, and
831 * promiscuous flags unconditionally.
833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
834 struct slave *old_active)
836 struct netdev_hw_addr *ha;
838 if (!USES_PRIMARY(bond->params.mode))
839 /* nothing to do - mc list is already up-to-date on
845 if (bond->dev->flags & IFF_PROMISC)
846 dev_set_promiscuity(old_active->dev, -1);
848 if (bond->dev->flags & IFF_ALLMULTI)
849 dev_set_allmulti(old_active->dev, -1);
851 netdev_for_each_mc_addr(ha, bond->dev)
852 dev_mc_del(old_active->dev, ha->addr);
856 /* FIXME: Signal errors upstream. */
857 if (bond->dev->flags & IFF_PROMISC)
858 dev_set_promiscuity(new_active->dev, 1);
860 if (bond->dev->flags & IFF_ALLMULTI)
861 dev_set_allmulti(new_active->dev, 1);
863 netdev_for_each_mc_addr(ha, bond->dev)
864 dev_mc_add(new_active->dev, ha->addr);
869 * bond_do_fail_over_mac
871 * Perform special MAC address swapping for fail_over_mac settings
873 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
875 static void bond_do_fail_over_mac(struct bonding *bond,
876 struct slave *new_active,
877 struct slave *old_active)
878 __releases(&bond->curr_slave_lock)
879 __releases(&bond->lock)
880 __acquires(&bond->lock)
881 __acquires(&bond->curr_slave_lock)
883 u8 tmp_mac[ETH_ALEN];
884 struct sockaddr saddr;
887 switch (bond->params.fail_over_mac) {
888 case BOND_FOM_ACTIVE:
890 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
891 new_active->dev->addr_len);
892 write_unlock_bh(&bond->curr_slave_lock);
893 read_unlock(&bond->lock);
894 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
895 read_lock(&bond->lock);
896 write_lock_bh(&bond->curr_slave_lock);
899 case BOND_FOM_FOLLOW:
901 * if new_active && old_active, swap them
902 * if just old_active, do nothing (going to no active slave)
903 * if just new_active, set new_active to bond's MAC
908 write_unlock_bh(&bond->curr_slave_lock);
909 read_unlock(&bond->lock);
912 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
913 memcpy(saddr.sa_data, old_active->dev->dev_addr,
915 saddr.sa_family = new_active->dev->type;
917 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
918 saddr.sa_family = bond->dev->type;
921 rv = dev_set_mac_address(new_active->dev, &saddr);
923 pr_err("%s: Error %d setting MAC of slave %s\n",
924 bond->dev->name, -rv, new_active->dev->name);
931 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
932 saddr.sa_family = old_active->dev->type;
934 rv = dev_set_mac_address(old_active->dev, &saddr);
936 pr_err("%s: Error %d setting MAC of slave %s\n",
937 bond->dev->name, -rv, new_active->dev->name);
939 read_lock(&bond->lock);
940 write_lock_bh(&bond->curr_slave_lock);
943 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
944 bond->dev->name, bond->params.fail_over_mac);
950 static bool bond_should_change_active(struct bonding *bond)
952 struct slave *prim = bond->primary_slave;
953 struct slave *curr = bond->curr_active_slave;
955 if (!prim || !curr || curr->link != BOND_LINK_UP)
957 if (bond->force_primary) {
958 bond->force_primary = false;
961 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
962 (prim->speed < curr->speed ||
963 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
965 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
971 * find_best_interface - select the best available slave to be the active one
972 * @bond: our bonding struct
974 * Warning: Caller must hold curr_slave_lock for writing.
976 static struct slave *bond_find_best_slave(struct bonding *bond)
978 struct slave *new_active, *old_active;
979 struct slave *bestslave = NULL;
980 int mintime = bond->params.updelay;
983 new_active = bond->curr_active_slave;
985 if (!new_active) { /* there were no active slaves left */
986 if (bond->slave_cnt > 0) /* found one slave */
987 new_active = bond->first_slave;
989 return NULL; /* still no slave, return NULL */
992 if ((bond->primary_slave) &&
993 bond->primary_slave->link == BOND_LINK_UP &&
994 bond_should_change_active(bond)) {
995 new_active = bond->primary_slave;
998 /* remember where to stop iterating over the slaves */
999 old_active = new_active;
1001 bond_for_each_slave_from(bond, new_active, i, old_active) {
1002 if (new_active->link == BOND_LINK_UP) {
1004 } else if (new_active->link == BOND_LINK_BACK &&
1005 IS_UP(new_active->dev)) {
1006 /* link up, but waiting for stabilization */
1007 if (new_active->delay < mintime) {
1008 mintime = new_active->delay;
1009 bestslave = new_active;
1017 static bool bond_should_notify_peers(struct bonding *bond)
1019 struct slave *slave = bond->curr_active_slave;
1021 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1022 bond->dev->name, slave ? slave->dev->name : "NULL");
1024 if (!slave || !bond->send_peer_notif ||
1025 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1028 bond->send_peer_notif--;
1033 * change_active_interface - change the active slave into the specified one
1034 * @bond: our bonding struct
1035 * @new: the new slave to make the active one
1037 * Set the new slave to the bond's settings and unset them on the old
1038 * curr_active_slave.
1039 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1041 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1042 * because it is apparently the best available slave we have, even though its
1043 * updelay hasn't timed out yet.
1045 * If new_active is not NULL, caller must hold bond->lock for read and
1046 * curr_slave_lock for write_bh.
1048 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1050 struct slave *old_active = bond->curr_active_slave;
1052 if (old_active == new_active)
1056 new_active->jiffies = jiffies;
1058 if (new_active->link == BOND_LINK_BACK) {
1059 if (USES_PRIMARY(bond->params.mode)) {
1060 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1061 bond->dev->name, new_active->dev->name,
1062 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1065 new_active->delay = 0;
1066 new_active->link = BOND_LINK_UP;
1068 if (bond->params.mode == BOND_MODE_8023AD)
1069 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1071 if (bond_is_lb(bond))
1072 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1074 if (USES_PRIMARY(bond->params.mode)) {
1075 pr_info("%s: making interface %s the new active one.\n",
1076 bond->dev->name, new_active->dev->name);
1081 if (USES_PRIMARY(bond->params.mode))
1082 bond_mc_swap(bond, new_active, old_active);
1084 if (bond_is_lb(bond)) {
1085 bond_alb_handle_active_change(bond, new_active);
1087 bond_set_slave_inactive_flags(old_active);
1089 bond_set_slave_active_flags(new_active);
1091 bond->curr_active_slave = new_active;
1094 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1096 bond_set_slave_inactive_flags(old_active);
1099 bool should_notify_peers = false;
1101 bond_set_slave_active_flags(new_active);
1103 if (bond->params.fail_over_mac)
1104 bond_do_fail_over_mac(bond, new_active,
1107 if (netif_running(bond->dev)) {
1108 bond->send_peer_notif =
1109 bond->params.num_peer_notif;
1110 should_notify_peers =
1111 bond_should_notify_peers(bond);
1114 write_unlock_bh(&bond->curr_slave_lock);
1115 read_unlock(&bond->lock);
1117 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1118 if (should_notify_peers)
1119 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1122 read_lock(&bond->lock);
1123 write_lock_bh(&bond->curr_slave_lock);
1127 /* resend IGMP joins since active slave has changed or
1128 * all were sent on curr_active_slave.
1129 * resend only if bond is brought up with the affected
1130 * bonding modes and the retransmission is enabled */
1131 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1132 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1133 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1134 bond->igmp_retrans = bond->params.resend_igmp;
1135 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1140 * bond_select_active_slave - select a new active slave, if needed
1141 * @bond: our bonding struct
1143 * This functions should be called when one of the following occurs:
1144 * - The old curr_active_slave has been released or lost its link.
1145 * - The primary_slave has got its link back.
1146 * - A slave has got its link back and there's no old curr_active_slave.
1148 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1150 void bond_select_active_slave(struct bonding *bond)
1152 struct slave *best_slave;
1155 best_slave = bond_find_best_slave(bond);
1156 if (best_slave != bond->curr_active_slave) {
1157 bond_change_active_slave(bond, best_slave);
1158 rv = bond_set_carrier(bond);
1162 if (netif_carrier_ok(bond->dev)) {
1163 pr_info("%s: first active interface up!\n",
1166 pr_info("%s: now running without any active interface !\n",
1172 /*--------------------------- slave list handling ---------------------------*/
1175 * This function attaches the slave to the end of list.
1177 * bond->lock held for writing by caller.
1179 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1181 if (bond->first_slave == NULL) { /* attaching the first slave */
1182 new_slave->next = new_slave;
1183 new_slave->prev = new_slave;
1184 bond->first_slave = new_slave;
1186 new_slave->next = bond->first_slave;
1187 new_slave->prev = bond->first_slave->prev;
1188 new_slave->next->prev = new_slave;
1189 new_slave->prev->next = new_slave;
1196 * This function detaches the slave from the list.
1197 * WARNING: no check is made to verify if the slave effectively
1198 * belongs to <bond>.
1199 * Nothing is freed on return, structures are just unchained.
1200 * If any slave pointer in bond was pointing to <slave>,
1201 * it should be changed by the calling function.
1203 * bond->lock held for writing by caller.
1205 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1208 slave->next->prev = slave->prev;
1211 slave->prev->next = slave->next;
1213 if (bond->first_slave == slave) { /* slave is the first slave */
1214 if (bond->slave_cnt > 1) { /* there are more slave */
1215 bond->first_slave = slave->next;
1217 bond->first_slave = NULL; /* slave was the last one */
1226 #ifdef CONFIG_NET_POLL_CONTROLLER
1227 static inline int slave_enable_netpoll(struct slave *slave)
1232 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1237 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1246 static inline void slave_disable_netpoll(struct slave *slave)
1248 struct netpoll *np = slave->np;
1254 __netpoll_free_async(np);
1256 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1258 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1260 if (!slave_dev->netdev_ops->ndo_poll_controller)
1265 static void bond_poll_controller(struct net_device *bond_dev)
1269 static void __bond_netpoll_cleanup(struct bonding *bond)
1271 struct slave *slave;
1274 bond_for_each_slave(bond, slave, i)
1275 if (IS_UP(slave->dev))
1276 slave_disable_netpoll(slave);
1278 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1280 struct bonding *bond = netdev_priv(bond_dev);
1282 read_lock(&bond->lock);
1283 __bond_netpoll_cleanup(bond);
1284 read_unlock(&bond->lock);
1287 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1289 struct bonding *bond = netdev_priv(dev);
1290 struct slave *slave;
1293 read_lock(&bond->lock);
1294 bond_for_each_slave(bond, slave, i) {
1295 err = slave_enable_netpoll(slave);
1297 __bond_netpoll_cleanup(bond);
1301 read_unlock(&bond->lock);
1305 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1307 return bond->dev->npinfo;
1311 static inline int slave_enable_netpoll(struct slave *slave)
1315 static inline void slave_disable_netpoll(struct slave *slave)
1318 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1323 /*---------------------------------- IOCTL ----------------------------------*/
1325 static void bond_set_dev_addr(struct net_device *bond_dev,
1326 struct net_device *slave_dev)
1328 pr_debug("bond_dev=%p\n", bond_dev);
1329 pr_debug("slave_dev=%p\n", slave_dev);
1330 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1331 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1332 bond_dev->addr_assign_type = NET_ADDR_SET;
1333 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1336 static netdev_features_t bond_fix_features(struct net_device *dev,
1337 netdev_features_t features)
1339 struct slave *slave;
1340 struct bonding *bond = netdev_priv(dev);
1341 netdev_features_t mask;
1344 read_lock(&bond->lock);
1346 if (!bond->first_slave) {
1347 /* Disable adding VLANs to empty bond. But why? --mq */
1348 features |= NETIF_F_VLAN_CHALLENGED;
1353 features &= ~NETIF_F_ONE_FOR_ALL;
1354 features |= NETIF_F_ALL_FOR_ALL;
1356 bond_for_each_slave(bond, slave, i) {
1357 features = netdev_increment_features(features,
1358 slave->dev->features,
1363 read_unlock(&bond->lock);
1367 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1368 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1369 NETIF_F_HIGHDMA | NETIF_F_LRO)
1371 static void bond_compute_features(struct bonding *bond)
1373 struct slave *slave;
1374 struct net_device *bond_dev = bond->dev;
1375 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1376 unsigned short max_hard_header_len = ETH_HLEN;
1377 unsigned int gso_max_size = GSO_MAX_SIZE;
1378 u16 gso_max_segs = GSO_MAX_SEGS;
1380 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1382 read_lock(&bond->lock);
1384 if (!bond->first_slave)
1387 bond_for_each_slave(bond, slave, i) {
1388 vlan_features = netdev_increment_features(vlan_features,
1389 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1391 dst_release_flag &= slave->dev->priv_flags;
1392 if (slave->dev->hard_header_len > max_hard_header_len)
1393 max_hard_header_len = slave->dev->hard_header_len;
1395 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1396 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1400 bond_dev->vlan_features = vlan_features;
1401 bond_dev->hard_header_len = max_hard_header_len;
1402 bond_dev->gso_max_segs = gso_max_segs;
1403 netif_set_gso_max_size(bond_dev, gso_max_size);
1405 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1406 bond_dev->priv_flags = flags | dst_release_flag;
1408 read_unlock(&bond->lock);
1410 netdev_change_features(bond_dev);
1413 static void bond_setup_by_slave(struct net_device *bond_dev,
1414 struct net_device *slave_dev)
1416 struct bonding *bond = netdev_priv(bond_dev);
1418 bond_dev->header_ops = slave_dev->header_ops;
1420 bond_dev->type = slave_dev->type;
1421 bond_dev->hard_header_len = slave_dev->hard_header_len;
1422 bond_dev->addr_len = slave_dev->addr_len;
1424 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1425 slave_dev->addr_len);
1426 bond->setup_by_slave = 1;
1429 /* On bonding slaves other than the currently active slave, suppress
1430 * duplicates except for alb non-mcast/bcast.
1432 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1433 struct slave *slave,
1434 struct bonding *bond)
1436 if (bond_is_slave_inactive(slave)) {
1437 if (bond->params.mode == BOND_MODE_ALB &&
1438 skb->pkt_type != PACKET_BROADCAST &&
1439 skb->pkt_type != PACKET_MULTICAST)
1446 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1448 struct sk_buff *skb = *pskb;
1449 struct slave *slave;
1450 struct bonding *bond;
1451 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1453 int ret = RX_HANDLER_ANOTHER;
1455 skb = skb_share_check(skb, GFP_ATOMIC);
1457 return RX_HANDLER_CONSUMED;
1461 slave = bond_slave_get_rcu(skb->dev);
1464 if (bond->params.arp_interval)
1465 slave->dev->last_rx = jiffies;
1467 recv_probe = ACCESS_ONCE(bond->recv_probe);
1469 ret = recv_probe(skb, bond, slave);
1470 if (ret == RX_HANDLER_CONSUMED) {
1476 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1477 return RX_HANDLER_EXACT;
1480 skb->dev = bond->dev;
1482 if (bond->params.mode == BOND_MODE_ALB &&
1483 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1484 skb->pkt_type == PACKET_HOST) {
1486 if (unlikely(skb_cow_head(skb,
1487 skb->data - skb_mac_header(skb)))) {
1489 return RX_HANDLER_CONSUMED;
1491 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1497 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1498 struct net_device *slave_dev)
1502 err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1505 slave_dev->flags |= IFF_SLAVE;
1506 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1510 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1511 struct net_device *slave_dev)
1513 netdev_upper_dev_unlink(slave_dev, bond_dev);
1514 slave_dev->flags &= ~IFF_SLAVE;
1515 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1518 /* enslave device <slave> to bond device <master> */
1519 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1521 struct bonding *bond = netdev_priv(bond_dev);
1522 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1523 struct slave *new_slave = NULL;
1524 struct netdev_hw_addr *ha;
1525 struct sockaddr addr;
1529 if (!bond->params.use_carrier &&
1530 slave_dev->ethtool_ops->get_link == NULL &&
1531 slave_ops->ndo_do_ioctl == NULL) {
1532 pr_warning("%s: Warning: no link monitoring support for %s\n",
1533 bond_dev->name, slave_dev->name);
1536 /* already enslaved */
1537 if (slave_dev->flags & IFF_SLAVE) {
1538 pr_debug("Error, Device was already enslaved\n");
1542 /* vlan challenged mutual exclusion */
1543 /* no need to lock since we're protected by rtnl_lock */
1544 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1545 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1546 if (vlan_uses_dev(bond_dev)) {
1547 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1548 bond_dev->name, slave_dev->name, bond_dev->name);
1551 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1552 bond_dev->name, slave_dev->name,
1553 slave_dev->name, bond_dev->name);
1556 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1560 * Old ifenslave binaries are no longer supported. These can
1561 * be identified with moderate accuracy by the state of the slave:
1562 * the current ifenslave will set the interface down prior to
1563 * enslaving it; the old ifenslave will not.
1565 if ((slave_dev->flags & IFF_UP)) {
1566 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1569 goto err_undo_flags;
1572 /* set bonding device ether type by slave - bonding netdevices are
1573 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1574 * there is a need to override some of the type dependent attribs/funcs.
1576 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1577 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1579 if (bond->slave_cnt == 0) {
1580 if (bond_dev->type != slave_dev->type) {
1581 pr_debug("%s: change device type from %d to %d\n",
1583 bond_dev->type, slave_dev->type);
1585 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1587 res = notifier_to_errno(res);
1589 pr_err("%s: refused to change device type\n",
1592 goto err_undo_flags;
1595 /* Flush unicast and multicast addresses */
1596 dev_uc_flush(bond_dev);
1597 dev_mc_flush(bond_dev);
1599 if (slave_dev->type != ARPHRD_ETHER)
1600 bond_setup_by_slave(bond_dev, slave_dev);
1602 ether_setup(bond_dev);
1603 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1606 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1609 } else if (bond_dev->type != slave_dev->type) {
1610 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1612 slave_dev->type, bond_dev->type);
1614 goto err_undo_flags;
1617 if (slave_ops->ndo_set_mac_address == NULL) {
1618 if (bond->slave_cnt == 0) {
1619 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1621 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1622 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1623 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",
1626 goto err_undo_flags;
1630 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1632 /* If this is the first slave, then we need to set the master's hardware
1633 * address to be the same as the slave's. */
1634 if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1635 bond_set_dev_addr(bond->dev, slave_dev);
1637 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1640 goto err_undo_flags;
1644 * Set the new_slave's queue_id to be zero. Queue ID mapping
1645 * is set via sysfs or module option if desired.
1647 new_slave->queue_id = 0;
1649 /* Save slave's original mtu and then set it to match the bond */
1650 new_slave->original_mtu = slave_dev->mtu;
1651 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1653 pr_debug("Error %d calling dev_set_mtu\n", res);
1658 * Save slave's original ("permanent") mac address for modes
1659 * that need it, and for restoring it upon release, and then
1660 * set it to the master's address
1662 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1664 if (!bond->params.fail_over_mac) {
1666 * Set slave to master's mac address. The application already
1667 * set the master's mac address to that of the first slave
1669 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1670 addr.sa_family = slave_dev->type;
1671 res = dev_set_mac_address(slave_dev, &addr);
1673 pr_debug("Error %d calling set_mac_address\n", res);
1674 goto err_restore_mtu;
1678 res = bond_master_upper_dev_link(bond_dev, slave_dev);
1680 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1681 goto err_restore_mac;
1684 /* open the slave since the application closed it */
1685 res = dev_open(slave_dev);
1687 pr_debug("Opening slave %s failed\n", slave_dev->name);
1688 goto err_unset_master;
1691 new_slave->bond = bond;
1692 new_slave->dev = slave_dev;
1693 slave_dev->priv_flags |= IFF_BONDING;
1695 if (bond_is_lb(bond)) {
1696 /* bond_alb_init_slave() must be called before all other stages since
1697 * it might fail and we do not want to have to undo everything
1699 res = bond_alb_init_slave(bond, new_slave);
1704 /* If the mode USES_PRIMARY, then the new slave gets the
1705 * master's promisc (and mc) settings only if it becomes the
1706 * curr_active_slave, and that is taken care of later when calling
1707 * bond_change_active()
1709 if (!USES_PRIMARY(bond->params.mode)) {
1710 /* set promiscuity level to new slave */
1711 if (bond_dev->flags & IFF_PROMISC) {
1712 res = dev_set_promiscuity(slave_dev, 1);
1717 /* set allmulti level to new slave */
1718 if (bond_dev->flags & IFF_ALLMULTI) {
1719 res = dev_set_allmulti(slave_dev, 1);
1724 netif_addr_lock_bh(bond_dev);
1725 /* upload master's mc_list to new slave */
1726 netdev_for_each_mc_addr(ha, bond_dev)
1727 dev_mc_add(slave_dev, ha->addr);
1728 netif_addr_unlock_bh(bond_dev);
1731 if (bond->params.mode == BOND_MODE_8023AD) {
1732 /* add lacpdu mc addr to mc list */
1733 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1735 dev_mc_add(slave_dev, lacpdu_multicast);
1738 bond_add_vlans_on_slave(bond, slave_dev);
1740 write_lock_bh(&bond->lock);
1742 bond_attach_slave(bond, new_slave);
1744 new_slave->delay = 0;
1745 new_slave->link_failure_count = 0;
1747 write_unlock_bh(&bond->lock);
1749 bond_compute_features(bond);
1751 bond_update_speed_duplex(new_slave);
1753 read_lock(&bond->lock);
1755 new_slave->last_arp_rx = jiffies -
1756 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1758 if (bond->params.miimon && !bond->params.use_carrier) {
1759 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1761 if ((link_reporting == -1) && !bond->params.arp_interval) {
1763 * miimon is set but a bonded network driver
1764 * does not support ETHTOOL/MII and
1765 * arp_interval is not set. Note: if
1766 * use_carrier is enabled, we will never go
1767 * here (because netif_carrier is always
1768 * supported); thus, we don't need to change
1769 * the messages for netif_carrier.
1771 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",
1772 bond_dev->name, slave_dev->name);
1773 } else if (link_reporting == -1) {
1774 /* unable get link status using mii/ethtool */
1775 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",
1776 bond_dev->name, slave_dev->name);
1780 /* check for initial state */
1781 if (bond->params.miimon) {
1782 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1783 if (bond->params.updelay) {
1784 new_slave->link = BOND_LINK_BACK;
1785 new_slave->delay = bond->params.updelay;
1787 new_slave->link = BOND_LINK_UP;
1790 new_slave->link = BOND_LINK_DOWN;
1792 } else if (bond->params.arp_interval) {
1793 new_slave->link = (netif_carrier_ok(slave_dev) ?
1794 BOND_LINK_UP : BOND_LINK_DOWN);
1796 new_slave->link = BOND_LINK_UP;
1799 if (new_slave->link != BOND_LINK_DOWN)
1800 new_slave->jiffies = jiffies;
1801 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1802 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1803 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1805 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1806 /* if there is a primary slave, remember it */
1807 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1808 bond->primary_slave = new_slave;
1809 bond->force_primary = true;
1813 write_lock_bh(&bond->curr_slave_lock);
1815 switch (bond->params.mode) {
1816 case BOND_MODE_ACTIVEBACKUP:
1817 bond_set_slave_inactive_flags(new_slave);
1818 bond_select_active_slave(bond);
1820 case BOND_MODE_8023AD:
1821 /* in 802.3ad mode, the internal mechanism
1822 * will activate the slaves in the selected
1825 bond_set_slave_inactive_flags(new_slave);
1826 /* if this is the first slave */
1827 if (bond->slave_cnt == 1) {
1828 SLAVE_AD_INFO(new_slave).id = 1;
1829 /* Initialize AD with the number of times that the AD timer is called in 1 second
1830 * can be called only after the mac address of the bond is set
1832 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1834 SLAVE_AD_INFO(new_slave).id =
1835 SLAVE_AD_INFO(new_slave->prev).id + 1;
1838 bond_3ad_bind_slave(new_slave);
1842 bond_set_active_slave(new_slave);
1843 bond_set_slave_inactive_flags(new_slave);
1844 bond_select_active_slave(bond);
1847 pr_debug("This slave is always active in trunk mode\n");
1849 /* always active in trunk mode */
1850 bond_set_active_slave(new_slave);
1852 /* In trunking mode there is little meaning to curr_active_slave
1853 * anyway (it holds no special properties of the bond device),
1854 * so we can change it without calling change_active_interface()
1856 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1857 bond->curr_active_slave = new_slave;
1860 } /* switch(bond_mode) */
1862 write_unlock_bh(&bond->curr_slave_lock);
1864 bond_set_carrier(bond);
1866 #ifdef CONFIG_NET_POLL_CONTROLLER
1867 slave_dev->npinfo = bond_netpoll_info(bond);
1868 if (slave_dev->npinfo) {
1869 if (slave_enable_netpoll(new_slave)) {
1870 read_unlock(&bond->lock);
1871 pr_info("Error, %s: master_dev is using netpoll, "
1872 "but new slave device does not support netpoll.\n",
1880 read_unlock(&bond->lock);
1882 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1886 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1889 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1890 goto err_dest_symlinks;
1893 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1894 bond_dev->name, slave_dev->name,
1895 bond_is_active_slave(new_slave) ? "n active" : " backup",
1896 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1898 /* enslave is successful */
1901 /* Undo stages on error */
1903 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1906 write_lock_bh(&bond->lock);
1907 bond_detach_slave(bond, new_slave);
1908 write_unlock_bh(&bond->lock);
1911 dev_close(slave_dev);
1914 bond_upper_dev_unlink(bond_dev, slave_dev);
1917 if (!bond->params.fail_over_mac) {
1918 /* XXX TODO - fom follow mode needs to change master's
1919 * MAC if this slave's MAC is in use by the bond, or at
1920 * least print a warning.
1922 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1923 addr.sa_family = slave_dev->type;
1924 dev_set_mac_address(slave_dev, &addr);
1928 dev_set_mtu(slave_dev, new_slave->original_mtu);
1934 bond_compute_features(bond);
1940 * Try to release the slave device <slave> from the bond device <master>
1941 * It is legal to access curr_active_slave without a lock because all the function
1942 * is write-locked. If "all" is true it means that the function is being called
1943 * while destroying a bond interface and all slaves are being released.
1945 * The rules for slave state should be:
1946 * for Active/Backup:
1947 * Active stays on all backups go down
1948 * for Bonded connections:
1949 * The first up interface should be left on and all others downed.
1951 static int __bond_release_one(struct net_device *bond_dev,
1952 struct net_device *slave_dev,
1955 struct bonding *bond = netdev_priv(bond_dev);
1956 struct slave *slave, *oldcurrent;
1957 struct sockaddr addr;
1958 netdev_features_t old_features = bond_dev->features;
1960 /* slave is not a slave or master is not master of this slave */
1961 if (!(slave_dev->flags & IFF_SLAVE) ||
1962 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1963 pr_err("%s: Error: cannot release %s.\n",
1964 bond_dev->name, slave_dev->name);
1969 write_lock_bh(&bond->lock);
1971 slave = bond_get_slave_by_dev(bond, slave_dev);
1973 /* not a slave of this bond */
1974 pr_info("%s: %s not enslaved\n",
1975 bond_dev->name, slave_dev->name);
1976 write_unlock_bh(&bond->lock);
1977 unblock_netpoll_tx();
1981 write_unlock_bh(&bond->lock);
1982 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1983 * for this slave anymore.
1985 netdev_rx_handler_unregister(slave_dev);
1986 write_lock_bh(&bond->lock);
1988 if (!all && !bond->params.fail_over_mac) {
1989 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1990 bond->slave_cnt > 1)
1991 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",
1992 bond_dev->name, slave_dev->name,
1994 bond_dev->name, slave_dev->name);
1997 /* Inform AD package of unbinding of slave. */
1998 if (bond->params.mode == BOND_MODE_8023AD) {
1999 /* must be called before the slave is
2000 * detached from the list
2002 bond_3ad_unbind_slave(slave);
2005 pr_info("%s: releasing %s interface %s\n",
2007 bond_is_active_slave(slave) ? "active" : "backup",
2010 oldcurrent = bond->curr_active_slave;
2012 bond->current_arp_slave = NULL;
2014 /* release the slave from its bond */
2015 bond_detach_slave(bond, slave);
2017 if (bond->primary_slave == slave)
2018 bond->primary_slave = NULL;
2020 if (oldcurrent == slave)
2021 bond_change_active_slave(bond, NULL);
2023 if (bond_is_lb(bond)) {
2024 /* Must be called only after the slave has been
2025 * detached from the list and the curr_active_slave
2026 * has been cleared (if our_slave == old_current),
2027 * but before a new active slave is selected.
2029 write_unlock_bh(&bond->lock);
2030 bond_alb_deinit_slave(bond, slave);
2031 write_lock_bh(&bond->lock);
2035 bond->curr_active_slave = NULL;
2036 } else if (oldcurrent == slave) {
2038 * Note that we hold RTNL over this sequence, so there
2039 * is no concern that another slave add/remove event
2042 write_unlock_bh(&bond->lock);
2043 read_lock(&bond->lock);
2044 write_lock_bh(&bond->curr_slave_lock);
2046 bond_select_active_slave(bond);
2048 write_unlock_bh(&bond->curr_slave_lock);
2049 read_unlock(&bond->lock);
2050 write_lock_bh(&bond->lock);
2053 if (bond->slave_cnt == 0) {
2054 bond_set_carrier(bond);
2055 eth_hw_addr_random(bond_dev);
2056 bond->dev_addr_from_first = true;
2058 if (bond_vlan_used(bond)) {
2059 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2060 bond_dev->name, bond_dev->name);
2061 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2066 write_unlock_bh(&bond->lock);
2067 unblock_netpoll_tx();
2069 if (bond->slave_cnt == 0) {
2070 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2071 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2074 bond_compute_features(bond);
2075 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2076 (old_features & NETIF_F_VLAN_CHALLENGED))
2077 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2078 bond_dev->name, slave_dev->name, bond_dev->name);
2080 /* must do this from outside any spinlocks */
2081 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2083 bond_del_vlans_from_slave(bond, slave_dev);
2085 /* If the mode USES_PRIMARY, then we should only remove its
2086 * promisc and mc settings if it was the curr_active_slave, but that was
2087 * already taken care of above when we detached the slave
2089 if (!USES_PRIMARY(bond->params.mode)) {
2090 /* unset promiscuity level from slave */
2091 if (bond_dev->flags & IFF_PROMISC)
2092 dev_set_promiscuity(slave_dev, -1);
2094 /* unset allmulti level from slave */
2095 if (bond_dev->flags & IFF_ALLMULTI)
2096 dev_set_allmulti(slave_dev, -1);
2098 /* flush master's mc_list from slave */
2099 netif_addr_lock_bh(bond_dev);
2100 bond_mc_list_flush(bond_dev, slave_dev);
2101 netif_addr_unlock_bh(bond_dev);
2104 bond_upper_dev_unlink(bond_dev, slave_dev);
2106 slave_disable_netpoll(slave);
2108 /* close slave before restoring its mac address */
2109 dev_close(slave_dev);
2111 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2112 /* restore original ("permanent") mac address */
2113 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2114 addr.sa_family = slave_dev->type;
2115 dev_set_mac_address(slave_dev, &addr);
2118 dev_set_mtu(slave_dev, slave->original_mtu);
2120 slave_dev->priv_flags &= ~IFF_BONDING;
2124 return 0; /* deletion OK */
2127 /* A wrapper used because of ndo_del_link */
2128 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2130 return __bond_release_one(bond_dev, slave_dev, false);
2134 * First release a slave and then destroy the bond if no more slaves are left.
2135 * Must be under rtnl_lock when this function is called.
2137 static int bond_release_and_destroy(struct net_device *bond_dev,
2138 struct net_device *slave_dev)
2140 struct bonding *bond = netdev_priv(bond_dev);
2143 ret = bond_release(bond_dev, slave_dev);
2144 if ((ret == 0) && (bond->slave_cnt == 0)) {
2145 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2146 pr_info("%s: destroying bond %s.\n",
2147 bond_dev->name, bond_dev->name);
2148 unregister_netdevice(bond_dev);
2154 * This function changes the active slave to slave <slave_dev>.
2155 * It returns -EINVAL in the following cases.
2156 * - <slave_dev> is not found in the list.
2157 * - There is not active slave now.
2158 * - <slave_dev> is already active.
2159 * - The link state of <slave_dev> is not BOND_LINK_UP.
2160 * - <slave_dev> is not running.
2161 * In these cases, this function does nothing.
2162 * In the other cases, current_slave pointer is changed and 0 is returned.
2164 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2166 struct bonding *bond = netdev_priv(bond_dev);
2167 struct slave *old_active = NULL;
2168 struct slave *new_active = NULL;
2171 if (!USES_PRIMARY(bond->params.mode))
2174 /* Verify that bond_dev is indeed the master of slave_dev */
2175 if (!(slave_dev->flags & IFF_SLAVE) ||
2176 !netdev_has_upper_dev(slave_dev, bond_dev))
2179 read_lock(&bond->lock);
2181 read_lock(&bond->curr_slave_lock);
2182 old_active = bond->curr_active_slave;
2183 read_unlock(&bond->curr_slave_lock);
2185 new_active = bond_get_slave_by_dev(bond, slave_dev);
2188 * Changing to the current active: do nothing; return success.
2190 if (new_active && (new_active == old_active)) {
2191 read_unlock(&bond->lock);
2197 (new_active->link == BOND_LINK_UP) &&
2198 IS_UP(new_active->dev)) {
2200 write_lock_bh(&bond->curr_slave_lock);
2201 bond_change_active_slave(bond, new_active);
2202 write_unlock_bh(&bond->curr_slave_lock);
2203 unblock_netpoll_tx();
2207 read_unlock(&bond->lock);
2212 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2214 struct bonding *bond = netdev_priv(bond_dev);
2216 info->bond_mode = bond->params.mode;
2217 info->miimon = bond->params.miimon;
2219 read_lock(&bond->lock);
2220 info->num_slaves = bond->slave_cnt;
2221 read_unlock(&bond->lock);
2226 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2228 struct bonding *bond = netdev_priv(bond_dev);
2229 struct slave *slave;
2230 int i, res = -ENODEV;
2232 read_lock(&bond->lock);
2234 bond_for_each_slave(bond, slave, i) {
2235 if (i == (int)info->slave_id) {
2237 strcpy(info->slave_name, slave->dev->name);
2238 info->link = slave->link;
2239 info->state = bond_slave_state(slave);
2240 info->link_failure_count = slave->link_failure_count;
2245 read_unlock(&bond->lock);
2250 /*-------------------------------- Monitoring -------------------------------*/
2253 static int bond_miimon_inspect(struct bonding *bond)
2255 struct slave *slave;
2256 int i, link_state, commit = 0;
2257 bool ignore_updelay;
2259 ignore_updelay = !bond->curr_active_slave ? true : false;
2261 bond_for_each_slave(bond, slave, i) {
2262 slave->new_link = BOND_LINK_NOCHANGE;
2264 link_state = bond_check_dev_link(bond, slave->dev, 0);
2266 switch (slave->link) {
2271 slave->link = BOND_LINK_FAIL;
2272 slave->delay = bond->params.downdelay;
2274 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2276 (bond->params.mode ==
2277 BOND_MODE_ACTIVEBACKUP) ?
2278 (bond_is_active_slave(slave) ?
2279 "active " : "backup ") : "",
2281 bond->params.downdelay * bond->params.miimon);
2284 case BOND_LINK_FAIL:
2287 * recovered before downdelay expired
2289 slave->link = BOND_LINK_UP;
2290 slave->jiffies = jiffies;
2291 pr_info("%s: link status up again after %d ms for interface %s.\n",
2293 (bond->params.downdelay - slave->delay) *
2294 bond->params.miimon,
2299 if (slave->delay <= 0) {
2300 slave->new_link = BOND_LINK_DOWN;
2308 case BOND_LINK_DOWN:
2312 slave->link = BOND_LINK_BACK;
2313 slave->delay = bond->params.updelay;
2316 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2317 bond->dev->name, slave->dev->name,
2318 ignore_updelay ? 0 :
2319 bond->params.updelay *
2320 bond->params.miimon);
2323 case BOND_LINK_BACK:
2325 slave->link = BOND_LINK_DOWN;
2326 pr_info("%s: link status down again after %d ms for interface %s.\n",
2328 (bond->params.updelay - slave->delay) *
2329 bond->params.miimon,
2338 if (slave->delay <= 0) {
2339 slave->new_link = BOND_LINK_UP;
2341 ignore_updelay = false;
2353 static void bond_miimon_commit(struct bonding *bond)
2355 struct slave *slave;
2358 bond_for_each_slave(bond, slave, i) {
2359 switch (slave->new_link) {
2360 case BOND_LINK_NOCHANGE:
2364 slave->link = BOND_LINK_UP;
2365 slave->jiffies = jiffies;
2367 if (bond->params.mode == BOND_MODE_8023AD) {
2368 /* prevent it from being the active one */
2369 bond_set_backup_slave(slave);
2370 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2371 /* make it immediately active */
2372 bond_set_active_slave(slave);
2373 } else if (slave != bond->primary_slave) {
2374 /* prevent it from being the active one */
2375 bond_set_backup_slave(slave);
2378 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2379 bond->dev->name, slave->dev->name,
2380 slave->speed, slave->duplex ? "full" : "half");
2382 /* notify ad that the link status has changed */
2383 if (bond->params.mode == BOND_MODE_8023AD)
2384 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2386 if (bond_is_lb(bond))
2387 bond_alb_handle_link_change(bond, slave,
2390 if (!bond->curr_active_slave ||
2391 (slave == bond->primary_slave))
2396 case BOND_LINK_DOWN:
2397 if (slave->link_failure_count < UINT_MAX)
2398 slave->link_failure_count++;
2400 slave->link = BOND_LINK_DOWN;
2402 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2403 bond->params.mode == BOND_MODE_8023AD)
2404 bond_set_slave_inactive_flags(slave);
2406 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2407 bond->dev->name, slave->dev->name);
2409 if (bond->params.mode == BOND_MODE_8023AD)
2410 bond_3ad_handle_link_change(slave,
2413 if (bond_is_lb(bond))
2414 bond_alb_handle_link_change(bond, slave,
2417 if (slave == bond->curr_active_slave)
2423 pr_err("%s: invalid new link %d on slave %s\n",
2424 bond->dev->name, slave->new_link,
2426 slave->new_link = BOND_LINK_NOCHANGE;
2434 write_lock_bh(&bond->curr_slave_lock);
2435 bond_select_active_slave(bond);
2436 write_unlock_bh(&bond->curr_slave_lock);
2437 unblock_netpoll_tx();
2440 bond_set_carrier(bond);
2446 * Really a wrapper that splits the mii monitor into two phases: an
2447 * inspection, then (if inspection indicates something needs to be done)
2448 * an acquisition of appropriate locks followed by a commit phase to
2449 * implement whatever link state changes are indicated.
2451 void bond_mii_monitor(struct work_struct *work)
2453 struct bonding *bond = container_of(work, struct bonding,
2455 bool should_notify_peers = false;
2456 unsigned long delay;
2458 read_lock(&bond->lock);
2460 delay = msecs_to_jiffies(bond->params.miimon);
2462 if (bond->slave_cnt == 0)
2465 should_notify_peers = bond_should_notify_peers(bond);
2467 if (bond_miimon_inspect(bond)) {
2468 read_unlock(&bond->lock);
2470 /* Race avoidance with bond_close cancel of workqueue */
2471 if (!rtnl_trylock()) {
2472 read_lock(&bond->lock);
2474 should_notify_peers = false;
2478 read_lock(&bond->lock);
2480 bond_miimon_commit(bond);
2482 read_unlock(&bond->lock);
2483 rtnl_unlock(); /* might sleep, hold no other locks */
2484 read_lock(&bond->lock);
2488 if (bond->params.miimon)
2489 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2491 read_unlock(&bond->lock);
2493 if (should_notify_peers) {
2494 if (!rtnl_trylock()) {
2495 read_lock(&bond->lock);
2496 bond->send_peer_notif++;
2497 read_unlock(&bond->lock);
2500 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2505 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2507 struct vlan_entry *vlan;
2508 struct net_device *vlan_dev;
2510 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2513 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2515 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2518 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2526 * We go to the (large) trouble of VLAN tagging ARP frames because
2527 * switches in VLAN mode (especially if ports are configured as
2528 * "native" to a VLAN) might not pass non-tagged frames.
2530 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2532 struct sk_buff *skb;
2534 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2535 slave_dev->name, dest_ip, src_ip, vlan_id);
2537 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2538 NULL, slave_dev->dev_addr, NULL);
2541 pr_err("ARP packet allocation failed\n");
2545 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2547 pr_err("failed to insert VLAN tag\n");
2555 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2558 __be32 *targets = bond->params.arp_targets;
2559 struct vlan_entry *vlan;
2560 struct net_device *vlan_dev = NULL;
2563 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2567 pr_debug("basa: target %x\n", targets[i]);
2568 if (!bond_vlan_used(bond)) {
2569 pr_debug("basa: empty vlan: arp_send\n");
2570 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2571 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2577 * If VLANs are configured, we do a route lookup to
2578 * determine which VLAN interface would be used, so we
2579 * can tag the ARP with the proper VLAN tag.
2581 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2584 if (net_ratelimit()) {
2585 pr_warning("%s: no route to arp_ip_target %pI4\n",
2586 bond->dev->name, &targets[i]);
2592 * This target is not on a VLAN
2594 if (rt->dst.dev == bond->dev) {
2596 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2597 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2598 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2604 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2606 vlan_dev = __vlan_find_dev_deep(bond->dev,
2610 if (vlan_dev == rt->dst.dev) {
2611 vlan_id = vlan->vlan_id;
2612 pr_debug("basa: vlan match on %s %d\n",
2613 vlan_dev->name, vlan_id);
2618 if (vlan_id && vlan_dev) {
2620 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2621 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2626 if (net_ratelimit()) {
2627 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2628 bond->dev->name, &targets[i],
2629 rt->dst.dev ? rt->dst.dev->name : "NULL");
2635 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2638 __be32 *targets = bond->params.arp_targets;
2640 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2641 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2642 &sip, &tip, i, &targets[i],
2643 bond_has_this_ip(bond, tip));
2644 if (sip == targets[i]) {
2645 if (bond_has_this_ip(bond, tip))
2646 slave->last_arp_rx = jiffies;
2652 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2653 struct slave *slave)
2655 struct arphdr *arp = (struct arphdr *)skb->data;
2656 unsigned char *arp_ptr;
2660 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2661 return RX_HANDLER_ANOTHER;
2663 read_lock(&bond->lock);
2664 alen = arp_hdr_len(bond->dev);
2666 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2667 bond->dev->name, skb->dev->name);
2669 if (alen > skb_headlen(skb)) {
2670 arp = kmalloc(alen, GFP_ATOMIC);
2673 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2677 if (arp->ar_hln != bond->dev->addr_len ||
2678 skb->pkt_type == PACKET_OTHERHOST ||
2679 skb->pkt_type == PACKET_LOOPBACK ||
2680 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2681 arp->ar_pro != htons(ETH_P_IP) ||
2685 arp_ptr = (unsigned char *)(arp + 1);
2686 arp_ptr += bond->dev->addr_len;
2687 memcpy(&sip, arp_ptr, 4);
2688 arp_ptr += 4 + bond->dev->addr_len;
2689 memcpy(&tip, arp_ptr, 4);
2691 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2692 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2693 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2697 * Backup slaves won't see the ARP reply, but do come through
2698 * here for each ARP probe (so we swap the sip/tip to validate
2699 * the probe). In a "redundant switch, common router" type of
2700 * configuration, the ARP probe will (hopefully) travel from
2701 * the active, through one switch, the router, then the other
2702 * switch before reaching the backup.
2704 if (bond_is_active_slave(slave))
2705 bond_validate_arp(bond, slave, sip, tip);
2707 bond_validate_arp(bond, slave, tip, sip);
2710 read_unlock(&bond->lock);
2711 if (arp != (struct arphdr *)skb->data)
2713 return RX_HANDLER_ANOTHER;
2717 * this function is called regularly to monitor each slave's link
2718 * ensuring that traffic is being sent and received when arp monitoring
2719 * is used in load-balancing mode. if the adapter has been dormant, then an
2720 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2721 * arp monitoring in active backup mode.
2723 void bond_loadbalance_arp_mon(struct work_struct *work)
2725 struct bonding *bond = container_of(work, struct bonding,
2727 struct slave *slave, *oldcurrent;
2728 int do_failover = 0;
2729 int delta_in_ticks, extra_ticks;
2732 read_lock(&bond->lock);
2734 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2735 extra_ticks = delta_in_ticks / 2;
2737 if (bond->slave_cnt == 0)
2740 read_lock(&bond->curr_slave_lock);
2741 oldcurrent = bond->curr_active_slave;
2742 read_unlock(&bond->curr_slave_lock);
2744 /* see if any of the previous devices are up now (i.e. they have
2745 * xmt and rcv traffic). the curr_active_slave does not come into
2746 * the picture unless it is null. also, slave->jiffies is not needed
2747 * here because we send an arp on each slave and give a slave as
2748 * long as it needs to get the tx/rx within the delta.
2749 * TODO: what about up/down delay in arp mode? it wasn't here before
2752 bond_for_each_slave(bond, slave, i) {
2753 unsigned long trans_start = dev_trans_start(slave->dev);
2755 if (slave->link != BOND_LINK_UP) {
2756 if (time_in_range(jiffies,
2757 trans_start - delta_in_ticks,
2758 trans_start + delta_in_ticks + extra_ticks) &&
2759 time_in_range(jiffies,
2760 slave->dev->last_rx - delta_in_ticks,
2761 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2763 slave->link = BOND_LINK_UP;
2764 bond_set_active_slave(slave);
2766 /* primary_slave has no meaning in round-robin
2767 * mode. the window of a slave being up and
2768 * curr_active_slave being null after enslaving
2772 pr_info("%s: link status definitely up for interface %s, ",
2777 pr_info("%s: interface %s is now up\n",
2783 /* slave->link == BOND_LINK_UP */
2785 /* not all switches will respond to an arp request
2786 * when the source ip is 0, so don't take the link down
2787 * if we don't know our ip yet
2789 if (!time_in_range(jiffies,
2790 trans_start - delta_in_ticks,
2791 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2792 !time_in_range(jiffies,
2793 slave->dev->last_rx - delta_in_ticks,
2794 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2796 slave->link = BOND_LINK_DOWN;
2797 bond_set_backup_slave(slave);
2799 if (slave->link_failure_count < UINT_MAX)
2800 slave->link_failure_count++;
2802 pr_info("%s: interface %s is now down.\n",
2806 if (slave == oldcurrent)
2811 /* note: if switch is in round-robin mode, all links
2812 * must tx arp to ensure all links rx an arp - otherwise
2813 * links may oscillate or not come up at all; if switch is
2814 * in something like xor mode, there is nothing we can
2815 * do - all replies will be rx'ed on same link causing slaves
2816 * to be unstable during low/no traffic periods
2818 if (IS_UP(slave->dev))
2819 bond_arp_send_all(bond, slave);
2824 write_lock_bh(&bond->curr_slave_lock);
2826 bond_select_active_slave(bond);
2828 write_unlock_bh(&bond->curr_slave_lock);
2829 unblock_netpoll_tx();
2833 if (bond->params.arp_interval)
2834 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2836 read_unlock(&bond->lock);
2840 * Called to inspect slaves for active-backup mode ARP monitor link state
2841 * changes. Sets new_link in slaves to specify what action should take
2842 * place for the slave. Returns 0 if no changes are found, >0 if changes
2843 * to link states must be committed.
2845 * Called with bond->lock held for read.
2847 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2849 struct slave *slave;
2851 unsigned long trans_start;
2854 /* All the time comparisons below need some extra time. Otherwise, on
2855 * fast networks the ARP probe/reply may arrive within the same jiffy
2856 * as it was sent. Then, the next time the ARP monitor is run, one
2857 * arp_interval will already have passed in the comparisons.
2859 extra_ticks = delta_in_ticks / 2;
2861 bond_for_each_slave(bond, slave, i) {
2862 slave->new_link = BOND_LINK_NOCHANGE;
2864 if (slave->link != BOND_LINK_UP) {
2865 if (time_in_range(jiffies,
2866 slave_last_rx(bond, slave) - delta_in_ticks,
2867 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2869 slave->new_link = BOND_LINK_UP;
2877 * Give slaves 2*delta after being enslaved or made
2878 * active. This avoids bouncing, as the last receive
2879 * times need a full ARP monitor cycle to be updated.
2881 if (time_in_range(jiffies,
2882 slave->jiffies - delta_in_ticks,
2883 slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2887 * Backup slave is down if:
2888 * - No current_arp_slave AND
2889 * - more than 3*delta since last receive AND
2890 * - the bond has an IP address
2892 * Note: a non-null current_arp_slave indicates
2893 * the curr_active_slave went down and we are
2894 * searching for a new one; under this condition
2895 * we only take the curr_active_slave down - this
2896 * gives each slave a chance to tx/rx traffic
2897 * before being taken out
2899 if (!bond_is_active_slave(slave) &&
2900 !bond->current_arp_slave &&
2901 !time_in_range(jiffies,
2902 slave_last_rx(bond, slave) - delta_in_ticks,
2903 slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2905 slave->new_link = BOND_LINK_DOWN;
2910 * Active slave is down if:
2911 * - more than 2*delta since transmitting OR
2912 * - (more than 2*delta since receive AND
2913 * the bond has an IP address)
2915 trans_start = dev_trans_start(slave->dev);
2916 if (bond_is_active_slave(slave) &&
2917 (!time_in_range(jiffies,
2918 trans_start - delta_in_ticks,
2919 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2920 !time_in_range(jiffies,
2921 slave_last_rx(bond, slave) - delta_in_ticks,
2922 slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2924 slave->new_link = BOND_LINK_DOWN;
2933 * Called to commit link state changes noted by inspection step of
2934 * active-backup mode ARP monitor.
2936 * Called with RTNL and bond->lock for read.
2938 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2940 struct slave *slave;
2942 unsigned long trans_start;
2944 bond_for_each_slave(bond, slave, i) {
2945 switch (slave->new_link) {
2946 case BOND_LINK_NOCHANGE:
2950 trans_start = dev_trans_start(slave->dev);
2951 if ((!bond->curr_active_slave &&
2952 time_in_range(jiffies,
2953 trans_start - delta_in_ticks,
2954 trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2955 bond->curr_active_slave != slave) {
2956 slave->link = BOND_LINK_UP;
2957 if (bond->current_arp_slave) {
2958 bond_set_slave_inactive_flags(
2959 bond->current_arp_slave);
2960 bond->current_arp_slave = NULL;
2963 pr_info("%s: link status definitely up for interface %s.\n",
2964 bond->dev->name, slave->dev->name);
2966 if (!bond->curr_active_slave ||
2967 (slave == bond->primary_slave))
2974 case BOND_LINK_DOWN:
2975 if (slave->link_failure_count < UINT_MAX)
2976 slave->link_failure_count++;
2978 slave->link = BOND_LINK_DOWN;
2979 bond_set_slave_inactive_flags(slave);
2981 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2982 bond->dev->name, slave->dev->name);
2984 if (slave == bond->curr_active_slave) {
2985 bond->current_arp_slave = NULL;
2992 pr_err("%s: impossible: new_link %d on slave %s\n",
2993 bond->dev->name, slave->new_link,
3001 write_lock_bh(&bond->curr_slave_lock);
3002 bond_select_active_slave(bond);
3003 write_unlock_bh(&bond->curr_slave_lock);
3004 unblock_netpoll_tx();
3007 bond_set_carrier(bond);
3011 * Send ARP probes for active-backup mode ARP monitor.
3013 * Called with bond->lock held for read.
3015 static void bond_ab_arp_probe(struct bonding *bond)
3017 struct slave *slave;
3020 read_lock(&bond->curr_slave_lock);
3022 if (bond->current_arp_slave && bond->curr_active_slave)
3023 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3024 bond->current_arp_slave->dev->name,
3025 bond->curr_active_slave->dev->name);
3027 if (bond->curr_active_slave) {
3028 bond_arp_send_all(bond, bond->curr_active_slave);
3029 read_unlock(&bond->curr_slave_lock);
3033 read_unlock(&bond->curr_slave_lock);
3035 /* if we don't have a curr_active_slave, search for the next available
3036 * backup slave from the current_arp_slave and make it the candidate
3037 * for becoming the curr_active_slave
3040 if (!bond->current_arp_slave) {
3041 bond->current_arp_slave = bond->first_slave;
3042 if (!bond->current_arp_slave)
3046 bond_set_slave_inactive_flags(bond->current_arp_slave);
3048 /* search for next candidate */
3049 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3050 if (IS_UP(slave->dev)) {
3051 slave->link = BOND_LINK_BACK;
3052 bond_set_slave_active_flags(slave);
3053 bond_arp_send_all(bond, slave);
3054 slave->jiffies = jiffies;
3055 bond->current_arp_slave = slave;
3059 /* if the link state is up at this point, we
3060 * mark it down - this can happen if we have
3061 * simultaneous link failures and
3062 * reselect_active_interface doesn't make this
3063 * one the current slave so it is still marked
3064 * up when it is actually down
3066 if (slave->link == BOND_LINK_UP) {
3067 slave->link = BOND_LINK_DOWN;
3068 if (slave->link_failure_count < UINT_MAX)
3069 slave->link_failure_count++;
3071 bond_set_slave_inactive_flags(slave);
3073 pr_info("%s: backup interface %s is now down.\n",
3074 bond->dev->name, slave->dev->name);
3079 void bond_activebackup_arp_mon(struct work_struct *work)
3081 struct bonding *bond = container_of(work, struct bonding,
3083 bool should_notify_peers = false;
3086 read_lock(&bond->lock);
3088 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3090 if (bond->slave_cnt == 0)
3093 should_notify_peers = bond_should_notify_peers(bond);
3095 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3096 read_unlock(&bond->lock);
3098 /* Race avoidance with bond_close flush of workqueue */
3099 if (!rtnl_trylock()) {
3100 read_lock(&bond->lock);
3102 should_notify_peers = false;
3106 read_lock(&bond->lock);
3108 bond_ab_arp_commit(bond, delta_in_ticks);
3110 read_unlock(&bond->lock);
3112 read_lock(&bond->lock);
3115 bond_ab_arp_probe(bond);
3118 if (bond->params.arp_interval)
3119 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3121 read_unlock(&bond->lock);
3123 if (should_notify_peers) {
3124 if (!rtnl_trylock()) {
3125 read_lock(&bond->lock);
3126 bond->send_peer_notif++;
3127 read_unlock(&bond->lock);
3130 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3135 /*-------------------------- netdev event handling --------------------------*/
3138 * Change device name
3140 static int bond_event_changename(struct bonding *bond)
3142 bond_remove_proc_entry(bond);
3143 bond_create_proc_entry(bond);
3145 bond_debug_reregister(bond);
3150 static int bond_master_netdev_event(unsigned long event,
3151 struct net_device *bond_dev)
3153 struct bonding *event_bond = netdev_priv(bond_dev);
3156 case NETDEV_CHANGENAME:
3157 return bond_event_changename(event_bond);
3158 case NETDEV_UNREGISTER:
3159 bond_remove_proc_entry(event_bond);
3161 case NETDEV_REGISTER:
3162 bond_create_proc_entry(event_bond);
3171 static int bond_slave_netdev_event(unsigned long event,
3172 struct net_device *slave_dev)
3174 struct slave *slave = bond_slave_get_rtnl(slave_dev);
3175 struct bonding *bond = slave->bond;
3176 struct net_device *bond_dev = slave->bond->dev;
3181 case NETDEV_UNREGISTER:
3182 if (bond->setup_by_slave)
3183 bond_release_and_destroy(bond_dev, slave_dev);
3185 bond_release(bond_dev, slave_dev);
3189 old_speed = slave->speed;
3190 old_duplex = slave->duplex;
3192 bond_update_speed_duplex(slave);
3194 if (bond->params.mode == BOND_MODE_8023AD) {
3195 if (old_speed != slave->speed)
3196 bond_3ad_adapter_speed_changed(slave);
3197 if (old_duplex != slave->duplex)
3198 bond_3ad_adapter_duplex_changed(slave);
3203 * ... Or is it this?
3206 case NETDEV_CHANGEMTU:
3208 * TODO: Should slaves be allowed to
3209 * independently alter their MTU? For
3210 * an active-backup bond, slaves need
3211 * not be the same type of device, so
3212 * MTUs may vary. For other modes,
3213 * slaves arguably should have the
3214 * same MTUs. To do this, we'd need to
3215 * take over the slave's change_mtu
3216 * function for the duration of their
3220 case NETDEV_CHANGENAME:
3222 * TODO: handle changing the primary's name
3225 case NETDEV_FEAT_CHANGE:
3226 bond_compute_features(bond);
3236 * bond_netdev_event: handle netdev notifier chain events.
3238 * This function receives events for the netdev chain. The caller (an
3239 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3240 * locks for us to safely manipulate the slave devices (RTNL lock,
3243 static int bond_netdev_event(struct notifier_block *this,
3244 unsigned long event, void *ptr)
3246 struct net_device *event_dev = (struct net_device *)ptr;
3248 pr_debug("event_dev: %s, event: %lx\n",
3249 event_dev ? event_dev->name : "None",
3252 if (!(event_dev->priv_flags & IFF_BONDING))
3255 if (event_dev->flags & IFF_MASTER) {
3256 pr_debug("IFF_MASTER\n");
3257 return bond_master_netdev_event(event, event_dev);
3260 if (event_dev->flags & IFF_SLAVE) {
3261 pr_debug("IFF_SLAVE\n");
3262 return bond_slave_netdev_event(event, event_dev);
3268 static struct notifier_block bond_netdev_notifier = {
3269 .notifier_call = bond_netdev_event,
3272 /*---------------------------- Hashing Policies -----------------------------*/
3275 * Hash for the output device based upon layer 2 data
3277 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3279 struct ethhdr *data = (struct ethhdr *)skb->data;
3281 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3282 return (data->h_dest[5] ^ data->h_source[5]) % count;
3288 * Hash for the output device based upon layer 2 and layer 3 data. If
3289 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3291 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3293 struct ethhdr *data = (struct ethhdr *)skb->data;
3295 struct ipv6hdr *ipv6h;
3299 if (skb->protocol == htons(ETH_P_IP) &&
3300 skb_network_header_len(skb) >= sizeof(*iph)) {
3302 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3303 (data->h_dest[5] ^ data->h_source[5])) % count;
3304 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3305 skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3306 ipv6h = ipv6_hdr(skb);
3307 s = &ipv6h->saddr.s6_addr32[0];
3308 d = &ipv6h->daddr.s6_addr32[0];
3309 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3310 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3311 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3314 return bond_xmit_hash_policy_l2(skb, count);
3318 * Hash for the output device based upon layer 3 and layer 4 data. If
3319 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3320 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3322 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3326 struct ipv6hdr *ipv6h;
3330 if (skb->protocol == htons(ETH_P_IP) &&
3331 skb_network_header_len(skb) >= sizeof(*iph)) {
3333 if (!ip_is_fragment(iph) &&
3334 (iph->protocol == IPPROTO_TCP ||
3335 iph->protocol == IPPROTO_UDP) &&
3336 (skb_headlen(skb) - skb_network_offset(skb) >=
3337 iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3338 layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3339 layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3341 return (layer4_xor ^
3342 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3343 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3344 skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3345 ipv6h = ipv6_hdr(skb);
3346 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3347 ipv6h->nexthdr == IPPROTO_UDP) &&
3348 (skb_headlen(skb) - skb_network_offset(skb) >=
3349 sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3350 layer4hdr = (__be16 *)(ipv6h + 1);
3351 layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3353 s = &ipv6h->saddr.s6_addr32[0];
3354 d = &ipv6h->daddr.s6_addr32[0];
3355 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3356 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3358 return layer4_xor % count;
3361 return bond_xmit_hash_policy_l2(skb, count);
3364 /*-------------------------- Device entry points ----------------------------*/
3366 static void bond_work_init_all(struct bonding *bond)
3368 INIT_DELAYED_WORK(&bond->mcast_work,
3369 bond_resend_igmp_join_requests_delayed);
3370 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3371 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3372 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3373 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3375 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3376 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3379 static void bond_work_cancel_all(struct bonding *bond)
3381 cancel_delayed_work_sync(&bond->mii_work);
3382 cancel_delayed_work_sync(&bond->arp_work);
3383 cancel_delayed_work_sync(&bond->alb_work);
3384 cancel_delayed_work_sync(&bond->ad_work);
3385 cancel_delayed_work_sync(&bond->mcast_work);
3388 static int bond_open(struct net_device *bond_dev)
3390 struct bonding *bond = netdev_priv(bond_dev);
3391 struct slave *slave;
3394 /* reset slave->backup and slave->inactive */
3395 read_lock(&bond->lock);
3396 if (bond->slave_cnt > 0) {
3397 read_lock(&bond->curr_slave_lock);
3398 bond_for_each_slave(bond, slave, i) {
3399 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3400 && (slave != bond->curr_active_slave)) {
3401 bond_set_slave_inactive_flags(slave);
3403 bond_set_slave_active_flags(slave);
3406 read_unlock(&bond->curr_slave_lock);
3408 read_unlock(&bond->lock);
3410 bond_work_init_all(bond);
3412 if (bond_is_lb(bond)) {
3413 /* bond_alb_initialize must be called before the timer
3416 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3418 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3421 if (bond->params.miimon) /* link check interval, in milliseconds. */
3422 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3424 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3425 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3426 if (bond->params.arp_validate)
3427 bond->recv_probe = bond_arp_rcv;
3430 if (bond->params.mode == BOND_MODE_8023AD) {
3431 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3432 /* register to receive LACPDUs */
3433 bond->recv_probe = bond_3ad_lacpdu_recv;
3434 bond_3ad_initiate_agg_selection(bond, 1);
3440 static int bond_close(struct net_device *bond_dev)
3442 struct bonding *bond = netdev_priv(bond_dev);
3444 write_lock_bh(&bond->lock);
3445 bond->send_peer_notif = 0;
3446 write_unlock_bh(&bond->lock);
3448 bond_work_cancel_all(bond);
3449 if (bond_is_lb(bond)) {
3450 /* Must be called only after all
3451 * slaves have been released
3453 bond_alb_deinitialize(bond);
3455 bond->recv_probe = NULL;
3460 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3461 struct rtnl_link_stats64 *stats)
3463 struct bonding *bond = netdev_priv(bond_dev);
3464 struct rtnl_link_stats64 temp;
3465 struct slave *slave;
3468 memset(stats, 0, sizeof(*stats));
3470 read_lock_bh(&bond->lock);
3472 bond_for_each_slave(bond, slave, i) {
3473 const struct rtnl_link_stats64 *sstats =
3474 dev_get_stats(slave->dev, &temp);
3476 stats->rx_packets += sstats->rx_packets;
3477 stats->rx_bytes += sstats->rx_bytes;
3478 stats->rx_errors += sstats->rx_errors;
3479 stats->rx_dropped += sstats->rx_dropped;
3481 stats->tx_packets += sstats->tx_packets;
3482 stats->tx_bytes += sstats->tx_bytes;
3483 stats->tx_errors += sstats->tx_errors;
3484 stats->tx_dropped += sstats->tx_dropped;
3486 stats->multicast += sstats->multicast;
3487 stats->collisions += sstats->collisions;
3489 stats->rx_length_errors += sstats->rx_length_errors;
3490 stats->rx_over_errors += sstats->rx_over_errors;
3491 stats->rx_crc_errors += sstats->rx_crc_errors;
3492 stats->rx_frame_errors += sstats->rx_frame_errors;
3493 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3494 stats->rx_missed_errors += sstats->rx_missed_errors;
3496 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3497 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3498 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3499 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3500 stats->tx_window_errors += sstats->tx_window_errors;
3503 read_unlock_bh(&bond->lock);
3508 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3510 struct net_device *slave_dev = NULL;
3511 struct ifbond k_binfo;
3512 struct ifbond __user *u_binfo = NULL;
3513 struct ifslave k_sinfo;
3514 struct ifslave __user *u_sinfo = NULL;
3515 struct mii_ioctl_data *mii = NULL;
3519 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3531 * We do this again just in case we were called by SIOCGMIIREG
3532 * instead of SIOCGMIIPHY.
3539 if (mii->reg_num == 1) {
3540 struct bonding *bond = netdev_priv(bond_dev);
3542 read_lock(&bond->lock);
3543 read_lock(&bond->curr_slave_lock);
3544 if (netif_carrier_ok(bond->dev))
3545 mii->val_out = BMSR_LSTATUS;
3547 read_unlock(&bond->curr_slave_lock);
3548 read_unlock(&bond->lock);
3552 case BOND_INFO_QUERY_OLD:
3553 case SIOCBONDINFOQUERY:
3554 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3556 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3559 res = bond_info_query(bond_dev, &k_binfo);
3561 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3565 case BOND_SLAVE_INFO_QUERY_OLD:
3566 case SIOCBONDSLAVEINFOQUERY:
3567 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3569 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3572 res = bond_slave_info_query(bond_dev, &k_sinfo);
3574 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3583 net = dev_net(bond_dev);
3585 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3588 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3590 pr_debug("slave_dev=%p:\n", slave_dev);
3595 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3597 case BOND_ENSLAVE_OLD:
3598 case SIOCBONDENSLAVE:
3599 res = bond_enslave(bond_dev, slave_dev);
3601 case BOND_RELEASE_OLD:
3602 case SIOCBONDRELEASE:
3603 res = bond_release(bond_dev, slave_dev);
3605 case BOND_SETHWADDR_OLD:
3606 case SIOCBONDSETHWADDR:
3607 bond_set_dev_addr(bond_dev, slave_dev);
3610 case BOND_CHANGE_ACTIVE_OLD:
3611 case SIOCBONDCHANGEACTIVE:
3612 res = bond_ioctl_change_active(bond_dev, slave_dev);
3624 static bool bond_addr_in_mc_list(unsigned char *addr,
3625 struct netdev_hw_addr_list *list,
3628 struct netdev_hw_addr *ha;
3630 netdev_hw_addr_list_for_each(ha, list)
3631 if (!memcmp(ha->addr, addr, addrlen))
3637 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3639 struct bonding *bond = netdev_priv(bond_dev);
3641 if (change & IFF_PROMISC)
3642 bond_set_promiscuity(bond,
3643 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3645 if (change & IFF_ALLMULTI)
3646 bond_set_allmulti(bond,
3647 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3650 static void bond_set_multicast_list(struct net_device *bond_dev)
3652 struct bonding *bond = netdev_priv(bond_dev);
3653 struct netdev_hw_addr *ha;
3656 read_lock(&bond->lock);
3658 /* looking for addresses to add to slaves' mc list */
3659 netdev_for_each_mc_addr(ha, bond_dev) {
3660 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3661 bond_dev->addr_len);
3663 bond_mc_add(bond, ha->addr);
3666 /* looking for addresses to delete from slaves' list */
3667 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3668 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3669 bond_dev->addr_len);
3671 bond_mc_del(bond, ha->addr);
3674 /* save master's multicast list */
3675 __hw_addr_flush(&bond->mc_list);
3676 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3677 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3679 read_unlock(&bond->lock);
3682 static int bond_neigh_init(struct neighbour *n)
3684 struct bonding *bond = netdev_priv(n->dev);
3685 struct slave *slave = bond->first_slave;
3686 const struct net_device_ops *slave_ops;
3687 struct neigh_parms parms;
3693 slave_ops = slave->dev->netdev_ops;
3695 if (!slave_ops->ndo_neigh_setup)
3698 parms.neigh_setup = NULL;
3699 parms.neigh_cleanup = NULL;
3700 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3705 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3706 * after the last slave has been detached. Assumes that all slaves
3707 * utilize the same neigh_cleanup (true at this writing as only user
3710 n->parms->neigh_cleanup = parms.neigh_cleanup;
3712 if (!parms.neigh_setup)
3715 return parms.neigh_setup(n);
3719 * The bonding ndo_neigh_setup is called at init time beofre any
3720 * slave exists. So we must declare proxy setup function which will
3721 * be used at run time to resolve the actual slave neigh param setup.
3723 static int bond_neigh_setup(struct net_device *dev,
3724 struct neigh_parms *parms)
3726 parms->neigh_setup = bond_neigh_init;
3732 * Change the MTU of all of a master's slaves to match the master
3734 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3736 struct bonding *bond = netdev_priv(bond_dev);
3737 struct slave *slave, *stop_at;
3741 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3742 (bond_dev ? bond_dev->name : "None"), new_mtu);
3744 /* Can't hold bond->lock with bh disabled here since
3745 * some base drivers panic. On the other hand we can't
3746 * hold bond->lock without bh disabled because we'll
3747 * deadlock. The only solution is to rely on the fact
3748 * that we're under rtnl_lock here, and the slaves
3749 * list won't change. This doesn't solve the problem
3750 * of setting the slave's MTU while it is
3751 * transmitting, but the assumption is that the base
3752 * driver can handle that.
3754 * TODO: figure out a way to safely iterate the slaves
3755 * list, but without holding a lock around the actual
3756 * call to the base driver.
3759 bond_for_each_slave(bond, slave, i) {
3760 pr_debug("s %p s->p %p c_m %p\n",
3763 slave->dev->netdev_ops->ndo_change_mtu);
3765 res = dev_set_mtu(slave->dev, new_mtu);
3768 /* If we failed to set the slave's mtu to the new value
3769 * we must abort the operation even in ACTIVE_BACKUP
3770 * mode, because if we allow the backup slaves to have
3771 * different mtu values than the active slave we'll
3772 * need to change their mtu when doing a failover. That
3773 * means changing their mtu from timer context, which
3774 * is probably not a good idea.
3776 pr_debug("err %d %s\n", res, slave->dev->name);
3781 bond_dev->mtu = new_mtu;
3786 /* unwind from head to the slave that failed */
3788 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3791 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3793 pr_debug("unwind err %d dev %s\n",
3794 tmp_res, slave->dev->name);
3804 * Note that many devices must be down to change the HW address, and
3805 * downing the master releases all slaves. We can make bonds full of
3806 * bonding devices to test this, however.
3808 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3810 struct bonding *bond = netdev_priv(bond_dev);
3811 struct sockaddr *sa = addr, tmp_sa;
3812 struct slave *slave, *stop_at;
3816 if (bond->params.mode == BOND_MODE_ALB)
3817 return bond_alb_set_mac_address(bond_dev, addr);
3820 pr_debug("bond=%p, name=%s\n",
3821 bond, bond_dev ? bond_dev->name : "None");
3824 * If fail_over_mac is set to active, do nothing and return
3825 * success. Returning an error causes ifenslave to fail.
3827 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3830 if (!is_valid_ether_addr(sa->sa_data))
3831 return -EADDRNOTAVAIL;
3833 /* Can't hold bond->lock with bh disabled here since
3834 * some base drivers panic. On the other hand we can't
3835 * hold bond->lock without bh disabled because we'll
3836 * deadlock. The only solution is to rely on the fact
3837 * that we're under rtnl_lock here, and the slaves
3838 * list won't change. This doesn't solve the problem
3839 * of setting the slave's hw address while it is
3840 * transmitting, but the assumption is that the base
3841 * driver can handle that.
3843 * TODO: figure out a way to safely iterate the slaves
3844 * list, but without holding a lock around the actual
3845 * call to the base driver.
3848 bond_for_each_slave(bond, slave, i) {
3849 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3850 pr_debug("slave %p %s\n", slave, slave->dev->name);
3852 if (slave_ops->ndo_set_mac_address == NULL) {
3854 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3858 res = dev_set_mac_address(slave->dev, addr);
3860 /* TODO: consider downing the slave
3862 * User should expect communications
3863 * breakage anyway until ARP finish
3866 pr_debug("err %d %s\n", res, slave->dev->name);
3872 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3876 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3877 tmp_sa.sa_family = bond_dev->type;
3879 /* unwind from head to the slave that failed */
3881 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3884 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3886 pr_debug("unwind err %d dev %s\n",
3887 tmp_res, slave->dev->name);
3894 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3896 struct bonding *bond = netdev_priv(bond_dev);
3897 struct slave *slave, *start_at;
3898 int i, slave_no, res = 1;
3899 struct iphdr *iph = ip_hdr(skb);
3902 * Start with the curr_active_slave that joined the bond as the
3903 * default for sending IGMP traffic. For failover purposes one
3904 * needs to maintain some consistency for the interface that will
3905 * send the join/membership reports. The curr_active_slave found
3906 * will send all of this type of traffic.
3908 if ((iph->protocol == IPPROTO_IGMP) &&
3909 (skb->protocol == htons(ETH_P_IP))) {
3911 read_lock(&bond->curr_slave_lock);
3912 slave = bond->curr_active_slave;
3913 read_unlock(&bond->curr_slave_lock);
3919 * Concurrent TX may collide on rr_tx_counter; we accept
3920 * that as being rare enough not to justify using an
3923 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3925 bond_for_each_slave(bond, slave, i) {
3933 bond_for_each_slave_from(bond, slave, i, start_at) {
3934 if (IS_UP(slave->dev) &&
3935 (slave->link == BOND_LINK_UP) &&
3936 bond_is_active_slave(slave)) {
3937 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3944 /* no suitable interface, frame not sent */
3948 return NETDEV_TX_OK;
3953 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3954 * the bond has a usable interface.
3956 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3958 struct bonding *bond = netdev_priv(bond_dev);
3961 read_lock(&bond->curr_slave_lock);
3963 if (bond->curr_active_slave)
3964 res = bond_dev_queue_xmit(bond, skb,
3965 bond->curr_active_slave->dev);
3967 read_unlock(&bond->curr_slave_lock);
3970 /* no suitable interface, frame not sent */
3973 return NETDEV_TX_OK;
3977 * In bond_xmit_xor() , we determine the output device by using a pre-
3978 * determined xmit_hash_policy(), If the selected device is not enabled,
3979 * find the next active slave.
3981 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3983 struct bonding *bond = netdev_priv(bond_dev);
3984 struct slave *slave, *start_at;
3989 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3991 bond_for_each_slave(bond, slave, i) {
3999 bond_for_each_slave_from(bond, slave, i, start_at) {
4000 if (IS_UP(slave->dev) &&
4001 (slave->link == BOND_LINK_UP) &&
4002 bond_is_active_slave(slave)) {
4003 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4009 /* no suitable interface, frame not sent */
4013 return NETDEV_TX_OK;
4017 * in broadcast mode, we send everything to all usable interfaces.
4019 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4021 struct bonding *bond = netdev_priv(bond_dev);
4022 struct slave *slave, *start_at;
4023 struct net_device *tx_dev = NULL;
4027 read_lock(&bond->curr_slave_lock);
4028 start_at = bond->curr_active_slave;
4029 read_unlock(&bond->curr_slave_lock);
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)) {
4039 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4041 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4046 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4052 tx_dev = slave->dev;
4057 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4061 /* no suitable interface, frame not sent */
4064 /* frame sent to all suitable interfaces */
4065 return NETDEV_TX_OK;
4068 /*------------------------- Device initialization ---------------------------*/
4070 static void bond_set_xmit_hash_policy(struct bonding *bond)
4072 switch (bond->params.xmit_policy) {
4073 case BOND_XMIT_POLICY_LAYER23:
4074 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4076 case BOND_XMIT_POLICY_LAYER34:
4077 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4079 case BOND_XMIT_POLICY_LAYER2:
4081 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4087 * Lookup the slave that corresponds to a qid
4089 static inline int bond_slave_override(struct bonding *bond,
4090 struct sk_buff *skb)
4093 struct slave *slave = NULL;
4094 struct slave *check_slave;
4096 if (!skb->queue_mapping)
4099 /* Find out if any slaves have the same mapping as this skb. */
4100 bond_for_each_slave(bond, check_slave, i) {
4101 if (check_slave->queue_id == skb->queue_mapping) {
4102 slave = check_slave;
4107 /* If the slave isn't UP, use default transmit policy. */
4108 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4109 (slave->link == BOND_LINK_UP)) {
4110 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4117 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4120 * This helper function exists to help dev_pick_tx get the correct
4121 * destination queue. Using a helper function skips a call to
4122 * skb_tx_hash and will put the skbs in the queue we expect on their
4123 * way down to the bonding driver.
4125 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4128 * Save the original txq to restore before passing to the driver
4130 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4132 if (unlikely(txq >= dev->real_num_tx_queues)) {
4134 txq -= dev->real_num_tx_queues;
4135 } while (txq >= dev->real_num_tx_queues);
4140 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4142 struct bonding *bond = netdev_priv(dev);
4144 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4145 if (!bond_slave_override(bond, skb))
4146 return NETDEV_TX_OK;
4149 switch (bond->params.mode) {
4150 case BOND_MODE_ROUNDROBIN:
4151 return bond_xmit_roundrobin(skb, dev);
4152 case BOND_MODE_ACTIVEBACKUP:
4153 return bond_xmit_activebackup(skb, dev);
4155 return bond_xmit_xor(skb, dev);
4156 case BOND_MODE_BROADCAST:
4157 return bond_xmit_broadcast(skb, dev);
4158 case BOND_MODE_8023AD:
4159 return bond_3ad_xmit_xor(skb, dev);
4162 return bond_alb_xmit(skb, dev);
4164 /* Should never happen, mode already checked */
4165 pr_err("%s: Error: Unknown bonding mode %d\n",
4166 dev->name, bond->params.mode);
4169 return NETDEV_TX_OK;
4173 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4175 struct bonding *bond = netdev_priv(dev);
4176 netdev_tx_t ret = NETDEV_TX_OK;
4179 * If we risk deadlock from transmitting this in the
4180 * netpoll path, tell netpoll to queue the frame for later tx
4182 if (is_netpoll_tx_blocked(dev))
4183 return NETDEV_TX_BUSY;
4185 read_lock(&bond->lock);
4187 if (bond->slave_cnt)
4188 ret = __bond_start_xmit(skb, dev);
4192 read_unlock(&bond->lock);
4198 * set bond mode specific net device operations
4200 void bond_set_mode_ops(struct bonding *bond, int mode)
4202 struct net_device *bond_dev = bond->dev;
4205 case BOND_MODE_ROUNDROBIN:
4207 case BOND_MODE_ACTIVEBACKUP:
4210 bond_set_xmit_hash_policy(bond);
4212 case BOND_MODE_BROADCAST:
4214 case BOND_MODE_8023AD:
4215 bond_set_xmit_hash_policy(bond);
4222 /* Should never happen, mode already checked */
4223 pr_err("%s: Error: Unknown bonding mode %d\n",
4224 bond_dev->name, mode);
4229 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4230 struct ethtool_cmd *ecmd)
4232 struct bonding *bond = netdev_priv(bond_dev);
4233 struct slave *slave;
4235 unsigned long speed = 0;
4237 ecmd->duplex = DUPLEX_UNKNOWN;
4238 ecmd->port = PORT_OTHER;
4240 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4241 * do not need to check mode. Though link speed might not represent
4242 * the true receive or transmit bandwidth (not all modes are symmetric)
4243 * this is an accurate maximum.
4245 read_lock(&bond->lock);
4246 bond_for_each_slave(bond, slave, i) {
4247 if (SLAVE_IS_OK(slave)) {
4248 if (slave->speed != SPEED_UNKNOWN)
4249 speed += slave->speed;
4250 if (ecmd->duplex == DUPLEX_UNKNOWN &&
4251 slave->duplex != DUPLEX_UNKNOWN)
4252 ecmd->duplex = slave->duplex;
4255 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4256 read_unlock(&bond->lock);
4260 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4261 struct ethtool_drvinfo *drvinfo)
4263 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4264 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4265 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4269 static const struct ethtool_ops bond_ethtool_ops = {
4270 .get_drvinfo = bond_ethtool_get_drvinfo,
4271 .get_settings = bond_ethtool_get_settings,
4272 .get_link = ethtool_op_get_link,
4275 static const struct net_device_ops bond_netdev_ops = {
4276 .ndo_init = bond_init,
4277 .ndo_uninit = bond_uninit,
4278 .ndo_open = bond_open,
4279 .ndo_stop = bond_close,
4280 .ndo_start_xmit = bond_start_xmit,
4281 .ndo_select_queue = bond_select_queue,
4282 .ndo_get_stats64 = bond_get_stats,
4283 .ndo_do_ioctl = bond_do_ioctl,
4284 .ndo_change_rx_flags = bond_change_rx_flags,
4285 .ndo_set_rx_mode = bond_set_multicast_list,
4286 .ndo_change_mtu = bond_change_mtu,
4287 .ndo_set_mac_address = bond_set_mac_address,
4288 .ndo_neigh_setup = bond_neigh_setup,
4289 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4290 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4291 #ifdef CONFIG_NET_POLL_CONTROLLER
4292 .ndo_netpoll_setup = bond_netpoll_setup,
4293 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4294 .ndo_poll_controller = bond_poll_controller,
4296 .ndo_add_slave = bond_enslave,
4297 .ndo_del_slave = bond_release,
4298 .ndo_fix_features = bond_fix_features,
4301 static const struct device_type bond_type = {
4305 static void bond_destructor(struct net_device *bond_dev)
4307 struct bonding *bond = netdev_priv(bond_dev);
4309 destroy_workqueue(bond->wq);
4310 free_netdev(bond_dev);
4313 static void bond_setup(struct net_device *bond_dev)
4315 struct bonding *bond = netdev_priv(bond_dev);
4317 /* initialize rwlocks */
4318 rwlock_init(&bond->lock);
4319 rwlock_init(&bond->curr_slave_lock);
4321 bond->params = bonding_defaults;
4323 /* Initialize pointers */
4324 bond->dev = bond_dev;
4325 INIT_LIST_HEAD(&bond->vlan_list);
4327 /* Initialize the device entry points */
4328 ether_setup(bond_dev);
4329 bond_dev->netdev_ops = &bond_netdev_ops;
4330 bond_dev->ethtool_ops = &bond_ethtool_ops;
4331 bond_set_mode_ops(bond, bond->params.mode);
4333 bond_dev->destructor = bond_destructor;
4335 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4337 /* Initialize the device options */
4338 bond_dev->tx_queue_len = 0;
4339 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4340 bond_dev->priv_flags |= IFF_BONDING;
4341 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4343 /* At first, we block adding VLANs. That's the only way to
4344 * prevent problems that occur when adding VLANs over an
4345 * empty bond. The block will be removed once non-challenged
4346 * slaves are enslaved.
4348 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4350 /* don't acquire bond device's netif_tx_lock when
4352 bond_dev->features |= NETIF_F_LLTX;
4354 /* By default, we declare the bond to be fully
4355 * VLAN hardware accelerated capable. Special
4356 * care is taken in the various xmit functions
4357 * when there are slaves that are not hw accel
4361 bond_dev->hw_features = BOND_VLAN_FEATURES |
4362 NETIF_F_HW_VLAN_CTAG_TX |
4363 NETIF_F_HW_VLAN_CTAG_RX |
4364 NETIF_F_HW_VLAN_CTAG_FILTER;
4366 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4367 bond_dev->features |= bond_dev->hw_features;
4371 * Destroy a bonding device.
4372 * Must be under rtnl_lock when this function is called.
4374 static void bond_uninit(struct net_device *bond_dev)
4376 struct bonding *bond = netdev_priv(bond_dev);
4377 struct vlan_entry *vlan, *tmp;
4379 bond_netpoll_cleanup(bond_dev);
4381 /* Release the bonded slaves */
4382 while (bond->first_slave != NULL)
4383 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4384 pr_info("%s: released all slaves\n", bond_dev->name);
4386 list_del(&bond->bond_list);
4388 bond_debug_unregister(bond);
4390 __hw_addr_flush(&bond->mc_list);
4392 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4393 list_del(&vlan->vlan_list);
4398 /*------------------------- Module initialization ---------------------------*/
4401 * Convert string input module parms. Accept either the
4402 * number of the mode or its string name. A bit complicated because
4403 * some mode names are substrings of other names, and calls from sysfs
4404 * may have whitespace in the name (trailing newlines, for example).
4406 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4408 int modeint = -1, i, rv;
4409 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4411 for (p = (char *)buf; *p; p++)
4412 if (!(isdigit(*p) || isspace(*p)))
4416 rv = sscanf(buf, "%20s", modestr);
4418 rv = sscanf(buf, "%d", &modeint);
4423 for (i = 0; tbl[i].modename; i++) {
4424 if (modeint == tbl[i].mode)
4426 if (strcmp(modestr, tbl[i].modename) == 0)
4433 static int bond_check_params(struct bond_params *params)
4435 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4438 * Convert string parameters.
4441 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4442 if (bond_mode == -1) {
4443 pr_err("Error: Invalid bonding mode \"%s\"\n",
4444 mode == NULL ? "NULL" : mode);
4449 if (xmit_hash_policy) {
4450 if ((bond_mode != BOND_MODE_XOR) &&
4451 (bond_mode != BOND_MODE_8023AD)) {
4452 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4453 bond_mode_name(bond_mode));
4455 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4457 if (xmit_hashtype == -1) {
4458 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4459 xmit_hash_policy == NULL ? "NULL" :
4467 if (bond_mode != BOND_MODE_8023AD) {
4468 pr_info("lacp_rate param is irrelevant in mode %s\n",
4469 bond_mode_name(bond_mode));
4471 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4472 if (lacp_fast == -1) {
4473 pr_err("Error: Invalid lacp rate \"%s\"\n",
4474 lacp_rate == NULL ? "NULL" : lacp_rate);
4481 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4482 if (params->ad_select == -1) {
4483 pr_err("Error: Invalid ad_select \"%s\"\n",
4484 ad_select == NULL ? "NULL" : ad_select);
4488 if (bond_mode != BOND_MODE_8023AD) {
4489 pr_warning("ad_select param only affects 802.3ad mode\n");
4492 params->ad_select = BOND_AD_STABLE;
4495 if (max_bonds < 0) {
4496 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4497 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4498 max_bonds = BOND_DEFAULT_MAX_BONDS;
4502 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4503 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4504 miimon = BOND_LINK_MON_INTERV;
4508 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4513 if (downdelay < 0) {
4514 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4515 downdelay, INT_MAX);
4519 if ((use_carrier != 0) && (use_carrier != 1)) {
4520 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4525 if (num_peer_notif < 0 || num_peer_notif > 255) {
4526 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4531 /* reset values for 802.3ad */
4532 if (bond_mode == BOND_MODE_8023AD) {
4534 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");
4535 pr_warning("Forcing miimon to 100msec\n");
4540 if (tx_queues < 1 || tx_queues > 255) {
4541 pr_warning("Warning: tx_queues (%d) should be between "
4542 "1 and 255, resetting to %d\n",
4543 tx_queues, BOND_DEFAULT_TX_QUEUES);
4544 tx_queues = BOND_DEFAULT_TX_QUEUES;
4547 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4548 pr_warning("Warning: all_slaves_active module parameter (%d), "
4549 "not of valid value (0/1), so it was set to "
4550 "0\n", all_slaves_active);
4551 all_slaves_active = 0;
4554 if (resend_igmp < 0 || resend_igmp > 255) {
4555 pr_warning("Warning: resend_igmp (%d) should be between "
4556 "0 and 255, resetting to %d\n",
4557 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4558 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4561 /* reset values for TLB/ALB */
4562 if ((bond_mode == BOND_MODE_TLB) ||
4563 (bond_mode == BOND_MODE_ALB)) {
4565 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");
4566 pr_warning("Forcing miimon to 100msec\n");
4571 if (bond_mode == BOND_MODE_ALB) {
4572 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",
4577 if (updelay || downdelay) {
4578 /* just warn the user the up/down delay will have
4579 * no effect since miimon is zero...
4581 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",
4582 updelay, downdelay);
4585 /* don't allow arp monitoring */
4587 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4588 miimon, arp_interval);
4592 if ((updelay % miimon) != 0) {
4593 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4595 (updelay / miimon) * miimon);
4600 if ((downdelay % miimon) != 0) {
4601 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4603 (downdelay / miimon) * miimon);
4606 downdelay /= miimon;
4609 if (arp_interval < 0) {
4610 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4611 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4612 arp_interval = BOND_LINK_ARP_INTERV;
4615 for (arp_ip_count = 0;
4616 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4618 /* not complete check, but should be good enough to
4620 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4621 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4622 ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4623 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4624 arp_ip_target[arp_ip_count]);
4627 arp_target[arp_ip_count] = ip;
4631 if (arp_interval && !arp_ip_count) {
4632 /* don't allow arping if no arp_ip_target given... */
4633 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4639 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4640 pr_err("arp_validate only supported in active-backup mode\n");
4643 if (!arp_interval) {
4644 pr_err("arp_validate requires arp_interval\n");
4648 arp_validate_value = bond_parse_parm(arp_validate,
4650 if (arp_validate_value == -1) {
4651 pr_err("Error: invalid arp_validate \"%s\"\n",
4652 arp_validate == NULL ? "NULL" : arp_validate);
4656 arp_validate_value = 0;
4659 pr_info("MII link monitoring set to %d ms\n", miimon);
4660 } else if (arp_interval) {
4663 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4665 arp_validate_tbl[arp_validate_value].modename,
4668 for (i = 0; i < arp_ip_count; i++)
4669 pr_info(" %s", arp_ip_target[i]);
4673 } else if (max_bonds) {
4674 /* miimon and arp_interval not set, we need one so things
4675 * work as expected, see bonding.txt for details
4677 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");
4680 if (primary && !USES_PRIMARY(bond_mode)) {
4681 /* currently, using a primary only makes sense
4682 * in active backup, TLB or ALB modes
4684 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4685 primary, bond_mode_name(bond_mode));
4689 if (primary && primary_reselect) {
4690 primary_reselect_value = bond_parse_parm(primary_reselect,
4692 if (primary_reselect_value == -1) {
4693 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4695 NULL ? "NULL" : primary_reselect);
4699 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4702 if (fail_over_mac) {
4703 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4705 if (fail_over_mac_value == -1) {
4706 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4707 arp_validate == NULL ? "NULL" : arp_validate);
4711 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4712 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4714 fail_over_mac_value = BOND_FOM_NONE;
4717 /* fill params struct with the proper values */
4718 params->mode = bond_mode;
4719 params->xmit_policy = xmit_hashtype;
4720 params->miimon = miimon;
4721 params->num_peer_notif = num_peer_notif;
4722 params->arp_interval = arp_interval;
4723 params->arp_validate = arp_validate_value;
4724 params->updelay = updelay;
4725 params->downdelay = downdelay;
4726 params->use_carrier = use_carrier;
4727 params->lacp_fast = lacp_fast;
4728 params->primary[0] = 0;
4729 params->primary_reselect = primary_reselect_value;
4730 params->fail_over_mac = fail_over_mac_value;
4731 params->tx_queues = tx_queues;
4732 params->all_slaves_active = all_slaves_active;
4733 params->resend_igmp = resend_igmp;
4734 params->min_links = min_links;
4737 strncpy(params->primary, primary, IFNAMSIZ);
4738 params->primary[IFNAMSIZ - 1] = 0;
4741 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4746 static struct lock_class_key bonding_netdev_xmit_lock_key;
4747 static struct lock_class_key bonding_netdev_addr_lock_key;
4748 static struct lock_class_key bonding_tx_busylock_key;
4750 static void bond_set_lockdep_class_one(struct net_device *dev,
4751 struct netdev_queue *txq,
4754 lockdep_set_class(&txq->_xmit_lock,
4755 &bonding_netdev_xmit_lock_key);
4758 static void bond_set_lockdep_class(struct net_device *dev)
4760 lockdep_set_class(&dev->addr_list_lock,
4761 &bonding_netdev_addr_lock_key);
4762 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4763 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4767 * Called from registration process
4769 static int bond_init(struct net_device *bond_dev)
4771 struct bonding *bond = netdev_priv(bond_dev);
4772 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4773 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4775 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4778 * Initialize locks that may be required during
4779 * en/deslave operations. All of the bond_open work
4780 * (of which this is part) should really be moved to
4781 * a phase prior to dev_open
4783 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4784 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4786 bond->wq = create_singlethread_workqueue(bond_dev->name);
4790 bond_set_lockdep_class(bond_dev);
4792 list_add_tail(&bond->bond_list, &bn->dev_list);
4794 bond_prepare_sysfs_group(bond);
4796 bond_debug_register(bond);
4798 /* Ensure valid dev_addr */
4799 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4800 bond_dev->addr_assign_type == NET_ADDR_PERM) {
4801 eth_hw_addr_random(bond_dev);
4802 bond->dev_addr_from_first = true;
4805 __hw_addr_init(&bond->mc_list);
4809 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4811 if (tb[IFLA_ADDRESS]) {
4812 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4814 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4815 return -EADDRNOTAVAIL;
4820 static unsigned int bond_get_num_tx_queues(void)
4825 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4827 .priv_size = sizeof(struct bonding),
4828 .setup = bond_setup,
4829 .validate = bond_validate,
4830 .get_num_tx_queues = bond_get_num_tx_queues,
4831 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4835 /* Create a new bond based on the specified name and bonding parameters.
4836 * If name is NULL, obtain a suitable "bond%d" name for us.
4837 * Caller must NOT hold rtnl_lock; we need to release it here before we
4838 * set up our sysfs entries.
4840 int bond_create(struct net *net, const char *name)
4842 struct net_device *bond_dev;
4847 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4848 name ? name : "bond%d",
4849 bond_setup, tx_queues);
4851 pr_err("%s: eek! can't alloc netdev!\n", name);
4856 dev_net_set(bond_dev, net);
4857 bond_dev->rtnl_link_ops = &bond_link_ops;
4859 res = register_netdevice(bond_dev);
4861 netif_carrier_off(bond_dev);
4865 bond_destructor(bond_dev);
4869 static int __net_init bond_net_init(struct net *net)
4871 struct bond_net *bn = net_generic(net, bond_net_id);
4874 INIT_LIST_HEAD(&bn->dev_list);
4876 bond_create_proc_dir(bn);
4877 bond_create_sysfs(bn);
4882 static void __net_exit bond_net_exit(struct net *net)
4884 struct bond_net *bn = net_generic(net, bond_net_id);
4886 bond_destroy_sysfs(bn);
4887 bond_destroy_proc_dir(bn);
4890 static struct pernet_operations bond_net_ops = {
4891 .init = bond_net_init,
4892 .exit = bond_net_exit,
4894 .size = sizeof(struct bond_net),
4897 static int __init bonding_init(void)
4902 pr_info("%s", bond_version);
4904 res = bond_check_params(&bonding_defaults);
4908 res = register_pernet_subsys(&bond_net_ops);
4912 res = rtnl_link_register(&bond_link_ops);
4916 bond_create_debugfs();
4918 for (i = 0; i < max_bonds; i++) {
4919 res = bond_create(&init_net, NULL);
4924 register_netdevice_notifier(&bond_netdev_notifier);
4928 rtnl_link_unregister(&bond_link_ops);
4930 unregister_pernet_subsys(&bond_net_ops);
4935 static void __exit bonding_exit(void)
4937 unregister_netdevice_notifier(&bond_netdev_notifier);
4939 bond_destroy_debugfs();
4941 unregister_pernet_subsys(&bond_net_ops);
4942 rtnl_link_unregister(&bond_link_ops);
4944 #ifdef CONFIG_NET_POLL_CONTROLLER
4946 * Make sure we don't have an imbalance on our netpoll blocking
4948 WARN_ON(atomic_read(&netpoll_block_tx));
4952 module_init(bonding_init);
4953 module_exit(bonding_exit);
4954 MODULE_LICENSE("GPL");
4955 MODULE_VERSION(DRV_VERSION);
4956 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4957 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4958 MODULE_ALIAS_RTNL_LINK("bond");