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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/icmp.h>
45 #include <linux/icmpv6.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>
80 #include <linux/rculist.h>
81 #include <net/flow_dissector.h>
83 #include <net/bonding.h>
84 #include <net/bond_3ad.h>
85 #include <net/bond_alb.h>
87 #include "bonding_priv.h"
89 /*---------------------------- Module parameters ----------------------------*/
91 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_peer_notif = 1;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static int min_links;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *arp_all_targets;
111 static char *fail_over_mac;
112 static int all_slaves_active;
113 static struct bond_params bonding_defaults;
114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 static int packets_per_slave = 1;
116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
118 module_param(max_bonds, int, 0);
119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
120 module_param(tx_queues, int, 0);
121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
122 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
124 "failover event (alias of num_unsol_na)");
125 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
127 "failover event (alias of num_grat_arp)");
128 module_param(miimon, int, 0);
129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
130 module_param(updelay, int, 0);
131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
132 module_param(downdelay, int, 0);
133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
135 module_param(use_carrier, int, 0);
136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
137 "0 for off, 1 for on (default)");
138 module_param(mode, charp, 0);
139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
140 "1 for active-backup, 2 for balance-xor, "
141 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
142 "6 for balance-alb");
143 module_param(primary, charp, 0);
144 MODULE_PARM_DESC(primary, "Primary network device to use");
145 module_param(primary_reselect, charp, 0);
146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
148 "0 for always (default), "
149 "1 for only if speed of primary is "
151 "2 for only on active slave "
153 module_param(lacp_rate, charp, 0);
154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
155 "0 for slow, 1 for fast");
156 module_param(ad_select, charp, 0);
157 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
158 "0 for stable (default), 1 for bandwidth, "
160 module_param(min_links, int, 0);
161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
163 module_param(xmit_hash_policy, charp, 0);
164 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
165 "0 for layer 2 (default), 1 for layer 3+4, "
166 "2 for layer 2+3, 3 for encap layer 2+3, "
167 "4 for encap layer 3+4");
168 module_param(arp_interval, int, 0);
169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
170 module_param_array(arp_ip_target, charp, NULL, 0);
171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
172 module_param(arp_validate, charp, 0);
173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
174 "0 for none (default), 1 for active, "
175 "2 for backup, 3 for all");
176 module_param(arp_all_targets, charp, 0);
177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
178 module_param(fail_over_mac, charp, 0);
179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
180 "the same MAC; 0 for none (default), "
181 "1 for active, 2 for follow");
182 module_param(all_slaves_active, int, 0);
183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
184 "by setting active flag for all slaves; "
185 "0 for never (default), 1 for always.");
186 module_param(resend_igmp, int, 0);
187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
189 module_param(packets_per_slave, int, 0);
190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
191 "mode; 0 for a random slave, 1 packet per "
192 "slave (default), >1 packets per slave.");
193 module_param(lp_interval, uint, 0);
194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
195 "the bonding driver sends learning packets to "
196 "each slaves peer switch. The default is 1.");
198 /*----------------------------- Global variables ----------------------------*/
200 #ifdef CONFIG_NET_POLL_CONTROLLER
201 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
204 unsigned int bond_net_id __read_mostly;
206 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
208 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
209 .offset = offsetof(struct flow_keys, control),
212 .key_id = FLOW_DISSECTOR_KEY_BASIC,
213 .offset = offsetof(struct flow_keys, basic),
216 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
217 .offset = offsetof(struct flow_keys, addrs.v4addrs),
220 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
221 .offset = offsetof(struct flow_keys, addrs.v6addrs),
224 .key_id = FLOW_DISSECTOR_KEY_TIPC,
225 .offset = offsetof(struct flow_keys, addrs.tipckey),
228 .key_id = FLOW_DISSECTOR_KEY_PORTS,
229 .offset = offsetof(struct flow_keys, ports),
232 .key_id = FLOW_DISSECTOR_KEY_ICMP,
233 .offset = offsetof(struct flow_keys, icmp),
236 .key_id = FLOW_DISSECTOR_KEY_VLAN,
237 .offset = offsetof(struct flow_keys, vlan),
240 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
241 .offset = offsetof(struct flow_keys, tags),
244 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
245 .offset = offsetof(struct flow_keys, keyid),
249 static struct flow_dissector flow_keys_bonding __read_mostly;
251 /*-------------------------- Forward declarations ---------------------------*/
253 static int bond_init(struct net_device *bond_dev);
254 static void bond_uninit(struct net_device *bond_dev);
255 static void bond_get_stats(struct net_device *bond_dev,
256 struct rtnl_link_stats64 *stats);
257 static void bond_slave_arr_handler(struct work_struct *work);
258 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
260 static void bond_netdev_notify_work(struct work_struct *work);
262 /*---------------------------- General routines -----------------------------*/
264 const char *bond_mode_name(int mode)
266 static const char *names[] = {
267 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
268 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
269 [BOND_MODE_XOR] = "load balancing (xor)",
270 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
271 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
272 [BOND_MODE_TLB] = "transmit load balancing",
273 [BOND_MODE_ALB] = "adaptive load balancing",
276 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
283 * bond_dev_queue_xmit - Prepare skb for xmit.
285 * @bond: bond device that got this skb for tx.
286 * @skb: hw accel VLAN tagged skb to transmit
287 * @slave_dev: slave that is supposed to xmit this skbuff
289 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
290 struct net_device *slave_dev)
292 skb->dev = slave_dev;
294 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
295 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
296 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
298 if (unlikely(netpoll_tx_running(bond->dev)))
299 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
301 return dev_queue_xmit(skb);
304 /*---------------------------------- VLAN -----------------------------------*/
306 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
307 * We don't protect the slave list iteration with a lock because:
308 * a. This operation is performed in IOCTL context,
309 * b. The operation is protected by the RTNL semaphore in the 8021q code,
310 * c. Holding a lock with BH disabled while directly calling a base driver
311 * entry point is generally a BAD idea.
313 * The design of synchronization/protection for this operation in the 8021q
314 * module is good for one or more VLAN devices over a single physical device
315 * and cannot be extended for a teaming solution like bonding, so there is a
316 * potential race condition here where a net device from the vlan group might
317 * be referenced (either by a base driver or the 8021q code) while it is being
318 * removed from the system. However, it turns out we're not making matters
319 * worse, and if it works for regular VLAN usage it will work here too.
323 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
324 * @bond_dev: bonding net device that got called
325 * @proto: network protocol ID
326 * @vid: vlan id being added
328 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
329 __be16 proto, u16 vid)
331 struct bonding *bond = netdev_priv(bond_dev);
332 struct slave *slave, *rollback_slave;
333 struct list_head *iter;
336 bond_for_each_slave(bond, slave, iter) {
337 res = vlan_vid_add(slave->dev, proto, vid);
345 /* unwind to the slave that failed */
346 bond_for_each_slave(bond, rollback_slave, iter) {
347 if (rollback_slave == slave)
350 vlan_vid_del(rollback_slave->dev, proto, vid);
357 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
358 * @bond_dev: bonding net device that got called
359 * @proto: network protocol ID
360 * @vid: vlan id being removed
362 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
363 __be16 proto, u16 vid)
365 struct bonding *bond = netdev_priv(bond_dev);
366 struct list_head *iter;
369 bond_for_each_slave(bond, slave, iter)
370 vlan_vid_del(slave->dev, proto, vid);
372 if (bond_is_lb(bond))
373 bond_alb_clear_vlan(bond, vid);
378 /*---------------------------------- XFRM -----------------------------------*/
380 #ifdef CONFIG_XFRM_OFFLOAD
382 * bond_ipsec_add_sa - program device with a security association
383 * @xs: pointer to transformer state struct
385 static int bond_ipsec_add_sa(struct xfrm_state *xs)
387 struct net_device *bond_dev = xs->xso.dev;
388 struct bond_ipsec *ipsec;
389 struct bonding *bond;
397 bond = netdev_priv(bond_dev);
398 slave = rcu_dereference(bond->curr_active_slave);
404 if (!slave->dev->xfrmdev_ops ||
405 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
406 netif_is_bond_master(slave->dev)) {
407 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
412 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
417 xs->xso.real_dev = slave->dev;
419 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
422 INIT_LIST_HEAD(&ipsec->list);
423 spin_lock_bh(&bond->ipsec_lock);
424 list_add(&ipsec->list, &bond->ipsec_list);
425 spin_unlock_bh(&bond->ipsec_lock);
433 static void bond_ipsec_add_sa_all(struct bonding *bond)
435 struct net_device *bond_dev = bond->dev;
436 struct bond_ipsec *ipsec;
440 slave = rcu_dereference(bond->curr_active_slave);
444 if (!slave->dev->xfrmdev_ops ||
445 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
446 netif_is_bond_master(slave->dev)) {
447 spin_lock_bh(&bond->ipsec_lock);
448 if (!list_empty(&bond->ipsec_list))
449 slave_warn(bond_dev, slave->dev,
450 "%s: no slave xdo_dev_state_add\n",
452 spin_unlock_bh(&bond->ipsec_lock);
456 spin_lock_bh(&bond->ipsec_lock);
457 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
458 ipsec->xs->xso.real_dev = slave->dev;
459 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
460 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
461 ipsec->xs->xso.real_dev = NULL;
464 spin_unlock_bh(&bond->ipsec_lock);
470 * bond_ipsec_del_sa - clear out this specific SA
471 * @xs: pointer to transformer state struct
473 static void bond_ipsec_del_sa(struct xfrm_state *xs)
475 struct net_device *bond_dev = xs->xso.dev;
476 struct bond_ipsec *ipsec;
477 struct bonding *bond;
484 bond = netdev_priv(bond_dev);
485 slave = rcu_dereference(bond->curr_active_slave);
490 if (!xs->xso.real_dev)
493 WARN_ON(xs->xso.real_dev != slave->dev);
495 if (!slave->dev->xfrmdev_ops ||
496 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
497 netif_is_bond_master(slave->dev)) {
498 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
502 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
504 spin_lock_bh(&bond->ipsec_lock);
505 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
506 if (ipsec->xs == xs) {
507 list_del(&ipsec->list);
512 spin_unlock_bh(&bond->ipsec_lock);
516 static void bond_ipsec_del_sa_all(struct bonding *bond)
518 struct net_device *bond_dev = bond->dev;
519 struct bond_ipsec *ipsec;
523 slave = rcu_dereference(bond->curr_active_slave);
529 spin_lock_bh(&bond->ipsec_lock);
530 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
531 if (!ipsec->xs->xso.real_dev)
534 if (!slave->dev->xfrmdev_ops ||
535 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
536 netif_is_bond_master(slave->dev)) {
537 slave_warn(bond_dev, slave->dev,
538 "%s: no slave xdo_dev_state_delete\n",
541 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
543 ipsec->xs->xso.real_dev = NULL;
545 spin_unlock_bh(&bond->ipsec_lock);
550 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
551 * @skb: current data packet
552 * @xs: pointer to transformer state struct
554 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
556 struct net_device *bond_dev = xs->xso.dev;
557 struct net_device *real_dev;
558 struct slave *curr_active;
559 struct bonding *bond;
562 bond = netdev_priv(bond_dev);
564 curr_active = rcu_dereference(bond->curr_active_slave);
565 real_dev = curr_active->dev;
567 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
572 if (!xs->xso.real_dev) {
577 if (!real_dev->xfrmdev_ops ||
578 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
579 netif_is_bond_master(real_dev)) {
584 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
590 static const struct xfrmdev_ops bond_xfrmdev_ops = {
591 .xdo_dev_state_add = bond_ipsec_add_sa,
592 .xdo_dev_state_delete = bond_ipsec_del_sa,
593 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
595 #endif /* CONFIG_XFRM_OFFLOAD */
597 /*------------------------------- Link status -------------------------------*/
599 /* Set the carrier state for the master according to the state of its
600 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
601 * do special 802.3ad magic.
603 * Returns zero if carrier state does not change, nonzero if it does.
605 int bond_set_carrier(struct bonding *bond)
607 struct list_head *iter;
610 if (!bond_has_slaves(bond))
613 if (BOND_MODE(bond) == BOND_MODE_8023AD)
614 return bond_3ad_set_carrier(bond);
616 bond_for_each_slave(bond, slave, iter) {
617 if (slave->link == BOND_LINK_UP) {
618 if (!netif_carrier_ok(bond->dev)) {
619 netif_carrier_on(bond->dev);
627 if (netif_carrier_ok(bond->dev)) {
628 netif_carrier_off(bond->dev);
634 /* Get link speed and duplex from the slave's base driver
635 * using ethtool. If for some reason the call fails or the
636 * values are invalid, set speed and duplex to -1,
637 * and return. Return 1 if speed or duplex settings are
638 * UNKNOWN; 0 otherwise.
640 static int bond_update_speed_duplex(struct slave *slave)
642 struct net_device *slave_dev = slave->dev;
643 struct ethtool_link_ksettings ecmd;
646 slave->speed = SPEED_UNKNOWN;
647 slave->duplex = DUPLEX_UNKNOWN;
649 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
652 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
654 switch (ecmd.base.duplex) {
662 slave->speed = ecmd.base.speed;
663 slave->duplex = ecmd.base.duplex;
668 const char *bond_slave_link_status(s8 link)
684 /* if <dev> supports MII link status reporting, check its link status.
686 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
687 * depending upon the setting of the use_carrier parameter.
689 * Return either BMSR_LSTATUS, meaning that the link is up (or we
690 * can't tell and just pretend it is), or 0, meaning that the link is
693 * If reporting is non-zero, instead of faking link up, return -1 if
694 * both ETHTOOL and MII ioctls fail (meaning the device does not
695 * support them). If use_carrier is set, return whatever it says.
696 * It'd be nice if there was a good way to tell if a driver supports
697 * netif_carrier, but there really isn't.
699 static int bond_check_dev_link(struct bonding *bond,
700 struct net_device *slave_dev, int reporting)
702 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
703 int (*ioctl)(struct net_device *, struct ifreq *, int);
705 struct mii_ioctl_data *mii;
707 if (!reporting && !netif_running(slave_dev))
710 if (bond->params.use_carrier)
711 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
713 /* Try to get link status using Ethtool first. */
714 if (slave_dev->ethtool_ops->get_link)
715 return slave_dev->ethtool_ops->get_link(slave_dev) ?
718 /* Ethtool can't be used, fallback to MII ioctls. */
719 ioctl = slave_ops->ndo_do_ioctl;
721 /* TODO: set pointer to correct ioctl on a per team member
722 * bases to make this more efficient. that is, once
723 * we determine the correct ioctl, we will always
724 * call it and not the others for that team
728 /* We cannot assume that SIOCGMIIPHY will also read a
729 * register; not all network drivers (e.g., e100)
733 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
734 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
736 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
737 mii->reg_num = MII_BMSR;
738 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
739 return mii->val_out & BMSR_LSTATUS;
743 /* If reporting, report that either there's no dev->do_ioctl,
744 * or both SIOCGMIIREG and get_link failed (meaning that we
745 * cannot report link status). If not reporting, pretend
748 return reporting ? -1 : BMSR_LSTATUS;
751 /*----------------------------- Multicast list ------------------------------*/
753 /* Push the promiscuity flag down to appropriate slaves */
754 static int bond_set_promiscuity(struct bonding *bond, int inc)
756 struct list_head *iter;
759 if (bond_uses_primary(bond)) {
760 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
763 err = dev_set_promiscuity(curr_active->dev, inc);
767 bond_for_each_slave(bond, slave, iter) {
768 err = dev_set_promiscuity(slave->dev, inc);
776 /* Push the allmulti flag down to all slaves */
777 static int bond_set_allmulti(struct bonding *bond, int inc)
779 struct list_head *iter;
782 if (bond_uses_primary(bond)) {
783 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
786 err = dev_set_allmulti(curr_active->dev, inc);
790 bond_for_each_slave(bond, slave, iter) {
791 err = dev_set_allmulti(slave->dev, inc);
799 /* Retrieve the list of registered multicast addresses for the bonding
800 * device and retransmit an IGMP JOIN request to the current active
803 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
805 struct bonding *bond = container_of(work, struct bonding,
808 if (!rtnl_trylock()) {
809 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
812 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
814 if (bond->igmp_retrans > 1) {
815 bond->igmp_retrans--;
816 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
821 /* Flush bond's hardware addresses from slave */
822 static void bond_hw_addr_flush(struct net_device *bond_dev,
823 struct net_device *slave_dev)
825 struct bonding *bond = netdev_priv(bond_dev);
827 dev_uc_unsync(slave_dev, bond_dev);
828 dev_mc_unsync(slave_dev, bond_dev);
830 if (BOND_MODE(bond) == BOND_MODE_8023AD)
831 dev_mc_del(slave_dev, lacpdu_mcast_addr);
834 /*--------------------------- Active slave change ---------------------------*/
836 /* Update the hardware address list and promisc/allmulti for the new and
837 * old active slaves (if any). Modes that are not using primary keep all
838 * slaves up date at all times; only the modes that use primary need to call
839 * this function to swap these settings during a failover.
841 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
842 struct slave *old_active)
845 if (bond->dev->flags & IFF_PROMISC)
846 dev_set_promiscuity(old_active->dev, -1);
848 if (bond->dev->flags & IFF_ALLMULTI)
849 dev_set_allmulti(old_active->dev, -1);
851 if (bond->dev->flags & IFF_UP)
852 bond_hw_addr_flush(bond->dev, old_active->dev);
856 /* FIXME: Signal errors upstream. */
857 if (bond->dev->flags & IFF_PROMISC)
858 dev_set_promiscuity(new_active->dev, 1);
860 if (bond->dev->flags & IFF_ALLMULTI)
861 dev_set_allmulti(new_active->dev, 1);
863 if (bond->dev->flags & IFF_UP) {
864 netif_addr_lock_bh(bond->dev);
865 dev_uc_sync(new_active->dev, bond->dev);
866 dev_mc_sync(new_active->dev, bond->dev);
867 netif_addr_unlock_bh(bond->dev);
873 * bond_set_dev_addr - clone slave's address to bond
874 * @bond_dev: bond net device
875 * @slave_dev: slave net device
877 * Should be called with RTNL held.
879 static int bond_set_dev_addr(struct net_device *bond_dev,
880 struct net_device *slave_dev)
884 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
885 bond_dev, slave_dev, slave_dev->addr_len);
886 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
890 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
891 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
892 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
896 static struct slave *bond_get_old_active(struct bonding *bond,
897 struct slave *new_active)
900 struct list_head *iter;
902 bond_for_each_slave(bond, slave, iter) {
903 if (slave == new_active)
906 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
913 /* bond_do_fail_over_mac
915 * Perform special MAC address swapping for fail_over_mac settings
919 static void bond_do_fail_over_mac(struct bonding *bond,
920 struct slave *new_active,
921 struct slave *old_active)
923 u8 tmp_mac[MAX_ADDR_LEN];
924 struct sockaddr_storage ss;
927 switch (bond->params.fail_over_mac) {
928 case BOND_FOM_ACTIVE:
930 rv = bond_set_dev_addr(bond->dev, new_active->dev);
932 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
936 case BOND_FOM_FOLLOW:
937 /* if new_active && old_active, swap them
938 * if just old_active, do nothing (going to no active slave)
939 * if just new_active, set new_active to bond's MAC
945 old_active = bond_get_old_active(bond, new_active);
948 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
949 new_active->dev->addr_len);
950 bond_hw_addr_copy(ss.__data,
951 old_active->dev->dev_addr,
952 old_active->dev->addr_len);
953 ss.ss_family = new_active->dev->type;
955 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
956 bond->dev->addr_len);
957 ss.ss_family = bond->dev->type;
960 rv = dev_set_mac_address(new_active->dev,
961 (struct sockaddr *)&ss, NULL);
963 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
971 bond_hw_addr_copy(ss.__data, tmp_mac,
972 new_active->dev->addr_len);
973 ss.ss_family = old_active->dev->type;
975 rv = dev_set_mac_address(old_active->dev,
976 (struct sockaddr *)&ss, NULL);
978 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
983 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
984 bond->params.fail_over_mac);
990 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
992 struct slave *prim = rtnl_dereference(bond->primary_slave);
993 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
995 if (!prim || prim->link != BOND_LINK_UP) {
996 if (!curr || curr->link != BOND_LINK_UP)
1001 if (bond->force_primary) {
1002 bond->force_primary = false;
1006 if (!curr || curr->link != BOND_LINK_UP)
1009 /* At this point, prim and curr are both up */
1010 switch (bond->params.primary_reselect) {
1011 case BOND_PRI_RESELECT_ALWAYS:
1013 case BOND_PRI_RESELECT_BETTER:
1014 if (prim->speed < curr->speed)
1016 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1019 case BOND_PRI_RESELECT_FAILURE:
1022 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1023 bond->params.primary_reselect);
1029 * bond_find_best_slave - select the best available slave to be the active one
1030 * @bond: our bonding struct
1032 static struct slave *bond_find_best_slave(struct bonding *bond)
1034 struct slave *slave, *bestslave = NULL;
1035 struct list_head *iter;
1036 int mintime = bond->params.updelay;
1038 slave = bond_choose_primary_or_current(bond);
1042 bond_for_each_slave(bond, slave, iter) {
1043 if (slave->link == BOND_LINK_UP)
1045 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1046 slave->delay < mintime) {
1047 mintime = slave->delay;
1055 static bool bond_should_notify_peers(struct bonding *bond)
1057 struct slave *slave;
1060 slave = rcu_dereference(bond->curr_active_slave);
1063 if (!slave || !bond->send_peer_notif ||
1064 bond->send_peer_notif %
1065 max(1, bond->params.peer_notif_delay) != 0 ||
1066 !netif_carrier_ok(bond->dev) ||
1067 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1070 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1071 slave ? slave->dev->name : "NULL");
1077 * change_active_interface - change the active slave into the specified one
1078 * @bond: our bonding struct
1079 * @new_active: the new slave to make the active one
1081 * Set the new slave to the bond's settings and unset them on the old
1082 * curr_active_slave.
1083 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1085 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1086 * because it is apparently the best available slave we have, even though its
1087 * updelay hasn't timed out yet.
1089 * Caller must hold RTNL.
1091 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1093 struct slave *old_active;
1097 old_active = rtnl_dereference(bond->curr_active_slave);
1099 if (old_active == new_active)
1102 #ifdef CONFIG_XFRM_OFFLOAD
1103 bond_ipsec_del_sa_all(bond);
1104 #endif /* CONFIG_XFRM_OFFLOAD */
1107 new_active->last_link_up = jiffies;
1109 if (new_active->link == BOND_LINK_BACK) {
1110 if (bond_uses_primary(bond)) {
1111 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1112 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1115 new_active->delay = 0;
1116 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1117 BOND_SLAVE_NOTIFY_NOW);
1119 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1120 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1122 if (bond_is_lb(bond))
1123 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1125 if (bond_uses_primary(bond)) {
1126 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1131 if (bond_uses_primary(bond))
1132 bond_hw_addr_swap(bond, new_active, old_active);
1134 if (bond_is_lb(bond)) {
1135 bond_alb_handle_active_change(bond, new_active);
1137 bond_set_slave_inactive_flags(old_active,
1138 BOND_SLAVE_NOTIFY_NOW);
1140 bond_set_slave_active_flags(new_active,
1141 BOND_SLAVE_NOTIFY_NOW);
1143 rcu_assign_pointer(bond->curr_active_slave, new_active);
1146 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1148 bond_set_slave_inactive_flags(old_active,
1149 BOND_SLAVE_NOTIFY_NOW);
1152 bool should_notify_peers = false;
1154 bond_set_slave_active_flags(new_active,
1155 BOND_SLAVE_NOTIFY_NOW);
1157 if (bond->params.fail_over_mac)
1158 bond_do_fail_over_mac(bond, new_active,
1161 if (netif_running(bond->dev)) {
1162 bond->send_peer_notif =
1163 bond->params.num_peer_notif *
1164 max(1, bond->params.peer_notif_delay);
1165 should_notify_peers =
1166 bond_should_notify_peers(bond);
1169 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1170 if (should_notify_peers) {
1171 bond->send_peer_notif--;
1172 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1178 #ifdef CONFIG_XFRM_OFFLOAD
1179 bond_ipsec_add_sa_all(bond);
1180 #endif /* CONFIG_XFRM_OFFLOAD */
1182 /* resend IGMP joins since active slave has changed or
1183 * all were sent on curr_active_slave.
1184 * resend only if bond is brought up with the affected
1185 * bonding modes and the retransmission is enabled
1187 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1188 ((bond_uses_primary(bond) && new_active) ||
1189 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1190 bond->igmp_retrans = bond->params.resend_igmp;
1191 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1196 * bond_select_active_slave - select a new active slave, if needed
1197 * @bond: our bonding struct
1199 * This functions should be called when one of the following occurs:
1200 * - The old curr_active_slave has been released or lost its link.
1201 * - The primary_slave has got its link back.
1202 * - A slave has got its link back and there's no old curr_active_slave.
1204 * Caller must hold RTNL.
1206 void bond_select_active_slave(struct bonding *bond)
1208 struct slave *best_slave;
1213 best_slave = bond_find_best_slave(bond);
1214 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1215 bond_change_active_slave(bond, best_slave);
1216 rv = bond_set_carrier(bond);
1220 if (netif_carrier_ok(bond->dev))
1221 netdev_info(bond->dev, "active interface up!\n");
1223 netdev_info(bond->dev, "now running without any active interface!\n");
1227 #ifdef CONFIG_NET_POLL_CONTROLLER
1228 static inline int slave_enable_netpoll(struct slave *slave)
1233 np = kzalloc(sizeof(*np), GFP_KERNEL);
1238 err = __netpoll_setup(np, slave->dev);
1247 static inline void slave_disable_netpoll(struct slave *slave)
1249 struct netpoll *np = slave->np;
1259 static void bond_poll_controller(struct net_device *bond_dev)
1261 struct bonding *bond = netdev_priv(bond_dev);
1262 struct slave *slave = NULL;
1263 struct list_head *iter;
1264 struct ad_info ad_info;
1266 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1267 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1270 bond_for_each_slave_rcu(bond, slave, iter) {
1271 if (!bond_slave_is_up(slave))
1274 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1275 struct aggregator *agg =
1276 SLAVE_AD_INFO(slave)->port.aggregator;
1279 agg->aggregator_identifier != ad_info.aggregator_id)
1283 netpoll_poll_dev(slave->dev);
1287 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1289 struct bonding *bond = netdev_priv(bond_dev);
1290 struct list_head *iter;
1291 struct slave *slave;
1293 bond_for_each_slave(bond, slave, iter)
1294 if (bond_slave_is_up(slave))
1295 slave_disable_netpoll(slave);
1298 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1300 struct bonding *bond = netdev_priv(dev);
1301 struct list_head *iter;
1302 struct slave *slave;
1305 bond_for_each_slave(bond, slave, iter) {
1306 err = slave_enable_netpoll(slave);
1308 bond_netpoll_cleanup(dev);
1315 static inline int slave_enable_netpoll(struct slave *slave)
1319 static inline void slave_disable_netpoll(struct slave *slave)
1322 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1327 /*---------------------------------- IOCTL ----------------------------------*/
1329 static netdev_features_t bond_fix_features(struct net_device *dev,
1330 netdev_features_t features)
1332 struct bonding *bond = netdev_priv(dev);
1333 struct list_head *iter;
1334 netdev_features_t mask;
1335 struct slave *slave;
1339 features &= ~NETIF_F_ONE_FOR_ALL;
1340 features |= NETIF_F_ALL_FOR_ALL;
1342 bond_for_each_slave(bond, slave, iter) {
1343 features = netdev_increment_features(features,
1344 slave->dev->features,
1347 features = netdev_add_tso_features(features, mask);
1352 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1353 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1354 NETIF_F_HIGHDMA | NETIF_F_LRO)
1356 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1357 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1359 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1363 static void bond_compute_features(struct bonding *bond)
1365 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1366 IFF_XMIT_DST_RELEASE_PERM;
1367 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1368 netdev_features_t enc_features = BOND_ENC_FEATURES;
1369 #ifdef CONFIG_XFRM_OFFLOAD
1370 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1371 #endif /* CONFIG_XFRM_OFFLOAD */
1372 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1373 struct net_device *bond_dev = bond->dev;
1374 struct list_head *iter;
1375 struct slave *slave;
1376 unsigned short max_hard_header_len = ETH_HLEN;
1377 unsigned int gso_max_size = GSO_MAX_SIZE;
1378 u16 gso_max_segs = GSO_MAX_SEGS;
1380 if (!bond_has_slaves(bond))
1382 vlan_features &= NETIF_F_ALL_FOR_ALL;
1383 mpls_features &= NETIF_F_ALL_FOR_ALL;
1385 bond_for_each_slave(bond, slave, iter) {
1386 vlan_features = netdev_increment_features(vlan_features,
1387 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1389 enc_features = netdev_increment_features(enc_features,
1390 slave->dev->hw_enc_features,
1393 #ifdef CONFIG_XFRM_OFFLOAD
1394 xfrm_features = netdev_increment_features(xfrm_features,
1395 slave->dev->hw_enc_features,
1396 BOND_XFRM_FEATURES);
1397 #endif /* CONFIG_XFRM_OFFLOAD */
1399 mpls_features = netdev_increment_features(mpls_features,
1400 slave->dev->mpls_features,
1401 BOND_MPLS_FEATURES);
1403 dst_release_flag &= slave->dev->priv_flags;
1404 if (slave->dev->hard_header_len > max_hard_header_len)
1405 max_hard_header_len = slave->dev->hard_header_len;
1407 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1408 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1410 bond_dev->hard_header_len = max_hard_header_len;
1413 bond_dev->vlan_features = vlan_features;
1414 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1415 NETIF_F_HW_VLAN_CTAG_TX |
1416 NETIF_F_HW_VLAN_STAG_TX |
1418 #ifdef CONFIG_XFRM_OFFLOAD
1419 bond_dev->hw_enc_features |= xfrm_features;
1420 #endif /* CONFIG_XFRM_OFFLOAD */
1421 bond_dev->mpls_features = mpls_features;
1422 bond_dev->gso_max_segs = gso_max_segs;
1423 netif_set_gso_max_size(bond_dev, gso_max_size);
1425 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1426 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1427 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1428 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1430 netdev_change_features(bond_dev);
1433 static void bond_setup_by_slave(struct net_device *bond_dev,
1434 struct net_device *slave_dev)
1436 bool was_up = !!(bond_dev->flags & IFF_UP);
1438 dev_close(bond_dev);
1440 bond_dev->header_ops = slave_dev->header_ops;
1442 bond_dev->type = slave_dev->type;
1443 bond_dev->hard_header_len = slave_dev->hard_header_len;
1444 bond_dev->needed_headroom = slave_dev->needed_headroom;
1445 bond_dev->addr_len = slave_dev->addr_len;
1447 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1448 slave_dev->addr_len);
1450 if (slave_dev->flags & IFF_POINTOPOINT) {
1451 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1452 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1455 dev_open(bond_dev, NULL);
1458 /* On bonding slaves other than the currently active slave, suppress
1459 * duplicates except for alb non-mcast/bcast.
1461 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1462 struct slave *slave,
1463 struct bonding *bond)
1465 if (bond_is_slave_inactive(slave)) {
1466 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1467 skb->pkt_type != PACKET_BROADCAST &&
1468 skb->pkt_type != PACKET_MULTICAST)
1475 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1477 struct sk_buff *skb = *pskb;
1478 struct slave *slave;
1479 struct bonding *bond;
1480 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1482 int ret = RX_HANDLER_ANOTHER;
1484 skb = skb_share_check(skb, GFP_ATOMIC);
1486 return RX_HANDLER_CONSUMED;
1490 slave = bond_slave_get_rcu(skb->dev);
1493 recv_probe = READ_ONCE(bond->recv_probe);
1495 ret = recv_probe(skb, bond, slave);
1496 if (ret == RX_HANDLER_CONSUMED) {
1503 * For packets determined by bond_should_deliver_exact_match() call to
1504 * be suppressed we want to make an exception for link-local packets.
1505 * This is necessary for e.g. LLDP daemons to be able to monitor
1506 * inactive slave links without being forced to bind to them
1509 * At the same time, packets that are passed to the bonding master
1510 * (including link-local ones) can have their originating interface
1511 * determined via PACKET_ORIGDEV socket option.
1513 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1514 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1515 return RX_HANDLER_PASS;
1516 return RX_HANDLER_EXACT;
1519 skb->dev = bond->dev;
1521 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1522 netif_is_bridge_port(bond->dev) &&
1523 skb->pkt_type == PACKET_HOST) {
1525 if (unlikely(skb_cow_head(skb,
1526 skb->data - skb_mac_header(skb)))) {
1528 return RX_HANDLER_CONSUMED;
1530 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1531 bond->dev->addr_len);
1537 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1539 switch (BOND_MODE(bond)) {
1540 case BOND_MODE_ROUNDROBIN:
1541 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1542 case BOND_MODE_ACTIVEBACKUP:
1543 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1544 case BOND_MODE_BROADCAST:
1545 return NETDEV_LAG_TX_TYPE_BROADCAST;
1547 case BOND_MODE_8023AD:
1548 return NETDEV_LAG_TX_TYPE_HASH;
1550 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1554 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1555 enum netdev_lag_tx_type type)
1557 if (type != NETDEV_LAG_TX_TYPE_HASH)
1558 return NETDEV_LAG_HASH_NONE;
1560 switch (bond->params.xmit_policy) {
1561 case BOND_XMIT_POLICY_LAYER2:
1562 return NETDEV_LAG_HASH_L2;
1563 case BOND_XMIT_POLICY_LAYER34:
1564 return NETDEV_LAG_HASH_L34;
1565 case BOND_XMIT_POLICY_LAYER23:
1566 return NETDEV_LAG_HASH_L23;
1567 case BOND_XMIT_POLICY_ENCAP23:
1568 return NETDEV_LAG_HASH_E23;
1569 case BOND_XMIT_POLICY_ENCAP34:
1570 return NETDEV_LAG_HASH_E34;
1572 return NETDEV_LAG_HASH_UNKNOWN;
1576 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1577 struct netlink_ext_ack *extack)
1579 struct netdev_lag_upper_info lag_upper_info;
1580 enum netdev_lag_tx_type type;
1582 type = bond_lag_tx_type(bond);
1583 lag_upper_info.tx_type = type;
1584 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1586 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1587 &lag_upper_info, extack);
1590 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1592 netdev_upper_dev_unlink(slave->dev, bond->dev);
1593 slave->dev->flags &= ~IFF_SLAVE;
1596 static void slave_kobj_release(struct kobject *kobj)
1598 struct slave *slave = to_slave(kobj);
1599 struct bonding *bond = bond_get_bond_by_slave(slave);
1601 cancel_delayed_work_sync(&slave->notify_work);
1602 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1603 kfree(SLAVE_AD_INFO(slave));
1608 static struct kobj_type slave_ktype = {
1609 .release = slave_kobj_release,
1611 .sysfs_ops = &slave_sysfs_ops,
1615 static int bond_kobj_init(struct slave *slave)
1619 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1620 &(slave->dev->dev.kobj), "bonding_slave");
1622 kobject_put(&slave->kobj);
1627 static struct slave *bond_alloc_slave(struct bonding *bond,
1628 struct net_device *slave_dev)
1630 struct slave *slave = NULL;
1632 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1637 slave->dev = slave_dev;
1638 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1640 if (bond_kobj_init(slave))
1643 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1644 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1646 if (!SLAVE_AD_INFO(slave)) {
1647 kobject_put(&slave->kobj);
1655 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1657 info->bond_mode = BOND_MODE(bond);
1658 info->miimon = bond->params.miimon;
1659 info->num_slaves = bond->slave_cnt;
1662 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1664 strcpy(info->slave_name, slave->dev->name);
1665 info->link = slave->link;
1666 info->state = bond_slave_state(slave);
1667 info->link_failure_count = slave->link_failure_count;
1670 static void bond_netdev_notify_work(struct work_struct *_work)
1672 struct slave *slave = container_of(_work, struct slave,
1675 if (rtnl_trylock()) {
1676 struct netdev_bonding_info binfo;
1678 bond_fill_ifslave(slave, &binfo.slave);
1679 bond_fill_ifbond(slave->bond, &binfo.master);
1680 netdev_bonding_info_change(slave->dev, &binfo);
1683 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1687 void bond_queue_slave_event(struct slave *slave)
1689 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1692 void bond_lower_state_changed(struct slave *slave)
1694 struct netdev_lag_lower_state_info info;
1696 info.link_up = slave->link == BOND_LINK_UP ||
1697 slave->link == BOND_LINK_FAIL;
1698 info.tx_enabled = bond_is_active_slave(slave);
1699 netdev_lower_state_changed(slave->dev, &info);
1702 /* enslave device <slave> to bond device <master> */
1703 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1704 struct netlink_ext_ack *extack)
1706 struct bonding *bond = netdev_priv(bond_dev);
1707 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1708 struct slave *new_slave = NULL, *prev_slave;
1709 struct sockaddr_storage ss;
1713 if (!bond->params.use_carrier &&
1714 slave_dev->ethtool_ops->get_link == NULL &&
1715 slave_ops->ndo_do_ioctl == NULL) {
1716 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1719 /* already in-use? */
1720 if (netdev_is_rx_handler_busy(slave_dev)) {
1721 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1722 slave_err(bond_dev, slave_dev,
1723 "Error: Device is in use and cannot be enslaved\n");
1727 if (bond_dev == slave_dev) {
1728 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1729 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1733 /* vlan challenged mutual exclusion */
1734 /* no need to lock since we're protected by rtnl_lock */
1735 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1736 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1737 if (vlan_uses_dev(bond_dev)) {
1738 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1739 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1742 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1745 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1748 if (slave_dev->features & NETIF_F_HW_ESP)
1749 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1751 /* Old ifenslave binaries are no longer supported. These can
1752 * be identified with moderate accuracy by the state of the slave:
1753 * the current ifenslave will set the interface down prior to
1754 * enslaving it; the old ifenslave will not.
1756 if (slave_dev->flags & IFF_UP) {
1757 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1758 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1762 /* set bonding device ether type by slave - bonding netdevices are
1763 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1764 * there is a need to override some of the type dependent attribs/funcs.
1766 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1767 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1769 if (!bond_has_slaves(bond)) {
1770 if (bond_dev->type != slave_dev->type) {
1771 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1772 bond_dev->type, slave_dev->type);
1774 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1776 res = notifier_to_errno(res);
1778 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1782 /* Flush unicast and multicast addresses */
1783 dev_uc_flush(bond_dev);
1784 dev_mc_flush(bond_dev);
1786 if (slave_dev->type != ARPHRD_ETHER)
1787 bond_setup_by_slave(bond_dev, slave_dev);
1789 ether_setup(bond_dev);
1790 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1793 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1796 } else if (bond_dev->type != slave_dev->type) {
1797 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1798 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1799 slave_dev->type, bond_dev->type);
1803 if (slave_dev->type == ARPHRD_INFINIBAND &&
1804 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1805 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1806 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1809 goto err_undo_flags;
1812 if (!slave_ops->ndo_set_mac_address ||
1813 slave_dev->type == ARPHRD_INFINIBAND) {
1814 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1815 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1816 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1817 if (!bond_has_slaves(bond)) {
1818 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1819 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1821 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1822 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1824 goto err_undo_flags;
1829 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1831 /* If this is the first slave, then we need to set the master's hardware
1832 * address to be the same as the slave's.
1834 if (!bond_has_slaves(bond) &&
1835 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1836 res = bond_set_dev_addr(bond->dev, slave_dev);
1838 goto err_undo_flags;
1841 new_slave = bond_alloc_slave(bond, slave_dev);
1844 goto err_undo_flags;
1847 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1848 * is set via sysfs or module option if desired.
1850 new_slave->queue_id = 0;
1852 /* Save slave's original mtu and then set it to match the bond */
1853 new_slave->original_mtu = slave_dev->mtu;
1854 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1856 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1860 /* Save slave's original ("permanent") mac address for modes
1861 * that need it, and for restoring it upon release, and then
1862 * set it to the master's address
1864 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1865 slave_dev->addr_len);
1867 if (!bond->params.fail_over_mac ||
1868 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1869 /* Set slave to master's mac address. The application already
1870 * set the master's mac address to that of the first slave
1872 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1873 ss.ss_family = slave_dev->type;
1874 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1877 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1878 goto err_restore_mtu;
1882 /* set slave flag before open to prevent IPv6 addrconf */
1883 slave_dev->flags |= IFF_SLAVE;
1885 /* open the slave since the application closed it */
1886 res = dev_open(slave_dev, extack);
1888 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1889 goto err_restore_mac;
1892 slave_dev->priv_flags |= IFF_BONDING;
1893 /* initialize slave stats */
1894 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1896 if (bond_is_lb(bond)) {
1897 /* bond_alb_init_slave() must be called before all other stages since
1898 * it might fail and we do not want to have to undo everything
1900 res = bond_alb_init_slave(bond, new_slave);
1905 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1907 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1911 prev_slave = bond_last_slave(bond);
1913 new_slave->delay = 0;
1914 new_slave->link_failure_count = 0;
1916 if (bond_update_speed_duplex(new_slave) &&
1917 bond_needs_speed_duplex(bond))
1918 new_slave->link = BOND_LINK_DOWN;
1920 new_slave->last_rx = jiffies -
1921 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1922 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1923 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1925 if (bond->params.miimon && !bond->params.use_carrier) {
1926 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1928 if ((link_reporting == -1) && !bond->params.arp_interval) {
1929 /* miimon is set but a bonded network driver
1930 * does not support ETHTOOL/MII and
1931 * arp_interval is not set. Note: if
1932 * use_carrier is enabled, we will never go
1933 * here (because netif_carrier is always
1934 * supported); thus, we don't need to change
1935 * the messages for netif_carrier.
1937 slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1938 } else if (link_reporting == -1) {
1939 /* unable get link status using mii/ethtool */
1940 slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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");
1944 /* check for initial state */
1945 new_slave->link = BOND_LINK_NOCHANGE;
1946 if (bond->params.miimon) {
1947 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1948 if (bond->params.updelay) {
1949 bond_set_slave_link_state(new_slave,
1951 BOND_SLAVE_NOTIFY_NOW);
1952 new_slave->delay = bond->params.updelay;
1954 bond_set_slave_link_state(new_slave,
1956 BOND_SLAVE_NOTIFY_NOW);
1959 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1960 BOND_SLAVE_NOTIFY_NOW);
1962 } else if (bond->params.arp_interval) {
1963 bond_set_slave_link_state(new_slave,
1964 (netif_carrier_ok(slave_dev) ?
1965 BOND_LINK_UP : BOND_LINK_DOWN),
1966 BOND_SLAVE_NOTIFY_NOW);
1968 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1969 BOND_SLAVE_NOTIFY_NOW);
1972 if (new_slave->link != BOND_LINK_DOWN)
1973 new_slave->last_link_up = jiffies;
1974 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1975 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1976 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1978 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1979 /* if there is a primary slave, remember it */
1980 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1981 rcu_assign_pointer(bond->primary_slave, new_slave);
1982 bond->force_primary = true;
1986 switch (BOND_MODE(bond)) {
1987 case BOND_MODE_ACTIVEBACKUP:
1988 bond_set_slave_inactive_flags(new_slave,
1989 BOND_SLAVE_NOTIFY_NOW);
1991 case BOND_MODE_8023AD:
1992 /* in 802.3ad mode, the internal mechanism
1993 * will activate the slaves in the selected
1996 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1997 /* if this is the first slave */
1999 SLAVE_AD_INFO(new_slave)->id = 1;
2000 /* Initialize AD with the number of times that the AD timer is called in 1 second
2001 * can be called only after the mac address of the bond is set
2003 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2005 SLAVE_AD_INFO(new_slave)->id =
2006 SLAVE_AD_INFO(prev_slave)->id + 1;
2009 bond_3ad_bind_slave(new_slave);
2013 bond_set_active_slave(new_slave);
2014 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2017 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2019 /* always active in trunk mode */
2020 bond_set_active_slave(new_slave);
2022 /* In trunking mode there is little meaning to curr_active_slave
2023 * anyway (it holds no special properties of the bond device),
2024 * so we can change it without calling change_active_interface()
2026 if (!rcu_access_pointer(bond->curr_active_slave) &&
2027 new_slave->link == BOND_LINK_UP)
2028 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2031 } /* switch(bond_mode) */
2033 #ifdef CONFIG_NET_POLL_CONTROLLER
2034 if (bond->dev->npinfo) {
2035 if (slave_enable_netpoll(new_slave)) {
2036 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2043 if (!(bond_dev->features & NETIF_F_LRO))
2044 dev_disable_lro(slave_dev);
2046 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2049 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2053 res = bond_master_upper_dev_link(bond, new_slave, extack);
2055 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2056 goto err_unregister;
2059 res = bond_sysfs_slave_add(new_slave);
2061 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2062 goto err_upper_unlink;
2065 /* If the mode uses primary, then the following is handled by
2066 * bond_change_active_slave().
2068 if (!bond_uses_primary(bond)) {
2069 /* set promiscuity level to new slave */
2070 if (bond_dev->flags & IFF_PROMISC) {
2071 res = dev_set_promiscuity(slave_dev, 1);
2076 /* set allmulti level to new slave */
2077 if (bond_dev->flags & IFF_ALLMULTI) {
2078 res = dev_set_allmulti(slave_dev, 1);
2080 if (bond_dev->flags & IFF_PROMISC)
2081 dev_set_promiscuity(slave_dev, -1);
2086 if (bond_dev->flags & IFF_UP) {
2087 netif_addr_lock_bh(bond_dev);
2088 dev_mc_sync_multiple(slave_dev, bond_dev);
2089 dev_uc_sync_multiple(slave_dev, bond_dev);
2090 netif_addr_unlock_bh(bond_dev);
2092 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2093 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2098 bond_compute_features(bond);
2099 bond_set_carrier(bond);
2101 if (bond_uses_primary(bond)) {
2103 bond_select_active_slave(bond);
2104 unblock_netpoll_tx();
2107 if (bond_mode_can_use_xmit_hash(bond))
2108 bond_update_slave_arr(bond, NULL);
2111 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2112 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2113 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2115 /* enslave is successful */
2116 bond_queue_slave_event(new_slave);
2119 /* Undo stages on error */
2121 bond_sysfs_slave_del(new_slave);
2124 bond_upper_dev_unlink(bond, new_slave);
2127 netdev_rx_handler_unregister(slave_dev);
2130 vlan_vids_del_by_dev(slave_dev, bond_dev);
2131 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2132 RCU_INIT_POINTER(bond->primary_slave, NULL);
2133 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2135 bond_change_active_slave(bond, NULL);
2136 bond_select_active_slave(bond);
2137 unblock_netpoll_tx();
2139 /* either primary_slave or curr_active_slave might've changed */
2141 slave_disable_netpoll(new_slave);
2144 if (!netif_is_bond_master(slave_dev))
2145 slave_dev->priv_flags &= ~IFF_BONDING;
2146 dev_close(slave_dev);
2149 slave_dev->flags &= ~IFF_SLAVE;
2150 if (!bond->params.fail_over_mac ||
2151 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2152 /* XXX TODO - fom follow mode needs to change master's
2153 * MAC if this slave's MAC is in use by the bond, or at
2154 * least print a warning.
2156 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2157 new_slave->dev->addr_len);
2158 ss.ss_family = slave_dev->type;
2159 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2163 dev_set_mtu(slave_dev, new_slave->original_mtu);
2166 kobject_put(&new_slave->kobj);
2169 /* Enslave of first slave has failed and we need to fix master's mac */
2170 if (!bond_has_slaves(bond)) {
2171 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2172 slave_dev->dev_addr))
2173 eth_hw_addr_random(bond_dev);
2174 if (bond_dev->type != ARPHRD_ETHER) {
2175 dev_close(bond_dev);
2176 ether_setup(bond_dev);
2177 bond_dev->flags |= IFF_MASTER;
2178 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2185 /* Try to release the slave device <slave> from the bond device <master>
2186 * It is legal to access curr_active_slave without a lock because all the function
2187 * is RTNL-locked. If "all" is true it means that the function is being called
2188 * while destroying a bond interface and all slaves are being released.
2190 * The rules for slave state should be:
2191 * for Active/Backup:
2192 * Active stays on all backups go down
2193 * for Bonded connections:
2194 * The first up interface should be left on and all others downed.
2196 static int __bond_release_one(struct net_device *bond_dev,
2197 struct net_device *slave_dev,
2198 bool all, bool unregister)
2200 struct bonding *bond = netdev_priv(bond_dev);
2201 struct slave *slave, *oldcurrent;
2202 struct sockaddr_storage ss;
2203 int old_flags = bond_dev->flags;
2204 netdev_features_t old_features = bond_dev->features;
2206 /* slave is not a slave or master is not master of this slave */
2207 if (!(slave_dev->flags & IFF_SLAVE) ||
2208 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2209 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2215 slave = bond_get_slave_by_dev(bond, slave_dev);
2217 /* not a slave of this bond */
2218 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2219 unblock_netpoll_tx();
2223 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2225 bond_sysfs_slave_del(slave);
2227 /* recompute stats just before removing the slave */
2228 bond_get_stats(bond->dev, &bond->bond_stats);
2230 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2231 * for this slave anymore.
2233 netdev_rx_handler_unregister(slave_dev);
2235 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2236 bond_3ad_unbind_slave(slave);
2238 bond_upper_dev_unlink(bond, slave);
2240 if (bond_mode_can_use_xmit_hash(bond))
2241 bond_update_slave_arr(bond, slave);
2243 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2244 bond_is_active_slave(slave) ? "active" : "backup");
2246 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2248 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2250 if (!all && (!bond->params.fail_over_mac ||
2251 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2252 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2253 bond_has_slaves(bond))
2254 slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
2255 slave->perm_hwaddr);
2258 if (rtnl_dereference(bond->primary_slave) == slave)
2259 RCU_INIT_POINTER(bond->primary_slave, NULL);
2261 if (oldcurrent == slave)
2262 bond_change_active_slave(bond, NULL);
2264 if (bond_is_lb(bond)) {
2265 /* Must be called only after the slave has been
2266 * detached from the list and the curr_active_slave
2267 * has been cleared (if our_slave == old_current),
2268 * but before a new active slave is selected.
2270 bond_alb_deinit_slave(bond, slave);
2274 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2275 } else if (oldcurrent == slave) {
2276 /* Note that we hold RTNL over this sequence, so there
2277 * is no concern that another slave add/remove event
2280 bond_select_active_slave(bond);
2283 bond_set_carrier(bond);
2284 if (!bond_has_slaves(bond))
2285 eth_hw_addr_random(bond_dev);
2287 unblock_netpoll_tx();
2291 if (!bond_has_slaves(bond)) {
2292 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2293 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2296 bond_compute_features(bond);
2297 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2298 (old_features & NETIF_F_VLAN_CHALLENGED))
2299 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2301 vlan_vids_del_by_dev(slave_dev, bond_dev);
2303 /* If the mode uses primary, then this case was handled above by
2304 * bond_change_active_slave(..., NULL)
2306 if (!bond_uses_primary(bond)) {
2307 /* unset promiscuity level from slave
2308 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2309 * of the IFF_PROMISC flag in the bond_dev, but we need the
2310 * value of that flag before that change, as that was the value
2311 * when this slave was attached, so we cache at the start of the
2312 * function and use it here. Same goes for ALLMULTI below
2314 if (old_flags & IFF_PROMISC)
2315 dev_set_promiscuity(slave_dev, -1);
2317 /* unset allmulti level from slave */
2318 if (old_flags & IFF_ALLMULTI)
2319 dev_set_allmulti(slave_dev, -1);
2321 if (old_flags & IFF_UP)
2322 bond_hw_addr_flush(bond_dev, slave_dev);
2325 slave_disable_netpoll(slave);
2327 /* close slave before restoring its mac address */
2328 dev_close(slave_dev);
2330 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2331 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2332 /* restore original ("permanent") mac address */
2333 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2334 slave->dev->addr_len);
2335 ss.ss_family = slave_dev->type;
2336 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2340 __dev_set_mtu(slave_dev, slave->original_mtu);
2342 dev_set_mtu(slave_dev, slave->original_mtu);
2344 if (!netif_is_bond_master(slave_dev))
2345 slave_dev->priv_flags &= ~IFF_BONDING;
2347 kobject_put(&slave->kobj);
2352 /* A wrapper used because of ndo_del_link */
2353 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2355 return __bond_release_one(bond_dev, slave_dev, false, false);
2358 /* First release a slave and then destroy the bond if no more slaves are left.
2359 * Must be under rtnl_lock when this function is called.
2361 static int bond_release_and_destroy(struct net_device *bond_dev,
2362 struct net_device *slave_dev)
2364 struct bonding *bond = netdev_priv(bond_dev);
2367 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2368 if (ret == 0 && !bond_has_slaves(bond) &&
2369 bond_dev->reg_state != NETREG_UNREGISTERING) {
2370 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2371 netdev_info(bond_dev, "Destroying bond\n");
2372 bond_remove_proc_entry(bond);
2373 unregister_netdevice(bond_dev);
2378 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2380 struct bonding *bond = netdev_priv(bond_dev);
2381 bond_fill_ifbond(bond, info);
2384 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2386 struct bonding *bond = netdev_priv(bond_dev);
2387 struct list_head *iter;
2388 int i = 0, res = -ENODEV;
2389 struct slave *slave;
2391 bond_for_each_slave(bond, slave, iter) {
2392 if (i++ == (int)info->slave_id) {
2394 bond_fill_ifslave(slave, info);
2402 /*-------------------------------- Monitoring -------------------------------*/
2404 /* called with rcu_read_lock() */
2405 static int bond_miimon_inspect(struct bonding *bond)
2407 bool ignore_updelay = false;
2408 int link_state, commit = 0;
2409 struct list_head *iter;
2410 struct slave *slave;
2412 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2413 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2415 struct bond_up_slave *usable_slaves;
2417 usable_slaves = rcu_dereference(bond->usable_slaves);
2419 if (usable_slaves && usable_slaves->count == 0)
2420 ignore_updelay = true;
2423 bond_for_each_slave_rcu(bond, slave, iter) {
2424 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2426 link_state = bond_check_dev_link(bond, slave->dev, 0);
2428 switch (slave->link) {
2433 bond_propose_link_state(slave, BOND_LINK_FAIL);
2435 slave->delay = bond->params.downdelay;
2437 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2439 BOND_MODE_ACTIVEBACKUP) ?
2440 (bond_is_active_slave(slave) ?
2441 "active " : "backup ") : "",
2442 bond->params.downdelay * bond->params.miimon);
2445 case BOND_LINK_FAIL:
2447 /* recovered before downdelay expired */
2448 bond_propose_link_state(slave, BOND_LINK_UP);
2449 slave->last_link_up = jiffies;
2450 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2451 (bond->params.downdelay - slave->delay) *
2452 bond->params.miimon);
2457 if (slave->delay <= 0) {
2458 bond_propose_link_state(slave, BOND_LINK_DOWN);
2466 case BOND_LINK_DOWN:
2470 bond_propose_link_state(slave, BOND_LINK_BACK);
2472 slave->delay = bond->params.updelay;
2475 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2476 ignore_updelay ? 0 :
2477 bond->params.updelay *
2478 bond->params.miimon);
2481 case BOND_LINK_BACK:
2483 bond_propose_link_state(slave, BOND_LINK_DOWN);
2484 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2485 (bond->params.updelay - slave->delay) *
2486 bond->params.miimon);
2494 if (slave->delay <= 0) {
2495 bond_propose_link_state(slave, BOND_LINK_UP);
2497 ignore_updelay = false;
2509 static void bond_miimon_link_change(struct bonding *bond,
2510 struct slave *slave,
2513 switch (BOND_MODE(bond)) {
2514 case BOND_MODE_8023AD:
2515 bond_3ad_handle_link_change(slave, link);
2519 bond_alb_handle_link_change(bond, slave, link);
2522 bond_update_slave_arr(bond, NULL);
2527 static void bond_miimon_commit(struct bonding *bond)
2529 struct list_head *iter;
2530 struct slave *slave, *primary;
2532 bond_for_each_slave(bond, slave, iter) {
2533 switch (slave->link_new_state) {
2534 case BOND_LINK_NOCHANGE:
2535 /* For 802.3ad mode, check current slave speed and
2536 * duplex again in case its port was disabled after
2537 * invalid speed/duplex reporting but recovered before
2538 * link monitoring could make a decision on the actual
2541 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2542 slave->link == BOND_LINK_UP)
2543 bond_3ad_adapter_speed_duplex_changed(slave);
2547 if (bond_update_speed_duplex(slave) &&
2548 bond_needs_speed_duplex(bond)) {
2549 slave->link = BOND_LINK_DOWN;
2550 if (net_ratelimit())
2551 slave_warn(bond->dev, slave->dev,
2552 "failed to get link speed/duplex\n");
2555 bond_set_slave_link_state(slave, BOND_LINK_UP,
2556 BOND_SLAVE_NOTIFY_NOW);
2557 slave->last_link_up = jiffies;
2559 primary = rtnl_dereference(bond->primary_slave);
2560 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2561 /* prevent it from being the active one */
2562 bond_set_backup_slave(slave);
2563 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2564 /* make it immediately active */
2565 bond_set_active_slave(slave);
2568 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2569 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2570 slave->duplex ? "full" : "half");
2572 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2574 if (!bond->curr_active_slave || slave == primary)
2579 case BOND_LINK_DOWN:
2580 if (slave->link_failure_count < UINT_MAX)
2581 slave->link_failure_count++;
2583 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2584 BOND_SLAVE_NOTIFY_NOW);
2586 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2587 BOND_MODE(bond) == BOND_MODE_8023AD)
2588 bond_set_slave_inactive_flags(slave,
2589 BOND_SLAVE_NOTIFY_NOW);
2591 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2593 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2595 if (slave == rcu_access_pointer(bond->curr_active_slave))
2601 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2602 slave->link_new_state);
2603 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2610 bond_select_active_slave(bond);
2611 unblock_netpoll_tx();
2614 bond_set_carrier(bond);
2619 * Really a wrapper that splits the mii monitor into two phases: an
2620 * inspection, then (if inspection indicates something needs to be done)
2621 * an acquisition of appropriate locks followed by a commit phase to
2622 * implement whatever link state changes are indicated.
2624 static void bond_mii_monitor(struct work_struct *work)
2626 struct bonding *bond = container_of(work, struct bonding,
2628 bool should_notify_peers = false;
2630 unsigned long delay;
2631 struct slave *slave;
2632 struct list_head *iter;
2634 delay = msecs_to_jiffies(bond->params.miimon);
2636 if (!bond_has_slaves(bond))
2640 should_notify_peers = bond_should_notify_peers(bond);
2641 commit = !!bond_miimon_inspect(bond);
2642 if (bond->send_peer_notif) {
2644 if (rtnl_trylock()) {
2645 bond->send_peer_notif--;
2653 /* Race avoidance with bond_close cancel of workqueue */
2654 if (!rtnl_trylock()) {
2656 should_notify_peers = false;
2660 bond_for_each_slave(bond, slave, iter) {
2661 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2663 bond_miimon_commit(bond);
2665 rtnl_unlock(); /* might sleep, hold no other locks */
2669 if (bond->params.miimon)
2670 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2672 if (should_notify_peers) {
2673 if (!rtnl_trylock())
2675 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2680 static int bond_upper_dev_walk(struct net_device *upper,
2681 struct netdev_nested_priv *priv)
2683 __be32 ip = *(__be32 *)priv->data;
2685 return ip == bond_confirm_addr(upper, 0, ip);
2688 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2690 struct netdev_nested_priv priv = {
2691 .data = (void *)&ip,
2695 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2699 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2706 /* We go to the (large) trouble of VLAN tagging ARP frames because
2707 * switches in VLAN mode (especially if ports are configured as
2708 * "native" to a VLAN) might not pass non-tagged frames.
2710 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2711 __be32 src_ip, struct bond_vlan_tag *tags)
2713 struct sk_buff *skb;
2714 struct bond_vlan_tag *outer_tag = tags;
2715 struct net_device *slave_dev = slave->dev;
2716 struct net_device *bond_dev = slave->bond->dev;
2718 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2719 arp_op, &dest_ip, &src_ip);
2721 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2722 NULL, slave_dev->dev_addr, NULL);
2725 net_err_ratelimited("ARP packet allocation failed\n");
2729 if (!tags || tags->vlan_proto == VLAN_N_VID)
2734 /* Go through all the tags backwards and add them to the packet */
2735 while (tags->vlan_proto != VLAN_N_VID) {
2736 if (!tags->vlan_id) {
2741 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2742 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2743 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2746 net_err_ratelimited("failed to insert inner VLAN tag\n");
2752 /* Set the outer tag */
2753 if (outer_tag->vlan_id) {
2754 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2755 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2756 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2757 outer_tag->vlan_id);
2764 /* Validate the device path between the @start_dev and the @end_dev.
2765 * The path is valid if the @end_dev is reachable through device
2767 * When the path is validated, collect any vlan information in the
2770 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2771 struct net_device *end_dev,
2774 struct bond_vlan_tag *tags;
2775 struct net_device *upper;
2776 struct list_head *iter;
2778 if (start_dev == end_dev) {
2779 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2781 return ERR_PTR(-ENOMEM);
2782 tags[level].vlan_proto = VLAN_N_VID;
2786 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2787 tags = bond_verify_device_path(upper, end_dev, level + 1);
2788 if (IS_ERR_OR_NULL(tags)) {
2793 if (is_vlan_dev(upper)) {
2794 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2795 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2804 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2807 struct bond_vlan_tag *tags;
2808 __be32 *targets = bond->params.arp_targets, addr;
2811 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2812 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2813 __func__, &targets[i]);
2816 /* Find out through which dev should the packet go */
2817 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2820 /* there's no route to target - try to send arp
2821 * probe to generate any traffic (arp_validate=0)
2823 if (bond->params.arp_validate)
2824 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2827 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2832 /* bond device itself */
2833 if (rt->dst.dev == bond->dev)
2837 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2840 if (!IS_ERR_OR_NULL(tags))
2843 /* Not our device - skip */
2844 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2845 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2851 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2853 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2858 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2862 if (!sip || !bond_has_this_ip(bond, tip)) {
2863 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2864 __func__, &sip, &tip);
2868 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2870 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2874 slave->last_rx = jiffies;
2875 slave->target_last_arp_rx[i] = jiffies;
2878 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2879 struct slave *slave)
2881 struct arphdr *arp = (struct arphdr *)skb->data;
2882 struct slave *curr_active_slave, *curr_arp_slave;
2883 unsigned char *arp_ptr;
2885 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2888 if (!slave_do_arp_validate(bond, slave)) {
2889 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2890 !slave_do_arp_validate_only(bond))
2891 slave->last_rx = jiffies;
2892 return RX_HANDLER_ANOTHER;
2893 } else if (!is_arp) {
2894 return RX_HANDLER_ANOTHER;
2897 alen = arp_hdr_len(bond->dev);
2899 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2900 __func__, skb->dev->name);
2902 if (alen > skb_headlen(skb)) {
2903 arp = kmalloc(alen, GFP_ATOMIC);
2906 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2910 if (arp->ar_hln != bond->dev->addr_len ||
2911 skb->pkt_type == PACKET_OTHERHOST ||
2912 skb->pkt_type == PACKET_LOOPBACK ||
2913 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2914 arp->ar_pro != htons(ETH_P_IP) ||
2918 arp_ptr = (unsigned char *)(arp + 1);
2919 arp_ptr += bond->dev->addr_len;
2920 memcpy(&sip, arp_ptr, 4);
2921 arp_ptr += 4 + bond->dev->addr_len;
2922 memcpy(&tip, arp_ptr, 4);
2924 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2925 __func__, slave->dev->name, bond_slave_state(slave),
2926 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2929 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2930 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2932 /* We 'trust' the received ARP enough to validate it if:
2934 * (a) the slave receiving the ARP is active (which includes the
2935 * current ARP slave, if any), or
2937 * (b) the receiving slave isn't active, but there is a currently
2938 * active slave and it received valid arp reply(s) after it became
2939 * the currently active slave, or
2941 * (c) there is an ARP slave that sent an ARP during the prior ARP
2942 * interval, and we receive an ARP reply on any slave. We accept
2943 * these because switch FDB update delays may deliver the ARP
2944 * reply to a slave other than the sender of the ARP request.
2946 * Note: for (b), backup slaves are receiving the broadcast ARP
2947 * request, not a reply. This request passes from the sending
2948 * slave through the L2 switch(es) to the receiving slave. Since
2949 * this is checking the request, sip/tip are swapped for
2952 * This is done to avoid endless looping when we can't reach the
2953 * arp_ip_target and fool ourselves with our own arp requests.
2955 if (bond_is_active_slave(slave))
2956 bond_validate_arp(bond, slave, sip, tip);
2957 else if (curr_active_slave &&
2958 time_after(slave_last_rx(bond, curr_active_slave),
2959 curr_active_slave->last_link_up))
2960 bond_validate_arp(bond, slave, tip, sip);
2961 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2962 bond_time_in_interval(bond,
2963 dev_trans_start(curr_arp_slave->dev), 1))
2964 bond_validate_arp(bond, slave, sip, tip);
2967 if (arp != (struct arphdr *)skb->data)
2969 return RX_HANDLER_ANOTHER;
2972 /* function to verify if we're in the arp_interval timeslice, returns true if
2973 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2974 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2976 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2979 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2981 return time_in_range(jiffies,
2982 last_act - delta_in_ticks,
2983 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2986 /* This function is called regularly to monitor each slave's link
2987 * ensuring that traffic is being sent and received when arp monitoring
2988 * is used in load-balancing mode. if the adapter has been dormant, then an
2989 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2990 * arp monitoring in active backup mode.
2992 static void bond_loadbalance_arp_mon(struct bonding *bond)
2994 struct slave *slave, *oldcurrent;
2995 struct list_head *iter;
2996 int do_failover = 0, slave_state_changed = 0;
2998 if (!bond_has_slaves(bond))
3003 oldcurrent = rcu_dereference(bond->curr_active_slave);
3004 /* see if any of the previous devices are up now (i.e. they have
3005 * xmt and rcv traffic). the curr_active_slave does not come into
3006 * the picture unless it is null. also, slave->last_link_up is not
3007 * needed here because we send an arp on each slave and give a slave
3008 * as long as it needs to get the tx/rx within the delta.
3009 * TODO: what about up/down delay in arp mode? it wasn't here before
3012 bond_for_each_slave_rcu(bond, slave, iter) {
3013 unsigned long trans_start = dev_trans_start(slave->dev);
3015 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3017 if (slave->link != BOND_LINK_UP) {
3018 if (bond_time_in_interval(bond, trans_start, 1) &&
3019 bond_time_in_interval(bond, slave->last_rx, 1)) {
3021 bond_propose_link_state(slave, BOND_LINK_UP);
3022 slave_state_changed = 1;
3024 /* primary_slave has no meaning in round-robin
3025 * mode. the window of a slave being up and
3026 * curr_active_slave being null after enslaving
3030 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3033 slave_info(bond->dev, slave->dev, "interface is now up\n");
3037 /* slave->link == BOND_LINK_UP */
3039 /* not all switches will respond to an arp request
3040 * when the source ip is 0, so don't take the link down
3041 * if we don't know our ip yet
3043 if (!bond_time_in_interval(bond, trans_start, 2) ||
3044 !bond_time_in_interval(bond, slave->last_rx, 2)) {
3046 bond_propose_link_state(slave, BOND_LINK_DOWN);
3047 slave_state_changed = 1;
3049 if (slave->link_failure_count < UINT_MAX)
3050 slave->link_failure_count++;
3052 slave_info(bond->dev, slave->dev, "interface is now down\n");
3054 if (slave == oldcurrent)
3059 /* note: if switch is in round-robin mode, all links
3060 * must tx arp to ensure all links rx an arp - otherwise
3061 * links may oscillate or not come up at all; if switch is
3062 * in something like xor mode, there is nothing we can
3063 * do - all replies will be rx'ed on same link causing slaves
3064 * to be unstable during low/no traffic periods
3066 if (bond_slave_is_up(slave))
3067 bond_arp_send_all(bond, slave);
3072 if (do_failover || slave_state_changed) {
3073 if (!rtnl_trylock())
3076 bond_for_each_slave(bond, slave, iter) {
3077 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3078 slave->link = slave->link_new_state;
3081 if (slave_state_changed) {
3082 bond_slave_state_change(bond);
3083 if (BOND_MODE(bond) == BOND_MODE_XOR)
3084 bond_update_slave_arr(bond, NULL);
3088 bond_select_active_slave(bond);
3089 unblock_netpoll_tx();
3095 if (bond->params.arp_interval)
3096 queue_delayed_work(bond->wq, &bond->arp_work,
3097 msecs_to_jiffies(bond->params.arp_interval));
3100 /* Called to inspect slaves for active-backup mode ARP monitor link state
3101 * changes. Sets proposed link state in slaves to specify what action
3102 * should take place for the slave. Returns 0 if no changes are found, >0
3103 * if changes to link states must be committed.
3105 * Called with rcu_read_lock held.
3107 static int bond_ab_arp_inspect(struct bonding *bond)
3109 unsigned long trans_start, last_rx;
3110 struct list_head *iter;
3111 struct slave *slave;
3114 bond_for_each_slave_rcu(bond, slave, iter) {
3115 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3116 last_rx = slave_last_rx(bond, slave);
3118 if (slave->link != BOND_LINK_UP) {
3119 if (bond_time_in_interval(bond, last_rx, 1)) {
3120 bond_propose_link_state(slave, BOND_LINK_UP);
3122 } else if (slave->link == BOND_LINK_BACK) {
3123 bond_propose_link_state(slave, BOND_LINK_FAIL);
3129 /* Give slaves 2*delta after being enslaved or made
3130 * active. This avoids bouncing, as the last receive
3131 * times need a full ARP monitor cycle to be updated.
3133 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3136 /* Backup slave is down if:
3137 * - No current_arp_slave AND
3138 * - more than 3*delta since last receive AND
3139 * - the bond has an IP address
3141 * Note: a non-null current_arp_slave indicates
3142 * the curr_active_slave went down and we are
3143 * searching for a new one; under this condition
3144 * we only take the curr_active_slave down - this
3145 * gives each slave a chance to tx/rx traffic
3146 * before being taken out
3148 if (!bond_is_active_slave(slave) &&
3149 !rcu_access_pointer(bond->current_arp_slave) &&
3150 !bond_time_in_interval(bond, last_rx, 3)) {
3151 bond_propose_link_state(slave, BOND_LINK_DOWN);
3155 /* Active slave is down if:
3156 * - more than 2*delta since transmitting OR
3157 * - (more than 2*delta since receive AND
3158 * the bond has an IP address)
3160 trans_start = dev_trans_start(slave->dev);
3161 if (bond_is_active_slave(slave) &&
3162 (!bond_time_in_interval(bond, trans_start, 2) ||
3163 !bond_time_in_interval(bond, last_rx, 2))) {
3164 bond_propose_link_state(slave, BOND_LINK_DOWN);
3172 /* Called to commit link state changes noted by inspection step of
3173 * active-backup mode ARP monitor.
3175 * Called with RTNL hold.
3177 static void bond_ab_arp_commit(struct bonding *bond)
3179 unsigned long trans_start;
3180 struct list_head *iter;
3181 struct slave *slave;
3183 bond_for_each_slave(bond, slave, iter) {
3184 switch (slave->link_new_state) {
3185 case BOND_LINK_NOCHANGE:
3189 trans_start = dev_trans_start(slave->dev);
3190 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3191 (!rtnl_dereference(bond->curr_active_slave) &&
3192 bond_time_in_interval(bond, trans_start, 1))) {
3193 struct slave *current_arp_slave;
3195 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3196 bond_set_slave_link_state(slave, BOND_LINK_UP,
3197 BOND_SLAVE_NOTIFY_NOW);
3198 if (current_arp_slave) {
3199 bond_set_slave_inactive_flags(
3201 BOND_SLAVE_NOTIFY_NOW);
3202 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3205 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3207 if (!rtnl_dereference(bond->curr_active_slave) ||
3208 slave == rtnl_dereference(bond->primary_slave))
3215 case BOND_LINK_DOWN:
3216 if (slave->link_failure_count < UINT_MAX)
3217 slave->link_failure_count++;
3219 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3220 BOND_SLAVE_NOTIFY_NOW);
3221 bond_set_slave_inactive_flags(slave,
3222 BOND_SLAVE_NOTIFY_NOW);
3224 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3226 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3227 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3233 case BOND_LINK_FAIL:
3234 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3235 BOND_SLAVE_NOTIFY_NOW);
3236 bond_set_slave_inactive_flags(slave,
3237 BOND_SLAVE_NOTIFY_NOW);
3239 /* A slave has just been enslaved and has become
3240 * the current active slave.
3242 if (rtnl_dereference(bond->curr_active_slave))
3243 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3247 slave_err(bond->dev, slave->dev,
3248 "impossible: link_new_state %d on slave\n",
3249 slave->link_new_state);
3255 bond_select_active_slave(bond);
3256 unblock_netpoll_tx();
3259 bond_set_carrier(bond);
3262 /* Send ARP probes for active-backup mode ARP monitor.
3264 * Called with rcu_read_lock held.
3266 static bool bond_ab_arp_probe(struct bonding *bond)
3268 struct slave *slave, *before = NULL, *new_slave = NULL,
3269 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3270 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3271 struct list_head *iter;
3273 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3275 if (curr_arp_slave && curr_active_slave)
3276 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3277 curr_arp_slave->dev->name,
3278 curr_active_slave->dev->name);
3280 if (curr_active_slave) {
3281 bond_arp_send_all(bond, curr_active_slave);
3282 return should_notify_rtnl;
3285 /* if we don't have a curr_active_slave, search for the next available
3286 * backup slave from the current_arp_slave and make it the candidate
3287 * for becoming the curr_active_slave
3290 if (!curr_arp_slave) {
3291 curr_arp_slave = bond_first_slave_rcu(bond);
3292 if (!curr_arp_slave)
3293 return should_notify_rtnl;
3296 bond_for_each_slave_rcu(bond, slave, iter) {
3297 if (!found && !before && bond_slave_is_up(slave))
3300 if (found && !new_slave && bond_slave_is_up(slave))
3302 /* if the link state is up at this point, we
3303 * mark it down - this can happen if we have
3304 * simultaneous link failures and
3305 * reselect_active_interface doesn't make this
3306 * one the current slave so it is still marked
3307 * up when it is actually down
3309 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3310 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3311 BOND_SLAVE_NOTIFY_LATER);
3312 if (slave->link_failure_count < UINT_MAX)
3313 slave->link_failure_count++;
3315 bond_set_slave_inactive_flags(slave,
3316 BOND_SLAVE_NOTIFY_LATER);
3318 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3320 if (slave == curr_arp_slave)
3324 if (!new_slave && before)
3330 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3331 BOND_SLAVE_NOTIFY_LATER);
3332 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3333 bond_arp_send_all(bond, new_slave);
3334 new_slave->last_link_up = jiffies;
3335 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3338 bond_for_each_slave_rcu(bond, slave, iter) {
3339 if (slave->should_notify || slave->should_notify_link) {
3340 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3344 return should_notify_rtnl;
3347 static void bond_activebackup_arp_mon(struct bonding *bond)
3349 bool should_notify_peers = false;
3350 bool should_notify_rtnl = false;
3353 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3355 if (!bond_has_slaves(bond))
3360 should_notify_peers = bond_should_notify_peers(bond);
3362 if (bond_ab_arp_inspect(bond)) {
3365 /* Race avoidance with bond_close flush of workqueue */
3366 if (!rtnl_trylock()) {
3368 should_notify_peers = false;
3372 bond_ab_arp_commit(bond);
3378 should_notify_rtnl = bond_ab_arp_probe(bond);
3382 if (bond->params.arp_interval)
3383 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3385 if (should_notify_peers || should_notify_rtnl) {
3386 if (!rtnl_trylock())
3389 if (should_notify_peers) {
3390 bond->send_peer_notif--;
3391 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3394 if (should_notify_rtnl) {
3395 bond_slave_state_notify(bond);
3396 bond_slave_link_notify(bond);
3403 static void bond_arp_monitor(struct work_struct *work)
3405 struct bonding *bond = container_of(work, struct bonding,
3408 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3409 bond_activebackup_arp_mon(bond);
3411 bond_loadbalance_arp_mon(bond);
3414 /*-------------------------- netdev event handling --------------------------*/
3416 /* Change device name */
3417 static int bond_event_changename(struct bonding *bond)
3419 bond_remove_proc_entry(bond);
3420 bond_create_proc_entry(bond);
3422 bond_debug_reregister(bond);
3427 static int bond_master_netdev_event(unsigned long event,
3428 struct net_device *bond_dev)
3430 struct bonding *event_bond = netdev_priv(bond_dev);
3432 netdev_dbg(bond_dev, "%s called\n", __func__);
3435 case NETDEV_CHANGENAME:
3436 return bond_event_changename(event_bond);
3437 case NETDEV_UNREGISTER:
3438 bond_remove_proc_entry(event_bond);
3439 #ifdef CONFIG_XFRM_OFFLOAD
3440 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3441 #endif /* CONFIG_XFRM_OFFLOAD */
3443 case NETDEV_REGISTER:
3444 bond_create_proc_entry(event_bond);
3453 static int bond_slave_netdev_event(unsigned long event,
3454 struct net_device *slave_dev)
3456 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3457 struct bonding *bond;
3458 struct net_device *bond_dev;
3460 /* A netdev event can be generated while enslaving a device
3461 * before netdev_rx_handler_register is called in which case
3462 * slave will be NULL
3465 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3469 bond_dev = slave->bond->dev;
3471 primary = rtnl_dereference(bond->primary_slave);
3473 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3476 case NETDEV_UNREGISTER:
3477 if (bond_dev->type != ARPHRD_ETHER)
3478 bond_release_and_destroy(bond_dev, slave_dev);
3480 __bond_release_one(bond_dev, slave_dev, false, true);
3484 /* For 802.3ad mode only:
3485 * Getting invalid Speed/Duplex values here will put slave
3486 * in weird state. Mark it as link-fail if the link was
3487 * previously up or link-down if it hasn't yet come up, and
3488 * let link-monitoring (miimon) set it right when correct
3489 * speeds/duplex are available.
3491 if (bond_update_speed_duplex(slave) &&
3492 BOND_MODE(bond) == BOND_MODE_8023AD) {
3493 if (slave->last_link_up)
3494 slave->link = BOND_LINK_FAIL;
3496 slave->link = BOND_LINK_DOWN;
3499 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3500 bond_3ad_adapter_speed_duplex_changed(slave);
3503 /* Refresh slave-array if applicable!
3504 * If the setup does not use miimon or arpmon (mode-specific!),
3505 * then these events will not cause the slave-array to be
3506 * refreshed. This will cause xmit to use a slave that is not
3507 * usable. Avoid such situation by refeshing the array at these
3508 * events. If these (miimon/arpmon) parameters are configured
3509 * then array gets refreshed twice and that should be fine!
3511 if (bond_mode_can_use_xmit_hash(bond))
3512 bond_update_slave_arr(bond, NULL);
3514 case NETDEV_CHANGEMTU:
3515 /* TODO: Should slaves be allowed to
3516 * independently alter their MTU? For
3517 * an active-backup bond, slaves need
3518 * not be the same type of device, so
3519 * MTUs may vary. For other modes,
3520 * slaves arguably should have the
3521 * same MTUs. To do this, we'd need to
3522 * take over the slave's change_mtu
3523 * function for the duration of their
3527 case NETDEV_CHANGENAME:
3528 /* we don't care if we don't have primary set */
3529 if (!bond_uses_primary(bond) ||
3530 !bond->params.primary[0])
3533 if (slave == primary) {
3534 /* slave's name changed - he's no longer primary */
3535 RCU_INIT_POINTER(bond->primary_slave, NULL);
3536 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3537 /* we have a new primary slave */
3538 rcu_assign_pointer(bond->primary_slave, slave);
3539 } else { /* we didn't change primary - exit */
3543 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3544 primary ? slave_dev->name : "none");
3547 bond_select_active_slave(bond);
3548 unblock_netpoll_tx();
3550 case NETDEV_FEAT_CHANGE:
3551 if (!bond->notifier_ctx) {
3552 bond->notifier_ctx = true;
3553 bond_compute_features(bond);
3554 bond->notifier_ctx = false;
3557 case NETDEV_RESEND_IGMP:
3558 /* Propagate to master device */
3559 call_netdevice_notifiers(event, slave->bond->dev);
3568 /* bond_netdev_event: handle netdev notifier chain events.
3570 * This function receives events for the netdev chain. The caller (an
3571 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3572 * locks for us to safely manipulate the slave devices (RTNL lock,
3575 static int bond_netdev_event(struct notifier_block *this,
3576 unsigned long event, void *ptr)
3578 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3580 netdev_dbg(event_dev, "%s received %s\n",
3581 __func__, netdev_cmd_to_name(event));
3583 if (!(event_dev->priv_flags & IFF_BONDING))
3586 if (event_dev->flags & IFF_MASTER) {
3589 ret = bond_master_netdev_event(event, event_dev);
3590 if (ret != NOTIFY_DONE)
3594 if (event_dev->flags & IFF_SLAVE)
3595 return bond_slave_netdev_event(event, event_dev);
3600 static struct notifier_block bond_netdev_notifier = {
3601 .notifier_call = bond_netdev_event,
3604 /*---------------------------- Hashing Policies -----------------------------*/
3606 /* L2 hash helper */
3607 static inline u32 bond_eth_hash(struct sk_buff *skb)
3609 struct ethhdr *ep, hdr_tmp;
3611 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3613 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3617 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3618 int *noff, int *proto, bool l34)
3620 const struct ipv6hdr *iph6;
3621 const struct iphdr *iph;
3623 if (skb->protocol == htons(ETH_P_IP)) {
3624 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3626 iph = (const struct iphdr *)(skb->data + *noff);
3627 iph_to_flow_copy_v4addrs(fk, iph);
3628 *noff += iph->ihl << 2;
3629 if (!ip_is_fragment(iph))
3630 *proto = iph->protocol;
3631 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3632 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3634 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3635 iph_to_flow_copy_v6addrs(fk, iph6);
3636 *noff += sizeof(*iph6);
3637 *proto = iph6->nexthdr;
3642 if (l34 && *proto >= 0)
3643 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3648 /* Extract the appropriate headers based on bond's xmit policy */
3649 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3650 struct flow_keys *fk)
3652 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3653 int noff, proto = -1;
3655 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) {
3656 memset(fk, 0, sizeof(*fk));
3657 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3658 fk, NULL, 0, 0, 0, 0);
3661 fk->ports.ports = 0;
3662 memset(&fk->icmp, 0, sizeof(fk->icmp));
3663 noff = skb_network_offset(skb);
3664 if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3667 /* ICMP error packets contains at least 8 bytes of the header
3668 * of the packet which generated the error. Use this information
3669 * to correlate ICMP error packets within the same flow which
3670 * generated the error.
3672 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3673 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3674 skb_transport_offset(skb),
3676 if (proto == IPPROTO_ICMP) {
3677 if (!icmp_is_err(fk->icmp.type))
3680 noff += sizeof(struct icmphdr);
3681 } else if (proto == IPPROTO_ICMPV6) {
3682 if (!icmpv6_is_err(fk->icmp.type))
3685 noff += sizeof(struct icmp6hdr);
3687 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3694 * bond_xmit_hash - generate a hash value based on the xmit policy
3695 * @bond: bonding device
3696 * @skb: buffer to use for headers
3698 * This function will extract the necessary headers from the skb buffer and use
3699 * them to generate a hash based on the xmit_policy set in the bonding device
3701 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3703 struct flow_keys flow;
3706 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3710 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3711 !bond_flow_dissect(bond, skb, &flow))
3712 return bond_eth_hash(skb);
3714 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3715 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3716 hash = bond_eth_hash(skb);
3719 memcpy(&hash, &flow.icmp, sizeof(hash));
3721 memcpy(&hash, &flow.ports.ports, sizeof(hash));
3723 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3724 (__force u32)flow_get_u32_src(&flow);
3725 hash ^= (hash >> 16);
3726 hash ^= (hash >> 8);
3731 /*-------------------------- Device entry points ----------------------------*/
3733 void bond_work_init_all(struct bonding *bond)
3735 INIT_DELAYED_WORK(&bond->mcast_work,
3736 bond_resend_igmp_join_requests_delayed);
3737 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3738 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3739 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3740 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3741 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3744 static void bond_work_cancel_all(struct bonding *bond)
3746 cancel_delayed_work_sync(&bond->mii_work);
3747 cancel_delayed_work_sync(&bond->arp_work);
3748 cancel_delayed_work_sync(&bond->alb_work);
3749 cancel_delayed_work_sync(&bond->ad_work);
3750 cancel_delayed_work_sync(&bond->mcast_work);
3751 cancel_delayed_work_sync(&bond->slave_arr_work);
3754 static int bond_open(struct net_device *bond_dev)
3756 struct bonding *bond = netdev_priv(bond_dev);
3757 struct list_head *iter;
3758 struct slave *slave;
3760 /* reset slave->backup and slave->inactive */
3761 if (bond_has_slaves(bond)) {
3762 bond_for_each_slave(bond, slave, iter) {
3763 if (bond_uses_primary(bond) &&
3764 slave != rcu_access_pointer(bond->curr_active_slave)) {
3765 bond_set_slave_inactive_flags(slave,
3766 BOND_SLAVE_NOTIFY_NOW);
3767 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3768 bond_set_slave_active_flags(slave,
3769 BOND_SLAVE_NOTIFY_NOW);
3774 if (bond_is_lb(bond)) {
3775 /* bond_alb_initialize must be called before the timer
3778 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3780 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3781 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3784 if (bond->params.miimon) /* link check interval, in milliseconds. */
3785 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3787 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3788 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3789 bond->recv_probe = bond_arp_rcv;
3792 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3793 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3794 /* register to receive LACPDUs */
3795 bond->recv_probe = bond_3ad_lacpdu_recv;
3796 bond_3ad_initiate_agg_selection(bond, 1);
3798 bond_for_each_slave(bond, slave, iter)
3799 dev_mc_add(slave->dev, lacpdu_mcast_addr);
3802 if (bond_mode_can_use_xmit_hash(bond))
3803 bond_update_slave_arr(bond, NULL);
3808 static int bond_close(struct net_device *bond_dev)
3810 struct bonding *bond = netdev_priv(bond_dev);
3811 struct slave *slave;
3813 bond_work_cancel_all(bond);
3814 bond->send_peer_notif = 0;
3815 if (bond_is_lb(bond))
3816 bond_alb_deinitialize(bond);
3817 bond->recv_probe = NULL;
3819 if (bond_uses_primary(bond)) {
3821 slave = rcu_dereference(bond->curr_active_slave);
3823 bond_hw_addr_flush(bond_dev, slave->dev);
3826 struct list_head *iter;
3828 bond_for_each_slave(bond, slave, iter)
3829 bond_hw_addr_flush(bond_dev, slave->dev);
3835 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3836 * that some drivers can provide 32bit values only.
3838 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3839 const struct rtnl_link_stats64 *_new,
3840 const struct rtnl_link_stats64 *_old)
3842 const u64 *new = (const u64 *)_new;
3843 const u64 *old = (const u64 *)_old;
3844 u64 *res = (u64 *)_res;
3847 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3850 s64 delta = nv - ov;
3852 /* detects if this particular field is 32bit only */
3853 if (((nv | ov) >> 32) == 0)
3854 delta = (s64)(s32)((u32)nv - (u32)ov);
3856 /* filter anomalies, some drivers reset their stats
3857 * at down/up events.
3864 #ifdef CONFIG_LOCKDEP
3865 static int bond_get_lowest_level_rcu(struct net_device *dev)
3867 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3868 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3869 int cur = 0, max = 0;
3872 iter = &dev->adj_list.lower;
3877 ldev = netdev_next_lower_dev_rcu(now, &iter);
3882 niter = &ldev->adj_list.lower;
3883 dev_stack[cur] = now;
3884 iter_stack[cur++] = iter;
3893 next = dev_stack[--cur];
3894 niter = iter_stack[cur];
3905 static void bond_get_stats(struct net_device *bond_dev,
3906 struct rtnl_link_stats64 *stats)
3908 struct bonding *bond = netdev_priv(bond_dev);
3909 struct rtnl_link_stats64 temp;
3910 struct list_head *iter;
3911 struct slave *slave;
3916 #ifdef CONFIG_LOCKDEP
3917 nest_level = bond_get_lowest_level_rcu(bond_dev);
3920 spin_lock_nested(&bond->stats_lock, nest_level);
3921 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3923 bond_for_each_slave_rcu(bond, slave, iter) {
3924 const struct rtnl_link_stats64 *new =
3925 dev_get_stats(slave->dev, &temp);
3927 bond_fold_stats(stats, new, &slave->slave_stats);
3929 /* save off the slave stats for the next run */
3930 memcpy(&slave->slave_stats, new, sizeof(*new));
3933 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3934 spin_unlock(&bond->stats_lock);
3938 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3940 struct bonding *bond = netdev_priv(bond_dev);
3941 struct net_device *slave_dev = NULL;
3942 struct ifbond k_binfo;
3943 struct ifbond __user *u_binfo = NULL;
3944 struct ifslave k_sinfo;
3945 struct ifslave __user *u_sinfo = NULL;
3946 struct mii_ioctl_data *mii = NULL;
3947 struct bond_opt_value newval;
3951 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3962 /* We do this again just in case we were called by SIOCGMIIREG
3963 * instead of SIOCGMIIPHY.
3969 if (mii->reg_num == 1) {
3971 if (netif_carrier_ok(bond->dev))
3972 mii->val_out = BMSR_LSTATUS;
3976 case BOND_INFO_QUERY_OLD:
3977 case SIOCBONDINFOQUERY:
3978 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3980 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3983 bond_info_query(bond_dev, &k_binfo);
3984 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3988 case BOND_SLAVE_INFO_QUERY_OLD:
3989 case SIOCBONDSLAVEINFOQUERY:
3990 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3992 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3995 res = bond_slave_info_query(bond_dev, &k_sinfo);
3997 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4005 net = dev_net(bond_dev);
4007 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4010 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4012 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4018 case BOND_ENSLAVE_OLD:
4019 case SIOCBONDENSLAVE:
4020 res = bond_enslave(bond_dev, slave_dev, NULL);
4022 case BOND_RELEASE_OLD:
4023 case SIOCBONDRELEASE:
4024 res = bond_release(bond_dev, slave_dev);
4026 case BOND_SETHWADDR_OLD:
4027 case SIOCBONDSETHWADDR:
4028 res = bond_set_dev_addr(bond_dev, slave_dev);
4030 case BOND_CHANGE_ACTIVE_OLD:
4031 case SIOCBONDCHANGEACTIVE:
4032 bond_opt_initstr(&newval, slave_dev->name);
4033 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4043 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4045 struct bonding *bond = netdev_priv(bond_dev);
4047 if (change & IFF_PROMISC)
4048 bond_set_promiscuity(bond,
4049 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4051 if (change & IFF_ALLMULTI)
4052 bond_set_allmulti(bond,
4053 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4056 static void bond_set_rx_mode(struct net_device *bond_dev)
4058 struct bonding *bond = netdev_priv(bond_dev);
4059 struct list_head *iter;
4060 struct slave *slave;
4063 if (bond_uses_primary(bond)) {
4064 slave = rcu_dereference(bond->curr_active_slave);
4066 dev_uc_sync(slave->dev, bond_dev);
4067 dev_mc_sync(slave->dev, bond_dev);
4070 bond_for_each_slave_rcu(bond, slave, iter) {
4071 dev_uc_sync_multiple(slave->dev, bond_dev);
4072 dev_mc_sync_multiple(slave->dev, bond_dev);
4078 static int bond_neigh_init(struct neighbour *n)
4080 struct bonding *bond = netdev_priv(n->dev);
4081 const struct net_device_ops *slave_ops;
4082 struct neigh_parms parms;
4083 struct slave *slave;
4087 slave = bond_first_slave_rcu(bond);
4090 slave_ops = slave->dev->netdev_ops;
4091 if (!slave_ops->ndo_neigh_setup)
4094 /* TODO: find another way [1] to implement this.
4095 * Passing a zeroed structure is fragile,
4096 * but at least we do not pass garbage.
4098 * [1] One way would be that ndo_neigh_setup() never touch
4099 * struct neigh_parms, but propagate the new neigh_setup()
4100 * back to ___neigh_create() / neigh_parms_alloc()
4102 memset(&parms, 0, sizeof(parms));
4103 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4108 if (parms.neigh_setup)
4109 ret = parms.neigh_setup(n);
4115 /* The bonding ndo_neigh_setup is called at init time beofre any
4116 * slave exists. So we must declare proxy setup function which will
4117 * be used at run time to resolve the actual slave neigh param setup.
4119 * It's also called by master devices (such as vlans) to setup their
4120 * underlying devices. In that case - do nothing, we're already set up from
4123 static int bond_neigh_setup(struct net_device *dev,
4124 struct neigh_parms *parms)
4126 /* modify only our neigh_parms */
4127 if (parms->dev == dev)
4128 parms->neigh_setup = bond_neigh_init;
4133 /* Change the MTU of all of a master's slaves to match the master */
4134 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4136 struct bonding *bond = netdev_priv(bond_dev);
4137 struct slave *slave, *rollback_slave;
4138 struct list_head *iter;
4141 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4143 bond_for_each_slave(bond, slave, iter) {
4144 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4145 slave, slave->dev->netdev_ops->ndo_change_mtu);
4147 res = dev_set_mtu(slave->dev, new_mtu);
4150 /* If we failed to set the slave's mtu to the new value
4151 * we must abort the operation even in ACTIVE_BACKUP
4152 * mode, because if we allow the backup slaves to have
4153 * different mtu values than the active slave we'll
4154 * need to change their mtu when doing a failover. That
4155 * means changing their mtu from timer context, which
4156 * is probably not a good idea.
4158 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4164 bond_dev->mtu = new_mtu;
4169 /* unwind from head to the slave that failed */
4170 bond_for_each_slave(bond, rollback_slave, iter) {
4173 if (rollback_slave == slave)
4176 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4178 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4185 /* Change HW address
4187 * Note that many devices must be down to change the HW address, and
4188 * downing the master releases all slaves. We can make bonds full of
4189 * bonding devices to test this, however.
4191 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4193 struct bonding *bond = netdev_priv(bond_dev);
4194 struct slave *slave, *rollback_slave;
4195 struct sockaddr_storage *ss = addr, tmp_ss;
4196 struct list_head *iter;
4199 if (BOND_MODE(bond) == BOND_MODE_ALB)
4200 return bond_alb_set_mac_address(bond_dev, addr);
4203 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4205 /* If fail_over_mac is enabled, do nothing and return success.
4206 * Returning an error causes ifenslave to fail.
4208 if (bond->params.fail_over_mac &&
4209 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4212 if (!is_valid_ether_addr(ss->__data))
4213 return -EADDRNOTAVAIL;
4215 bond_for_each_slave(bond, slave, iter) {
4216 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4218 res = dev_set_mac_address(slave->dev, addr, NULL);
4220 /* TODO: consider downing the slave
4222 * User should expect communications
4223 * breakage anyway until ARP finish
4226 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4233 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4237 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4238 tmp_ss.ss_family = bond_dev->type;
4240 /* unwind from head to the slave that failed */
4241 bond_for_each_slave(bond, rollback_slave, iter) {
4244 if (rollback_slave == slave)
4247 tmp_res = dev_set_mac_address(rollback_slave->dev,
4248 (struct sockaddr *)&tmp_ss, NULL);
4250 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4259 * bond_get_slave_by_id - get xmit slave with slave_id
4260 * @bond: bonding device that is transmitting
4261 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4263 * This function tries to get slave with slave_id but in case
4264 * it fails, it tries to find the first available slave for transmission.
4266 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4269 struct list_head *iter;
4270 struct slave *slave;
4273 /* Here we start from the slave with slave_id */
4274 bond_for_each_slave_rcu(bond, slave, iter) {
4276 if (bond_slave_can_tx(slave))
4281 /* Here we start from the first slave up to slave_id */
4283 bond_for_each_slave_rcu(bond, slave, iter) {
4286 if (bond_slave_can_tx(slave))
4289 /* no slave that can tx has been found */
4294 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4295 * @bond: bonding device to use
4297 * Based on the value of the bonding device's packets_per_slave parameter
4298 * this function generates a slave id, which is usually used as the next
4299 * slave to transmit through.
4301 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4304 struct reciprocal_value reciprocal_packets_per_slave;
4305 int packets_per_slave = bond->params.packets_per_slave;
4307 switch (packets_per_slave) {
4309 slave_id = prandom_u32();
4312 slave_id = bond->rr_tx_counter;
4315 reciprocal_packets_per_slave =
4316 bond->params.reciprocal_packets_per_slave;
4317 slave_id = reciprocal_divide(bond->rr_tx_counter,
4318 reciprocal_packets_per_slave);
4321 bond->rr_tx_counter++;
4326 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4327 struct sk_buff *skb)
4329 struct slave *slave;
4333 /* Start with the curr_active_slave that joined the bond as the
4334 * default for sending IGMP traffic. For failover purposes one
4335 * needs to maintain some consistency for the interface that will
4336 * send the join/membership reports. The curr_active_slave found
4337 * will send all of this type of traffic.
4339 if (skb->protocol == htons(ETH_P_IP)) {
4340 int noff = skb_network_offset(skb);
4343 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4347 if (iph->protocol == IPPROTO_IGMP) {
4348 slave = rcu_dereference(bond->curr_active_slave);
4351 return bond_get_slave_by_id(bond, 0);
4356 slave_cnt = READ_ONCE(bond->slave_cnt);
4357 if (likely(slave_cnt)) {
4358 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4359 return bond_get_slave_by_id(bond, slave_id);
4364 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4365 struct net_device *bond_dev)
4367 struct bonding *bond = netdev_priv(bond_dev);
4368 struct slave *slave;
4370 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4372 return bond_dev_queue_xmit(bond, skb, slave->dev);
4374 return bond_tx_drop(bond_dev, skb);
4377 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4378 struct sk_buff *skb)
4380 return rcu_dereference(bond->curr_active_slave);
4383 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4384 * the bond has a usable interface.
4386 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4387 struct net_device *bond_dev)
4389 struct bonding *bond = netdev_priv(bond_dev);
4390 struct slave *slave;
4392 slave = bond_xmit_activebackup_slave_get(bond, skb);
4394 return bond_dev_queue_xmit(bond, skb, slave->dev);
4396 return bond_tx_drop(bond_dev, skb);
4399 /* Use this to update slave_array when (a) it's not appropriate to update
4400 * slave_array right away (note that update_slave_array() may sleep)
4401 * and / or (b) RTNL is not held.
4403 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4405 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4408 /* Slave array work handler. Holds only RTNL */
4409 static void bond_slave_arr_handler(struct work_struct *work)
4411 struct bonding *bond = container_of(work, struct bonding,
4412 slave_arr_work.work);
4415 if (!rtnl_trylock())
4418 ret = bond_update_slave_arr(bond, NULL);
4421 pr_warn_ratelimited("Failed to update slave array from WT\n");
4427 bond_slave_arr_work_rearm(bond, 1);
4430 static void bond_skip_slave(struct bond_up_slave *slaves,
4431 struct slave *skipslave)
4435 /* Rare situation where caller has asked to skip a specific
4436 * slave but allocation failed (most likely!). BTW this is
4437 * only possible when the call is initiated from
4438 * __bond_release_one(). In this situation; overwrite the
4439 * skipslave entry in the array with the last entry from the
4440 * array to avoid a situation where the xmit path may choose
4441 * this to-be-skipped slave to send a packet out.
4443 for (idx = 0; slaves && idx < slaves->count; idx++) {
4444 if (skipslave == slaves->arr[idx]) {
4446 slaves->arr[slaves->count - 1];
4453 static void bond_set_slave_arr(struct bonding *bond,
4454 struct bond_up_slave *usable_slaves,
4455 struct bond_up_slave *all_slaves)
4457 struct bond_up_slave *usable, *all;
4459 usable = rtnl_dereference(bond->usable_slaves);
4460 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4461 kfree_rcu(usable, rcu);
4463 all = rtnl_dereference(bond->all_slaves);
4464 rcu_assign_pointer(bond->all_slaves, all_slaves);
4465 kfree_rcu(all, rcu);
4468 static void bond_reset_slave_arr(struct bonding *bond)
4470 struct bond_up_slave *usable, *all;
4472 usable = rtnl_dereference(bond->usable_slaves);
4474 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4475 kfree_rcu(usable, rcu);
4478 all = rtnl_dereference(bond->all_slaves);
4480 RCU_INIT_POINTER(bond->all_slaves, NULL);
4481 kfree_rcu(all, rcu);
4485 /* Build the usable slaves array in control path for modes that use xmit-hash
4486 * to determine the slave interface -
4487 * (a) BOND_MODE_8023AD
4489 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4491 * The caller is expected to hold RTNL only and NO other lock!
4493 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4495 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4496 struct slave *slave;
4497 struct list_head *iter;
4501 #ifdef CONFIG_LOCKDEP
4502 WARN_ON(lockdep_is_held(&bond->mode_lock));
4505 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4506 bond->slave_cnt), GFP_KERNEL);
4507 all_slaves = kzalloc(struct_size(all_slaves, arr,
4508 bond->slave_cnt), GFP_KERNEL);
4509 if (!usable_slaves || !all_slaves) {
4513 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4514 struct ad_info ad_info;
4516 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4517 pr_debug("bond_3ad_get_active_agg_info failed\n");
4518 /* No active aggragator means it's not safe to use
4519 * the previous array.
4521 bond_reset_slave_arr(bond);
4524 agg_id = ad_info.aggregator_id;
4526 bond_for_each_slave(bond, slave, iter) {
4527 if (skipslave == slave)
4530 all_slaves->arr[all_slaves->count++] = slave;
4531 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4532 struct aggregator *agg;
4534 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4535 if (!agg || agg->aggregator_identifier != agg_id)
4538 if (!bond_slave_can_tx(slave))
4541 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4542 usable_slaves->count);
4544 usable_slaves->arr[usable_slaves->count++] = slave;
4547 bond_set_slave_arr(bond, usable_slaves, all_slaves);
4550 if (ret != 0 && skipslave) {
4551 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4553 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4556 kfree_rcu(all_slaves, rcu);
4557 kfree_rcu(usable_slaves, rcu);
4562 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4563 struct sk_buff *skb,
4564 struct bond_up_slave *slaves)
4566 struct slave *slave;
4570 hash = bond_xmit_hash(bond, skb);
4571 count = slaves ? READ_ONCE(slaves->count) : 0;
4572 if (unlikely(!count))
4575 slave = slaves->arr[hash % count];
4579 /* Use this Xmit function for 3AD as well as XOR modes. The current
4580 * usable slave array is formed in the control path. The xmit function
4581 * just calculates hash and sends the packet out.
4583 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4584 struct net_device *dev)
4586 struct bonding *bond = netdev_priv(dev);
4587 struct bond_up_slave *slaves;
4588 struct slave *slave;
4590 slaves = rcu_dereference(bond->usable_slaves);
4591 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4593 return bond_dev_queue_xmit(bond, skb, slave->dev);
4595 return bond_tx_drop(dev, skb);
4598 /* in broadcast mode, we send everything to all usable interfaces. */
4599 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4600 struct net_device *bond_dev)
4602 struct bonding *bond = netdev_priv(bond_dev);
4603 struct slave *slave = NULL;
4604 struct list_head *iter;
4605 bool xmit_suc = false;
4606 bool skb_used = false;
4608 bond_for_each_slave_rcu(bond, slave, iter) {
4609 struct sk_buff *skb2;
4611 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
4614 if (bond_is_last_slave(bond, slave)) {
4618 skb2 = skb_clone(skb, GFP_ATOMIC);
4620 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4621 bond_dev->name, __func__);
4626 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
4631 dev_kfree_skb_any(skb);
4634 return NETDEV_TX_OK;
4636 atomic_long_inc(&bond_dev->tx_dropped);
4637 return NET_XMIT_DROP;
4640 /*------------------------- Device initialization ---------------------------*/
4642 /* Lookup the slave that corresponds to a qid */
4643 static inline int bond_slave_override(struct bonding *bond,
4644 struct sk_buff *skb)
4646 struct slave *slave = NULL;
4647 struct list_head *iter;
4649 if (!skb_rx_queue_recorded(skb))
4652 /* Find out if any slaves have the same mapping as this skb. */
4653 bond_for_each_slave_rcu(bond, slave, iter) {
4654 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4655 if (bond_slave_is_up(slave) &&
4656 slave->link == BOND_LINK_UP) {
4657 bond_dev_queue_xmit(bond, skb, slave->dev);
4660 /* If the slave isn't UP, use default transmit policy. */
4669 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4670 struct net_device *sb_dev)
4672 /* This helper function exists to help dev_pick_tx get the correct
4673 * destination queue. Using a helper function skips a call to
4674 * skb_tx_hash and will put the skbs in the queue we expect on their
4675 * way down to the bonding driver.
4677 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4679 /* Save the original txq to restore before passing to the driver */
4680 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4682 if (unlikely(txq >= dev->real_num_tx_queues)) {
4684 txq -= dev->real_num_tx_queues;
4685 } while (txq >= dev->real_num_tx_queues);
4690 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4691 struct sk_buff *skb,
4694 struct bonding *bond = netdev_priv(master_dev);
4695 struct bond_up_slave *slaves;
4696 struct slave *slave = NULL;
4698 switch (BOND_MODE(bond)) {
4699 case BOND_MODE_ROUNDROBIN:
4700 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4702 case BOND_MODE_ACTIVEBACKUP:
4703 slave = bond_xmit_activebackup_slave_get(bond, skb);
4705 case BOND_MODE_8023AD:
4708 slaves = rcu_dereference(bond->all_slaves);
4710 slaves = rcu_dereference(bond->usable_slaves);
4711 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4713 case BOND_MODE_BROADCAST:
4716 slave = bond_xmit_alb_slave_get(bond, skb);
4719 slave = bond_xmit_tlb_slave_get(bond, skb);
4722 /* Should never happen, mode already checked */
4723 WARN_ONCE(true, "Unknown bonding mode");
4732 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4734 struct bonding *bond = netdev_priv(dev);
4736 if (bond_should_override_tx_queue(bond) &&
4737 !bond_slave_override(bond, skb))
4738 return NETDEV_TX_OK;
4740 switch (BOND_MODE(bond)) {
4741 case BOND_MODE_ROUNDROBIN:
4742 return bond_xmit_roundrobin(skb, dev);
4743 case BOND_MODE_ACTIVEBACKUP:
4744 return bond_xmit_activebackup(skb, dev);
4745 case BOND_MODE_8023AD:
4747 return bond_3ad_xor_xmit(skb, dev);
4748 case BOND_MODE_BROADCAST:
4749 return bond_xmit_broadcast(skb, dev);
4751 return bond_alb_xmit(skb, dev);
4753 return bond_tlb_xmit(skb, dev);
4755 /* Should never happen, mode already checked */
4756 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4758 return bond_tx_drop(dev, skb);
4762 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4764 struct bonding *bond = netdev_priv(dev);
4765 netdev_tx_t ret = NETDEV_TX_OK;
4767 /* If we risk deadlock from transmitting this in the
4768 * netpoll path, tell netpoll to queue the frame for later tx
4770 if (unlikely(is_netpoll_tx_blocked(dev)))
4771 return NETDEV_TX_BUSY;
4774 if (bond_has_slaves(bond))
4775 ret = __bond_start_xmit(skb, dev);
4777 ret = bond_tx_drop(dev, skb);
4783 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4785 if (speed == 0 || speed == SPEED_UNKNOWN)
4786 speed = slave->speed;
4788 speed = min(speed, slave->speed);
4793 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4794 struct ethtool_link_ksettings *cmd)
4796 struct bonding *bond = netdev_priv(bond_dev);
4797 struct list_head *iter;
4798 struct slave *slave;
4801 cmd->base.duplex = DUPLEX_UNKNOWN;
4802 cmd->base.port = PORT_OTHER;
4804 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4805 * do not need to check mode. Though link speed might not represent
4806 * the true receive or transmit bandwidth (not all modes are symmetric)
4807 * this is an accurate maximum.
4809 bond_for_each_slave(bond, slave, iter) {
4810 if (bond_slave_can_tx(slave)) {
4811 if (slave->speed != SPEED_UNKNOWN) {
4812 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4813 speed = bond_mode_bcast_speed(slave,
4816 speed += slave->speed;
4818 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4819 slave->duplex != DUPLEX_UNKNOWN)
4820 cmd->base.duplex = slave->duplex;
4823 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4828 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4829 struct ethtool_drvinfo *drvinfo)
4831 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4832 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4836 static const struct ethtool_ops bond_ethtool_ops = {
4837 .get_drvinfo = bond_ethtool_get_drvinfo,
4838 .get_link = ethtool_op_get_link,
4839 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4842 static const struct net_device_ops bond_netdev_ops = {
4843 .ndo_init = bond_init,
4844 .ndo_uninit = bond_uninit,
4845 .ndo_open = bond_open,
4846 .ndo_stop = bond_close,
4847 .ndo_start_xmit = bond_start_xmit,
4848 .ndo_select_queue = bond_select_queue,
4849 .ndo_get_stats64 = bond_get_stats,
4850 .ndo_do_ioctl = bond_do_ioctl,
4851 .ndo_change_rx_flags = bond_change_rx_flags,
4852 .ndo_set_rx_mode = bond_set_rx_mode,
4853 .ndo_change_mtu = bond_change_mtu,
4854 .ndo_set_mac_address = bond_set_mac_address,
4855 .ndo_neigh_setup = bond_neigh_setup,
4856 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4857 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4858 #ifdef CONFIG_NET_POLL_CONTROLLER
4859 .ndo_netpoll_setup = bond_netpoll_setup,
4860 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4861 .ndo_poll_controller = bond_poll_controller,
4863 .ndo_add_slave = bond_enslave,
4864 .ndo_del_slave = bond_release,
4865 .ndo_fix_features = bond_fix_features,
4866 .ndo_features_check = passthru_features_check,
4867 .ndo_get_xmit_slave = bond_xmit_get_slave,
4870 static const struct device_type bond_type = {
4874 static void bond_destructor(struct net_device *bond_dev)
4876 struct bonding *bond = netdev_priv(bond_dev);
4878 destroy_workqueue(bond->wq);
4881 void bond_setup(struct net_device *bond_dev)
4883 struct bonding *bond = netdev_priv(bond_dev);
4885 spin_lock_init(&bond->mode_lock);
4886 bond->params = bonding_defaults;
4888 /* Initialize pointers */
4889 bond->dev = bond_dev;
4891 /* Initialize the device entry points */
4892 ether_setup(bond_dev);
4893 bond_dev->max_mtu = ETH_MAX_MTU;
4894 bond_dev->netdev_ops = &bond_netdev_ops;
4895 bond_dev->ethtool_ops = &bond_ethtool_ops;
4897 bond_dev->needs_free_netdev = true;
4898 bond_dev->priv_destructor = bond_destructor;
4900 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4902 /* Initialize the device options */
4903 bond_dev->flags |= IFF_MASTER;
4904 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4905 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4907 #ifdef CONFIG_XFRM_OFFLOAD
4908 /* set up xfrm device ops (only supported in active-backup right now) */
4909 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
4910 INIT_LIST_HEAD(&bond->ipsec_list);
4911 spin_lock_init(&bond->ipsec_lock);
4912 #endif /* CONFIG_XFRM_OFFLOAD */
4914 /* don't acquire bond device's netif_tx_lock when transmitting */
4915 bond_dev->features |= NETIF_F_LLTX;
4917 /* By default, we declare the bond to be fully
4918 * VLAN hardware accelerated capable. Special
4919 * care is taken in the various xmit functions
4920 * when there are slaves that are not hw accel
4924 /* Don't allow bond devices to change network namespaces. */
4925 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4927 bond_dev->hw_features = BOND_VLAN_FEATURES |
4928 NETIF_F_HW_VLAN_CTAG_RX |
4929 NETIF_F_HW_VLAN_CTAG_FILTER |
4930 NETIF_F_HW_VLAN_STAG_RX |
4931 NETIF_F_HW_VLAN_STAG_FILTER;
4933 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4934 #ifdef CONFIG_XFRM_OFFLOAD
4935 bond_dev->hw_features |= BOND_XFRM_FEATURES;
4936 #endif /* CONFIG_XFRM_OFFLOAD */
4937 bond_dev->features |= bond_dev->hw_features;
4938 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4939 #ifdef CONFIG_XFRM_OFFLOAD
4940 /* Disable XFRM features if this isn't an active-backup config */
4941 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
4942 bond_dev->features &= ~BOND_XFRM_FEATURES;
4943 #endif /* CONFIG_XFRM_OFFLOAD */
4946 /* Destroy a bonding device.
4947 * Must be under rtnl_lock when this function is called.
4949 static void bond_uninit(struct net_device *bond_dev)
4951 struct bonding *bond = netdev_priv(bond_dev);
4952 struct bond_up_slave *usable, *all;
4953 struct list_head *iter;
4954 struct slave *slave;
4956 bond_netpoll_cleanup(bond_dev);
4958 /* Release the bonded slaves */
4959 bond_for_each_slave(bond, slave, iter)
4960 __bond_release_one(bond_dev, slave->dev, true, true);
4961 netdev_info(bond_dev, "Released all slaves\n");
4963 usable = rtnl_dereference(bond->usable_slaves);
4965 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4966 kfree_rcu(usable, rcu);
4969 all = rtnl_dereference(bond->all_slaves);
4971 RCU_INIT_POINTER(bond->all_slaves, NULL);
4972 kfree_rcu(all, rcu);
4975 list_del(&bond->bond_list);
4977 bond_debug_unregister(bond);
4980 /*------------------------- Module initialization ---------------------------*/
4982 static int bond_check_params(struct bond_params *params)
4984 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4985 struct bond_opt_value newval;
4986 const struct bond_opt_value *valptr;
4987 int arp_all_targets_value = 0;
4988 u16 ad_actor_sys_prio = 0;
4989 u16 ad_user_port_key = 0;
4990 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4992 int bond_mode = BOND_MODE_ROUNDROBIN;
4993 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4997 /* Convert string parameters. */
4999 bond_opt_initstr(&newval, mode);
5000 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5002 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5005 bond_mode = valptr->value;
5008 if (xmit_hash_policy) {
5009 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5010 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5011 bond_mode == BOND_MODE_BROADCAST) {
5012 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5013 bond_mode_name(bond_mode));
5015 bond_opt_initstr(&newval, xmit_hash_policy);
5016 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5019 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5023 xmit_hashtype = valptr->value;
5028 if (bond_mode != BOND_MODE_8023AD) {
5029 pr_info("lacp_rate param is irrelevant in mode %s\n",
5030 bond_mode_name(bond_mode));
5032 bond_opt_initstr(&newval, lacp_rate);
5033 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5036 pr_err("Error: Invalid lacp rate \"%s\"\n",
5040 lacp_fast = valptr->value;
5045 bond_opt_initstr(&newval, ad_select);
5046 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5049 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5052 params->ad_select = valptr->value;
5053 if (bond_mode != BOND_MODE_8023AD)
5054 pr_warn("ad_select param only affects 802.3ad mode\n");
5056 params->ad_select = BOND_AD_STABLE;
5059 if (max_bonds < 0) {
5060 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5061 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5062 max_bonds = BOND_DEFAULT_MAX_BONDS;
5066 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5072 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5077 if (downdelay < 0) {
5078 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5079 downdelay, INT_MAX);
5083 if ((use_carrier != 0) && (use_carrier != 1)) {
5084 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5089 if (num_peer_notif < 0 || num_peer_notif > 255) {
5090 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5095 /* reset values for 802.3ad/TLB/ALB */
5096 if (!bond_mode_uses_arp(bond_mode)) {
5098 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
5099 pr_warn("Forcing miimon to 100msec\n");
5100 miimon = BOND_DEFAULT_MIIMON;
5104 if (tx_queues < 1 || tx_queues > 255) {
5105 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5106 tx_queues, BOND_DEFAULT_TX_QUEUES);
5107 tx_queues = BOND_DEFAULT_TX_QUEUES;
5110 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5111 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5113 all_slaves_active = 0;
5116 if (resend_igmp < 0 || resend_igmp > 255) {
5117 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5118 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5119 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5122 bond_opt_initval(&newval, packets_per_slave);
5123 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5124 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5125 packets_per_slave, USHRT_MAX);
5126 packets_per_slave = 1;
5129 if (bond_mode == BOND_MODE_ALB) {
5130 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",
5135 if (updelay || downdelay) {
5136 /* just warn the user the up/down delay will have
5137 * no effect since miimon is zero...
5139 pr_warn("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",
5140 updelay, downdelay);
5143 /* don't allow arp monitoring */
5145 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5146 miimon, arp_interval);
5150 if ((updelay % miimon) != 0) {
5151 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5152 updelay, miimon, (updelay / miimon) * miimon);
5157 if ((downdelay % miimon) != 0) {
5158 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5160 (downdelay / miimon) * miimon);
5163 downdelay /= miimon;
5166 if (arp_interval < 0) {
5167 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5168 arp_interval, INT_MAX);
5172 for (arp_ip_count = 0, i = 0;
5173 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5176 /* not a complete check, but good enough to catch mistakes */
5177 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5178 !bond_is_ip_target_ok(ip)) {
5179 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5183 if (bond_get_targets_ip(arp_target, ip) == -1)
5184 arp_target[arp_ip_count++] = ip;
5186 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5191 if (arp_interval && !arp_ip_count) {
5192 /* don't allow arping if no arp_ip_target given... */
5193 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5199 if (!arp_interval) {
5200 pr_err("arp_validate requires arp_interval\n");
5204 bond_opt_initstr(&newval, arp_validate);
5205 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5208 pr_err("Error: invalid arp_validate \"%s\"\n",
5212 arp_validate_value = valptr->value;
5214 arp_validate_value = 0;
5217 if (arp_all_targets) {
5218 bond_opt_initstr(&newval, arp_all_targets);
5219 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5222 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5224 arp_all_targets_value = 0;
5226 arp_all_targets_value = valptr->value;
5231 pr_info("MII link monitoring set to %d ms\n", miimon);
5232 } else if (arp_interval) {
5233 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5234 arp_validate_value);
5235 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5236 arp_interval, valptr->string, arp_ip_count);
5238 for (i = 0; i < arp_ip_count; i++)
5239 pr_cont(" %s", arp_ip_target[i]);
5243 } else if (max_bonds) {
5244 /* miimon and arp_interval not set, we need one so things
5245 * work as expected, see bonding.txt for details
5247 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");
5250 if (primary && !bond_mode_uses_primary(bond_mode)) {
5251 /* currently, using a primary only makes sense
5252 * in active backup, TLB or ALB modes
5254 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5255 primary, bond_mode_name(bond_mode));
5259 if (primary && primary_reselect) {
5260 bond_opt_initstr(&newval, primary_reselect);
5261 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5264 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5268 primary_reselect_value = valptr->value;
5270 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5273 if (fail_over_mac) {
5274 bond_opt_initstr(&newval, fail_over_mac);
5275 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5278 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5282 fail_over_mac_value = valptr->value;
5283 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5284 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5286 fail_over_mac_value = BOND_FOM_NONE;
5289 bond_opt_initstr(&newval, "default");
5290 valptr = bond_opt_parse(
5291 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5294 pr_err("Error: No ad_actor_sys_prio default value");
5297 ad_actor_sys_prio = valptr->value;
5299 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5302 pr_err("Error: No ad_user_port_key default value");
5305 ad_user_port_key = valptr->value;
5307 bond_opt_initstr(&newval, "default");
5308 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5310 pr_err("Error: No tlb_dynamic_lb default value");
5313 tlb_dynamic_lb = valptr->value;
5315 if (lp_interval == 0) {
5316 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5317 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5318 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5321 /* fill params struct with the proper values */
5322 params->mode = bond_mode;
5323 params->xmit_policy = xmit_hashtype;
5324 params->miimon = miimon;
5325 params->num_peer_notif = num_peer_notif;
5326 params->arp_interval = arp_interval;
5327 params->arp_validate = arp_validate_value;
5328 params->arp_all_targets = arp_all_targets_value;
5329 params->updelay = updelay;
5330 params->downdelay = downdelay;
5331 params->peer_notif_delay = 0;
5332 params->use_carrier = use_carrier;
5333 params->lacp_fast = lacp_fast;
5334 params->primary[0] = 0;
5335 params->primary_reselect = primary_reselect_value;
5336 params->fail_over_mac = fail_over_mac_value;
5337 params->tx_queues = tx_queues;
5338 params->all_slaves_active = all_slaves_active;
5339 params->resend_igmp = resend_igmp;
5340 params->min_links = min_links;
5341 params->lp_interval = lp_interval;
5342 params->packets_per_slave = packets_per_slave;
5343 params->tlb_dynamic_lb = tlb_dynamic_lb;
5344 params->ad_actor_sys_prio = ad_actor_sys_prio;
5345 eth_zero_addr(params->ad_actor_system);
5346 params->ad_user_port_key = ad_user_port_key;
5347 if (packets_per_slave > 0) {
5348 params->reciprocal_packets_per_slave =
5349 reciprocal_value(packets_per_slave);
5351 /* reciprocal_packets_per_slave is unused if
5352 * packets_per_slave is 0 or 1, just initialize it
5354 params->reciprocal_packets_per_slave =
5355 (struct reciprocal_value) { 0 };
5359 strncpy(params->primary, primary, IFNAMSIZ);
5360 params->primary[IFNAMSIZ - 1] = 0;
5363 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5368 /* Called from registration process */
5369 static int bond_init(struct net_device *bond_dev)
5371 struct bonding *bond = netdev_priv(bond_dev);
5372 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5374 netdev_dbg(bond_dev, "Begin bond_init\n");
5376 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5380 bond->notifier_ctx = false;
5382 spin_lock_init(&bond->stats_lock);
5383 netdev_lockdep_set_classes(bond_dev);
5385 list_add_tail(&bond->bond_list, &bn->dev_list);
5387 bond_prepare_sysfs_group(bond);
5389 bond_debug_register(bond);
5391 /* Ensure valid dev_addr */
5392 if (is_zero_ether_addr(bond_dev->dev_addr) &&
5393 bond_dev->addr_assign_type == NET_ADDR_PERM)
5394 eth_hw_addr_random(bond_dev);
5399 unsigned int bond_get_num_tx_queues(void)
5404 /* Create a new bond based on the specified name and bonding parameters.
5405 * If name is NULL, obtain a suitable "bond%d" name for us.
5406 * Caller must NOT hold rtnl_lock; we need to release it here before we
5407 * set up our sysfs entries.
5409 int bond_create(struct net *net, const char *name)
5411 struct net_device *bond_dev;
5412 struct bonding *bond;
5413 struct alb_bond_info *bond_info;
5418 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5419 name ? name : "bond%d", NET_NAME_UNKNOWN,
5420 bond_setup, tx_queues);
5422 pr_err("%s: eek! can't alloc netdev!\n", name);
5428 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5429 * It is set to 0 by default which is wrong.
5431 bond = netdev_priv(bond_dev);
5432 bond_info = &(BOND_ALB_INFO(bond));
5433 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5435 dev_net_set(bond_dev, net);
5436 bond_dev->rtnl_link_ops = &bond_link_ops;
5438 res = register_netdevice(bond_dev);
5440 free_netdev(bond_dev);
5446 netif_carrier_off(bond_dev);
5448 bond_work_init_all(bond);
5454 static int __net_init bond_net_init(struct net *net)
5456 struct bond_net *bn = net_generic(net, bond_net_id);
5459 INIT_LIST_HEAD(&bn->dev_list);
5461 bond_create_proc_dir(bn);
5462 bond_create_sysfs(bn);
5467 static void __net_exit bond_net_exit(struct net *net)
5469 struct bond_net *bn = net_generic(net, bond_net_id);
5470 struct bonding *bond, *tmp_bond;
5473 bond_destroy_sysfs(bn);
5475 /* Kill off any bonds created after unregistering bond rtnl ops */
5477 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5478 unregister_netdevice_queue(bond->dev, &list);
5479 unregister_netdevice_many(&list);
5482 bond_destroy_proc_dir(bn);
5485 static struct pernet_operations bond_net_ops = {
5486 .init = bond_net_init,
5487 .exit = bond_net_exit,
5489 .size = sizeof(struct bond_net),
5492 static int __init bonding_init(void)
5497 res = bond_check_params(&bonding_defaults);
5501 res = register_pernet_subsys(&bond_net_ops);
5505 res = bond_netlink_init();
5509 bond_create_debugfs();
5511 for (i = 0; i < max_bonds; i++) {
5512 res = bond_create(&init_net, NULL);
5517 skb_flow_dissector_init(&flow_keys_bonding,
5518 flow_keys_bonding_keys,
5519 ARRAY_SIZE(flow_keys_bonding_keys));
5521 register_netdevice_notifier(&bond_netdev_notifier);
5525 bond_destroy_debugfs();
5526 bond_netlink_fini();
5528 unregister_pernet_subsys(&bond_net_ops);
5533 static void __exit bonding_exit(void)
5535 unregister_netdevice_notifier(&bond_netdev_notifier);
5537 bond_destroy_debugfs();
5539 bond_netlink_fini();
5540 unregister_pernet_subsys(&bond_net_ops);
5542 #ifdef CONFIG_NET_POLL_CONTROLLER
5543 /* Make sure we don't have an imbalance on our netpoll blocking */
5544 WARN_ON(atomic_read(&netpoll_block_tx));
5548 module_init(bonding_init);
5549 module_exit(bonding_exit);
5550 MODULE_LICENSE("GPL");
5551 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5552 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");