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)->bond_queue_mapping));
399 skb->queue_mapping = qdisc_skb_cb(skb)->bond_queue_mapping;
401 if (unlikely(netpoll_tx_running(slave_dev)))
402 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
410 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411 * We don't protect the slave list iteration with a lock because:
412 * a. This operation is performed in IOCTL context,
413 * b. The operation is protected by the RTNL semaphore in the 8021q code,
414 * c. Holding a lock with BH disabled while directly calling a base driver
415 * entry point is generally a BAD idea.
417 * The design of synchronization/protection for this operation in the 8021q
418 * module is good for one or more VLAN devices over a single physical device
419 * and cannot be extended for a teaming solution like bonding, so there is a
420 * potential race condition here where a net device from the vlan group might
421 * be referenced (either by a base driver or the 8021q code) while it is being
422 * removed from the system. However, it turns out we're not making matters
423 * worse, and if it works for regular VLAN usage it will work here too.
427 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428 * @bond_dev: bonding net device that got called
429 * @vid: vlan id being added
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
433 struct bonding *bond = netdev_priv(bond_dev);
434 struct slave *slave, *stop_at;
437 bond_for_each_slave(bond, slave, i) {
438 res = vlan_vid_add(slave->dev, vid);
443 res = bond_add_vlan(bond, vid);
445 pr_err("%s: Error: Failed to add vlan id %d\n",
446 bond_dev->name, vid);
453 /* unwind from head to the slave that failed */
455 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
456 vlan_vid_del(slave->dev, vid);
462 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463 * @bond_dev: bonding net device that got called
464 * @vid: vlan id being removed
466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
468 struct bonding *bond = netdev_priv(bond_dev);
472 bond_for_each_slave(bond, slave, i)
473 vlan_vid_del(slave->dev, vid);
475 res = bond_del_vlan(bond, vid);
477 pr_err("%s: Error: Failed to remove vlan id %d\n",
478 bond_dev->name, vid);
485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
487 struct vlan_entry *vlan;
490 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
491 res = vlan_vid_add(slave_dev, vlan->vlan_id);
493 pr_warning("%s: Failed to add vlan id %d to device %s\n",
494 bond->dev->name, vlan->vlan_id,
499 static void bond_del_vlans_from_slave(struct bonding *bond,
500 struct net_device *slave_dev)
502 struct vlan_entry *vlan;
504 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
507 vlan_vid_del(slave_dev, vlan->vlan_id);
511 /*------------------------------- Link status -------------------------------*/
514 * Set the carrier state for the master according to the state of its
515 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
516 * do special 802.3ad magic.
518 * Returns zero if carrier state does not change, nonzero if it does.
520 static int bond_set_carrier(struct bonding *bond)
525 if (bond->slave_cnt == 0)
528 if (bond->params.mode == BOND_MODE_8023AD)
529 return bond_3ad_set_carrier(bond);
531 bond_for_each_slave(bond, slave, i) {
532 if (slave->link == BOND_LINK_UP) {
533 if (!netif_carrier_ok(bond->dev)) {
534 netif_carrier_on(bond->dev);
542 if (netif_carrier_ok(bond->dev)) {
543 netif_carrier_off(bond->dev);
550 * Get link speed and duplex from the slave's base driver
551 * using ethtool. If for some reason the call fails or the
552 * values are invalid, set speed and duplex to -1,
555 static void bond_update_speed_duplex(struct slave *slave)
557 struct net_device *slave_dev = slave->dev;
558 struct ethtool_cmd ecmd;
562 slave->speed = SPEED_UNKNOWN;
563 slave->duplex = DUPLEX_UNKNOWN;
565 res = __ethtool_get_settings(slave_dev, &ecmd);
569 slave_speed = ethtool_cmd_speed(&ecmd);
570 if (slave_speed == 0 || slave_speed == ((__u32) -1))
573 switch (ecmd.duplex) {
581 slave->speed = slave_speed;
582 slave->duplex = ecmd.duplex;
588 * if <dev> supports MII link status reporting, check its link status.
590 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591 * depending upon the setting of the use_carrier parameter.
593 * Return either BMSR_LSTATUS, meaning that the link is up (or we
594 * can't tell and just pretend it is), or 0, meaning that the link is
597 * If reporting is non-zero, instead of faking link up, return -1 if
598 * both ETHTOOL and MII ioctls fail (meaning the device does not
599 * support them). If use_carrier is set, return whatever it says.
600 * It'd be nice if there was a good way to tell if a driver supports
601 * netif_carrier, but there really isn't.
603 static int bond_check_dev_link(struct bonding *bond,
604 struct net_device *slave_dev, int reporting)
606 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
607 int (*ioctl)(struct net_device *, struct ifreq *, int);
609 struct mii_ioctl_data *mii;
611 if (!reporting && !netif_running(slave_dev))
614 if (bond->params.use_carrier)
615 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
617 /* Try to get link status using Ethtool first. */
618 if (slave_dev->ethtool_ops) {
619 if (slave_dev->ethtool_ops->get_link) {
622 link = slave_dev->ethtool_ops->get_link(slave_dev);
624 return link ? BMSR_LSTATUS : 0;
628 /* Ethtool can't be used, fallback to MII ioctls. */
629 ioctl = slave_ops->ndo_do_ioctl;
631 /* TODO: set pointer to correct ioctl on a per team member */
632 /* bases to make this more efficient. that is, once */
633 /* we determine the correct ioctl, we will always */
634 /* call it and not the others for that team */
638 * We cannot assume that SIOCGMIIPHY will also read a
639 * register; not all network drivers (e.g., e100)
643 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
644 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
646 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
647 mii->reg_num = MII_BMSR;
648 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
649 return mii->val_out & BMSR_LSTATUS;
654 * If reporting, report that either there's no dev->do_ioctl,
655 * or both SIOCGMIIREG and get_link failed (meaning that we
656 * cannot report link status). If not reporting, pretend
659 return reporting ? -1 : BMSR_LSTATUS;
662 /*----------------------------- Multicast list ------------------------------*/
665 * Push the promiscuity flag down to appropriate slaves
667 static int bond_set_promiscuity(struct bonding *bond, int inc)
670 if (USES_PRIMARY(bond->params.mode)) {
671 /* write lock already acquired */
672 if (bond->curr_active_slave) {
673 err = dev_set_promiscuity(bond->curr_active_slave->dev,
679 bond_for_each_slave(bond, slave, i) {
680 err = dev_set_promiscuity(slave->dev, inc);
689 * Push the allmulti flag down to all slaves
691 static int bond_set_allmulti(struct bonding *bond, int inc)
694 if (USES_PRIMARY(bond->params.mode)) {
695 /* write lock already acquired */
696 if (bond->curr_active_slave) {
697 err = dev_set_allmulti(bond->curr_active_slave->dev,
703 bond_for_each_slave(bond, slave, i) {
704 err = dev_set_allmulti(slave->dev, inc);
713 * Add a Multicast address to slaves
716 static void bond_mc_add(struct bonding *bond, void *addr)
718 if (USES_PRIMARY(bond->params.mode)) {
719 /* write lock already acquired */
720 if (bond->curr_active_slave)
721 dev_mc_add(bond->curr_active_slave->dev, addr);
726 bond_for_each_slave(bond, slave, i)
727 dev_mc_add(slave->dev, addr);
732 * Remove a multicast address from slave
735 static void bond_mc_del(struct bonding *bond, void *addr)
737 if (USES_PRIMARY(bond->params.mode)) {
738 /* write lock already acquired */
739 if (bond->curr_active_slave)
740 dev_mc_del(bond->curr_active_slave->dev, addr);
744 bond_for_each_slave(bond, slave, i) {
745 dev_mc_del(slave->dev, addr);
751 static void __bond_resend_igmp_join_requests(struct net_device *dev)
753 struct in_device *in_dev;
756 in_dev = __in_dev_get_rcu(dev);
758 ip_mc_rejoin_groups(in_dev);
763 * Retrieve the list of registered multicast addresses for the bonding
764 * device and retransmit an IGMP JOIN request to the current active
767 static void bond_resend_igmp_join_requests(struct bonding *bond)
769 struct net_device *bond_dev, *vlan_dev, *master_dev;
770 struct vlan_entry *vlan;
772 read_lock(&bond->lock);
774 bond_dev = bond->dev;
776 /* rejoin all groups on bond device */
777 __bond_resend_igmp_join_requests(bond_dev);
780 * if bond is enslaved to a bridge,
781 * then rejoin all groups on its master
783 master_dev = bond_dev->master;
785 if ((master_dev->priv_flags & IFF_EBRIDGE)
786 && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
787 __bond_resend_igmp_join_requests(master_dev);
789 /* rejoin all groups on vlan devices */
790 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
792 vlan_dev = __vlan_find_dev_deep(bond_dev,
796 __bond_resend_igmp_join_requests(vlan_dev);
799 if (--bond->igmp_retrans > 0)
800 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
802 read_unlock(&bond->lock);
805 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
807 struct bonding *bond = container_of(work, struct bonding,
809 bond_resend_igmp_join_requests(bond);
813 * flush all members of flush->mc_list from device dev->mc_list
815 static void bond_mc_list_flush(struct net_device *bond_dev,
816 struct net_device *slave_dev)
818 struct bonding *bond = netdev_priv(bond_dev);
819 struct netdev_hw_addr *ha;
821 netdev_for_each_mc_addr(ha, bond_dev)
822 dev_mc_del(slave_dev, ha->addr);
824 if (bond->params.mode == BOND_MODE_8023AD) {
825 /* del lacpdu mc addr from mc list */
826 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
828 dev_mc_del(slave_dev, lacpdu_multicast);
832 /*--------------------------- Active slave change ---------------------------*/
835 * Update the mc list and multicast-related flags for the new and
836 * old active slaves (if any) according to the multicast mode, and
837 * promiscuous flags unconditionally.
839 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
840 struct slave *old_active)
842 struct netdev_hw_addr *ha;
844 if (!USES_PRIMARY(bond->params.mode))
845 /* nothing to do - mc list is already up-to-date on
851 if (bond->dev->flags & IFF_PROMISC)
852 dev_set_promiscuity(old_active->dev, -1);
854 if (bond->dev->flags & IFF_ALLMULTI)
855 dev_set_allmulti(old_active->dev, -1);
857 netdev_for_each_mc_addr(ha, bond->dev)
858 dev_mc_del(old_active->dev, ha->addr);
862 /* FIXME: Signal errors upstream. */
863 if (bond->dev->flags & IFF_PROMISC)
864 dev_set_promiscuity(new_active->dev, 1);
866 if (bond->dev->flags & IFF_ALLMULTI)
867 dev_set_allmulti(new_active->dev, 1);
869 netdev_for_each_mc_addr(ha, bond->dev)
870 dev_mc_add(new_active->dev, ha->addr);
875 * bond_do_fail_over_mac
877 * Perform special MAC address swapping for fail_over_mac settings
879 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
881 static void bond_do_fail_over_mac(struct bonding *bond,
882 struct slave *new_active,
883 struct slave *old_active)
884 __releases(&bond->curr_slave_lock)
885 __releases(&bond->lock)
886 __acquires(&bond->lock)
887 __acquires(&bond->curr_slave_lock)
889 u8 tmp_mac[ETH_ALEN];
890 struct sockaddr saddr;
893 switch (bond->params.fail_over_mac) {
894 case BOND_FOM_ACTIVE:
896 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
897 new_active->dev->addr_len);
898 write_unlock_bh(&bond->curr_slave_lock);
899 read_unlock(&bond->lock);
900 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
901 read_lock(&bond->lock);
902 write_lock_bh(&bond->curr_slave_lock);
905 case BOND_FOM_FOLLOW:
907 * if new_active && old_active, swap them
908 * if just old_active, do nothing (going to no active slave)
909 * if just new_active, set new_active to bond's MAC
914 write_unlock_bh(&bond->curr_slave_lock);
915 read_unlock(&bond->lock);
918 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
919 memcpy(saddr.sa_data, old_active->dev->dev_addr,
921 saddr.sa_family = new_active->dev->type;
923 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
924 saddr.sa_family = bond->dev->type;
927 rv = dev_set_mac_address(new_active->dev, &saddr);
929 pr_err("%s: Error %d setting MAC of slave %s\n",
930 bond->dev->name, -rv, new_active->dev->name);
937 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
938 saddr.sa_family = old_active->dev->type;
940 rv = dev_set_mac_address(old_active->dev, &saddr);
942 pr_err("%s: Error %d setting MAC of slave %s\n",
943 bond->dev->name, -rv, new_active->dev->name);
945 read_lock(&bond->lock);
946 write_lock_bh(&bond->curr_slave_lock);
949 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
950 bond->dev->name, bond->params.fail_over_mac);
956 static bool bond_should_change_active(struct bonding *bond)
958 struct slave *prim = bond->primary_slave;
959 struct slave *curr = bond->curr_active_slave;
961 if (!prim || !curr || curr->link != BOND_LINK_UP)
963 if (bond->force_primary) {
964 bond->force_primary = false;
967 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
968 (prim->speed < curr->speed ||
969 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
971 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
977 * find_best_interface - select the best available slave to be the active one
978 * @bond: our bonding struct
980 * Warning: Caller must hold curr_slave_lock for writing.
982 static struct slave *bond_find_best_slave(struct bonding *bond)
984 struct slave *new_active, *old_active;
985 struct slave *bestslave = NULL;
986 int mintime = bond->params.updelay;
989 new_active = bond->curr_active_slave;
991 if (!new_active) { /* there were no active slaves left */
992 if (bond->slave_cnt > 0) /* found one slave */
993 new_active = bond->first_slave;
995 return NULL; /* still no slave, return NULL */
998 if ((bond->primary_slave) &&
999 bond->primary_slave->link == BOND_LINK_UP &&
1000 bond_should_change_active(bond)) {
1001 new_active = bond->primary_slave;
1004 /* remember where to stop iterating over the slaves */
1005 old_active = new_active;
1007 bond_for_each_slave_from(bond, new_active, i, old_active) {
1008 if (new_active->link == BOND_LINK_UP) {
1010 } else if (new_active->link == BOND_LINK_BACK &&
1011 IS_UP(new_active->dev)) {
1012 /* link up, but waiting for stabilization */
1013 if (new_active->delay < mintime) {
1014 mintime = new_active->delay;
1015 bestslave = new_active;
1023 static bool bond_should_notify_peers(struct bonding *bond)
1025 struct slave *slave = bond->curr_active_slave;
1027 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028 bond->dev->name, slave ? slave->dev->name : "NULL");
1030 if (!slave || !bond->send_peer_notif ||
1031 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1034 bond->send_peer_notif--;
1039 * change_active_interface - change the active slave into the specified one
1040 * @bond: our bonding struct
1041 * @new: the new slave to make the active one
1043 * Set the new slave to the bond's settings and unset them on the old
1044 * curr_active_slave.
1045 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1047 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048 * because it is apparently the best available slave we have, even though its
1049 * updelay hasn't timed out yet.
1051 * If new_active is not NULL, caller must hold bond->lock for read and
1052 * curr_slave_lock for write_bh.
1054 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1056 struct slave *old_active = bond->curr_active_slave;
1058 if (old_active == new_active)
1062 new_active->jiffies = jiffies;
1064 if (new_active->link == BOND_LINK_BACK) {
1065 if (USES_PRIMARY(bond->params.mode)) {
1066 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067 bond->dev->name, new_active->dev->name,
1068 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1071 new_active->delay = 0;
1072 new_active->link = BOND_LINK_UP;
1074 if (bond->params.mode == BOND_MODE_8023AD)
1075 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1077 if (bond_is_lb(bond))
1078 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1080 if (USES_PRIMARY(bond->params.mode)) {
1081 pr_info("%s: making interface %s the new active one.\n",
1082 bond->dev->name, new_active->dev->name);
1087 if (USES_PRIMARY(bond->params.mode))
1088 bond_mc_swap(bond, new_active, old_active);
1090 if (bond_is_lb(bond)) {
1091 bond_alb_handle_active_change(bond, new_active);
1093 bond_set_slave_inactive_flags(old_active);
1095 bond_set_slave_active_flags(new_active);
1097 bond->curr_active_slave = new_active;
1100 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1102 bond_set_slave_inactive_flags(old_active);
1105 bool should_notify_peers = false;
1107 bond_set_slave_active_flags(new_active);
1109 if (bond->params.fail_over_mac)
1110 bond_do_fail_over_mac(bond, new_active,
1113 if (netif_running(bond->dev)) {
1114 bond->send_peer_notif =
1115 bond->params.num_peer_notif;
1116 should_notify_peers =
1117 bond_should_notify_peers(bond);
1120 write_unlock_bh(&bond->curr_slave_lock);
1121 read_unlock(&bond->lock);
1123 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1124 if (should_notify_peers)
1125 netdev_bonding_change(bond->dev,
1126 NETDEV_NOTIFY_PEERS);
1128 read_lock(&bond->lock);
1129 write_lock_bh(&bond->curr_slave_lock);
1133 /* resend IGMP joins since active slave has changed or
1134 * all were sent on curr_active_slave.
1135 * resend only if bond is brought up with the affected
1136 * bonding modes and the retransmission is enabled */
1137 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1138 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1139 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1140 bond->igmp_retrans = bond->params.resend_igmp;
1141 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1146 * bond_select_active_slave - select a new active slave, if needed
1147 * @bond: our bonding struct
1149 * This functions should be called when one of the following occurs:
1150 * - The old curr_active_slave has been released or lost its link.
1151 * - The primary_slave has got its link back.
1152 * - A slave has got its link back and there's no old curr_active_slave.
1154 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1156 void bond_select_active_slave(struct bonding *bond)
1158 struct slave *best_slave;
1161 best_slave = bond_find_best_slave(bond);
1162 if (best_slave != bond->curr_active_slave) {
1163 bond_change_active_slave(bond, best_slave);
1164 rv = bond_set_carrier(bond);
1168 if (netif_carrier_ok(bond->dev)) {
1169 pr_info("%s: first active interface up!\n",
1172 pr_info("%s: now running without any active interface !\n",
1178 /*--------------------------- slave list handling ---------------------------*/
1181 * This function attaches the slave to the end of list.
1183 * bond->lock held for writing by caller.
1185 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1187 if (bond->first_slave == NULL) { /* attaching the first slave */
1188 new_slave->next = new_slave;
1189 new_slave->prev = new_slave;
1190 bond->first_slave = new_slave;
1192 new_slave->next = bond->first_slave;
1193 new_slave->prev = bond->first_slave->prev;
1194 new_slave->next->prev = new_slave;
1195 new_slave->prev->next = new_slave;
1202 * This function detaches the slave from the list.
1203 * WARNING: no check is made to verify if the slave effectively
1204 * belongs to <bond>.
1205 * Nothing is freed on return, structures are just unchained.
1206 * If any slave pointer in bond was pointing to <slave>,
1207 * it should be changed by the calling function.
1209 * bond->lock held for writing by caller.
1211 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1214 slave->next->prev = slave->prev;
1217 slave->prev->next = slave->next;
1219 if (bond->first_slave == slave) { /* slave is the first slave */
1220 if (bond->slave_cnt > 1) { /* there are more slave */
1221 bond->first_slave = slave->next;
1223 bond->first_slave = NULL; /* slave was the last one */
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233 static inline int slave_enable_netpoll(struct slave *slave)
1238 np = kzalloc(sizeof(*np), GFP_KERNEL);
1243 np->dev = slave->dev;
1244 strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1245 err = __netpoll_setup(np);
1254 static inline void slave_disable_netpoll(struct slave *slave)
1256 struct netpoll *np = slave->np;
1262 synchronize_rcu_bh();
1263 __netpoll_cleanup(np);
1266 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1268 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1270 if (!slave_dev->netdev_ops->ndo_poll_controller)
1275 static void bond_poll_controller(struct net_device *bond_dev)
1279 static void __bond_netpoll_cleanup(struct bonding *bond)
1281 struct slave *slave;
1284 bond_for_each_slave(bond, slave, i)
1285 if (IS_UP(slave->dev))
1286 slave_disable_netpoll(slave);
1288 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1290 struct bonding *bond = netdev_priv(bond_dev);
1292 read_lock(&bond->lock);
1293 __bond_netpoll_cleanup(bond);
1294 read_unlock(&bond->lock);
1297 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1299 struct bonding *bond = netdev_priv(dev);
1300 struct slave *slave;
1303 read_lock(&bond->lock);
1304 bond_for_each_slave(bond, slave, i) {
1305 err = slave_enable_netpoll(slave);
1307 __bond_netpoll_cleanup(bond);
1311 read_unlock(&bond->lock);
1315 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1317 return bond->dev->npinfo;
1321 static inline int slave_enable_netpoll(struct slave *slave)
1325 static inline void slave_disable_netpoll(struct slave *slave)
1328 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1333 /*---------------------------------- IOCTL ----------------------------------*/
1335 static int bond_sethwaddr(struct net_device *bond_dev,
1336 struct net_device *slave_dev)
1338 pr_debug("bond_dev=%p\n", bond_dev);
1339 pr_debug("slave_dev=%p\n", slave_dev);
1340 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1341 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1345 static netdev_features_t bond_fix_features(struct net_device *dev,
1346 netdev_features_t features)
1348 struct slave *slave;
1349 struct bonding *bond = netdev_priv(dev);
1350 netdev_features_t mask;
1353 read_lock(&bond->lock);
1355 if (!bond->first_slave) {
1356 /* Disable adding VLANs to empty bond. But why? --mq */
1357 features |= NETIF_F_VLAN_CHALLENGED;
1362 features &= ~NETIF_F_ONE_FOR_ALL;
1363 features |= NETIF_F_ALL_FOR_ALL;
1365 bond_for_each_slave(bond, slave, i) {
1366 features = netdev_increment_features(features,
1367 slave->dev->features,
1372 read_unlock(&bond->lock);
1376 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1377 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1378 NETIF_F_HIGHDMA | NETIF_F_LRO)
1380 static void bond_compute_features(struct bonding *bond)
1382 struct slave *slave;
1383 struct net_device *bond_dev = bond->dev;
1384 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1385 unsigned short max_hard_header_len = ETH_HLEN;
1388 read_lock(&bond->lock);
1390 if (!bond->first_slave)
1393 bond_for_each_slave(bond, slave, i) {
1394 vlan_features = netdev_increment_features(vlan_features,
1395 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1397 if (slave->dev->hard_header_len > max_hard_header_len)
1398 max_hard_header_len = slave->dev->hard_header_len;
1402 bond_dev->vlan_features = vlan_features;
1403 bond_dev->hard_header_len = max_hard_header_len;
1405 read_unlock(&bond->lock);
1407 netdev_change_features(bond_dev);
1410 static void bond_setup_by_slave(struct net_device *bond_dev,
1411 struct net_device *slave_dev)
1413 struct bonding *bond = netdev_priv(bond_dev);
1415 bond_dev->header_ops = slave_dev->header_ops;
1417 bond_dev->type = slave_dev->type;
1418 bond_dev->hard_header_len = slave_dev->hard_header_len;
1419 bond_dev->addr_len = slave_dev->addr_len;
1421 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422 slave_dev->addr_len);
1423 bond->setup_by_slave = 1;
1426 /* On bonding slaves other than the currently active slave, suppress
1427 * duplicates except for alb non-mcast/bcast.
1429 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1430 struct slave *slave,
1431 struct bonding *bond)
1433 if (bond_is_slave_inactive(slave)) {
1434 if (bond->params.mode == BOND_MODE_ALB &&
1435 skb->pkt_type != PACKET_BROADCAST &&
1436 skb->pkt_type != PACKET_MULTICAST)
1443 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1445 struct sk_buff *skb = *pskb;
1446 struct slave *slave;
1447 struct bonding *bond;
1448 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1450 int ret = RX_HANDLER_ANOTHER;
1452 skb = skb_share_check(skb, GFP_ATOMIC);
1454 return RX_HANDLER_CONSUMED;
1458 slave = bond_slave_get_rcu(skb->dev);
1461 if (bond->params.arp_interval)
1462 slave->dev->last_rx = jiffies;
1464 recv_probe = ACCESS_ONCE(bond->recv_probe);
1466 ret = recv_probe(skb, bond, slave);
1467 if (ret == RX_HANDLER_CONSUMED) {
1473 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1474 return RX_HANDLER_EXACT;
1477 skb->dev = bond->dev;
1479 if (bond->params.mode == BOND_MODE_ALB &&
1480 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1481 skb->pkt_type == PACKET_HOST) {
1483 if (unlikely(skb_cow_head(skb,
1484 skb->data - skb_mac_header(skb)))) {
1486 return RX_HANDLER_CONSUMED;
1488 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494 /* enslave device <slave> to bond device <master> */
1495 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1497 struct bonding *bond = netdev_priv(bond_dev);
1498 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1499 struct slave *new_slave = NULL;
1500 struct netdev_hw_addr *ha;
1501 struct sockaddr addr;
1505 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1506 slave_ops->ndo_do_ioctl == NULL) {
1507 pr_warning("%s: Warning: no link monitoring support for %s\n",
1508 bond_dev->name, slave_dev->name);
1511 /* already enslaved */
1512 if (slave_dev->flags & IFF_SLAVE) {
1513 pr_debug("Error, Device was already enslaved\n");
1517 /* vlan challenged mutual exclusion */
1518 /* no need to lock since we're protected by rtnl_lock */
1519 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1520 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1521 if (bond_vlan_used(bond)) {
1522 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1523 bond_dev->name, slave_dev->name, bond_dev->name);
1526 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1527 bond_dev->name, slave_dev->name,
1528 slave_dev->name, bond_dev->name);
1531 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1535 * Old ifenslave binaries are no longer supported. These can
1536 * be identified with moderate accuracy by the state of the slave:
1537 * the current ifenslave will set the interface down prior to
1538 * enslaving it; the old ifenslave will not.
1540 if ((slave_dev->flags & IFF_UP)) {
1541 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1544 goto err_undo_flags;
1547 /* set bonding device ether type by slave - bonding netdevices are
1548 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1549 * there is a need to override some of the type dependent attribs/funcs.
1551 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1552 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1554 if (bond->slave_cnt == 0) {
1555 if (bond_dev->type != slave_dev->type) {
1556 pr_debug("%s: change device type from %d to %d\n",
1558 bond_dev->type, slave_dev->type);
1560 res = netdev_bonding_change(bond_dev,
1561 NETDEV_PRE_TYPE_CHANGE);
1562 res = notifier_to_errno(res);
1564 pr_err("%s: refused to change device type\n",
1567 goto err_undo_flags;
1570 /* Flush unicast and multicast addresses */
1571 dev_uc_flush(bond_dev);
1572 dev_mc_flush(bond_dev);
1574 if (slave_dev->type != ARPHRD_ETHER)
1575 bond_setup_by_slave(bond_dev, slave_dev);
1577 ether_setup(bond_dev);
1578 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1581 netdev_bonding_change(bond_dev,
1582 NETDEV_POST_TYPE_CHANGE);
1584 } else if (bond_dev->type != slave_dev->type) {
1585 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1587 slave_dev->type, bond_dev->type);
1589 goto err_undo_flags;
1592 if (slave_ops->ndo_set_mac_address == NULL) {
1593 if (bond->slave_cnt == 0) {
1594 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1596 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1597 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1598 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",
1601 goto err_undo_flags;
1605 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1607 /* If this is the first slave, then we need to set the master's hardware
1608 * address to be the same as the slave's. */
1609 if (is_zero_ether_addr(bond->dev->dev_addr))
1610 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1611 slave_dev->addr_len);
1614 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1617 goto err_undo_flags;
1621 * Set the new_slave's queue_id to be zero. Queue ID mapping
1622 * is set via sysfs or module option if desired.
1624 new_slave->queue_id = 0;
1626 /* Save slave's original mtu and then set it to match the bond */
1627 new_slave->original_mtu = slave_dev->mtu;
1628 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1630 pr_debug("Error %d calling dev_set_mtu\n", res);
1635 * Save slave's original ("permanent") mac address for modes
1636 * that need it, and for restoring it upon release, and then
1637 * set it to the master's address
1639 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1641 if (!bond->params.fail_over_mac) {
1643 * Set slave to master's mac address. The application already
1644 * set the master's mac address to that of the first slave
1646 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1647 addr.sa_family = slave_dev->type;
1648 res = dev_set_mac_address(slave_dev, &addr);
1650 pr_debug("Error %d calling set_mac_address\n", res);
1651 goto err_restore_mtu;
1655 res = netdev_set_bond_master(slave_dev, bond_dev);
1657 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1658 goto err_restore_mac;
1661 /* open the slave since the application closed it */
1662 res = dev_open(slave_dev);
1664 pr_debug("Opening slave %s failed\n", slave_dev->name);
1665 goto err_unset_master;
1668 new_slave->bond = bond;
1669 new_slave->dev = slave_dev;
1670 slave_dev->priv_flags |= IFF_BONDING;
1672 if (bond_is_lb(bond)) {
1673 /* bond_alb_init_slave() must be called before all other stages since
1674 * it might fail and we do not want to have to undo everything
1676 res = bond_alb_init_slave(bond, new_slave);
1681 /* If the mode USES_PRIMARY, then the new slave gets the
1682 * master's promisc (and mc) settings only if it becomes the
1683 * curr_active_slave, and that is taken care of later when calling
1684 * bond_change_active()
1686 if (!USES_PRIMARY(bond->params.mode)) {
1687 /* set promiscuity level to new slave */
1688 if (bond_dev->flags & IFF_PROMISC) {
1689 res = dev_set_promiscuity(slave_dev, 1);
1694 /* set allmulti level to new slave */
1695 if (bond_dev->flags & IFF_ALLMULTI) {
1696 res = dev_set_allmulti(slave_dev, 1);
1701 netif_addr_lock_bh(bond_dev);
1702 /* upload master's mc_list to new slave */
1703 netdev_for_each_mc_addr(ha, bond_dev)
1704 dev_mc_add(slave_dev, ha->addr);
1705 netif_addr_unlock_bh(bond_dev);
1708 if (bond->params.mode == BOND_MODE_8023AD) {
1709 /* add lacpdu mc addr to mc list */
1710 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1712 dev_mc_add(slave_dev, lacpdu_multicast);
1715 bond_add_vlans_on_slave(bond, slave_dev);
1717 write_lock_bh(&bond->lock);
1719 bond_attach_slave(bond, new_slave);
1721 new_slave->delay = 0;
1722 new_slave->link_failure_count = 0;
1724 write_unlock_bh(&bond->lock);
1726 bond_compute_features(bond);
1728 read_lock(&bond->lock);
1730 new_slave->last_arp_rx = jiffies -
1731 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1733 if (bond->params.miimon && !bond->params.use_carrier) {
1734 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1736 if ((link_reporting == -1) && !bond->params.arp_interval) {
1738 * miimon is set but a bonded network driver
1739 * does not support ETHTOOL/MII and
1740 * arp_interval is not set. Note: if
1741 * use_carrier is enabled, we will never go
1742 * here (because netif_carrier is always
1743 * supported); thus, we don't need to change
1744 * the messages for netif_carrier.
1746 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",
1747 bond_dev->name, slave_dev->name);
1748 } else if (link_reporting == -1) {
1749 /* unable get link status using mii/ethtool */
1750 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",
1751 bond_dev->name, slave_dev->name);
1755 /* check for initial state */
1756 if (bond->params.miimon) {
1757 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1758 if (bond->params.updelay) {
1759 new_slave->link = BOND_LINK_BACK;
1760 new_slave->delay = bond->params.updelay;
1762 new_slave->link = BOND_LINK_UP;
1765 new_slave->link = BOND_LINK_DOWN;
1767 } else if (bond->params.arp_interval) {
1768 new_slave->link = (netif_carrier_ok(slave_dev) ?
1769 BOND_LINK_UP : BOND_LINK_DOWN);
1771 new_slave->link = BOND_LINK_UP;
1774 if (new_slave->link != BOND_LINK_DOWN)
1775 new_slave->jiffies = jiffies;
1776 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1777 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1778 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1780 bond_update_speed_duplex(new_slave);
1782 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1783 /* if there is a primary slave, remember it */
1784 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1785 bond->primary_slave = new_slave;
1786 bond->force_primary = true;
1790 write_lock_bh(&bond->curr_slave_lock);
1792 switch (bond->params.mode) {
1793 case BOND_MODE_ACTIVEBACKUP:
1794 bond_set_slave_inactive_flags(new_slave);
1795 bond_select_active_slave(bond);
1797 case BOND_MODE_8023AD:
1798 /* in 802.3ad mode, the internal mechanism
1799 * will activate the slaves in the selected
1802 bond_set_slave_inactive_flags(new_slave);
1803 /* if this is the first slave */
1804 if (bond->slave_cnt == 1) {
1805 SLAVE_AD_INFO(new_slave).id = 1;
1806 /* Initialize AD with the number of times that the AD timer is called in 1 second
1807 * can be called only after the mac address of the bond is set
1809 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1811 SLAVE_AD_INFO(new_slave).id =
1812 SLAVE_AD_INFO(new_slave->prev).id + 1;
1815 bond_3ad_bind_slave(new_slave);
1819 bond_set_active_slave(new_slave);
1820 bond_set_slave_inactive_flags(new_slave);
1821 bond_select_active_slave(bond);
1824 pr_debug("This slave is always active in trunk mode\n");
1826 /* always active in trunk mode */
1827 bond_set_active_slave(new_slave);
1829 /* In trunking mode there is little meaning to curr_active_slave
1830 * anyway (it holds no special properties of the bond device),
1831 * so we can change it without calling change_active_interface()
1833 if (!bond->curr_active_slave)
1834 bond->curr_active_slave = new_slave;
1837 } /* switch(bond_mode) */
1839 write_unlock_bh(&bond->curr_slave_lock);
1841 bond_set_carrier(bond);
1843 #ifdef CONFIG_NET_POLL_CONTROLLER
1844 slave_dev->npinfo = bond_netpoll_info(bond);
1845 if (slave_dev->npinfo) {
1846 if (slave_enable_netpoll(new_slave)) {
1847 read_unlock(&bond->lock);
1848 pr_info("Error, %s: master_dev is using netpoll, "
1849 "but new slave device does not support netpoll.\n",
1857 read_unlock(&bond->lock);
1859 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1863 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1866 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1867 goto err_dest_symlinks;
1870 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1871 bond_dev->name, slave_dev->name,
1872 bond_is_active_slave(new_slave) ? "n active" : " backup",
1873 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1875 /* enslave is successful */
1878 /* Undo stages on error */
1880 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1883 write_lock_bh(&bond->lock);
1884 bond_detach_slave(bond, new_slave);
1885 write_unlock_bh(&bond->lock);
1888 dev_close(slave_dev);
1891 netdev_set_bond_master(slave_dev, NULL);
1894 if (!bond->params.fail_over_mac) {
1895 /* XXX TODO - fom follow mode needs to change master's
1896 * MAC if this slave's MAC is in use by the bond, or at
1897 * least print a warning.
1899 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1900 addr.sa_family = slave_dev->type;
1901 dev_set_mac_address(slave_dev, &addr);
1905 dev_set_mtu(slave_dev, new_slave->original_mtu);
1911 bond_compute_features(bond);
1917 * Try to release the slave device <slave> from the bond device <master>
1918 * It is legal to access curr_active_slave without a lock because all the function
1921 * The rules for slave state should be:
1922 * for Active/Backup:
1923 * Active stays on all backups go down
1924 * for Bonded connections:
1925 * The first up interface should be left on and all others downed.
1927 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1929 struct bonding *bond = netdev_priv(bond_dev);
1930 struct slave *slave, *oldcurrent;
1931 struct sockaddr addr;
1932 netdev_features_t old_features = bond_dev->features;
1934 /* slave is not a slave or master is not master of this slave */
1935 if (!(slave_dev->flags & IFF_SLAVE) ||
1936 (slave_dev->master != bond_dev)) {
1937 pr_err("%s: Error: cannot release %s.\n",
1938 bond_dev->name, slave_dev->name);
1943 netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1944 write_lock_bh(&bond->lock);
1946 slave = bond_get_slave_by_dev(bond, slave_dev);
1948 /* not a slave of this bond */
1949 pr_info("%s: %s not enslaved\n",
1950 bond_dev->name, slave_dev->name);
1951 write_unlock_bh(&bond->lock);
1952 unblock_netpoll_tx();
1956 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1957 * for this slave anymore.
1959 netdev_rx_handler_unregister(slave_dev);
1960 write_unlock_bh(&bond->lock);
1962 write_lock_bh(&bond->lock);
1964 if (!bond->params.fail_over_mac) {
1965 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1966 bond->slave_cnt > 1)
1967 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",
1968 bond_dev->name, slave_dev->name,
1970 bond_dev->name, slave_dev->name);
1973 /* Inform AD package of unbinding of slave. */
1974 if (bond->params.mode == BOND_MODE_8023AD) {
1975 /* must be called before the slave is
1976 * detached from the list
1978 bond_3ad_unbind_slave(slave);
1981 pr_info("%s: releasing %s interface %s\n",
1983 bond_is_active_slave(slave) ? "active" : "backup",
1986 oldcurrent = bond->curr_active_slave;
1988 bond->current_arp_slave = NULL;
1990 /* release the slave from its bond */
1991 bond_detach_slave(bond, slave);
1993 if (bond->primary_slave == slave)
1994 bond->primary_slave = NULL;
1996 if (oldcurrent == slave)
1997 bond_change_active_slave(bond, NULL);
1999 if (bond_is_lb(bond)) {
2000 /* Must be called only after the slave has been
2001 * detached from the list and the curr_active_slave
2002 * has been cleared (if our_slave == old_current),
2003 * but before a new active slave is selected.
2005 write_unlock_bh(&bond->lock);
2006 bond_alb_deinit_slave(bond, slave);
2007 write_lock_bh(&bond->lock);
2010 if (oldcurrent == slave) {
2012 * Note that we hold RTNL over this sequence, so there
2013 * is no concern that another slave add/remove event
2016 write_unlock_bh(&bond->lock);
2017 read_lock(&bond->lock);
2018 write_lock_bh(&bond->curr_slave_lock);
2020 bond_select_active_slave(bond);
2022 write_unlock_bh(&bond->curr_slave_lock);
2023 read_unlock(&bond->lock);
2024 write_lock_bh(&bond->lock);
2027 if (bond->slave_cnt == 0) {
2028 bond_set_carrier(bond);
2030 /* if the last slave was removed, zero the mac address
2031 * of the master so it will be set by the application
2032 * to the mac address of the first slave
2034 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2036 if (bond_vlan_used(bond)) {
2037 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2038 bond_dev->name, bond_dev->name);
2039 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2044 write_unlock_bh(&bond->lock);
2045 unblock_netpoll_tx();
2047 if (bond->slave_cnt == 0)
2048 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2050 bond_compute_features(bond);
2051 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2052 (old_features & NETIF_F_VLAN_CHALLENGED))
2053 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2054 bond_dev->name, slave_dev->name, bond_dev->name);
2056 /* must do this from outside any spinlocks */
2057 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2059 bond_del_vlans_from_slave(bond, slave_dev);
2061 /* If the mode USES_PRIMARY, then we should only remove its
2062 * promisc and mc settings if it was the curr_active_slave, but that was
2063 * already taken care of above when we detached the slave
2065 if (!USES_PRIMARY(bond->params.mode)) {
2066 /* unset promiscuity level from slave */
2067 if (bond_dev->flags & IFF_PROMISC)
2068 dev_set_promiscuity(slave_dev, -1);
2070 /* unset allmulti level from slave */
2071 if (bond_dev->flags & IFF_ALLMULTI)
2072 dev_set_allmulti(slave_dev, -1);
2074 /* flush master's mc_list from slave */
2075 netif_addr_lock_bh(bond_dev);
2076 bond_mc_list_flush(bond_dev, slave_dev);
2077 netif_addr_unlock_bh(bond_dev);
2080 netdev_set_bond_master(slave_dev, NULL);
2082 slave_disable_netpoll(slave);
2084 /* close slave before restoring its mac address */
2085 dev_close(slave_dev);
2087 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2088 /* restore original ("permanent") mac address */
2089 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2090 addr.sa_family = slave_dev->type;
2091 dev_set_mac_address(slave_dev, &addr);
2094 dev_set_mtu(slave_dev, slave->original_mtu);
2096 slave_dev->priv_flags &= ~IFF_BONDING;
2100 return 0; /* deletion OK */
2104 * First release a slave and then destroy the bond if no more slaves are left.
2105 * Must be under rtnl_lock when this function is called.
2107 static int bond_release_and_destroy(struct net_device *bond_dev,
2108 struct net_device *slave_dev)
2110 struct bonding *bond = netdev_priv(bond_dev);
2113 ret = bond_release(bond_dev, slave_dev);
2114 if ((ret == 0) && (bond->slave_cnt == 0)) {
2115 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2116 pr_info("%s: destroying bond %s.\n",
2117 bond_dev->name, bond_dev->name);
2118 unregister_netdevice(bond_dev);
2124 * This function releases all slaves.
2126 static int bond_release_all(struct net_device *bond_dev)
2128 struct bonding *bond = netdev_priv(bond_dev);
2129 struct slave *slave;
2130 struct net_device *slave_dev;
2131 struct sockaddr addr;
2133 write_lock_bh(&bond->lock);
2135 netif_carrier_off(bond_dev);
2137 if (bond->slave_cnt == 0)
2140 bond->current_arp_slave = NULL;
2141 bond->primary_slave = NULL;
2142 bond_change_active_slave(bond, NULL);
2144 while ((slave = bond->first_slave) != NULL) {
2145 /* Inform AD package of unbinding of slave
2146 * before slave is detached from the list.
2148 if (bond->params.mode == BOND_MODE_8023AD)
2149 bond_3ad_unbind_slave(slave);
2151 slave_dev = slave->dev;
2152 bond_detach_slave(bond, slave);
2154 /* now that the slave is detached, unlock and perform
2155 * all the undo steps that should not be called from
2158 write_unlock_bh(&bond->lock);
2160 /* unregister rx_handler early so bond_handle_frame wouldn't
2161 * be called for this slave anymore.
2163 netdev_rx_handler_unregister(slave_dev);
2166 if (bond_is_lb(bond)) {
2167 /* must be called only after the slave
2168 * has been detached from the list
2170 bond_alb_deinit_slave(bond, slave);
2173 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2174 bond_del_vlans_from_slave(bond, slave_dev);
2176 /* If the mode USES_PRIMARY, then we should only remove its
2177 * promisc and mc settings if it was the curr_active_slave, but that was
2178 * already taken care of above when we detached the slave
2180 if (!USES_PRIMARY(bond->params.mode)) {
2181 /* unset promiscuity level from slave */
2182 if (bond_dev->flags & IFF_PROMISC)
2183 dev_set_promiscuity(slave_dev, -1);
2185 /* unset allmulti level from slave */
2186 if (bond_dev->flags & IFF_ALLMULTI)
2187 dev_set_allmulti(slave_dev, -1);
2189 /* flush master's mc_list from slave */
2190 netif_addr_lock_bh(bond_dev);
2191 bond_mc_list_flush(bond_dev, slave_dev);
2192 netif_addr_unlock_bh(bond_dev);
2195 netdev_set_bond_master(slave_dev, NULL);
2197 slave_disable_netpoll(slave);
2199 /* close slave before restoring its mac address */
2200 dev_close(slave_dev);
2202 if (!bond->params.fail_over_mac) {
2203 /* restore original ("permanent") mac address*/
2204 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2205 addr.sa_family = slave_dev->type;
2206 dev_set_mac_address(slave_dev, &addr);
2211 /* re-acquire the lock before getting the next slave */
2212 write_lock_bh(&bond->lock);
2215 /* zero the mac address of the master so it will be
2216 * set by the application to the mac address of the
2219 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2221 if (bond_vlan_used(bond)) {
2222 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2223 bond_dev->name, bond_dev->name);
2224 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2228 pr_info("%s: released all slaves\n", bond_dev->name);
2231 write_unlock_bh(&bond->lock);
2233 bond_compute_features(bond);
2239 * This function changes the active slave to slave <slave_dev>.
2240 * It returns -EINVAL in the following cases.
2241 * - <slave_dev> is not found in the list.
2242 * - There is not active slave now.
2243 * - <slave_dev> is already active.
2244 * - The link state of <slave_dev> is not BOND_LINK_UP.
2245 * - <slave_dev> is not running.
2246 * In these cases, this function does nothing.
2247 * In the other cases, current_slave pointer is changed and 0 is returned.
2249 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2251 struct bonding *bond = netdev_priv(bond_dev);
2252 struct slave *old_active = NULL;
2253 struct slave *new_active = NULL;
2256 if (!USES_PRIMARY(bond->params.mode))
2259 /* Verify that master_dev is indeed the master of slave_dev */
2260 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2263 read_lock(&bond->lock);
2265 read_lock(&bond->curr_slave_lock);
2266 old_active = bond->curr_active_slave;
2267 read_unlock(&bond->curr_slave_lock);
2269 new_active = bond_get_slave_by_dev(bond, slave_dev);
2272 * Changing to the current active: do nothing; return success.
2274 if (new_active && (new_active == old_active)) {
2275 read_unlock(&bond->lock);
2281 (new_active->link == BOND_LINK_UP) &&
2282 IS_UP(new_active->dev)) {
2284 write_lock_bh(&bond->curr_slave_lock);
2285 bond_change_active_slave(bond, new_active);
2286 write_unlock_bh(&bond->curr_slave_lock);
2287 unblock_netpoll_tx();
2291 read_unlock(&bond->lock);
2296 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2298 struct bonding *bond = netdev_priv(bond_dev);
2300 info->bond_mode = bond->params.mode;
2301 info->miimon = bond->params.miimon;
2303 read_lock(&bond->lock);
2304 info->num_slaves = bond->slave_cnt;
2305 read_unlock(&bond->lock);
2310 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2312 struct bonding *bond = netdev_priv(bond_dev);
2313 struct slave *slave;
2314 int i, res = -ENODEV;
2316 read_lock(&bond->lock);
2318 bond_for_each_slave(bond, slave, i) {
2319 if (i == (int)info->slave_id) {
2321 strcpy(info->slave_name, slave->dev->name);
2322 info->link = slave->link;
2323 info->state = bond_slave_state(slave);
2324 info->link_failure_count = slave->link_failure_count;
2329 read_unlock(&bond->lock);
2334 /*-------------------------------- Monitoring -------------------------------*/
2337 static int bond_miimon_inspect(struct bonding *bond)
2339 struct slave *slave;
2340 int i, link_state, commit = 0;
2341 bool ignore_updelay;
2343 ignore_updelay = !bond->curr_active_slave ? true : false;
2345 bond_for_each_slave(bond, slave, i) {
2346 slave->new_link = BOND_LINK_NOCHANGE;
2348 link_state = bond_check_dev_link(bond, slave->dev, 0);
2350 switch (slave->link) {
2355 slave->link = BOND_LINK_FAIL;
2356 slave->delay = bond->params.downdelay;
2358 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2360 (bond->params.mode ==
2361 BOND_MODE_ACTIVEBACKUP) ?
2362 (bond_is_active_slave(slave) ?
2363 "active " : "backup ") : "",
2365 bond->params.downdelay * bond->params.miimon);
2368 case BOND_LINK_FAIL:
2371 * recovered before downdelay expired
2373 slave->link = BOND_LINK_UP;
2374 slave->jiffies = jiffies;
2375 pr_info("%s: link status up again after %d ms for interface %s.\n",
2377 (bond->params.downdelay - slave->delay) *
2378 bond->params.miimon,
2383 if (slave->delay <= 0) {
2384 slave->new_link = BOND_LINK_DOWN;
2392 case BOND_LINK_DOWN:
2396 slave->link = BOND_LINK_BACK;
2397 slave->delay = bond->params.updelay;
2400 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2401 bond->dev->name, slave->dev->name,
2402 ignore_updelay ? 0 :
2403 bond->params.updelay *
2404 bond->params.miimon);
2407 case BOND_LINK_BACK:
2409 slave->link = BOND_LINK_DOWN;
2410 pr_info("%s: link status down again after %d ms for interface %s.\n",
2412 (bond->params.updelay - slave->delay) *
2413 bond->params.miimon,
2422 if (slave->delay <= 0) {
2423 slave->new_link = BOND_LINK_UP;
2425 ignore_updelay = false;
2437 static void bond_miimon_commit(struct bonding *bond)
2439 struct slave *slave;
2442 bond_for_each_slave(bond, slave, i) {
2443 switch (slave->new_link) {
2444 case BOND_LINK_NOCHANGE:
2448 slave->link = BOND_LINK_UP;
2449 slave->jiffies = jiffies;
2451 if (bond->params.mode == BOND_MODE_8023AD) {
2452 /* prevent it from being the active one */
2453 bond_set_backup_slave(slave);
2454 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2455 /* make it immediately active */
2456 bond_set_active_slave(slave);
2457 } else if (slave != bond->primary_slave) {
2458 /* prevent it from being the active one */
2459 bond_set_backup_slave(slave);
2462 bond_update_speed_duplex(slave);
2464 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2465 bond->dev->name, slave->dev->name,
2466 slave->speed, slave->duplex ? "full" : "half");
2468 /* notify ad that the link status has changed */
2469 if (bond->params.mode == BOND_MODE_8023AD)
2470 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2472 if (bond_is_lb(bond))
2473 bond_alb_handle_link_change(bond, slave,
2476 if (!bond->curr_active_slave ||
2477 (slave == bond->primary_slave))
2482 case BOND_LINK_DOWN:
2483 if (slave->link_failure_count < UINT_MAX)
2484 slave->link_failure_count++;
2486 slave->link = BOND_LINK_DOWN;
2488 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2489 bond->params.mode == BOND_MODE_8023AD)
2490 bond_set_slave_inactive_flags(slave);
2492 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2493 bond->dev->name, slave->dev->name);
2495 if (bond->params.mode == BOND_MODE_8023AD)
2496 bond_3ad_handle_link_change(slave,
2499 if (bond_is_lb(bond))
2500 bond_alb_handle_link_change(bond, slave,
2503 if (slave == bond->curr_active_slave)
2509 pr_err("%s: invalid new link %d on slave %s\n",
2510 bond->dev->name, slave->new_link,
2512 slave->new_link = BOND_LINK_NOCHANGE;
2520 write_lock_bh(&bond->curr_slave_lock);
2521 bond_select_active_slave(bond);
2522 write_unlock_bh(&bond->curr_slave_lock);
2523 unblock_netpoll_tx();
2526 bond_set_carrier(bond);
2532 * Really a wrapper that splits the mii monitor into two phases: an
2533 * inspection, then (if inspection indicates something needs to be done)
2534 * an acquisition of appropriate locks followed by a commit phase to
2535 * implement whatever link state changes are indicated.
2537 void bond_mii_monitor(struct work_struct *work)
2539 struct bonding *bond = container_of(work, struct bonding,
2541 bool should_notify_peers = false;
2542 unsigned long delay;
2544 read_lock(&bond->lock);
2546 delay = msecs_to_jiffies(bond->params.miimon);
2548 if (bond->slave_cnt == 0)
2551 should_notify_peers = bond_should_notify_peers(bond);
2553 if (bond_miimon_inspect(bond)) {
2554 read_unlock(&bond->lock);
2556 /* Race avoidance with bond_close cancel of workqueue */
2557 if (!rtnl_trylock()) {
2558 read_lock(&bond->lock);
2560 should_notify_peers = false;
2564 read_lock(&bond->lock);
2566 bond_miimon_commit(bond);
2568 read_unlock(&bond->lock);
2569 rtnl_unlock(); /* might sleep, hold no other locks */
2570 read_lock(&bond->lock);
2574 if (bond->params.miimon)
2575 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2577 read_unlock(&bond->lock);
2579 if (should_notify_peers) {
2580 if (!rtnl_trylock()) {
2581 read_lock(&bond->lock);
2582 bond->send_peer_notif++;
2583 read_unlock(&bond->lock);
2586 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2591 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2593 struct vlan_entry *vlan;
2594 struct net_device *vlan_dev;
2596 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2599 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2601 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2603 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2611 * We go to the (large) trouble of VLAN tagging ARP frames because
2612 * switches in VLAN mode (especially if ports are configured as
2613 * "native" to a VLAN) might not pass non-tagged frames.
2615 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2617 struct sk_buff *skb;
2619 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2620 slave_dev->name, dest_ip, src_ip, vlan_id);
2622 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2623 NULL, slave_dev->dev_addr, NULL);
2626 pr_err("ARP packet allocation failed\n");
2630 skb = vlan_put_tag(skb, vlan_id);
2632 pr_err("failed to insert VLAN tag\n");
2640 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2643 __be32 *targets = bond->params.arp_targets;
2644 struct vlan_entry *vlan;
2645 struct net_device *vlan_dev = NULL;
2648 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2652 pr_debug("basa: target %x\n", targets[i]);
2653 if (!bond_vlan_used(bond)) {
2654 pr_debug("basa: empty vlan: arp_send\n");
2655 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2656 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2662 * If VLANs are configured, we do a route lookup to
2663 * determine which VLAN interface would be used, so we
2664 * can tag the ARP with the proper VLAN tag.
2666 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2669 if (net_ratelimit()) {
2670 pr_warning("%s: no route to arp_ip_target %pI4\n",
2671 bond->dev->name, &targets[i]);
2677 * This target is not on a VLAN
2679 if (rt->dst.dev == bond->dev) {
2681 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2682 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2683 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2689 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2691 vlan_dev = __vlan_find_dev_deep(bond->dev,
2694 if (vlan_dev == rt->dst.dev) {
2695 vlan_id = vlan->vlan_id;
2696 pr_debug("basa: vlan match on %s %d\n",
2697 vlan_dev->name, vlan_id);
2702 if (vlan_id && vlan_dev) {
2704 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2705 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2710 if (net_ratelimit()) {
2711 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2712 bond->dev->name, &targets[i],
2713 rt->dst.dev ? rt->dst.dev->name : "NULL");
2719 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2722 __be32 *targets = bond->params.arp_targets;
2724 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2725 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2726 &sip, &tip, i, &targets[i],
2727 bond_has_this_ip(bond, tip));
2728 if (sip == targets[i]) {
2729 if (bond_has_this_ip(bond, tip))
2730 slave->last_arp_rx = jiffies;
2736 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2737 struct slave *slave)
2739 struct arphdr *arp = (struct arphdr *)skb->data;
2740 unsigned char *arp_ptr;
2744 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2745 return RX_HANDLER_ANOTHER;
2747 read_lock(&bond->lock);
2748 alen = arp_hdr_len(bond->dev);
2750 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2751 bond->dev->name, skb->dev->name);
2753 if (alen > skb_headlen(skb)) {
2754 arp = kmalloc(alen, GFP_ATOMIC);
2757 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2761 if (arp->ar_hln != bond->dev->addr_len ||
2762 skb->pkt_type == PACKET_OTHERHOST ||
2763 skb->pkt_type == PACKET_LOOPBACK ||
2764 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2765 arp->ar_pro != htons(ETH_P_IP) ||
2769 arp_ptr = (unsigned char *)(arp + 1);
2770 arp_ptr += bond->dev->addr_len;
2771 memcpy(&sip, arp_ptr, 4);
2772 arp_ptr += 4 + bond->dev->addr_len;
2773 memcpy(&tip, arp_ptr, 4);
2775 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2776 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2777 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2781 * Backup slaves won't see the ARP reply, but do come through
2782 * here for each ARP probe (so we swap the sip/tip to validate
2783 * the probe). In a "redundant switch, common router" type of
2784 * configuration, the ARP probe will (hopefully) travel from
2785 * the active, through one switch, the router, then the other
2786 * switch before reaching the backup.
2788 if (bond_is_active_slave(slave))
2789 bond_validate_arp(bond, slave, sip, tip);
2791 bond_validate_arp(bond, slave, tip, sip);
2794 read_unlock(&bond->lock);
2795 if (arp != (struct arphdr *)skb->data)
2797 return RX_HANDLER_ANOTHER;
2801 * this function is called regularly to monitor each slave's link
2802 * ensuring that traffic is being sent and received when arp monitoring
2803 * is used in load-balancing mode. if the adapter has been dormant, then an
2804 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2805 * arp monitoring in active backup mode.
2807 void bond_loadbalance_arp_mon(struct work_struct *work)
2809 struct bonding *bond = container_of(work, struct bonding,
2811 struct slave *slave, *oldcurrent;
2812 int do_failover = 0;
2816 read_lock(&bond->lock);
2818 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2820 if (bond->slave_cnt == 0)
2823 read_lock(&bond->curr_slave_lock);
2824 oldcurrent = bond->curr_active_slave;
2825 read_unlock(&bond->curr_slave_lock);
2827 /* see if any of the previous devices are up now (i.e. they have
2828 * xmt and rcv traffic). the curr_active_slave does not come into
2829 * the picture unless it is null. also, slave->jiffies is not needed
2830 * here because we send an arp on each slave and give a slave as
2831 * long as it needs to get the tx/rx within the delta.
2832 * TODO: what about up/down delay in arp mode? it wasn't here before
2835 bond_for_each_slave(bond, slave, i) {
2836 unsigned long trans_start = dev_trans_start(slave->dev);
2838 if (slave->link != BOND_LINK_UP) {
2839 if (time_in_range(jiffies,
2840 trans_start - delta_in_ticks,
2841 trans_start + delta_in_ticks) &&
2842 time_in_range(jiffies,
2843 slave->dev->last_rx - delta_in_ticks,
2844 slave->dev->last_rx + delta_in_ticks)) {
2846 slave->link = BOND_LINK_UP;
2847 bond_set_active_slave(slave);
2849 /* primary_slave has no meaning in round-robin
2850 * mode. the window of a slave being up and
2851 * curr_active_slave being null after enslaving
2855 pr_info("%s: link status definitely up for interface %s, ",
2860 pr_info("%s: interface %s is now up\n",
2866 /* slave->link == BOND_LINK_UP */
2868 /* not all switches will respond to an arp request
2869 * when the source ip is 0, so don't take the link down
2870 * if we don't know our ip yet
2872 if (!time_in_range(jiffies,
2873 trans_start - delta_in_ticks,
2874 trans_start + 2 * delta_in_ticks) ||
2875 !time_in_range(jiffies,
2876 slave->dev->last_rx - delta_in_ticks,
2877 slave->dev->last_rx + 2 * delta_in_ticks)) {
2879 slave->link = BOND_LINK_DOWN;
2880 bond_set_backup_slave(slave);
2882 if (slave->link_failure_count < UINT_MAX)
2883 slave->link_failure_count++;
2885 pr_info("%s: interface %s is now down.\n",
2889 if (slave == oldcurrent)
2894 /* note: if switch is in round-robin mode, all links
2895 * must tx arp to ensure all links rx an arp - otherwise
2896 * links may oscillate or not come up at all; if switch is
2897 * in something like xor mode, there is nothing we can
2898 * do - all replies will be rx'ed on same link causing slaves
2899 * to be unstable during low/no traffic periods
2901 if (IS_UP(slave->dev))
2902 bond_arp_send_all(bond, slave);
2907 write_lock_bh(&bond->curr_slave_lock);
2909 bond_select_active_slave(bond);
2911 write_unlock_bh(&bond->curr_slave_lock);
2912 unblock_netpoll_tx();
2916 if (bond->params.arp_interval)
2917 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2919 read_unlock(&bond->lock);
2923 * Called to inspect slaves for active-backup mode ARP monitor link state
2924 * changes. Sets new_link in slaves to specify what action should take
2925 * place for the slave. Returns 0 if no changes are found, >0 if changes
2926 * to link states must be committed.
2928 * Called with bond->lock held for read.
2930 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2932 struct slave *slave;
2934 unsigned long trans_start;
2936 bond_for_each_slave(bond, slave, i) {
2937 slave->new_link = BOND_LINK_NOCHANGE;
2939 if (slave->link != BOND_LINK_UP) {
2940 if (time_in_range(jiffies,
2941 slave_last_rx(bond, slave) - delta_in_ticks,
2942 slave_last_rx(bond, slave) + delta_in_ticks)) {
2944 slave->new_link = BOND_LINK_UP;
2952 * Give slaves 2*delta after being enslaved or made
2953 * active. This avoids bouncing, as the last receive
2954 * times need a full ARP monitor cycle to be updated.
2956 if (time_in_range(jiffies,
2957 slave->jiffies - delta_in_ticks,
2958 slave->jiffies + 2 * delta_in_ticks))
2962 * Backup slave is down if:
2963 * - No current_arp_slave AND
2964 * - more than 3*delta since last receive AND
2965 * - the bond has an IP address
2967 * Note: a non-null current_arp_slave indicates
2968 * the curr_active_slave went down and we are
2969 * searching for a new one; under this condition
2970 * we only take the curr_active_slave down - this
2971 * gives each slave a chance to tx/rx traffic
2972 * before being taken out
2974 if (!bond_is_active_slave(slave) &&
2975 !bond->current_arp_slave &&
2976 !time_in_range(jiffies,
2977 slave_last_rx(bond, slave) - delta_in_ticks,
2978 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2980 slave->new_link = BOND_LINK_DOWN;
2985 * Active slave is down if:
2986 * - more than 2*delta since transmitting OR
2987 * - (more than 2*delta since receive AND
2988 * the bond has an IP address)
2990 trans_start = dev_trans_start(slave->dev);
2991 if (bond_is_active_slave(slave) &&
2992 (!time_in_range(jiffies,
2993 trans_start - delta_in_ticks,
2994 trans_start + 2 * delta_in_ticks) ||
2995 !time_in_range(jiffies,
2996 slave_last_rx(bond, slave) - delta_in_ticks,
2997 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2999 slave->new_link = BOND_LINK_DOWN;
3008 * Called to commit link state changes noted by inspection step of
3009 * active-backup mode ARP monitor.
3011 * Called with RTNL and bond->lock for read.
3013 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3015 struct slave *slave;
3017 unsigned long trans_start;
3019 bond_for_each_slave(bond, slave, i) {
3020 switch (slave->new_link) {
3021 case BOND_LINK_NOCHANGE:
3025 trans_start = dev_trans_start(slave->dev);
3026 if ((!bond->curr_active_slave &&
3027 time_in_range(jiffies,
3028 trans_start - delta_in_ticks,
3029 trans_start + delta_in_ticks)) ||
3030 bond->curr_active_slave != slave) {
3031 slave->link = BOND_LINK_UP;
3032 if (bond->current_arp_slave) {
3033 bond_set_slave_inactive_flags(
3034 bond->current_arp_slave);
3035 bond->current_arp_slave = NULL;
3038 pr_info("%s: link status definitely up for interface %s.\n",
3039 bond->dev->name, slave->dev->name);
3041 if (!bond->curr_active_slave ||
3042 (slave == bond->primary_slave))
3049 case BOND_LINK_DOWN:
3050 if (slave->link_failure_count < UINT_MAX)
3051 slave->link_failure_count++;
3053 slave->link = BOND_LINK_DOWN;
3054 bond_set_slave_inactive_flags(slave);
3056 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3057 bond->dev->name, slave->dev->name);
3059 if (slave == bond->curr_active_slave) {
3060 bond->current_arp_slave = NULL;
3067 pr_err("%s: impossible: new_link %d on slave %s\n",
3068 bond->dev->name, slave->new_link,
3076 write_lock_bh(&bond->curr_slave_lock);
3077 bond_select_active_slave(bond);
3078 write_unlock_bh(&bond->curr_slave_lock);
3079 unblock_netpoll_tx();
3082 bond_set_carrier(bond);
3086 * Send ARP probes for active-backup mode ARP monitor.
3088 * Called with bond->lock held for read.
3090 static void bond_ab_arp_probe(struct bonding *bond)
3092 struct slave *slave;
3095 read_lock(&bond->curr_slave_lock);
3097 if (bond->current_arp_slave && bond->curr_active_slave)
3098 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3099 bond->current_arp_slave->dev->name,
3100 bond->curr_active_slave->dev->name);
3102 if (bond->curr_active_slave) {
3103 bond_arp_send_all(bond, bond->curr_active_slave);
3104 read_unlock(&bond->curr_slave_lock);
3108 read_unlock(&bond->curr_slave_lock);
3110 /* if we don't have a curr_active_slave, search for the next available
3111 * backup slave from the current_arp_slave and make it the candidate
3112 * for becoming the curr_active_slave
3115 if (!bond->current_arp_slave) {
3116 bond->current_arp_slave = bond->first_slave;
3117 if (!bond->current_arp_slave)
3121 bond_set_slave_inactive_flags(bond->current_arp_slave);
3123 /* search for next candidate */
3124 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3125 if (IS_UP(slave->dev)) {
3126 slave->link = BOND_LINK_BACK;
3127 bond_set_slave_active_flags(slave);
3128 bond_arp_send_all(bond, slave);
3129 slave->jiffies = jiffies;
3130 bond->current_arp_slave = slave;
3134 /* if the link state is up at this point, we
3135 * mark it down - this can happen if we have
3136 * simultaneous link failures and
3137 * reselect_active_interface doesn't make this
3138 * one the current slave so it is still marked
3139 * up when it is actually down
3141 if (slave->link == BOND_LINK_UP) {
3142 slave->link = BOND_LINK_DOWN;
3143 if (slave->link_failure_count < UINT_MAX)
3144 slave->link_failure_count++;
3146 bond_set_slave_inactive_flags(slave);
3148 pr_info("%s: backup interface %s is now down.\n",
3149 bond->dev->name, slave->dev->name);
3154 void bond_activebackup_arp_mon(struct work_struct *work)
3156 struct bonding *bond = container_of(work, struct bonding,
3158 bool should_notify_peers = false;
3161 read_lock(&bond->lock);
3163 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3165 if (bond->slave_cnt == 0)
3168 should_notify_peers = bond_should_notify_peers(bond);
3170 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3171 read_unlock(&bond->lock);
3173 /* Race avoidance with bond_close flush of workqueue */
3174 if (!rtnl_trylock()) {
3175 read_lock(&bond->lock);
3177 should_notify_peers = false;
3181 read_lock(&bond->lock);
3183 bond_ab_arp_commit(bond, delta_in_ticks);
3185 read_unlock(&bond->lock);
3187 read_lock(&bond->lock);
3190 bond_ab_arp_probe(bond);
3193 if (bond->params.arp_interval)
3194 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3196 read_unlock(&bond->lock);
3198 if (should_notify_peers) {
3199 if (!rtnl_trylock()) {
3200 read_lock(&bond->lock);
3201 bond->send_peer_notif++;
3202 read_unlock(&bond->lock);
3205 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3210 /*-------------------------- netdev event handling --------------------------*/
3213 * Change device name
3215 static int bond_event_changename(struct bonding *bond)
3217 bond_remove_proc_entry(bond);
3218 bond_create_proc_entry(bond);
3220 bond_debug_reregister(bond);
3225 static int bond_master_netdev_event(unsigned long event,
3226 struct net_device *bond_dev)
3228 struct bonding *event_bond = netdev_priv(bond_dev);
3231 case NETDEV_CHANGENAME:
3232 return bond_event_changename(event_bond);
3240 static int bond_slave_netdev_event(unsigned long event,
3241 struct net_device *slave_dev)
3243 struct net_device *bond_dev = slave_dev->master;
3244 struct bonding *bond = netdev_priv(bond_dev);
3245 struct slave *slave = NULL;
3248 case NETDEV_UNREGISTER:
3250 if (bond->setup_by_slave)
3251 bond_release_and_destroy(bond_dev, slave_dev);
3253 bond_release(bond_dev, slave_dev);
3258 slave = bond_get_slave_by_dev(bond, slave_dev);
3260 u32 old_speed = slave->speed;
3261 u8 old_duplex = slave->duplex;
3263 bond_update_speed_duplex(slave);
3265 if (bond->params.mode == BOND_MODE_8023AD) {
3266 if (old_speed != slave->speed)
3267 bond_3ad_adapter_speed_changed(slave);
3268 if (old_duplex != slave->duplex)
3269 bond_3ad_adapter_duplex_changed(slave);
3276 * ... Or is it this?
3279 case NETDEV_CHANGEMTU:
3281 * TODO: Should slaves be allowed to
3282 * independently alter their MTU? For
3283 * an active-backup bond, slaves need
3284 * not be the same type of device, so
3285 * MTUs may vary. For other modes,
3286 * slaves arguably should have the
3287 * same MTUs. To do this, we'd need to
3288 * take over the slave's change_mtu
3289 * function for the duration of their
3293 case NETDEV_CHANGENAME:
3295 * TODO: handle changing the primary's name
3298 case NETDEV_FEAT_CHANGE:
3299 bond_compute_features(bond);
3309 * bond_netdev_event: handle netdev notifier chain events.
3311 * This function receives events for the netdev chain. The caller (an
3312 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3313 * locks for us to safely manipulate the slave devices (RTNL lock,
3316 static int bond_netdev_event(struct notifier_block *this,
3317 unsigned long event, void *ptr)
3319 struct net_device *event_dev = (struct net_device *)ptr;
3321 pr_debug("event_dev: %s, event: %lx\n",
3322 event_dev ? event_dev->name : "None",
3325 if (!(event_dev->priv_flags & IFF_BONDING))
3328 if (event_dev->flags & IFF_MASTER) {
3329 pr_debug("IFF_MASTER\n");
3330 return bond_master_netdev_event(event, event_dev);
3333 if (event_dev->flags & IFF_SLAVE) {
3334 pr_debug("IFF_SLAVE\n");
3335 return bond_slave_netdev_event(event, event_dev);
3341 static struct notifier_block bond_netdev_notifier = {
3342 .notifier_call = bond_netdev_event,
3345 /*---------------------------- Hashing Policies -----------------------------*/
3348 * Hash for the output device based upon layer 2 and layer 3 data. If
3349 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3351 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3353 struct ethhdr *data = (struct ethhdr *)skb->data;
3354 struct iphdr *iph = ip_hdr(skb);
3356 if (skb->protocol == htons(ETH_P_IP)) {
3357 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3358 (data->h_dest[5] ^ data->h_source[5])) % count;
3361 return (data->h_dest[5] ^ data->h_source[5]) % count;
3365 * Hash for the output device based upon layer 3 and layer 4 data. If
3366 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3367 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3369 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3371 struct ethhdr *data = (struct ethhdr *)skb->data;
3372 struct iphdr *iph = ip_hdr(skb);
3373 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3376 if (skb->protocol == htons(ETH_P_IP)) {
3377 if (!ip_is_fragment(iph) &&
3378 (iph->protocol == IPPROTO_TCP ||
3379 iph->protocol == IPPROTO_UDP)) {
3380 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3382 return (layer4_xor ^
3383 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3387 return (data->h_dest[5] ^ data->h_source[5]) % count;
3391 * Hash for the output device based upon layer 2 data
3393 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3395 struct ethhdr *data = (struct ethhdr *)skb->data;
3397 return (data->h_dest[5] ^ data->h_source[5]) % count;
3400 /*-------------------------- Device entry points ----------------------------*/
3402 static int bond_open(struct net_device *bond_dev)
3404 struct bonding *bond = netdev_priv(bond_dev);
3405 struct slave *slave;
3408 /* reset slave->backup and slave->inactive */
3409 read_lock(&bond->lock);
3410 if (bond->slave_cnt > 0) {
3411 read_lock(&bond->curr_slave_lock);
3412 bond_for_each_slave(bond, slave, i) {
3413 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3414 && (slave != bond->curr_active_slave)) {
3415 bond_set_slave_inactive_flags(slave);
3417 bond_set_slave_active_flags(slave);
3420 read_unlock(&bond->curr_slave_lock);
3422 read_unlock(&bond->lock);
3424 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3426 if (bond_is_lb(bond)) {
3427 /* bond_alb_initialize must be called before the timer
3430 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3431 /* something went wrong - fail the open operation */
3435 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3436 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3439 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3440 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3441 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3444 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3445 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3446 INIT_DELAYED_WORK(&bond->arp_work,
3447 bond_activebackup_arp_mon);
3449 INIT_DELAYED_WORK(&bond->arp_work,
3450 bond_loadbalance_arp_mon);
3452 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3453 if (bond->params.arp_validate)
3454 bond->recv_probe = bond_arp_rcv;
3457 if (bond->params.mode == BOND_MODE_8023AD) {
3458 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3459 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3460 /* register to receive LACPDUs */
3461 bond->recv_probe = bond_3ad_lacpdu_recv;
3462 bond_3ad_initiate_agg_selection(bond, 1);
3468 static int bond_close(struct net_device *bond_dev)
3470 struct bonding *bond = netdev_priv(bond_dev);
3472 write_lock_bh(&bond->lock);
3474 bond->send_peer_notif = 0;
3476 write_unlock_bh(&bond->lock);
3478 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3479 cancel_delayed_work_sync(&bond->mii_work);
3482 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3483 cancel_delayed_work_sync(&bond->arp_work);
3486 switch (bond->params.mode) {
3487 case BOND_MODE_8023AD:
3488 cancel_delayed_work_sync(&bond->ad_work);
3492 cancel_delayed_work_sync(&bond->alb_work);
3498 if (delayed_work_pending(&bond->mcast_work))
3499 cancel_delayed_work_sync(&bond->mcast_work);
3501 if (bond_is_lb(bond)) {
3502 /* Must be called only after all
3503 * slaves have been released
3505 bond_alb_deinitialize(bond);
3507 bond->recv_probe = NULL;
3512 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3513 struct rtnl_link_stats64 *stats)
3515 struct bonding *bond = netdev_priv(bond_dev);
3516 struct rtnl_link_stats64 temp;
3517 struct slave *slave;
3520 memset(stats, 0, sizeof(*stats));
3522 read_lock_bh(&bond->lock);
3524 bond_for_each_slave(bond, slave, i) {
3525 const struct rtnl_link_stats64 *sstats =
3526 dev_get_stats(slave->dev, &temp);
3528 stats->rx_packets += sstats->rx_packets;
3529 stats->rx_bytes += sstats->rx_bytes;
3530 stats->rx_errors += sstats->rx_errors;
3531 stats->rx_dropped += sstats->rx_dropped;
3533 stats->tx_packets += sstats->tx_packets;
3534 stats->tx_bytes += sstats->tx_bytes;
3535 stats->tx_errors += sstats->tx_errors;
3536 stats->tx_dropped += sstats->tx_dropped;
3538 stats->multicast += sstats->multicast;
3539 stats->collisions += sstats->collisions;
3541 stats->rx_length_errors += sstats->rx_length_errors;
3542 stats->rx_over_errors += sstats->rx_over_errors;
3543 stats->rx_crc_errors += sstats->rx_crc_errors;
3544 stats->rx_frame_errors += sstats->rx_frame_errors;
3545 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3546 stats->rx_missed_errors += sstats->rx_missed_errors;
3548 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3549 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3550 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3551 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3552 stats->tx_window_errors += sstats->tx_window_errors;
3555 read_unlock_bh(&bond->lock);
3560 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3562 struct net_device *slave_dev = NULL;
3563 struct ifbond k_binfo;
3564 struct ifbond __user *u_binfo = NULL;
3565 struct ifslave k_sinfo;
3566 struct ifslave __user *u_sinfo = NULL;
3567 struct mii_ioctl_data *mii = NULL;
3570 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3582 * We do this again just in case we were called by SIOCGMIIREG
3583 * instead of SIOCGMIIPHY.
3590 if (mii->reg_num == 1) {
3591 struct bonding *bond = netdev_priv(bond_dev);
3593 read_lock(&bond->lock);
3594 read_lock(&bond->curr_slave_lock);
3595 if (netif_carrier_ok(bond->dev))
3596 mii->val_out = BMSR_LSTATUS;
3598 read_unlock(&bond->curr_slave_lock);
3599 read_unlock(&bond->lock);
3603 case BOND_INFO_QUERY_OLD:
3604 case SIOCBONDINFOQUERY:
3605 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3607 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3610 res = bond_info_query(bond_dev, &k_binfo);
3612 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3616 case BOND_SLAVE_INFO_QUERY_OLD:
3617 case SIOCBONDSLAVEINFOQUERY:
3618 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3620 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3623 res = bond_slave_info_query(bond_dev, &k_sinfo);
3625 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3634 if (!capable(CAP_NET_ADMIN))
3637 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3639 pr_debug("slave_dev=%p:\n", slave_dev);
3644 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3646 case BOND_ENSLAVE_OLD:
3647 case SIOCBONDENSLAVE:
3648 res = bond_enslave(bond_dev, slave_dev);
3650 case BOND_RELEASE_OLD:
3651 case SIOCBONDRELEASE:
3652 res = bond_release(bond_dev, slave_dev);
3654 case BOND_SETHWADDR_OLD:
3655 case SIOCBONDSETHWADDR:
3656 res = bond_sethwaddr(bond_dev, slave_dev);
3658 case BOND_CHANGE_ACTIVE_OLD:
3659 case SIOCBONDCHANGEACTIVE:
3660 res = bond_ioctl_change_active(bond_dev, slave_dev);
3672 static bool bond_addr_in_mc_list(unsigned char *addr,
3673 struct netdev_hw_addr_list *list,
3676 struct netdev_hw_addr *ha;
3678 netdev_hw_addr_list_for_each(ha, list)
3679 if (!memcmp(ha->addr, addr, addrlen))
3685 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3687 struct bonding *bond = netdev_priv(bond_dev);
3689 if (change & IFF_PROMISC)
3690 bond_set_promiscuity(bond,
3691 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3693 if (change & IFF_ALLMULTI)
3694 bond_set_allmulti(bond,
3695 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3698 static void bond_set_multicast_list(struct net_device *bond_dev)
3700 struct bonding *bond = netdev_priv(bond_dev);
3701 struct netdev_hw_addr *ha;
3704 read_lock(&bond->lock);
3706 /* looking for addresses to add to slaves' mc list */
3707 netdev_for_each_mc_addr(ha, bond_dev) {
3708 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3709 bond_dev->addr_len);
3711 bond_mc_add(bond, ha->addr);
3714 /* looking for addresses to delete from slaves' list */
3715 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3716 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3717 bond_dev->addr_len);
3719 bond_mc_del(bond, ha->addr);
3722 /* save master's multicast list */
3723 __hw_addr_flush(&bond->mc_list);
3724 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3725 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3727 read_unlock(&bond->lock);
3730 static int bond_neigh_init(struct neighbour *n)
3732 struct bonding *bond = netdev_priv(n->dev);
3733 struct slave *slave = bond->first_slave;
3734 const struct net_device_ops *slave_ops;
3735 struct neigh_parms parms;
3741 slave_ops = slave->dev->netdev_ops;
3743 if (!slave_ops->ndo_neigh_setup)
3746 parms.neigh_setup = NULL;
3747 parms.neigh_cleanup = NULL;
3748 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3753 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3754 * after the last slave has been detached. Assumes that all slaves
3755 * utilize the same neigh_cleanup (true at this writing as only user
3758 n->parms->neigh_cleanup = parms.neigh_cleanup;
3760 if (!parms.neigh_setup)
3763 return parms.neigh_setup(n);
3767 * The bonding ndo_neigh_setup is called at init time beofre any
3768 * slave exists. So we must declare proxy setup function which will
3769 * be used at run time to resolve the actual slave neigh param setup.
3771 static int bond_neigh_setup(struct net_device *dev,
3772 struct neigh_parms *parms)
3774 parms->neigh_setup = bond_neigh_init;
3780 * Change the MTU of all of a master's slaves to match the master
3782 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3784 struct bonding *bond = netdev_priv(bond_dev);
3785 struct slave *slave, *stop_at;
3789 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3790 (bond_dev ? bond_dev->name : "None"), new_mtu);
3792 /* Can't hold bond->lock with bh disabled here since
3793 * some base drivers panic. On the other hand we can't
3794 * hold bond->lock without bh disabled because we'll
3795 * deadlock. The only solution is to rely on the fact
3796 * that we're under rtnl_lock here, and the slaves
3797 * list won't change. This doesn't solve the problem
3798 * of setting the slave's MTU while it is
3799 * transmitting, but the assumption is that the base
3800 * driver can handle that.
3802 * TODO: figure out a way to safely iterate the slaves
3803 * list, but without holding a lock around the actual
3804 * call to the base driver.
3807 bond_for_each_slave(bond, slave, i) {
3808 pr_debug("s %p s->p %p c_m %p\n",
3811 slave->dev->netdev_ops->ndo_change_mtu);
3813 res = dev_set_mtu(slave->dev, new_mtu);
3816 /* If we failed to set the slave's mtu to the new value
3817 * we must abort the operation even in ACTIVE_BACKUP
3818 * mode, because if we allow the backup slaves to have
3819 * different mtu values than the active slave we'll
3820 * need to change their mtu when doing a failover. That
3821 * means changing their mtu from timer context, which
3822 * is probably not a good idea.
3824 pr_debug("err %d %s\n", res, slave->dev->name);
3829 bond_dev->mtu = new_mtu;
3834 /* unwind from head to the slave that failed */
3836 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3839 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3841 pr_debug("unwind err %d dev %s\n",
3842 tmp_res, slave->dev->name);
3852 * Note that many devices must be down to change the HW address, and
3853 * downing the master releases all slaves. We can make bonds full of
3854 * bonding devices to test this, however.
3856 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3858 struct bonding *bond = netdev_priv(bond_dev);
3859 struct sockaddr *sa = addr, tmp_sa;
3860 struct slave *slave, *stop_at;
3864 if (bond->params.mode == BOND_MODE_ALB)
3865 return bond_alb_set_mac_address(bond_dev, addr);
3868 pr_debug("bond=%p, name=%s\n",
3869 bond, bond_dev ? bond_dev->name : "None");
3872 * If fail_over_mac is set to active, do nothing and return
3873 * success. Returning an error causes ifenslave to fail.
3875 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3878 if (!is_valid_ether_addr(sa->sa_data))
3879 return -EADDRNOTAVAIL;
3881 /* Can't hold bond->lock with bh disabled here since
3882 * some base drivers panic. On the other hand we can't
3883 * hold bond->lock without bh disabled because we'll
3884 * deadlock. The only solution is to rely on the fact
3885 * that we're under rtnl_lock here, and the slaves
3886 * list won't change. This doesn't solve the problem
3887 * of setting the slave's hw address while it is
3888 * transmitting, but the assumption is that the base
3889 * driver can handle that.
3891 * TODO: figure out a way to safely iterate the slaves
3892 * list, but without holding a lock around the actual
3893 * call to the base driver.
3896 bond_for_each_slave(bond, slave, i) {
3897 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3898 pr_debug("slave %p %s\n", slave, slave->dev->name);
3900 if (slave_ops->ndo_set_mac_address == NULL) {
3902 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3906 res = dev_set_mac_address(slave->dev, addr);
3908 /* TODO: consider downing the slave
3910 * User should expect communications
3911 * breakage anyway until ARP finish
3914 pr_debug("err %d %s\n", res, slave->dev->name);
3920 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3924 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3925 tmp_sa.sa_family = bond_dev->type;
3927 /* unwind from head to the slave that failed */
3929 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3932 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3934 pr_debug("unwind err %d dev %s\n",
3935 tmp_res, slave->dev->name);
3942 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3944 struct bonding *bond = netdev_priv(bond_dev);
3945 struct slave *slave, *start_at;
3946 int i, slave_no, res = 1;
3947 struct iphdr *iph = ip_hdr(skb);
3950 * Start with the curr_active_slave that joined the bond as the
3951 * default for sending IGMP traffic. For failover purposes one
3952 * needs to maintain some consistency for the interface that will
3953 * send the join/membership reports. The curr_active_slave found
3954 * will send all of this type of traffic.
3956 if ((iph->protocol == IPPROTO_IGMP) &&
3957 (skb->protocol == htons(ETH_P_IP))) {
3959 read_lock(&bond->curr_slave_lock);
3960 slave = bond->curr_active_slave;
3961 read_unlock(&bond->curr_slave_lock);
3967 * Concurrent TX may collide on rr_tx_counter; we accept
3968 * that as being rare enough not to justify using an
3971 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3973 bond_for_each_slave(bond, slave, i) {
3981 bond_for_each_slave_from(bond, slave, i, start_at) {
3982 if (IS_UP(slave->dev) &&
3983 (slave->link == BOND_LINK_UP) &&
3984 bond_is_active_slave(slave)) {
3985 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3992 /* no suitable interface, frame not sent */
3996 return NETDEV_TX_OK;
4001 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4002 * the bond has a usable interface.
4004 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4006 struct bonding *bond = netdev_priv(bond_dev);
4009 read_lock(&bond->curr_slave_lock);
4011 if (bond->curr_active_slave)
4012 res = bond_dev_queue_xmit(bond, skb,
4013 bond->curr_active_slave->dev);
4015 read_unlock(&bond->curr_slave_lock);
4018 /* no suitable interface, frame not sent */
4021 return NETDEV_TX_OK;
4025 * In bond_xmit_xor() , we determine the output device by using a pre-
4026 * determined xmit_hash_policy(), If the selected device is not enabled,
4027 * find the next active slave.
4029 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4031 struct bonding *bond = netdev_priv(bond_dev);
4032 struct slave *slave, *start_at;
4037 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4039 bond_for_each_slave(bond, slave, i) {
4047 bond_for_each_slave_from(bond, slave, i, start_at) {
4048 if (IS_UP(slave->dev) &&
4049 (slave->link == BOND_LINK_UP) &&
4050 bond_is_active_slave(slave)) {
4051 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4057 /* no suitable interface, frame not sent */
4061 return NETDEV_TX_OK;
4065 * in broadcast mode, we send everything to all usable interfaces.
4067 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4069 struct bonding *bond = netdev_priv(bond_dev);
4070 struct slave *slave, *start_at;
4071 struct net_device *tx_dev = NULL;
4075 read_lock(&bond->curr_slave_lock);
4076 start_at = bond->curr_active_slave;
4077 read_unlock(&bond->curr_slave_lock);
4082 bond_for_each_slave_from(bond, slave, i, start_at) {
4083 if (IS_UP(slave->dev) &&
4084 (slave->link == BOND_LINK_UP) &&
4085 bond_is_active_slave(slave)) {
4087 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4089 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4094 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4100 tx_dev = slave->dev;
4105 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4109 /* no suitable interface, frame not sent */
4112 /* frame sent to all suitable interfaces */
4113 return NETDEV_TX_OK;
4116 /*------------------------- Device initialization ---------------------------*/
4118 static void bond_set_xmit_hash_policy(struct bonding *bond)
4120 switch (bond->params.xmit_policy) {
4121 case BOND_XMIT_POLICY_LAYER23:
4122 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4124 case BOND_XMIT_POLICY_LAYER34:
4125 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4127 case BOND_XMIT_POLICY_LAYER2:
4129 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4135 * Lookup the slave that corresponds to a qid
4137 static inline int bond_slave_override(struct bonding *bond,
4138 struct sk_buff *skb)
4141 struct slave *slave = NULL;
4142 struct slave *check_slave;
4144 if (!skb->queue_mapping)
4147 /* Find out if any slaves have the same mapping as this skb. */
4148 bond_for_each_slave(bond, check_slave, i) {
4149 if (check_slave->queue_id == skb->queue_mapping) {
4150 slave = check_slave;
4155 /* If the slave isn't UP, use default transmit policy. */
4156 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4157 (slave->link == BOND_LINK_UP)) {
4158 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4165 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4168 * This helper function exists to help dev_pick_tx get the correct
4169 * destination queue. Using a helper function skips a call to
4170 * skb_tx_hash and will put the skbs in the queue we expect on their
4171 * way down to the bonding driver.
4173 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4176 * Save the original txq to restore before passing to the driver
4178 qdisc_skb_cb(skb)->bond_queue_mapping = skb->queue_mapping;
4180 if (unlikely(txq >= dev->real_num_tx_queues)) {
4182 txq -= dev->real_num_tx_queues;
4183 } while (txq >= dev->real_num_tx_queues);
4188 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4190 struct bonding *bond = netdev_priv(dev);
4192 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4193 if (!bond_slave_override(bond, skb))
4194 return NETDEV_TX_OK;
4197 switch (bond->params.mode) {
4198 case BOND_MODE_ROUNDROBIN:
4199 return bond_xmit_roundrobin(skb, dev);
4200 case BOND_MODE_ACTIVEBACKUP:
4201 return bond_xmit_activebackup(skb, dev);
4203 return bond_xmit_xor(skb, dev);
4204 case BOND_MODE_BROADCAST:
4205 return bond_xmit_broadcast(skb, dev);
4206 case BOND_MODE_8023AD:
4207 return bond_3ad_xmit_xor(skb, dev);
4210 return bond_alb_xmit(skb, dev);
4212 /* Should never happen, mode already checked */
4213 pr_err("%s: Error: Unknown bonding mode %d\n",
4214 dev->name, bond->params.mode);
4217 return NETDEV_TX_OK;
4221 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4223 struct bonding *bond = netdev_priv(dev);
4224 netdev_tx_t ret = NETDEV_TX_OK;
4227 * If we risk deadlock from transmitting this in the
4228 * netpoll path, tell netpoll to queue the frame for later tx
4230 if (is_netpoll_tx_blocked(dev))
4231 return NETDEV_TX_BUSY;
4233 read_lock(&bond->lock);
4235 if (bond->slave_cnt)
4236 ret = __bond_start_xmit(skb, dev);
4240 read_unlock(&bond->lock);
4246 * set bond mode specific net device operations
4248 void bond_set_mode_ops(struct bonding *bond, int mode)
4250 struct net_device *bond_dev = bond->dev;
4253 case BOND_MODE_ROUNDROBIN:
4255 case BOND_MODE_ACTIVEBACKUP:
4258 bond_set_xmit_hash_policy(bond);
4260 case BOND_MODE_BROADCAST:
4262 case BOND_MODE_8023AD:
4263 bond_set_xmit_hash_policy(bond);
4270 /* Should never happen, mode already checked */
4271 pr_err("%s: Error: Unknown bonding mode %d\n",
4272 bond_dev->name, mode);
4277 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4278 struct ethtool_drvinfo *drvinfo)
4280 strncpy(drvinfo->driver, DRV_NAME, 32);
4281 strncpy(drvinfo->version, DRV_VERSION, 32);
4282 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4285 static const struct ethtool_ops bond_ethtool_ops = {
4286 .get_drvinfo = bond_ethtool_get_drvinfo,
4287 .get_link = ethtool_op_get_link,
4290 static const struct net_device_ops bond_netdev_ops = {
4291 .ndo_init = bond_init,
4292 .ndo_uninit = bond_uninit,
4293 .ndo_open = bond_open,
4294 .ndo_stop = bond_close,
4295 .ndo_start_xmit = bond_start_xmit,
4296 .ndo_select_queue = bond_select_queue,
4297 .ndo_get_stats64 = bond_get_stats,
4298 .ndo_do_ioctl = bond_do_ioctl,
4299 .ndo_change_rx_flags = bond_change_rx_flags,
4300 .ndo_set_rx_mode = bond_set_multicast_list,
4301 .ndo_change_mtu = bond_change_mtu,
4302 .ndo_set_mac_address = bond_set_mac_address,
4303 .ndo_neigh_setup = bond_neigh_setup,
4304 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4305 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4306 #ifdef CONFIG_NET_POLL_CONTROLLER
4307 .ndo_netpoll_setup = bond_netpoll_setup,
4308 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4309 .ndo_poll_controller = bond_poll_controller,
4311 .ndo_add_slave = bond_enslave,
4312 .ndo_del_slave = bond_release,
4313 .ndo_fix_features = bond_fix_features,
4316 static void bond_destructor(struct net_device *bond_dev)
4318 struct bonding *bond = netdev_priv(bond_dev);
4320 destroy_workqueue(bond->wq);
4321 free_netdev(bond_dev);
4324 static void bond_setup(struct net_device *bond_dev)
4326 struct bonding *bond = netdev_priv(bond_dev);
4328 /* initialize rwlocks */
4329 rwlock_init(&bond->lock);
4330 rwlock_init(&bond->curr_slave_lock);
4332 bond->params = bonding_defaults;
4334 /* Initialize pointers */
4335 bond->dev = bond_dev;
4336 INIT_LIST_HEAD(&bond->vlan_list);
4338 /* Initialize the device entry points */
4339 ether_setup(bond_dev);
4340 bond_dev->netdev_ops = &bond_netdev_ops;
4341 bond_dev->ethtool_ops = &bond_ethtool_ops;
4342 bond_set_mode_ops(bond, bond->params.mode);
4344 bond_dev->destructor = bond_destructor;
4346 /* Initialize the device options */
4347 bond_dev->tx_queue_len = 0;
4348 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4349 bond_dev->priv_flags |= IFF_BONDING;
4350 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4352 /* At first, we block adding VLANs. That's the only way to
4353 * prevent problems that occur when adding VLANs over an
4354 * empty bond. The block will be removed once non-challenged
4355 * slaves are enslaved.
4357 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4359 /* don't acquire bond device's netif_tx_lock when
4361 bond_dev->features |= NETIF_F_LLTX;
4363 /* By default, we declare the bond to be fully
4364 * VLAN hardware accelerated capable. Special
4365 * care is taken in the various xmit functions
4366 * when there are slaves that are not hw accel
4370 bond_dev->hw_features = BOND_VLAN_FEATURES |
4371 NETIF_F_HW_VLAN_TX |
4372 NETIF_F_HW_VLAN_RX |
4373 NETIF_F_HW_VLAN_FILTER;
4375 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4376 bond_dev->features |= bond_dev->hw_features;
4379 static void bond_work_cancel_all(struct bonding *bond)
4381 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4382 cancel_delayed_work_sync(&bond->mii_work);
4384 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4385 cancel_delayed_work_sync(&bond->arp_work);
4387 if (bond->params.mode == BOND_MODE_ALB &&
4388 delayed_work_pending(&bond->alb_work))
4389 cancel_delayed_work_sync(&bond->alb_work);
4391 if (bond->params.mode == BOND_MODE_8023AD &&
4392 delayed_work_pending(&bond->ad_work))
4393 cancel_delayed_work_sync(&bond->ad_work);
4395 if (delayed_work_pending(&bond->mcast_work))
4396 cancel_delayed_work_sync(&bond->mcast_work);
4400 * Destroy a bonding device.
4401 * Must be under rtnl_lock when this function is called.
4403 static void bond_uninit(struct net_device *bond_dev)
4405 struct bonding *bond = netdev_priv(bond_dev);
4406 struct vlan_entry *vlan, *tmp;
4408 bond_netpoll_cleanup(bond_dev);
4410 /* Release the bonded slaves */
4411 bond_release_all(bond_dev);
4413 list_del(&bond->bond_list);
4415 bond_work_cancel_all(bond);
4417 bond_remove_proc_entry(bond);
4419 bond_debug_unregister(bond);
4421 __hw_addr_flush(&bond->mc_list);
4423 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4424 list_del(&vlan->vlan_list);
4429 /*------------------------- Module initialization ---------------------------*/
4432 * Convert string input module parms. Accept either the
4433 * number of the mode or its string name. A bit complicated because
4434 * some mode names are substrings of other names, and calls from sysfs
4435 * may have whitespace in the name (trailing newlines, for example).
4437 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4439 int modeint = -1, i, rv;
4440 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4442 for (p = (char *)buf; *p; p++)
4443 if (!(isdigit(*p) || isspace(*p)))
4447 rv = sscanf(buf, "%20s", modestr);
4449 rv = sscanf(buf, "%d", &modeint);
4454 for (i = 0; tbl[i].modename; i++) {
4455 if (modeint == tbl[i].mode)
4457 if (strcmp(modestr, tbl[i].modename) == 0)
4464 static int bond_check_params(struct bond_params *params)
4466 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4469 * Convert string parameters.
4472 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4473 if (bond_mode == -1) {
4474 pr_err("Error: Invalid bonding mode \"%s\"\n",
4475 mode == NULL ? "NULL" : mode);
4480 if (xmit_hash_policy) {
4481 if ((bond_mode != BOND_MODE_XOR) &&
4482 (bond_mode != BOND_MODE_8023AD)) {
4483 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4484 bond_mode_name(bond_mode));
4486 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4488 if (xmit_hashtype == -1) {
4489 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4490 xmit_hash_policy == NULL ? "NULL" :
4498 if (bond_mode != BOND_MODE_8023AD) {
4499 pr_info("lacp_rate param is irrelevant in mode %s\n",
4500 bond_mode_name(bond_mode));
4502 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4503 if (lacp_fast == -1) {
4504 pr_err("Error: Invalid lacp rate \"%s\"\n",
4505 lacp_rate == NULL ? "NULL" : lacp_rate);
4512 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4513 if (params->ad_select == -1) {
4514 pr_err("Error: Invalid ad_select \"%s\"\n",
4515 ad_select == NULL ? "NULL" : ad_select);
4519 if (bond_mode != BOND_MODE_8023AD) {
4520 pr_warning("ad_select param only affects 802.3ad mode\n");
4523 params->ad_select = BOND_AD_STABLE;
4526 if (max_bonds < 0) {
4527 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4528 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4529 max_bonds = BOND_DEFAULT_MAX_BONDS;
4533 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4534 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4535 miimon = BOND_LINK_MON_INTERV;
4539 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4544 if (downdelay < 0) {
4545 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4546 downdelay, INT_MAX);
4550 if ((use_carrier != 0) && (use_carrier != 1)) {
4551 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4556 if (num_peer_notif < 0 || num_peer_notif > 255) {
4557 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4562 /* reset values for 802.3ad */
4563 if (bond_mode == BOND_MODE_8023AD) {
4565 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");
4566 pr_warning("Forcing miimon to 100msec\n");
4571 if (tx_queues < 1 || tx_queues > 255) {
4572 pr_warning("Warning: tx_queues (%d) should be between "
4573 "1 and 255, resetting to %d\n",
4574 tx_queues, BOND_DEFAULT_TX_QUEUES);
4575 tx_queues = BOND_DEFAULT_TX_QUEUES;
4578 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4579 pr_warning("Warning: all_slaves_active module parameter (%d), "
4580 "not of valid value (0/1), so it was set to "
4581 "0\n", all_slaves_active);
4582 all_slaves_active = 0;
4585 if (resend_igmp < 0 || resend_igmp > 255) {
4586 pr_warning("Warning: resend_igmp (%d) should be between "
4587 "0 and 255, resetting to %d\n",
4588 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4589 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4592 /* reset values for TLB/ALB */
4593 if ((bond_mode == BOND_MODE_TLB) ||
4594 (bond_mode == BOND_MODE_ALB)) {
4596 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");
4597 pr_warning("Forcing miimon to 100msec\n");
4602 if (bond_mode == BOND_MODE_ALB) {
4603 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",
4608 if (updelay || downdelay) {
4609 /* just warn the user the up/down delay will have
4610 * no effect since miimon is zero...
4612 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",
4613 updelay, downdelay);
4616 /* don't allow arp monitoring */
4618 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4619 miimon, arp_interval);
4623 if ((updelay % miimon) != 0) {
4624 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4626 (updelay / miimon) * miimon);
4631 if ((downdelay % miimon) != 0) {
4632 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4634 (downdelay / miimon) * miimon);
4637 downdelay /= miimon;
4640 if (arp_interval < 0) {
4641 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4642 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4643 arp_interval = BOND_LINK_ARP_INTERV;
4646 for (arp_ip_count = 0;
4647 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4649 /* not complete check, but should be good enough to
4651 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4652 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4653 arp_ip_target[arp_ip_count]);
4656 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4657 arp_target[arp_ip_count] = ip;
4661 if (arp_interval && !arp_ip_count) {
4662 /* don't allow arping if no arp_ip_target given... */
4663 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4669 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4670 pr_err("arp_validate only supported in active-backup mode\n");
4673 if (!arp_interval) {
4674 pr_err("arp_validate requires arp_interval\n");
4678 arp_validate_value = bond_parse_parm(arp_validate,
4680 if (arp_validate_value == -1) {
4681 pr_err("Error: invalid arp_validate \"%s\"\n",
4682 arp_validate == NULL ? "NULL" : arp_validate);
4686 arp_validate_value = 0;
4689 pr_info("MII link monitoring set to %d ms\n", miimon);
4690 } else if (arp_interval) {
4693 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4695 arp_validate_tbl[arp_validate_value].modename,
4698 for (i = 0; i < arp_ip_count; i++)
4699 pr_info(" %s", arp_ip_target[i]);
4703 } else if (max_bonds) {
4704 /* miimon and arp_interval not set, we need one so things
4705 * work as expected, see bonding.txt for details
4707 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");
4710 if (primary && !USES_PRIMARY(bond_mode)) {
4711 /* currently, using a primary only makes sense
4712 * in active backup, TLB or ALB modes
4714 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4715 primary, bond_mode_name(bond_mode));
4719 if (primary && primary_reselect) {
4720 primary_reselect_value = bond_parse_parm(primary_reselect,
4722 if (primary_reselect_value == -1) {
4723 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4725 NULL ? "NULL" : primary_reselect);
4729 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4732 if (fail_over_mac) {
4733 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4735 if (fail_over_mac_value == -1) {
4736 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4737 arp_validate == NULL ? "NULL" : arp_validate);
4741 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4742 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4744 fail_over_mac_value = BOND_FOM_NONE;
4747 /* fill params struct with the proper values */
4748 params->mode = bond_mode;
4749 params->xmit_policy = xmit_hashtype;
4750 params->miimon = miimon;
4751 params->num_peer_notif = num_peer_notif;
4752 params->arp_interval = arp_interval;
4753 params->arp_validate = arp_validate_value;
4754 params->updelay = updelay;
4755 params->downdelay = downdelay;
4756 params->use_carrier = use_carrier;
4757 params->lacp_fast = lacp_fast;
4758 params->primary[0] = 0;
4759 params->primary_reselect = primary_reselect_value;
4760 params->fail_over_mac = fail_over_mac_value;
4761 params->tx_queues = tx_queues;
4762 params->all_slaves_active = all_slaves_active;
4763 params->resend_igmp = resend_igmp;
4764 params->min_links = min_links;
4767 strncpy(params->primary, primary, IFNAMSIZ);
4768 params->primary[IFNAMSIZ - 1] = 0;
4771 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4776 static struct lock_class_key bonding_netdev_xmit_lock_key;
4777 static struct lock_class_key bonding_netdev_addr_lock_key;
4779 static void bond_set_lockdep_class_one(struct net_device *dev,
4780 struct netdev_queue *txq,
4783 lockdep_set_class(&txq->_xmit_lock,
4784 &bonding_netdev_xmit_lock_key);
4787 static void bond_set_lockdep_class(struct net_device *dev)
4789 lockdep_set_class(&dev->addr_list_lock,
4790 &bonding_netdev_addr_lock_key);
4791 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4795 * Called from registration process
4797 static int bond_init(struct net_device *bond_dev)
4799 struct bonding *bond = netdev_priv(bond_dev);
4800 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4801 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4803 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4806 * Initialize locks that may be required during
4807 * en/deslave operations. All of the bond_open work
4808 * (of which this is part) should really be moved to
4809 * a phase prior to dev_open
4811 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4812 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4814 bond->wq = create_singlethread_workqueue(bond_dev->name);
4818 bond_set_lockdep_class(bond_dev);
4820 bond_create_proc_entry(bond);
4821 list_add_tail(&bond->bond_list, &bn->dev_list);
4823 bond_prepare_sysfs_group(bond);
4825 bond_debug_register(bond);
4827 __hw_addr_init(&bond->mc_list);
4831 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4833 if (tb[IFLA_ADDRESS]) {
4834 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4836 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4837 return -EADDRNOTAVAIL;
4842 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[])
4847 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4849 .priv_size = sizeof(struct bonding),
4850 .setup = bond_setup,
4851 .validate = bond_validate,
4852 .get_tx_queues = bond_get_tx_queues,
4855 /* Create a new bond based on the specified name and bonding parameters.
4856 * If name is NULL, obtain a suitable "bond%d" name for us.
4857 * Caller must NOT hold rtnl_lock; we need to release it here before we
4858 * set up our sysfs entries.
4860 int bond_create(struct net *net, const char *name)
4862 struct net_device *bond_dev;
4867 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4868 name ? name : "bond%d",
4869 bond_setup, tx_queues);
4871 pr_err("%s: eek! can't alloc netdev!\n", name);
4876 dev_net_set(bond_dev, net);
4877 bond_dev->rtnl_link_ops = &bond_link_ops;
4879 res = register_netdevice(bond_dev);
4881 netif_carrier_off(bond_dev);
4885 bond_destructor(bond_dev);
4889 static int __net_init bond_net_init(struct net *net)
4891 struct bond_net *bn = net_generic(net, bond_net_id);
4894 INIT_LIST_HEAD(&bn->dev_list);
4896 bond_create_proc_dir(bn);
4897 bond_create_sysfs(bn);
4902 static void __net_exit bond_net_exit(struct net *net)
4904 struct bond_net *bn = net_generic(net, bond_net_id);
4906 bond_destroy_sysfs(bn);
4907 bond_destroy_proc_dir(bn);
4910 static struct pernet_operations bond_net_ops = {
4911 .init = bond_net_init,
4912 .exit = bond_net_exit,
4914 .size = sizeof(struct bond_net),
4917 static int __init bonding_init(void)
4922 pr_info("%s", bond_version);
4924 res = bond_check_params(&bonding_defaults);
4928 res = register_pernet_subsys(&bond_net_ops);
4932 res = rtnl_link_register(&bond_link_ops);
4936 bond_create_debugfs();
4938 for (i = 0; i < max_bonds; i++) {
4939 res = bond_create(&init_net, NULL);
4944 register_netdevice_notifier(&bond_netdev_notifier);
4948 rtnl_link_unregister(&bond_link_ops);
4950 unregister_pernet_subsys(&bond_net_ops);
4955 static void __exit bonding_exit(void)
4957 unregister_netdevice_notifier(&bond_netdev_notifier);
4959 bond_destroy_debugfs();
4961 rtnl_link_unregister(&bond_link_ops);
4962 unregister_pernet_subsys(&bond_net_ops);
4964 #ifdef CONFIG_NET_POLL_CONTROLLER
4966 * Make sure we don't have an imbalance on our netpoll blocking
4968 WARN_ON(atomic_read(&netpoll_block_tx));
4972 module_init(bonding_init);
4973 module_exit(bonding_exit);
4974 MODULE_LICENSE("GPL");
4975 MODULE_VERSION(DRV_VERSION);
4976 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4977 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4978 MODULE_ALIAS_RTNL_LINK("bond");