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>
86 #if IS_ENABLED(CONFIG_TLS_DEVICE)
90 #include "bonding_priv.h"
92 /*---------------------------- Module parameters ----------------------------*/
94 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
96 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
97 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
98 static int num_peer_notif = 1;
101 static int downdelay;
102 static int use_carrier = 1;
104 static char *primary;
105 static char *primary_reselect;
106 static char *lacp_rate;
107 static int min_links;
108 static char *ad_select;
109 static char *xmit_hash_policy;
110 static int arp_interval;
111 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
112 static char *arp_validate;
113 static char *arp_all_targets;
114 static char *fail_over_mac;
115 static int all_slaves_active;
116 static struct bond_params bonding_defaults;
117 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
118 static int packets_per_slave = 1;
119 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
121 module_param(max_bonds, int, 0);
122 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
123 module_param(tx_queues, int, 0);
124 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
125 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
126 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
127 "failover event (alias of num_unsol_na)");
128 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
129 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
130 "failover event (alias of num_grat_arp)");
131 module_param(miimon, int, 0);
132 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
133 module_param(updelay, int, 0);
134 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
135 module_param(downdelay, int, 0);
136 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
138 module_param(use_carrier, int, 0);
139 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
140 "0 for off, 1 for on (default)");
141 module_param(mode, charp, 0);
142 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
143 "1 for active-backup, 2 for balance-xor, "
144 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
145 "6 for balance-alb");
146 module_param(primary, charp, 0);
147 MODULE_PARM_DESC(primary, "Primary network device to use");
148 module_param(primary_reselect, charp, 0);
149 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
151 "0 for always (default), "
152 "1 for only if speed of primary is "
154 "2 for only on active slave "
156 module_param(lacp_rate, charp, 0);
157 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
158 "0 for slow, 1 for fast");
159 module_param(ad_select, charp, 0);
160 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
161 "0 for stable (default), 1 for bandwidth, "
163 module_param(min_links, int, 0);
164 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
166 module_param(xmit_hash_policy, charp, 0);
167 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
168 "0 for layer 2 (default), 1 for layer 3+4, "
169 "2 for layer 2+3, 3 for encap layer 2+3, "
170 "4 for encap layer 3+4, 5 for vlan+srcmac");
171 module_param(arp_interval, int, 0);
172 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
173 module_param_array(arp_ip_target, charp, NULL, 0);
174 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
175 module_param(arp_validate, charp, 0);
176 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
177 "0 for none (default), 1 for active, "
178 "2 for backup, 3 for all");
179 module_param(arp_all_targets, charp, 0);
180 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
181 module_param(fail_over_mac, charp, 0);
182 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
183 "the same MAC; 0 for none (default), "
184 "1 for active, 2 for follow");
185 module_param(all_slaves_active, int, 0);
186 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
187 "by setting active flag for all slaves; "
188 "0 for never (default), 1 for always.");
189 module_param(resend_igmp, int, 0);
190 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
192 module_param(packets_per_slave, int, 0);
193 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
194 "mode; 0 for a random slave, 1 packet per "
195 "slave (default), >1 packets per slave.");
196 module_param(lp_interval, uint, 0);
197 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
198 "the bonding driver sends learning packets to "
199 "each slaves peer switch. The default is 1.");
201 /*----------------------------- Global variables ----------------------------*/
203 #ifdef CONFIG_NET_POLL_CONTROLLER
204 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
207 unsigned int bond_net_id __read_mostly;
209 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
211 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
212 .offset = offsetof(struct flow_keys, control),
215 .key_id = FLOW_DISSECTOR_KEY_BASIC,
216 .offset = offsetof(struct flow_keys, basic),
219 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
220 .offset = offsetof(struct flow_keys, addrs.v4addrs),
223 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
224 .offset = offsetof(struct flow_keys, addrs.v6addrs),
227 .key_id = FLOW_DISSECTOR_KEY_TIPC,
228 .offset = offsetof(struct flow_keys, addrs.tipckey),
231 .key_id = FLOW_DISSECTOR_KEY_PORTS,
232 .offset = offsetof(struct flow_keys, ports),
235 .key_id = FLOW_DISSECTOR_KEY_ICMP,
236 .offset = offsetof(struct flow_keys, icmp),
239 .key_id = FLOW_DISSECTOR_KEY_VLAN,
240 .offset = offsetof(struct flow_keys, vlan),
243 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
244 .offset = offsetof(struct flow_keys, tags),
247 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
248 .offset = offsetof(struct flow_keys, keyid),
252 static struct flow_dissector flow_keys_bonding __read_mostly;
254 /*-------------------------- Forward declarations ---------------------------*/
256 static int bond_init(struct net_device *bond_dev);
257 static void bond_uninit(struct net_device *bond_dev);
258 static void bond_get_stats(struct net_device *bond_dev,
259 struct rtnl_link_stats64 *stats);
260 static void bond_slave_arr_handler(struct work_struct *work);
261 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
263 static void bond_netdev_notify_work(struct work_struct *work);
265 /*---------------------------- General routines -----------------------------*/
267 const char *bond_mode_name(int mode)
269 static const char *names[] = {
270 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
271 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
272 [BOND_MODE_XOR] = "load balancing (xor)",
273 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
274 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
275 [BOND_MODE_TLB] = "transmit load balancing",
276 [BOND_MODE_ALB] = "adaptive load balancing",
279 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
295 skb->dev = slave_dev;
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
304 return dev_queue_xmit(skb);
307 bool bond_sk_check(struct bonding *bond)
309 switch (BOND_MODE(bond)) {
310 case BOND_MODE_8023AD:
312 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
320 /*---------------------------------- VLAN -----------------------------------*/
322 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
323 * We don't protect the slave list iteration with a lock because:
324 * a. This operation is performed in IOCTL context,
325 * b. The operation is protected by the RTNL semaphore in the 8021q code,
326 * c. Holding a lock with BH disabled while directly calling a base driver
327 * entry point is generally a BAD idea.
329 * The design of synchronization/protection for this operation in the 8021q
330 * module is good for one or more VLAN devices over a single physical device
331 * and cannot be extended for a teaming solution like bonding, so there is a
332 * potential race condition here where a net device from the vlan group might
333 * be referenced (either by a base driver or the 8021q code) while it is being
334 * removed from the system. However, it turns out we're not making matters
335 * worse, and if it works for regular VLAN usage it will work here too.
339 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
340 * @bond_dev: bonding net device that got called
341 * @proto: network protocol ID
342 * @vid: vlan id being added
344 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
345 __be16 proto, u16 vid)
347 struct bonding *bond = netdev_priv(bond_dev);
348 struct slave *slave, *rollback_slave;
349 struct list_head *iter;
352 bond_for_each_slave(bond, slave, iter) {
353 res = vlan_vid_add(slave->dev, proto, vid);
361 /* unwind to the slave that failed */
362 bond_for_each_slave(bond, rollback_slave, iter) {
363 if (rollback_slave == slave)
366 vlan_vid_del(rollback_slave->dev, proto, vid);
373 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
374 * @bond_dev: bonding net device that got called
375 * @proto: network protocol ID
376 * @vid: vlan id being removed
378 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
379 __be16 proto, u16 vid)
381 struct bonding *bond = netdev_priv(bond_dev);
382 struct list_head *iter;
385 bond_for_each_slave(bond, slave, iter)
386 vlan_vid_del(slave->dev, proto, vid);
388 if (bond_is_lb(bond))
389 bond_alb_clear_vlan(bond, vid);
394 /*---------------------------------- XFRM -----------------------------------*/
396 #ifdef CONFIG_XFRM_OFFLOAD
398 * bond_ipsec_add_sa - program device with a security association
399 * @xs: pointer to transformer state struct
401 static int bond_ipsec_add_sa(struct xfrm_state *xs)
403 struct net_device *bond_dev = xs->xso.dev;
404 struct bond_ipsec *ipsec;
405 struct bonding *bond;
413 bond = netdev_priv(bond_dev);
414 slave = rcu_dereference(bond->curr_active_slave);
420 if (!slave->dev->xfrmdev_ops ||
421 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
422 netif_is_bond_master(slave->dev)) {
423 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
428 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
433 xs->xso.real_dev = slave->dev;
435 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
438 INIT_LIST_HEAD(&ipsec->list);
439 spin_lock_bh(&bond->ipsec_lock);
440 list_add(&ipsec->list, &bond->ipsec_list);
441 spin_unlock_bh(&bond->ipsec_lock);
449 static void bond_ipsec_add_sa_all(struct bonding *bond)
451 struct net_device *bond_dev = bond->dev;
452 struct bond_ipsec *ipsec;
456 slave = rcu_dereference(bond->curr_active_slave);
460 if (!slave->dev->xfrmdev_ops ||
461 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
462 netif_is_bond_master(slave->dev)) {
463 spin_lock_bh(&bond->ipsec_lock);
464 if (!list_empty(&bond->ipsec_list))
465 slave_warn(bond_dev, slave->dev,
466 "%s: no slave xdo_dev_state_add\n",
468 spin_unlock_bh(&bond->ipsec_lock);
472 spin_lock_bh(&bond->ipsec_lock);
473 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
474 ipsec->xs->xso.real_dev = slave->dev;
475 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
476 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
477 ipsec->xs->xso.real_dev = NULL;
480 spin_unlock_bh(&bond->ipsec_lock);
486 * bond_ipsec_del_sa - clear out this specific SA
487 * @xs: pointer to transformer state struct
489 static void bond_ipsec_del_sa(struct xfrm_state *xs)
491 struct net_device *bond_dev = xs->xso.dev;
492 struct bond_ipsec *ipsec;
493 struct bonding *bond;
500 bond = netdev_priv(bond_dev);
501 slave = rcu_dereference(bond->curr_active_slave);
506 if (!xs->xso.real_dev)
509 WARN_ON(xs->xso.real_dev != slave->dev);
511 if (!slave->dev->xfrmdev_ops ||
512 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
513 netif_is_bond_master(slave->dev)) {
514 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
518 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
520 spin_lock_bh(&bond->ipsec_lock);
521 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
522 if (ipsec->xs == xs) {
523 list_del(&ipsec->list);
528 spin_unlock_bh(&bond->ipsec_lock);
532 static void bond_ipsec_del_sa_all(struct bonding *bond)
534 struct net_device *bond_dev = bond->dev;
535 struct bond_ipsec *ipsec;
539 slave = rcu_dereference(bond->curr_active_slave);
545 spin_lock_bh(&bond->ipsec_lock);
546 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
547 if (!ipsec->xs->xso.real_dev)
550 if (!slave->dev->xfrmdev_ops ||
551 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
552 netif_is_bond_master(slave->dev)) {
553 slave_warn(bond_dev, slave->dev,
554 "%s: no slave xdo_dev_state_delete\n",
557 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
559 ipsec->xs->xso.real_dev = NULL;
561 spin_unlock_bh(&bond->ipsec_lock);
566 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
567 * @skb: current data packet
568 * @xs: pointer to transformer state struct
570 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
572 struct net_device *bond_dev = xs->xso.dev;
573 struct net_device *real_dev;
574 struct slave *curr_active;
575 struct bonding *bond;
578 bond = netdev_priv(bond_dev);
580 curr_active = rcu_dereference(bond->curr_active_slave);
581 real_dev = curr_active->dev;
583 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
588 if (!xs->xso.real_dev) {
593 if (!real_dev->xfrmdev_ops ||
594 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
595 netif_is_bond_master(real_dev)) {
600 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
606 static const struct xfrmdev_ops bond_xfrmdev_ops = {
607 .xdo_dev_state_add = bond_ipsec_add_sa,
608 .xdo_dev_state_delete = bond_ipsec_del_sa,
609 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
611 #endif /* CONFIG_XFRM_OFFLOAD */
613 /*------------------------------- Link status -------------------------------*/
615 /* Set the carrier state for the master according to the state of its
616 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
617 * do special 802.3ad magic.
619 * Returns zero if carrier state does not change, nonzero if it does.
621 int bond_set_carrier(struct bonding *bond)
623 struct list_head *iter;
626 if (!bond_has_slaves(bond))
629 if (BOND_MODE(bond) == BOND_MODE_8023AD)
630 return bond_3ad_set_carrier(bond);
632 bond_for_each_slave(bond, slave, iter) {
633 if (slave->link == BOND_LINK_UP) {
634 if (!netif_carrier_ok(bond->dev)) {
635 netif_carrier_on(bond->dev);
643 if (netif_carrier_ok(bond->dev)) {
644 netif_carrier_off(bond->dev);
650 /* Get link speed and duplex from the slave's base driver
651 * using ethtool. If for some reason the call fails or the
652 * values are invalid, set speed and duplex to -1,
653 * and return. Return 1 if speed or duplex settings are
654 * UNKNOWN; 0 otherwise.
656 static int bond_update_speed_duplex(struct slave *slave)
658 struct net_device *slave_dev = slave->dev;
659 struct ethtool_link_ksettings ecmd;
662 slave->speed = SPEED_UNKNOWN;
663 slave->duplex = DUPLEX_UNKNOWN;
665 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
668 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
670 switch (ecmd.base.duplex) {
678 slave->speed = ecmd.base.speed;
679 slave->duplex = ecmd.base.duplex;
684 const char *bond_slave_link_status(s8 link)
700 /* if <dev> supports MII link status reporting, check its link status.
702 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
703 * depending upon the setting of the use_carrier parameter.
705 * Return either BMSR_LSTATUS, meaning that the link is up (or we
706 * can't tell and just pretend it is), or 0, meaning that the link is
709 * If reporting is non-zero, instead of faking link up, return -1 if
710 * both ETHTOOL and MII ioctls fail (meaning the device does not
711 * support them). If use_carrier is set, return whatever it says.
712 * It'd be nice if there was a good way to tell if a driver supports
713 * netif_carrier, but there really isn't.
715 static int bond_check_dev_link(struct bonding *bond,
716 struct net_device *slave_dev, int reporting)
718 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
719 int (*ioctl)(struct net_device *, struct ifreq *, int);
721 struct mii_ioctl_data *mii;
723 if (!reporting && !netif_running(slave_dev))
726 if (bond->params.use_carrier)
727 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
729 /* Try to get link status using Ethtool first. */
730 if (slave_dev->ethtool_ops->get_link)
731 return slave_dev->ethtool_ops->get_link(slave_dev) ?
734 /* Ethtool can't be used, fallback to MII ioctls. */
735 ioctl = slave_ops->ndo_do_ioctl;
737 /* TODO: set pointer to correct ioctl on a per team member
738 * bases to make this more efficient. that is, once
739 * we determine the correct ioctl, we will always
740 * call it and not the others for that team
744 /* We cannot assume that SIOCGMIIPHY will also read a
745 * register; not all network drivers (e.g., e100)
749 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
750 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
752 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
753 mii->reg_num = MII_BMSR;
754 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
755 return mii->val_out & BMSR_LSTATUS;
759 /* If reporting, report that either there's no dev->do_ioctl,
760 * or both SIOCGMIIREG and get_link failed (meaning that we
761 * cannot report link status). If not reporting, pretend
764 return reporting ? -1 : BMSR_LSTATUS;
767 /*----------------------------- Multicast list ------------------------------*/
769 /* Push the promiscuity flag down to appropriate slaves */
770 static int bond_set_promiscuity(struct bonding *bond, int inc)
772 struct list_head *iter;
775 if (bond_uses_primary(bond)) {
776 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
779 err = dev_set_promiscuity(curr_active->dev, inc);
783 bond_for_each_slave(bond, slave, iter) {
784 err = dev_set_promiscuity(slave->dev, inc);
792 /* Push the allmulti flag down to all slaves */
793 static int bond_set_allmulti(struct bonding *bond, int inc)
795 struct list_head *iter;
798 if (bond_uses_primary(bond)) {
799 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
802 err = dev_set_allmulti(curr_active->dev, inc);
806 bond_for_each_slave(bond, slave, iter) {
807 err = dev_set_allmulti(slave->dev, inc);
815 /* Retrieve the list of registered multicast addresses for the bonding
816 * device and retransmit an IGMP JOIN request to the current active
819 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
821 struct bonding *bond = container_of(work, struct bonding,
824 if (!rtnl_trylock()) {
825 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
828 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
830 if (bond->igmp_retrans > 1) {
831 bond->igmp_retrans--;
832 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
837 /* Flush bond's hardware addresses from slave */
838 static void bond_hw_addr_flush(struct net_device *bond_dev,
839 struct net_device *slave_dev)
841 struct bonding *bond = netdev_priv(bond_dev);
843 dev_uc_unsync(slave_dev, bond_dev);
844 dev_mc_unsync(slave_dev, bond_dev);
846 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
847 /* del lacpdu mc addr from mc list */
848 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
850 dev_mc_del(slave_dev, lacpdu_multicast);
854 /*--------------------------- Active slave change ---------------------------*/
856 /* Update the hardware address list and promisc/allmulti for the new and
857 * old active slaves (if any). Modes that are not using primary keep all
858 * slaves up date at all times; only the modes that use primary need to call
859 * this function to swap these settings during a failover.
861 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
862 struct slave *old_active)
865 if (bond->dev->flags & IFF_PROMISC)
866 dev_set_promiscuity(old_active->dev, -1);
868 if (bond->dev->flags & IFF_ALLMULTI)
869 dev_set_allmulti(old_active->dev, -1);
871 bond_hw_addr_flush(bond->dev, old_active->dev);
875 /* FIXME: Signal errors upstream. */
876 if (bond->dev->flags & IFF_PROMISC)
877 dev_set_promiscuity(new_active->dev, 1);
879 if (bond->dev->flags & IFF_ALLMULTI)
880 dev_set_allmulti(new_active->dev, 1);
882 netif_addr_lock_bh(bond->dev);
883 dev_uc_sync(new_active->dev, bond->dev);
884 dev_mc_sync(new_active->dev, bond->dev);
885 netif_addr_unlock_bh(bond->dev);
890 * bond_set_dev_addr - clone slave's address to bond
891 * @bond_dev: bond net device
892 * @slave_dev: slave net device
894 * Should be called with RTNL held.
896 static int bond_set_dev_addr(struct net_device *bond_dev,
897 struct net_device *slave_dev)
901 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
902 bond_dev, slave_dev, slave_dev->addr_len);
903 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
907 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
908 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
909 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
913 static struct slave *bond_get_old_active(struct bonding *bond,
914 struct slave *new_active)
917 struct list_head *iter;
919 bond_for_each_slave(bond, slave, iter) {
920 if (slave == new_active)
923 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
930 /* bond_do_fail_over_mac
932 * Perform special MAC address swapping for fail_over_mac settings
936 static void bond_do_fail_over_mac(struct bonding *bond,
937 struct slave *new_active,
938 struct slave *old_active)
940 u8 tmp_mac[MAX_ADDR_LEN];
941 struct sockaddr_storage ss;
944 switch (bond->params.fail_over_mac) {
945 case BOND_FOM_ACTIVE:
947 rv = bond_set_dev_addr(bond->dev, new_active->dev);
949 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
953 case BOND_FOM_FOLLOW:
954 /* if new_active && old_active, swap them
955 * if just old_active, do nothing (going to no active slave)
956 * if just new_active, set new_active to bond's MAC
962 old_active = bond_get_old_active(bond, new_active);
965 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
966 new_active->dev->addr_len);
967 bond_hw_addr_copy(ss.__data,
968 old_active->dev->dev_addr,
969 old_active->dev->addr_len);
970 ss.ss_family = new_active->dev->type;
972 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
973 bond->dev->addr_len);
974 ss.ss_family = bond->dev->type;
977 rv = dev_set_mac_address(new_active->dev,
978 (struct sockaddr *)&ss, NULL);
980 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
988 bond_hw_addr_copy(ss.__data, tmp_mac,
989 new_active->dev->addr_len);
990 ss.ss_family = old_active->dev->type;
992 rv = dev_set_mac_address(old_active->dev,
993 (struct sockaddr *)&ss, NULL);
995 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1000 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1001 bond->params.fail_over_mac);
1007 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1009 struct slave *prim = rtnl_dereference(bond->primary_slave);
1010 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1012 if (!prim || prim->link != BOND_LINK_UP) {
1013 if (!curr || curr->link != BOND_LINK_UP)
1018 if (bond->force_primary) {
1019 bond->force_primary = false;
1023 if (!curr || curr->link != BOND_LINK_UP)
1026 /* At this point, prim and curr are both up */
1027 switch (bond->params.primary_reselect) {
1028 case BOND_PRI_RESELECT_ALWAYS:
1030 case BOND_PRI_RESELECT_BETTER:
1031 if (prim->speed < curr->speed)
1033 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1036 case BOND_PRI_RESELECT_FAILURE:
1039 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1040 bond->params.primary_reselect);
1046 * bond_find_best_slave - select the best available slave to be the active one
1047 * @bond: our bonding struct
1049 static struct slave *bond_find_best_slave(struct bonding *bond)
1051 struct slave *slave, *bestslave = NULL;
1052 struct list_head *iter;
1053 int mintime = bond->params.updelay;
1055 slave = bond_choose_primary_or_current(bond);
1059 bond_for_each_slave(bond, slave, iter) {
1060 if (slave->link == BOND_LINK_UP)
1062 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1063 slave->delay < mintime) {
1064 mintime = slave->delay;
1072 static bool bond_should_notify_peers(struct bonding *bond)
1074 struct slave *slave;
1077 slave = rcu_dereference(bond->curr_active_slave);
1080 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1081 slave ? slave->dev->name : "NULL");
1083 if (!slave || !bond->send_peer_notif ||
1084 bond->send_peer_notif %
1085 max(1, bond->params.peer_notif_delay) != 0 ||
1086 !netif_carrier_ok(bond->dev) ||
1087 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1094 * bond_change_active_slave - change the active slave into the specified one
1095 * @bond: our bonding struct
1096 * @new_active: the new slave to make the active one
1098 * Set the new slave to the bond's settings and unset them on the old
1099 * curr_active_slave.
1100 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1102 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1103 * because it is apparently the best available slave we have, even though its
1104 * updelay hasn't timed out yet.
1106 * Caller must hold RTNL.
1108 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1110 struct slave *old_active;
1114 old_active = rtnl_dereference(bond->curr_active_slave);
1116 if (old_active == new_active)
1119 #ifdef CONFIG_XFRM_OFFLOAD
1120 bond_ipsec_del_sa_all(bond);
1121 #endif /* CONFIG_XFRM_OFFLOAD */
1124 new_active->last_link_up = jiffies;
1126 if (new_active->link == BOND_LINK_BACK) {
1127 if (bond_uses_primary(bond)) {
1128 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1129 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1132 new_active->delay = 0;
1133 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1134 BOND_SLAVE_NOTIFY_NOW);
1136 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1137 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1139 if (bond_is_lb(bond))
1140 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1142 if (bond_uses_primary(bond))
1143 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1147 if (bond_uses_primary(bond))
1148 bond_hw_addr_swap(bond, new_active, old_active);
1150 if (bond_is_lb(bond)) {
1151 bond_alb_handle_active_change(bond, new_active);
1153 bond_set_slave_inactive_flags(old_active,
1154 BOND_SLAVE_NOTIFY_NOW);
1156 bond_set_slave_active_flags(new_active,
1157 BOND_SLAVE_NOTIFY_NOW);
1159 rcu_assign_pointer(bond->curr_active_slave, new_active);
1162 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1164 bond_set_slave_inactive_flags(old_active,
1165 BOND_SLAVE_NOTIFY_NOW);
1168 bool should_notify_peers = false;
1170 bond_set_slave_active_flags(new_active,
1171 BOND_SLAVE_NOTIFY_NOW);
1173 if (bond->params.fail_over_mac)
1174 bond_do_fail_over_mac(bond, new_active,
1177 if (netif_running(bond->dev)) {
1178 bond->send_peer_notif =
1179 bond->params.num_peer_notif *
1180 max(1, bond->params.peer_notif_delay);
1181 should_notify_peers =
1182 bond_should_notify_peers(bond);
1185 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1186 if (should_notify_peers) {
1187 bond->send_peer_notif--;
1188 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1194 #ifdef CONFIG_XFRM_OFFLOAD
1195 bond_ipsec_add_sa_all(bond);
1196 #endif /* CONFIG_XFRM_OFFLOAD */
1198 /* resend IGMP joins since active slave has changed or
1199 * all were sent on curr_active_slave.
1200 * resend only if bond is brought up with the affected
1201 * bonding modes and the retransmission is enabled
1203 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1204 ((bond_uses_primary(bond) && new_active) ||
1205 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1206 bond->igmp_retrans = bond->params.resend_igmp;
1207 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1212 * bond_select_active_slave - select a new active slave, if needed
1213 * @bond: our bonding struct
1215 * This functions should be called when one of the following occurs:
1216 * - The old curr_active_slave has been released or lost its link.
1217 * - The primary_slave has got its link back.
1218 * - A slave has got its link back and there's no old curr_active_slave.
1220 * Caller must hold RTNL.
1222 void bond_select_active_slave(struct bonding *bond)
1224 struct slave *best_slave;
1229 best_slave = bond_find_best_slave(bond);
1230 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1231 bond_change_active_slave(bond, best_slave);
1232 rv = bond_set_carrier(bond);
1236 if (netif_carrier_ok(bond->dev))
1237 netdev_info(bond->dev, "active interface up!\n");
1239 netdev_info(bond->dev, "now running without any active interface!\n");
1243 #ifdef CONFIG_NET_POLL_CONTROLLER
1244 static inline int slave_enable_netpoll(struct slave *slave)
1249 np = kzalloc(sizeof(*np), GFP_KERNEL);
1254 err = __netpoll_setup(np, slave->dev);
1263 static inline void slave_disable_netpoll(struct slave *slave)
1265 struct netpoll *np = slave->np;
1275 static void bond_poll_controller(struct net_device *bond_dev)
1277 struct bonding *bond = netdev_priv(bond_dev);
1278 struct slave *slave = NULL;
1279 struct list_head *iter;
1280 struct ad_info ad_info;
1282 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1283 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1286 bond_for_each_slave_rcu(bond, slave, iter) {
1287 if (!bond_slave_is_up(slave))
1290 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1291 struct aggregator *agg =
1292 SLAVE_AD_INFO(slave)->port.aggregator;
1295 agg->aggregator_identifier != ad_info.aggregator_id)
1299 netpoll_poll_dev(slave->dev);
1303 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1305 struct bonding *bond = netdev_priv(bond_dev);
1306 struct list_head *iter;
1307 struct slave *slave;
1309 bond_for_each_slave(bond, slave, iter)
1310 if (bond_slave_is_up(slave))
1311 slave_disable_netpoll(slave);
1314 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1316 struct bonding *bond = netdev_priv(dev);
1317 struct list_head *iter;
1318 struct slave *slave;
1321 bond_for_each_slave(bond, slave, iter) {
1322 err = slave_enable_netpoll(slave);
1324 bond_netpoll_cleanup(dev);
1331 static inline int slave_enable_netpoll(struct slave *slave)
1335 static inline void slave_disable_netpoll(struct slave *slave)
1338 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1343 /*---------------------------------- IOCTL ----------------------------------*/
1345 static netdev_features_t bond_fix_features(struct net_device *dev,
1346 netdev_features_t features)
1348 struct bonding *bond = netdev_priv(dev);
1349 struct list_head *iter;
1350 netdev_features_t mask;
1351 struct slave *slave;
1353 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1354 if (bond_sk_check(bond))
1355 features |= BOND_TLS_FEATURES;
1357 features &= ~BOND_TLS_FEATURES;
1362 features &= ~NETIF_F_ONE_FOR_ALL;
1363 features |= NETIF_F_ALL_FOR_ALL;
1365 bond_for_each_slave(bond, slave, iter) {
1366 features = netdev_increment_features(features,
1367 slave->dev->features,
1370 features = netdev_add_tso_features(features, mask);
1375 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1376 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1377 NETIF_F_HIGHDMA | NETIF_F_LRO)
1379 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1380 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1382 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1383 NETIF_F_GSO_SOFTWARE)
1386 static void bond_compute_features(struct bonding *bond)
1388 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1389 IFF_XMIT_DST_RELEASE_PERM;
1390 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1391 netdev_features_t enc_features = BOND_ENC_FEATURES;
1392 #ifdef CONFIG_XFRM_OFFLOAD
1393 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1394 #endif /* CONFIG_XFRM_OFFLOAD */
1395 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1396 struct net_device *bond_dev = bond->dev;
1397 struct list_head *iter;
1398 struct slave *slave;
1399 unsigned short max_hard_header_len = ETH_HLEN;
1400 unsigned int gso_max_size = GSO_MAX_SIZE;
1401 u16 gso_max_segs = GSO_MAX_SEGS;
1403 if (!bond_has_slaves(bond))
1405 vlan_features &= NETIF_F_ALL_FOR_ALL;
1406 mpls_features &= NETIF_F_ALL_FOR_ALL;
1408 bond_for_each_slave(bond, slave, iter) {
1409 vlan_features = netdev_increment_features(vlan_features,
1410 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1412 enc_features = netdev_increment_features(enc_features,
1413 slave->dev->hw_enc_features,
1416 #ifdef CONFIG_XFRM_OFFLOAD
1417 xfrm_features = netdev_increment_features(xfrm_features,
1418 slave->dev->hw_enc_features,
1419 BOND_XFRM_FEATURES);
1420 #endif /* CONFIG_XFRM_OFFLOAD */
1422 mpls_features = netdev_increment_features(mpls_features,
1423 slave->dev->mpls_features,
1424 BOND_MPLS_FEATURES);
1426 dst_release_flag &= slave->dev->priv_flags;
1427 if (slave->dev->hard_header_len > max_hard_header_len)
1428 max_hard_header_len = slave->dev->hard_header_len;
1430 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1431 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1433 bond_dev->hard_header_len = max_hard_header_len;
1436 bond_dev->vlan_features = vlan_features;
1437 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1438 NETIF_F_HW_VLAN_CTAG_TX |
1439 NETIF_F_HW_VLAN_STAG_TX;
1440 #ifdef CONFIG_XFRM_OFFLOAD
1441 bond_dev->hw_enc_features |= xfrm_features;
1442 #endif /* CONFIG_XFRM_OFFLOAD */
1443 bond_dev->mpls_features = mpls_features;
1444 bond_dev->gso_max_segs = gso_max_segs;
1445 netif_set_gso_max_size(bond_dev, gso_max_size);
1447 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1448 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1449 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1450 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1452 netdev_change_features(bond_dev);
1455 static void bond_setup_by_slave(struct net_device *bond_dev,
1456 struct net_device *slave_dev)
1458 bond_dev->header_ops = slave_dev->header_ops;
1460 bond_dev->type = slave_dev->type;
1461 bond_dev->hard_header_len = slave_dev->hard_header_len;
1462 bond_dev->needed_headroom = slave_dev->needed_headroom;
1463 bond_dev->addr_len = slave_dev->addr_len;
1465 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1466 slave_dev->addr_len);
1469 /* On bonding slaves other than the currently active slave, suppress
1470 * duplicates except for alb non-mcast/bcast.
1472 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1473 struct slave *slave,
1474 struct bonding *bond)
1476 if (bond_is_slave_inactive(slave)) {
1477 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1478 skb->pkt_type != PACKET_BROADCAST &&
1479 skb->pkt_type != PACKET_MULTICAST)
1486 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1488 struct sk_buff *skb = *pskb;
1489 struct slave *slave;
1490 struct bonding *bond;
1491 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1493 int ret = RX_HANDLER_ANOTHER;
1495 skb = skb_share_check(skb, GFP_ATOMIC);
1497 return RX_HANDLER_CONSUMED;
1501 slave = bond_slave_get_rcu(skb->dev);
1504 recv_probe = READ_ONCE(bond->recv_probe);
1506 ret = recv_probe(skb, bond, slave);
1507 if (ret == RX_HANDLER_CONSUMED) {
1514 * For packets determined by bond_should_deliver_exact_match() call to
1515 * be suppressed we want to make an exception for link-local packets.
1516 * This is necessary for e.g. LLDP daemons to be able to monitor
1517 * inactive slave links without being forced to bind to them
1520 * At the same time, packets that are passed to the bonding master
1521 * (including link-local ones) can have their originating interface
1522 * determined via PACKET_ORIGDEV socket option.
1524 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1525 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1526 return RX_HANDLER_PASS;
1527 return RX_HANDLER_EXACT;
1530 skb->dev = bond->dev;
1532 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1533 netif_is_bridge_port(bond->dev) &&
1534 skb->pkt_type == PACKET_HOST) {
1536 if (unlikely(skb_cow_head(skb,
1537 skb->data - skb_mac_header(skb)))) {
1539 return RX_HANDLER_CONSUMED;
1541 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1542 bond->dev->addr_len);
1548 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1550 switch (BOND_MODE(bond)) {
1551 case BOND_MODE_ROUNDROBIN:
1552 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1553 case BOND_MODE_ACTIVEBACKUP:
1554 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1555 case BOND_MODE_BROADCAST:
1556 return NETDEV_LAG_TX_TYPE_BROADCAST;
1558 case BOND_MODE_8023AD:
1559 return NETDEV_LAG_TX_TYPE_HASH;
1561 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1565 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1566 enum netdev_lag_tx_type type)
1568 if (type != NETDEV_LAG_TX_TYPE_HASH)
1569 return NETDEV_LAG_HASH_NONE;
1571 switch (bond->params.xmit_policy) {
1572 case BOND_XMIT_POLICY_LAYER2:
1573 return NETDEV_LAG_HASH_L2;
1574 case BOND_XMIT_POLICY_LAYER34:
1575 return NETDEV_LAG_HASH_L34;
1576 case BOND_XMIT_POLICY_LAYER23:
1577 return NETDEV_LAG_HASH_L23;
1578 case BOND_XMIT_POLICY_ENCAP23:
1579 return NETDEV_LAG_HASH_E23;
1580 case BOND_XMIT_POLICY_ENCAP34:
1581 return NETDEV_LAG_HASH_E34;
1582 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1583 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1585 return NETDEV_LAG_HASH_UNKNOWN;
1589 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1590 struct netlink_ext_ack *extack)
1592 struct netdev_lag_upper_info lag_upper_info;
1593 enum netdev_lag_tx_type type;
1595 type = bond_lag_tx_type(bond);
1596 lag_upper_info.tx_type = type;
1597 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1599 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1600 &lag_upper_info, extack);
1603 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1605 netdev_upper_dev_unlink(slave->dev, bond->dev);
1606 slave->dev->flags &= ~IFF_SLAVE;
1609 static void slave_kobj_release(struct kobject *kobj)
1611 struct slave *slave = to_slave(kobj);
1612 struct bonding *bond = bond_get_bond_by_slave(slave);
1614 cancel_delayed_work_sync(&slave->notify_work);
1615 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1616 kfree(SLAVE_AD_INFO(slave));
1621 static struct kobj_type slave_ktype = {
1622 .release = slave_kobj_release,
1624 .sysfs_ops = &slave_sysfs_ops,
1628 static int bond_kobj_init(struct slave *slave)
1632 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1633 &(slave->dev->dev.kobj), "bonding_slave");
1635 kobject_put(&slave->kobj);
1640 static struct slave *bond_alloc_slave(struct bonding *bond,
1641 struct net_device *slave_dev)
1643 struct slave *slave = NULL;
1645 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1650 slave->dev = slave_dev;
1651 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1653 if (bond_kobj_init(slave))
1656 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1657 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1659 if (!SLAVE_AD_INFO(slave)) {
1660 kobject_put(&slave->kobj);
1668 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1670 info->bond_mode = BOND_MODE(bond);
1671 info->miimon = bond->params.miimon;
1672 info->num_slaves = bond->slave_cnt;
1675 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1677 strcpy(info->slave_name, slave->dev->name);
1678 info->link = slave->link;
1679 info->state = bond_slave_state(slave);
1680 info->link_failure_count = slave->link_failure_count;
1683 static void bond_netdev_notify_work(struct work_struct *_work)
1685 struct slave *slave = container_of(_work, struct slave,
1688 if (rtnl_trylock()) {
1689 struct netdev_bonding_info binfo;
1691 bond_fill_ifslave(slave, &binfo.slave);
1692 bond_fill_ifbond(slave->bond, &binfo.master);
1693 netdev_bonding_info_change(slave->dev, &binfo);
1696 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1700 void bond_queue_slave_event(struct slave *slave)
1702 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1705 void bond_lower_state_changed(struct slave *slave)
1707 struct netdev_lag_lower_state_info info;
1709 info.link_up = slave->link == BOND_LINK_UP ||
1710 slave->link == BOND_LINK_FAIL;
1711 info.tx_enabled = bond_is_active_slave(slave);
1712 netdev_lower_state_changed(slave->dev, &info);
1715 /* enslave device <slave> to bond device <master> */
1716 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1717 struct netlink_ext_ack *extack)
1719 struct bonding *bond = netdev_priv(bond_dev);
1720 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1721 struct slave *new_slave = NULL, *prev_slave;
1722 struct sockaddr_storage ss;
1726 if (slave_dev->flags & IFF_MASTER &&
1727 !netif_is_bond_master(slave_dev)) {
1728 NL_SET_ERR_MSG(extack, "Device with IFF_MASTER cannot be enslaved");
1729 netdev_err(bond_dev,
1730 "Error: Device with IFF_MASTER cannot be enslaved\n");
1734 if (!bond->params.use_carrier &&
1735 slave_dev->ethtool_ops->get_link == NULL &&
1736 slave_ops->ndo_do_ioctl == NULL) {
1737 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1740 /* already in-use? */
1741 if (netdev_is_rx_handler_busy(slave_dev)) {
1742 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1743 slave_err(bond_dev, slave_dev,
1744 "Error: Device is in use and cannot be enslaved\n");
1748 if (bond_dev == slave_dev) {
1749 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1750 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1754 /* vlan challenged mutual exclusion */
1755 /* no need to lock since we're protected by rtnl_lock */
1756 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1757 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1758 if (vlan_uses_dev(bond_dev)) {
1759 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1760 slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1763 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1766 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1769 if (slave_dev->features & NETIF_F_HW_ESP)
1770 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1772 /* Old ifenslave binaries are no longer supported. These can
1773 * be identified with moderate accuracy by the state of the slave:
1774 * the current ifenslave will set the interface down prior to
1775 * enslaving it; the old ifenslave will not.
1777 if (slave_dev->flags & IFF_UP) {
1778 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1779 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1783 /* set bonding device ether type by slave - bonding netdevices are
1784 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1785 * there is a need to override some of the type dependent attribs/funcs.
1787 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1788 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1790 if (!bond_has_slaves(bond)) {
1791 if (bond_dev->type != slave_dev->type) {
1792 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1793 bond_dev->type, slave_dev->type);
1795 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1797 res = notifier_to_errno(res);
1799 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1803 /* Flush unicast and multicast addresses */
1804 dev_uc_flush(bond_dev);
1805 dev_mc_flush(bond_dev);
1807 if (slave_dev->type != ARPHRD_ETHER)
1808 bond_setup_by_slave(bond_dev, slave_dev);
1810 ether_setup(bond_dev);
1811 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1814 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1817 } else if (bond_dev->type != slave_dev->type) {
1818 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1819 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1820 slave_dev->type, bond_dev->type);
1824 if (slave_dev->type == ARPHRD_INFINIBAND &&
1825 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1826 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1827 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1830 goto err_undo_flags;
1833 if (!slave_ops->ndo_set_mac_address ||
1834 slave_dev->type == ARPHRD_INFINIBAND) {
1835 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1836 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1837 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1838 if (!bond_has_slaves(bond)) {
1839 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1840 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1842 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1843 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");
1845 goto err_undo_flags;
1850 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1852 /* If this is the first slave, then we need to set the master's hardware
1853 * address to be the same as the slave's.
1855 if (!bond_has_slaves(bond) &&
1856 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1857 res = bond_set_dev_addr(bond->dev, slave_dev);
1859 goto err_undo_flags;
1862 new_slave = bond_alloc_slave(bond, slave_dev);
1865 goto err_undo_flags;
1868 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1869 * is set via sysfs or module option if desired.
1871 new_slave->queue_id = 0;
1873 /* Save slave's original mtu and then set it to match the bond */
1874 new_slave->original_mtu = slave_dev->mtu;
1875 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1877 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1881 /* Save slave's original ("permanent") mac address for modes
1882 * that need it, and for restoring it upon release, and then
1883 * set it to the master's address
1885 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1886 slave_dev->addr_len);
1888 if (!bond->params.fail_over_mac ||
1889 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1890 /* Set slave to master's mac address. The application already
1891 * set the master's mac address to that of the first slave
1893 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1894 ss.ss_family = slave_dev->type;
1895 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1898 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1899 goto err_restore_mtu;
1903 /* set slave flag before open to prevent IPv6 addrconf */
1904 slave_dev->flags |= IFF_SLAVE;
1906 /* open the slave since the application closed it */
1907 res = dev_open(slave_dev, extack);
1909 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1910 goto err_restore_mac;
1913 slave_dev->priv_flags |= IFF_BONDING;
1914 /* initialize slave stats */
1915 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1917 if (bond_is_lb(bond)) {
1918 /* bond_alb_init_slave() must be called before all other stages since
1919 * it might fail and we do not want to have to undo everything
1921 res = bond_alb_init_slave(bond, new_slave);
1926 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1928 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1932 prev_slave = bond_last_slave(bond);
1934 new_slave->delay = 0;
1935 new_slave->link_failure_count = 0;
1937 if (bond_update_speed_duplex(new_slave) &&
1938 bond_needs_speed_duplex(bond))
1939 new_slave->link = BOND_LINK_DOWN;
1941 new_slave->last_rx = jiffies -
1942 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1943 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1944 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1946 if (bond->params.miimon && !bond->params.use_carrier) {
1947 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1949 if ((link_reporting == -1) && !bond->params.arp_interval) {
1950 /* miimon is set but a bonded network driver
1951 * does not support ETHTOOL/MII and
1952 * arp_interval is not set. Note: if
1953 * use_carrier is enabled, we will never go
1954 * here (because netif_carrier is always
1955 * supported); thus, we don't need to change
1956 * the messages for netif_carrier.
1958 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");
1959 } else if (link_reporting == -1) {
1960 /* unable get link status using mii/ethtool */
1961 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");
1965 /* check for initial state */
1966 new_slave->link = BOND_LINK_NOCHANGE;
1967 if (bond->params.miimon) {
1968 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1969 if (bond->params.updelay) {
1970 bond_set_slave_link_state(new_slave,
1972 BOND_SLAVE_NOTIFY_NOW);
1973 new_slave->delay = bond->params.updelay;
1975 bond_set_slave_link_state(new_slave,
1977 BOND_SLAVE_NOTIFY_NOW);
1980 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1981 BOND_SLAVE_NOTIFY_NOW);
1983 } else if (bond->params.arp_interval) {
1984 bond_set_slave_link_state(new_slave,
1985 (netif_carrier_ok(slave_dev) ?
1986 BOND_LINK_UP : BOND_LINK_DOWN),
1987 BOND_SLAVE_NOTIFY_NOW);
1989 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1990 BOND_SLAVE_NOTIFY_NOW);
1993 if (new_slave->link != BOND_LINK_DOWN)
1994 new_slave->last_link_up = jiffies;
1995 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1996 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1997 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1999 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2000 /* if there is a primary slave, remember it */
2001 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2002 rcu_assign_pointer(bond->primary_slave, new_slave);
2003 bond->force_primary = true;
2007 switch (BOND_MODE(bond)) {
2008 case BOND_MODE_ACTIVEBACKUP:
2009 bond_set_slave_inactive_flags(new_slave,
2010 BOND_SLAVE_NOTIFY_NOW);
2012 case BOND_MODE_8023AD:
2013 /* in 802.3ad mode, the internal mechanism
2014 * will activate the slaves in the selected
2017 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2018 /* if this is the first slave */
2020 SLAVE_AD_INFO(new_slave)->id = 1;
2021 /* Initialize AD with the number of times that the AD timer is called in 1 second
2022 * can be called only after the mac address of the bond is set
2024 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2026 SLAVE_AD_INFO(new_slave)->id =
2027 SLAVE_AD_INFO(prev_slave)->id + 1;
2030 bond_3ad_bind_slave(new_slave);
2034 bond_set_active_slave(new_slave);
2035 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2038 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2040 /* always active in trunk mode */
2041 bond_set_active_slave(new_slave);
2043 /* In trunking mode there is little meaning to curr_active_slave
2044 * anyway (it holds no special properties of the bond device),
2045 * so we can change it without calling change_active_interface()
2047 if (!rcu_access_pointer(bond->curr_active_slave) &&
2048 new_slave->link == BOND_LINK_UP)
2049 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2052 } /* switch(bond_mode) */
2054 #ifdef CONFIG_NET_POLL_CONTROLLER
2055 if (bond->dev->npinfo) {
2056 if (slave_enable_netpoll(new_slave)) {
2057 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2064 if (!(bond_dev->features & NETIF_F_LRO))
2065 dev_disable_lro(slave_dev);
2067 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2070 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2074 res = bond_master_upper_dev_link(bond, new_slave, extack);
2076 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2077 goto err_unregister;
2080 bond_lower_state_changed(new_slave);
2082 res = bond_sysfs_slave_add(new_slave);
2084 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2085 goto err_upper_unlink;
2088 /* If the mode uses primary, then the following is handled by
2089 * bond_change_active_slave().
2091 if (!bond_uses_primary(bond)) {
2092 /* set promiscuity level to new slave */
2093 if (bond_dev->flags & IFF_PROMISC) {
2094 res = dev_set_promiscuity(slave_dev, 1);
2099 /* set allmulti level to new slave */
2100 if (bond_dev->flags & IFF_ALLMULTI) {
2101 res = dev_set_allmulti(slave_dev, 1);
2103 if (bond_dev->flags & IFF_PROMISC)
2104 dev_set_promiscuity(slave_dev, -1);
2109 netif_addr_lock_bh(bond_dev);
2110 dev_mc_sync_multiple(slave_dev, bond_dev);
2111 dev_uc_sync_multiple(slave_dev, bond_dev);
2112 netif_addr_unlock_bh(bond_dev);
2114 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2115 /* add lacpdu mc addr to mc list */
2116 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2118 dev_mc_add(slave_dev, lacpdu_multicast);
2123 bond_compute_features(bond);
2124 bond_set_carrier(bond);
2126 if (bond_uses_primary(bond)) {
2128 bond_select_active_slave(bond);
2129 unblock_netpoll_tx();
2132 if (bond_mode_can_use_xmit_hash(bond))
2133 bond_update_slave_arr(bond, NULL);
2136 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2137 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2138 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2140 /* enslave is successful */
2141 bond_queue_slave_event(new_slave);
2144 /* Undo stages on error */
2146 bond_sysfs_slave_del(new_slave);
2149 bond_upper_dev_unlink(bond, new_slave);
2152 netdev_rx_handler_unregister(slave_dev);
2155 vlan_vids_del_by_dev(slave_dev, bond_dev);
2156 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2157 RCU_INIT_POINTER(bond->primary_slave, NULL);
2158 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2160 bond_change_active_slave(bond, NULL);
2161 bond_select_active_slave(bond);
2162 unblock_netpoll_tx();
2164 /* either primary_slave or curr_active_slave might've changed */
2166 slave_disable_netpoll(new_slave);
2169 if (!netif_is_bond_master(slave_dev))
2170 slave_dev->priv_flags &= ~IFF_BONDING;
2171 dev_close(slave_dev);
2174 slave_dev->flags &= ~IFF_SLAVE;
2175 if (!bond->params.fail_over_mac ||
2176 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2177 /* XXX TODO - fom follow mode needs to change master's
2178 * MAC if this slave's MAC is in use by the bond, or at
2179 * least print a warning.
2181 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2182 new_slave->dev->addr_len);
2183 ss.ss_family = slave_dev->type;
2184 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2188 dev_set_mtu(slave_dev, new_slave->original_mtu);
2191 kobject_put(&new_slave->kobj);
2194 /* Enslave of first slave has failed and we need to fix master's mac */
2195 if (!bond_has_slaves(bond)) {
2196 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2197 slave_dev->dev_addr))
2198 eth_hw_addr_random(bond_dev);
2199 if (bond_dev->type != ARPHRD_ETHER) {
2200 dev_close(bond_dev);
2201 ether_setup(bond_dev);
2202 bond_dev->flags |= IFF_MASTER;
2203 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2210 /* Try to release the slave device <slave> from the bond device <master>
2211 * It is legal to access curr_active_slave without a lock because all the function
2212 * is RTNL-locked. If "all" is true it means that the function is being called
2213 * while destroying a bond interface and all slaves are being released.
2215 * The rules for slave state should be:
2216 * for Active/Backup:
2217 * Active stays on all backups go down
2218 * for Bonded connections:
2219 * The first up interface should be left on and all others downed.
2221 static int __bond_release_one(struct net_device *bond_dev,
2222 struct net_device *slave_dev,
2223 bool all, bool unregister)
2225 struct bonding *bond = netdev_priv(bond_dev);
2226 struct slave *slave, *oldcurrent;
2227 struct sockaddr_storage ss;
2228 int old_flags = bond_dev->flags;
2229 netdev_features_t old_features = bond_dev->features;
2231 /* slave is not a slave or master is not master of this slave */
2232 if (!(slave_dev->flags & IFF_SLAVE) ||
2233 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2234 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2240 slave = bond_get_slave_by_dev(bond, slave_dev);
2242 /* not a slave of this bond */
2243 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2244 unblock_netpoll_tx();
2248 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2250 bond_sysfs_slave_del(slave);
2252 /* recompute stats just before removing the slave */
2253 bond_get_stats(bond->dev, &bond->bond_stats);
2255 bond_upper_dev_unlink(bond, slave);
2256 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2257 * for this slave anymore.
2259 netdev_rx_handler_unregister(slave_dev);
2261 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2262 bond_3ad_unbind_slave(slave);
2264 if (bond_mode_can_use_xmit_hash(bond))
2265 bond_update_slave_arr(bond, slave);
2267 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2268 bond_is_active_slave(slave) ? "active" : "backup");
2270 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2272 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2274 if (!all && (!bond->params.fail_over_mac ||
2275 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2276 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2277 bond_has_slaves(bond))
2278 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",
2279 slave->perm_hwaddr);
2282 if (rtnl_dereference(bond->primary_slave) == slave)
2283 RCU_INIT_POINTER(bond->primary_slave, NULL);
2285 if (oldcurrent == slave)
2286 bond_change_active_slave(bond, NULL);
2288 if (bond_is_lb(bond)) {
2289 /* Must be called only after the slave has been
2290 * detached from the list and the curr_active_slave
2291 * has been cleared (if our_slave == old_current),
2292 * but before a new active slave is selected.
2294 bond_alb_deinit_slave(bond, slave);
2298 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2299 } else if (oldcurrent == slave) {
2300 /* Note that we hold RTNL over this sequence, so there
2301 * is no concern that another slave add/remove event
2304 bond_select_active_slave(bond);
2307 if (!bond_has_slaves(bond)) {
2308 bond_set_carrier(bond);
2309 eth_hw_addr_random(bond_dev);
2312 unblock_netpoll_tx();
2316 if (!bond_has_slaves(bond)) {
2317 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2318 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2321 bond_compute_features(bond);
2322 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2323 (old_features & NETIF_F_VLAN_CHALLENGED))
2324 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2326 vlan_vids_del_by_dev(slave_dev, bond_dev);
2328 /* If the mode uses primary, then this case was handled above by
2329 * bond_change_active_slave(..., NULL)
2331 if (!bond_uses_primary(bond)) {
2332 /* unset promiscuity level from slave
2333 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2334 * of the IFF_PROMISC flag in the bond_dev, but we need the
2335 * value of that flag before that change, as that was the value
2336 * when this slave was attached, so we cache at the start of the
2337 * function and use it here. Same goes for ALLMULTI below
2339 if (old_flags & IFF_PROMISC)
2340 dev_set_promiscuity(slave_dev, -1);
2342 /* unset allmulti level from slave */
2343 if (old_flags & IFF_ALLMULTI)
2344 dev_set_allmulti(slave_dev, -1);
2346 bond_hw_addr_flush(bond_dev, slave_dev);
2349 slave_disable_netpoll(slave);
2351 /* close slave before restoring its mac address */
2352 dev_close(slave_dev);
2354 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2355 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2356 /* restore original ("permanent") mac address */
2357 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2358 slave->dev->addr_len);
2359 ss.ss_family = slave_dev->type;
2360 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2364 __dev_set_mtu(slave_dev, slave->original_mtu);
2366 dev_set_mtu(slave_dev, slave->original_mtu);
2368 if (!netif_is_bond_master(slave_dev))
2369 slave_dev->priv_flags &= ~IFF_BONDING;
2371 kobject_put(&slave->kobj);
2376 /* A wrapper used because of ndo_del_link */
2377 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2379 return __bond_release_one(bond_dev, slave_dev, false, false);
2382 /* First release a slave and then destroy the bond if no more slaves are left.
2383 * Must be under rtnl_lock when this function is called.
2385 static int bond_release_and_destroy(struct net_device *bond_dev,
2386 struct net_device *slave_dev)
2388 struct bonding *bond = netdev_priv(bond_dev);
2391 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2392 if (ret == 0 && !bond_has_slaves(bond) &&
2393 bond_dev->reg_state != NETREG_UNREGISTERING) {
2394 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2395 netdev_info(bond_dev, "Destroying bond\n");
2396 bond_remove_proc_entry(bond);
2397 unregister_netdevice(bond_dev);
2402 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2404 struct bonding *bond = netdev_priv(bond_dev);
2406 bond_fill_ifbond(bond, info);
2409 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2411 struct bonding *bond = netdev_priv(bond_dev);
2412 struct list_head *iter;
2413 int i = 0, res = -ENODEV;
2414 struct slave *slave;
2416 bond_for_each_slave(bond, slave, iter) {
2417 if (i++ == (int)info->slave_id) {
2419 bond_fill_ifslave(slave, info);
2427 /*-------------------------------- Monitoring -------------------------------*/
2429 /* called with rcu_read_lock() */
2430 static int bond_miimon_inspect(struct bonding *bond)
2432 int link_state, commit = 0;
2433 struct list_head *iter;
2434 struct slave *slave;
2435 bool ignore_updelay;
2437 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2439 bond_for_each_slave_rcu(bond, slave, iter) {
2440 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2442 link_state = bond_check_dev_link(bond, slave->dev, 0);
2444 switch (slave->link) {
2449 bond_propose_link_state(slave, BOND_LINK_FAIL);
2451 slave->delay = bond->params.downdelay;
2453 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2455 BOND_MODE_ACTIVEBACKUP) ?
2456 (bond_is_active_slave(slave) ?
2457 "active " : "backup ") : "",
2458 bond->params.downdelay * bond->params.miimon);
2461 case BOND_LINK_FAIL:
2463 /* recovered before downdelay expired */
2464 bond_propose_link_state(slave, BOND_LINK_UP);
2465 slave->last_link_up = jiffies;
2466 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2467 (bond->params.downdelay - slave->delay) *
2468 bond->params.miimon);
2473 if (slave->delay <= 0) {
2474 bond_propose_link_state(slave, BOND_LINK_DOWN);
2482 case BOND_LINK_DOWN:
2486 bond_propose_link_state(slave, BOND_LINK_BACK);
2488 slave->delay = bond->params.updelay;
2491 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2492 ignore_updelay ? 0 :
2493 bond->params.updelay *
2494 bond->params.miimon);
2497 case BOND_LINK_BACK:
2499 bond_propose_link_state(slave, BOND_LINK_DOWN);
2500 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2501 (bond->params.updelay - slave->delay) *
2502 bond->params.miimon);
2510 if (slave->delay <= 0) {
2511 bond_propose_link_state(slave, BOND_LINK_UP);
2513 ignore_updelay = false;
2525 static void bond_miimon_link_change(struct bonding *bond,
2526 struct slave *slave,
2529 switch (BOND_MODE(bond)) {
2530 case BOND_MODE_8023AD:
2531 bond_3ad_handle_link_change(slave, link);
2535 bond_alb_handle_link_change(bond, slave, link);
2538 bond_update_slave_arr(bond, NULL);
2543 static void bond_miimon_commit(struct bonding *bond)
2545 struct list_head *iter;
2546 struct slave *slave, *primary;
2548 bond_for_each_slave(bond, slave, iter) {
2549 switch (slave->link_new_state) {
2550 case BOND_LINK_NOCHANGE:
2551 /* For 802.3ad mode, check current slave speed and
2552 * duplex again in case its port was disabled after
2553 * invalid speed/duplex reporting but recovered before
2554 * link monitoring could make a decision on the actual
2557 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2558 slave->link == BOND_LINK_UP)
2559 bond_3ad_adapter_speed_duplex_changed(slave);
2563 if (bond_update_speed_duplex(slave) &&
2564 bond_needs_speed_duplex(bond)) {
2565 slave->link = BOND_LINK_DOWN;
2566 if (net_ratelimit())
2567 slave_warn(bond->dev, slave->dev,
2568 "failed to get link speed/duplex\n");
2571 bond_set_slave_link_state(slave, BOND_LINK_UP,
2572 BOND_SLAVE_NOTIFY_NOW);
2573 slave->last_link_up = jiffies;
2575 primary = rtnl_dereference(bond->primary_slave);
2576 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2577 /* prevent it from being the active one */
2578 bond_set_backup_slave(slave);
2579 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2580 /* make it immediately active */
2581 bond_set_active_slave(slave);
2584 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2585 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2586 slave->duplex ? "full" : "half");
2588 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2590 if (!bond->curr_active_slave || slave == primary)
2595 case BOND_LINK_DOWN:
2596 if (slave->link_failure_count < UINT_MAX)
2597 slave->link_failure_count++;
2599 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2600 BOND_SLAVE_NOTIFY_NOW);
2602 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2603 BOND_MODE(bond) == BOND_MODE_8023AD)
2604 bond_set_slave_inactive_flags(slave,
2605 BOND_SLAVE_NOTIFY_NOW);
2607 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2609 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2611 if (slave == rcu_access_pointer(bond->curr_active_slave))
2617 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2618 slave->link_new_state);
2619 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2626 bond_select_active_slave(bond);
2627 unblock_netpoll_tx();
2630 bond_set_carrier(bond);
2635 * Really a wrapper that splits the mii monitor into two phases: an
2636 * inspection, then (if inspection indicates something needs to be done)
2637 * an acquisition of appropriate locks followed by a commit phase to
2638 * implement whatever link state changes are indicated.
2640 static void bond_mii_monitor(struct work_struct *work)
2642 struct bonding *bond = container_of(work, struct bonding,
2644 bool should_notify_peers = false;
2646 unsigned long delay;
2647 struct slave *slave;
2648 struct list_head *iter;
2650 delay = msecs_to_jiffies(bond->params.miimon);
2652 if (!bond_has_slaves(bond))
2656 should_notify_peers = bond_should_notify_peers(bond);
2657 commit = !!bond_miimon_inspect(bond);
2658 if (bond->send_peer_notif) {
2660 if (rtnl_trylock()) {
2661 bond->send_peer_notif--;
2669 /* Race avoidance with bond_close cancel of workqueue */
2670 if (!rtnl_trylock()) {
2672 should_notify_peers = false;
2676 bond_for_each_slave(bond, slave, iter) {
2677 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2679 bond_miimon_commit(bond);
2681 rtnl_unlock(); /* might sleep, hold no other locks */
2685 if (bond->params.miimon)
2686 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2688 if (should_notify_peers) {
2689 if (!rtnl_trylock())
2691 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2696 static int bond_upper_dev_walk(struct net_device *upper,
2697 struct netdev_nested_priv *priv)
2699 __be32 ip = *(__be32 *)priv->data;
2701 return ip == bond_confirm_addr(upper, 0, ip);
2704 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2706 struct netdev_nested_priv priv = {
2707 .data = (void *)&ip,
2711 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2715 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2722 /* We go to the (large) trouble of VLAN tagging ARP frames because
2723 * switches in VLAN mode (especially if ports are configured as
2724 * "native" to a VLAN) might not pass non-tagged frames.
2726 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2727 __be32 src_ip, struct bond_vlan_tag *tags)
2729 struct sk_buff *skb;
2730 struct bond_vlan_tag *outer_tag = tags;
2731 struct net_device *slave_dev = slave->dev;
2732 struct net_device *bond_dev = slave->bond->dev;
2734 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2735 arp_op, &dest_ip, &src_ip);
2737 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2738 NULL, slave_dev->dev_addr, NULL);
2741 net_err_ratelimited("ARP packet allocation failed\n");
2745 if (!tags || tags->vlan_proto == VLAN_N_VID)
2750 /* Go through all the tags backwards and add them to the packet */
2751 while (tags->vlan_proto != VLAN_N_VID) {
2752 if (!tags->vlan_id) {
2757 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2758 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2759 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2762 net_err_ratelimited("failed to insert inner VLAN tag\n");
2768 /* Set the outer tag */
2769 if (outer_tag->vlan_id) {
2770 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2771 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2772 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2773 outer_tag->vlan_id);
2780 /* Validate the device path between the @start_dev and the @end_dev.
2781 * The path is valid if the @end_dev is reachable through device
2783 * When the path is validated, collect any vlan information in the
2786 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2787 struct net_device *end_dev,
2790 struct bond_vlan_tag *tags;
2791 struct net_device *upper;
2792 struct list_head *iter;
2794 if (start_dev == end_dev) {
2795 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2797 return ERR_PTR(-ENOMEM);
2798 tags[level].vlan_proto = VLAN_N_VID;
2802 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2803 tags = bond_verify_device_path(upper, end_dev, level + 1);
2804 if (IS_ERR_OR_NULL(tags)) {
2809 if (is_vlan_dev(upper)) {
2810 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2811 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2820 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2823 struct bond_vlan_tag *tags;
2824 __be32 *targets = bond->params.arp_targets, addr;
2827 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2828 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2829 __func__, &targets[i]);
2832 /* Find out through which dev should the packet go */
2833 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2836 /* there's no route to target - try to send arp
2837 * probe to generate any traffic (arp_validate=0)
2839 if (bond->params.arp_validate)
2840 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2843 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2848 /* bond device itself */
2849 if (rt->dst.dev == bond->dev)
2853 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2856 if (!IS_ERR_OR_NULL(tags))
2859 /* Not our device - skip */
2860 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2861 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2867 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2869 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2874 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2878 if (!sip || !bond_has_this_ip(bond, tip)) {
2879 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2880 __func__, &sip, &tip);
2884 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2886 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2890 slave->last_rx = jiffies;
2891 slave->target_last_arp_rx[i] = jiffies;
2894 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2895 struct slave *slave)
2897 struct arphdr *arp = (struct arphdr *)skb->data;
2898 struct slave *curr_active_slave, *curr_arp_slave;
2899 unsigned char *arp_ptr;
2901 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2904 if (!slave_do_arp_validate(bond, slave)) {
2905 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2906 !slave_do_arp_validate_only(bond))
2907 slave->last_rx = jiffies;
2908 return RX_HANDLER_ANOTHER;
2909 } else if (!is_arp) {
2910 return RX_HANDLER_ANOTHER;
2913 alen = arp_hdr_len(bond->dev);
2915 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2916 __func__, skb->dev->name);
2918 if (alen > skb_headlen(skb)) {
2919 arp = kmalloc(alen, GFP_ATOMIC);
2922 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2926 if (arp->ar_hln != bond->dev->addr_len ||
2927 skb->pkt_type == PACKET_OTHERHOST ||
2928 skb->pkt_type == PACKET_LOOPBACK ||
2929 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2930 arp->ar_pro != htons(ETH_P_IP) ||
2934 arp_ptr = (unsigned char *)(arp + 1);
2935 arp_ptr += bond->dev->addr_len;
2936 memcpy(&sip, arp_ptr, 4);
2937 arp_ptr += 4 + bond->dev->addr_len;
2938 memcpy(&tip, arp_ptr, 4);
2940 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2941 __func__, slave->dev->name, bond_slave_state(slave),
2942 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2945 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2946 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2948 /* We 'trust' the received ARP enough to validate it if:
2950 * (a) the slave receiving the ARP is active (which includes the
2951 * current ARP slave, if any), or
2953 * (b) the receiving slave isn't active, but there is a currently
2954 * active slave and it received valid arp reply(s) after it became
2955 * the currently active slave, or
2957 * (c) there is an ARP slave that sent an ARP during the prior ARP
2958 * interval, and we receive an ARP reply on any slave. We accept
2959 * these because switch FDB update delays may deliver the ARP
2960 * reply to a slave other than the sender of the ARP request.
2962 * Note: for (b), backup slaves are receiving the broadcast ARP
2963 * request, not a reply. This request passes from the sending
2964 * slave through the L2 switch(es) to the receiving slave. Since
2965 * this is checking the request, sip/tip are swapped for
2968 * This is done to avoid endless looping when we can't reach the
2969 * arp_ip_target and fool ourselves with our own arp requests.
2971 if (bond_is_active_slave(slave))
2972 bond_validate_arp(bond, slave, sip, tip);
2973 else if (curr_active_slave &&
2974 time_after(slave_last_rx(bond, curr_active_slave),
2975 curr_active_slave->last_link_up))
2976 bond_validate_arp(bond, slave, tip, sip);
2977 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2978 bond_time_in_interval(bond,
2979 dev_trans_start(curr_arp_slave->dev), 1))
2980 bond_validate_arp(bond, slave, sip, tip);
2983 if (arp != (struct arphdr *)skb->data)
2985 return RX_HANDLER_ANOTHER;
2988 /* function to verify if we're in the arp_interval timeslice, returns true if
2989 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2990 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2992 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2995 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2997 return time_in_range(jiffies,
2998 last_act - delta_in_ticks,
2999 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3002 /* This function is called regularly to monitor each slave's link
3003 * ensuring that traffic is being sent and received when arp monitoring
3004 * is used in load-balancing mode. if the adapter has been dormant, then an
3005 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3006 * arp monitoring in active backup mode.
3008 static void bond_loadbalance_arp_mon(struct bonding *bond)
3010 struct slave *slave, *oldcurrent;
3011 struct list_head *iter;
3012 int do_failover = 0, slave_state_changed = 0;
3014 if (!bond_has_slaves(bond))
3019 oldcurrent = rcu_dereference(bond->curr_active_slave);
3020 /* see if any of the previous devices are up now (i.e. they have
3021 * xmt and rcv traffic). the curr_active_slave does not come into
3022 * the picture unless it is null. also, slave->last_link_up is not
3023 * needed here because we send an arp on each slave and give a slave
3024 * as long as it needs to get the tx/rx within the delta.
3025 * TODO: what about up/down delay in arp mode? it wasn't here before
3028 bond_for_each_slave_rcu(bond, slave, iter) {
3029 unsigned long trans_start = dev_trans_start(slave->dev);
3031 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3033 if (slave->link != BOND_LINK_UP) {
3034 if (bond_time_in_interval(bond, trans_start, 1) &&
3035 bond_time_in_interval(bond, slave->last_rx, 1)) {
3037 bond_propose_link_state(slave, BOND_LINK_UP);
3038 slave_state_changed = 1;
3040 /* primary_slave has no meaning in round-robin
3041 * mode. the window of a slave being up and
3042 * curr_active_slave being null after enslaving
3046 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3049 slave_info(bond->dev, slave->dev, "interface is now up\n");
3053 /* slave->link == BOND_LINK_UP */
3055 /* not all switches will respond to an arp request
3056 * when the source ip is 0, so don't take the link down
3057 * if we don't know our ip yet
3059 if (!bond_time_in_interval(bond, trans_start, 2) ||
3060 !bond_time_in_interval(bond, slave->last_rx, 2)) {
3062 bond_propose_link_state(slave, BOND_LINK_DOWN);
3063 slave_state_changed = 1;
3065 if (slave->link_failure_count < UINT_MAX)
3066 slave->link_failure_count++;
3068 slave_info(bond->dev, slave->dev, "interface is now down\n");
3070 if (slave == oldcurrent)
3075 /* note: if switch is in round-robin mode, all links
3076 * must tx arp to ensure all links rx an arp - otherwise
3077 * links may oscillate or not come up at all; if switch is
3078 * in something like xor mode, there is nothing we can
3079 * do - all replies will be rx'ed on same link causing slaves
3080 * to be unstable during low/no traffic periods
3082 if (bond_slave_is_up(slave))
3083 bond_arp_send_all(bond, slave);
3088 if (do_failover || slave_state_changed) {
3089 if (!rtnl_trylock())
3092 bond_for_each_slave(bond, slave, iter) {
3093 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3094 slave->link = slave->link_new_state;
3097 if (slave_state_changed) {
3098 bond_slave_state_change(bond);
3099 if (BOND_MODE(bond) == BOND_MODE_XOR)
3100 bond_update_slave_arr(bond, NULL);
3104 bond_select_active_slave(bond);
3105 unblock_netpoll_tx();
3111 if (bond->params.arp_interval)
3112 queue_delayed_work(bond->wq, &bond->arp_work,
3113 msecs_to_jiffies(bond->params.arp_interval));
3116 /* Called to inspect slaves for active-backup mode ARP monitor link state
3117 * changes. Sets proposed link state in slaves to specify what action
3118 * should take place for the slave. Returns 0 if no changes are found, >0
3119 * if changes to link states must be committed.
3121 * Called with rcu_read_lock held.
3123 static int bond_ab_arp_inspect(struct bonding *bond)
3125 unsigned long trans_start, last_rx;
3126 struct list_head *iter;
3127 struct slave *slave;
3130 bond_for_each_slave_rcu(bond, slave, iter) {
3131 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3132 last_rx = slave_last_rx(bond, slave);
3134 if (slave->link != BOND_LINK_UP) {
3135 if (bond_time_in_interval(bond, last_rx, 1)) {
3136 bond_propose_link_state(slave, BOND_LINK_UP);
3138 } else if (slave->link == BOND_LINK_BACK) {
3139 bond_propose_link_state(slave, BOND_LINK_FAIL);
3145 /* Give slaves 2*delta after being enslaved or made
3146 * active. This avoids bouncing, as the last receive
3147 * times need a full ARP monitor cycle to be updated.
3149 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3152 /* Backup slave is down if:
3153 * - No current_arp_slave AND
3154 * - more than 3*delta since last receive AND
3155 * - the bond has an IP address
3157 * Note: a non-null current_arp_slave indicates
3158 * the curr_active_slave went down and we are
3159 * searching for a new one; under this condition
3160 * we only take the curr_active_slave down - this
3161 * gives each slave a chance to tx/rx traffic
3162 * before being taken out
3164 if (!bond_is_active_slave(slave) &&
3165 !rcu_access_pointer(bond->current_arp_slave) &&
3166 !bond_time_in_interval(bond, last_rx, 3)) {
3167 bond_propose_link_state(slave, BOND_LINK_DOWN);
3171 /* Active slave is down if:
3172 * - more than 2*delta since transmitting OR
3173 * - (more than 2*delta since receive AND
3174 * the bond has an IP address)
3176 trans_start = dev_trans_start(slave->dev);
3177 if (bond_is_active_slave(slave) &&
3178 (!bond_time_in_interval(bond, trans_start, 2) ||
3179 !bond_time_in_interval(bond, last_rx, 2))) {
3180 bond_propose_link_state(slave, BOND_LINK_DOWN);
3188 /* Called to commit link state changes noted by inspection step of
3189 * active-backup mode ARP monitor.
3191 * Called with RTNL hold.
3193 static void bond_ab_arp_commit(struct bonding *bond)
3195 unsigned long trans_start;
3196 struct list_head *iter;
3197 struct slave *slave;
3199 bond_for_each_slave(bond, slave, iter) {
3200 switch (slave->link_new_state) {
3201 case BOND_LINK_NOCHANGE:
3205 trans_start = dev_trans_start(slave->dev);
3206 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3207 (!rtnl_dereference(bond->curr_active_slave) &&
3208 bond_time_in_interval(bond, trans_start, 1))) {
3209 struct slave *current_arp_slave;
3211 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3212 bond_set_slave_link_state(slave, BOND_LINK_UP,
3213 BOND_SLAVE_NOTIFY_NOW);
3214 if (current_arp_slave) {
3215 bond_set_slave_inactive_flags(
3217 BOND_SLAVE_NOTIFY_NOW);
3218 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3221 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3223 if (!rtnl_dereference(bond->curr_active_slave) ||
3224 slave == rtnl_dereference(bond->primary_slave))
3231 case BOND_LINK_DOWN:
3232 if (slave->link_failure_count < UINT_MAX)
3233 slave->link_failure_count++;
3235 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3236 BOND_SLAVE_NOTIFY_NOW);
3237 bond_set_slave_inactive_flags(slave,
3238 BOND_SLAVE_NOTIFY_NOW);
3240 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3242 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3243 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3249 case BOND_LINK_FAIL:
3250 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3251 BOND_SLAVE_NOTIFY_NOW);
3252 bond_set_slave_inactive_flags(slave,
3253 BOND_SLAVE_NOTIFY_NOW);
3255 /* A slave has just been enslaved and has become
3256 * the current active slave.
3258 if (rtnl_dereference(bond->curr_active_slave))
3259 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3263 slave_err(bond->dev, slave->dev,
3264 "impossible: link_new_state %d on slave\n",
3265 slave->link_new_state);
3271 bond_select_active_slave(bond);
3272 unblock_netpoll_tx();
3275 bond_set_carrier(bond);
3278 /* Send ARP probes for active-backup mode ARP monitor.
3280 * Called with rcu_read_lock held.
3282 static bool bond_ab_arp_probe(struct bonding *bond)
3284 struct slave *slave, *before = NULL, *new_slave = NULL,
3285 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3286 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3287 struct list_head *iter;
3289 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3291 if (curr_arp_slave && curr_active_slave)
3292 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3293 curr_arp_slave->dev->name,
3294 curr_active_slave->dev->name);
3296 if (curr_active_slave) {
3297 bond_arp_send_all(bond, curr_active_slave);
3298 return should_notify_rtnl;
3301 /* if we don't have a curr_active_slave, search for the next available
3302 * backup slave from the current_arp_slave and make it the candidate
3303 * for becoming the curr_active_slave
3306 if (!curr_arp_slave) {
3307 curr_arp_slave = bond_first_slave_rcu(bond);
3308 if (!curr_arp_slave)
3309 return should_notify_rtnl;
3312 bond_for_each_slave_rcu(bond, slave, iter) {
3313 if (!found && !before && bond_slave_is_up(slave))
3316 if (found && !new_slave && bond_slave_is_up(slave))
3318 /* if the link state is up at this point, we
3319 * mark it down - this can happen if we have
3320 * simultaneous link failures and
3321 * reselect_active_interface doesn't make this
3322 * one the current slave so it is still marked
3323 * up when it is actually down
3325 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3326 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3327 BOND_SLAVE_NOTIFY_LATER);
3328 if (slave->link_failure_count < UINT_MAX)
3329 slave->link_failure_count++;
3331 bond_set_slave_inactive_flags(slave,
3332 BOND_SLAVE_NOTIFY_LATER);
3334 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3336 if (slave == curr_arp_slave)
3340 if (!new_slave && before)
3346 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3347 BOND_SLAVE_NOTIFY_LATER);
3348 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3349 bond_arp_send_all(bond, new_slave);
3350 new_slave->last_link_up = jiffies;
3351 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3354 bond_for_each_slave_rcu(bond, slave, iter) {
3355 if (slave->should_notify || slave->should_notify_link) {
3356 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3360 return should_notify_rtnl;
3363 static void bond_activebackup_arp_mon(struct bonding *bond)
3365 bool should_notify_peers = false;
3366 bool should_notify_rtnl = false;
3369 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3371 if (!bond_has_slaves(bond))
3376 should_notify_peers = bond_should_notify_peers(bond);
3378 if (bond_ab_arp_inspect(bond)) {
3381 /* Race avoidance with bond_close flush of workqueue */
3382 if (!rtnl_trylock()) {
3384 should_notify_peers = false;
3388 bond_ab_arp_commit(bond);
3394 should_notify_rtnl = bond_ab_arp_probe(bond);
3398 if (bond->params.arp_interval)
3399 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3401 if (should_notify_peers || should_notify_rtnl) {
3402 if (!rtnl_trylock())
3405 if (should_notify_peers)
3406 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3408 if (should_notify_rtnl) {
3409 bond_slave_state_notify(bond);
3410 bond_slave_link_notify(bond);
3417 static void bond_arp_monitor(struct work_struct *work)
3419 struct bonding *bond = container_of(work, struct bonding,
3422 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3423 bond_activebackup_arp_mon(bond);
3425 bond_loadbalance_arp_mon(bond);
3428 /*-------------------------- netdev event handling --------------------------*/
3430 /* Change device name */
3431 static int bond_event_changename(struct bonding *bond)
3433 bond_remove_proc_entry(bond);
3434 bond_create_proc_entry(bond);
3436 bond_debug_reregister(bond);
3441 static int bond_master_netdev_event(unsigned long event,
3442 struct net_device *bond_dev)
3444 struct bonding *event_bond = netdev_priv(bond_dev);
3446 netdev_dbg(bond_dev, "%s called\n", __func__);
3449 case NETDEV_CHANGENAME:
3450 return bond_event_changename(event_bond);
3451 case NETDEV_UNREGISTER:
3452 bond_remove_proc_entry(event_bond);
3453 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3455 case NETDEV_REGISTER:
3456 bond_create_proc_entry(event_bond);
3465 static int bond_slave_netdev_event(unsigned long event,
3466 struct net_device *slave_dev)
3468 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3469 struct bonding *bond;
3470 struct net_device *bond_dev;
3472 /* A netdev event can be generated while enslaving a device
3473 * before netdev_rx_handler_register is called in which case
3474 * slave will be NULL
3477 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3481 bond_dev = slave->bond->dev;
3483 primary = rtnl_dereference(bond->primary_slave);
3485 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3488 case NETDEV_UNREGISTER:
3489 if (bond_dev->type != ARPHRD_ETHER)
3490 bond_release_and_destroy(bond_dev, slave_dev);
3492 __bond_release_one(bond_dev, slave_dev, false, true);
3496 /* For 802.3ad mode only:
3497 * Getting invalid Speed/Duplex values here will put slave
3498 * in weird state. Mark it as link-fail if the link was
3499 * previously up or link-down if it hasn't yet come up, and
3500 * let link-monitoring (miimon) set it right when correct
3501 * speeds/duplex are available.
3503 if (bond_update_speed_duplex(slave) &&
3504 BOND_MODE(bond) == BOND_MODE_8023AD) {
3505 if (slave->last_link_up)
3506 slave->link = BOND_LINK_FAIL;
3508 slave->link = BOND_LINK_DOWN;
3511 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3512 bond_3ad_adapter_speed_duplex_changed(slave);
3515 /* Refresh slave-array if applicable!
3516 * If the setup does not use miimon or arpmon (mode-specific!),
3517 * then these events will not cause the slave-array to be
3518 * refreshed. This will cause xmit to use a slave that is not
3519 * usable. Avoid such situation by refeshing the array at these
3520 * events. If these (miimon/arpmon) parameters are configured
3521 * then array gets refreshed twice and that should be fine!
3523 if (bond_mode_can_use_xmit_hash(bond))
3524 bond_update_slave_arr(bond, NULL);
3526 case NETDEV_CHANGEMTU:
3527 /* TODO: Should slaves be allowed to
3528 * independently alter their MTU? For
3529 * an active-backup bond, slaves need
3530 * not be the same type of device, so
3531 * MTUs may vary. For other modes,
3532 * slaves arguably should have the
3533 * same MTUs. To do this, we'd need to
3534 * take over the slave's change_mtu
3535 * function for the duration of their
3539 case NETDEV_CHANGENAME:
3540 /* we don't care if we don't have primary set */
3541 if (!bond_uses_primary(bond) ||
3542 !bond->params.primary[0])
3545 if (slave == primary) {
3546 /* slave's name changed - he's no longer primary */
3547 RCU_INIT_POINTER(bond->primary_slave, NULL);
3548 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3549 /* we have a new primary slave */
3550 rcu_assign_pointer(bond->primary_slave, slave);
3551 } else { /* we didn't change primary - exit */
3555 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3556 primary ? slave_dev->name : "none");
3559 bond_select_active_slave(bond);
3560 unblock_netpoll_tx();
3562 case NETDEV_FEAT_CHANGE:
3563 bond_compute_features(bond);
3565 case NETDEV_RESEND_IGMP:
3566 /* Propagate to master device */
3567 call_netdevice_notifiers(event, slave->bond->dev);
3576 /* bond_netdev_event: handle netdev notifier chain events.
3578 * This function receives events for the netdev chain. The caller (an
3579 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3580 * locks for us to safely manipulate the slave devices (RTNL lock,
3583 static int bond_netdev_event(struct notifier_block *this,
3584 unsigned long event, void *ptr)
3586 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3588 netdev_dbg(event_dev, "%s received %s\n",
3589 __func__, netdev_cmd_to_name(event));
3591 if (!(event_dev->priv_flags & IFF_BONDING))
3594 if (event_dev->flags & IFF_MASTER) {
3597 ret = bond_master_netdev_event(event, event_dev);
3598 if (ret != NOTIFY_DONE)
3602 if (event_dev->flags & IFF_SLAVE)
3603 return bond_slave_netdev_event(event, event_dev);
3608 static struct notifier_block bond_netdev_notifier = {
3609 .notifier_call = bond_netdev_event,
3612 /*---------------------------- Hashing Policies -----------------------------*/
3614 /* L2 hash helper */
3615 static inline u32 bond_eth_hash(struct sk_buff *skb)
3617 struct ethhdr *ep, hdr_tmp;
3619 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3621 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3625 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3626 int *noff, int *proto, bool l34)
3628 const struct ipv6hdr *iph6;
3629 const struct iphdr *iph;
3631 if (skb->protocol == htons(ETH_P_IP)) {
3632 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3634 iph = (const struct iphdr *)(skb->data + *noff);
3635 iph_to_flow_copy_v4addrs(fk, iph);
3636 *noff += iph->ihl << 2;
3637 if (!ip_is_fragment(iph))
3638 *proto = iph->protocol;
3639 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3640 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3642 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3643 iph_to_flow_copy_v6addrs(fk, iph6);
3644 *noff += sizeof(*iph6);
3645 *proto = iph6->nexthdr;
3650 if (l34 && *proto >= 0)
3651 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3656 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb)
3658 struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
3659 u32 srcmac_vendor = 0, srcmac_dev = 0;
3663 for (i = 0; i < 3; i++)
3664 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3666 for (i = 3; i < ETH_ALEN; i++)
3667 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3669 if (!skb_vlan_tag_present(skb))
3670 return srcmac_vendor ^ srcmac_dev;
3672 vlan = skb_vlan_tag_get(skb);
3674 return vlan ^ srcmac_vendor ^ srcmac_dev;
3677 /* Extract the appropriate headers based on bond's xmit policy */
3678 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3679 struct flow_keys *fk)
3681 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3682 int noff, proto = -1;
3684 switch (bond->params.xmit_policy) {
3685 case BOND_XMIT_POLICY_ENCAP23:
3686 case BOND_XMIT_POLICY_ENCAP34:
3687 memset(fk, 0, sizeof(*fk));
3688 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3689 fk, NULL, 0, 0, 0, 0);
3694 fk->ports.ports = 0;
3695 memset(&fk->icmp, 0, sizeof(fk->icmp));
3696 noff = skb_network_offset(skb);
3697 if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3700 /* ICMP error packets contains at least 8 bytes of the header
3701 * of the packet which generated the error. Use this information
3702 * to correlate ICMP error packets within the same flow which
3703 * generated the error.
3705 if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3706 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3707 skb_transport_offset(skb),
3709 if (proto == IPPROTO_ICMP) {
3710 if (!icmp_is_err(fk->icmp.type))
3713 noff += sizeof(struct icmphdr);
3714 } else if (proto == IPPROTO_ICMPV6) {
3715 if (!icmpv6_is_err(fk->icmp.type))
3718 noff += sizeof(struct icmp6hdr);
3720 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3726 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow)
3728 hash ^= (__force u32)flow_get_u32_dst(flow) ^
3729 (__force u32)flow_get_u32_src(flow);
3730 hash ^= (hash >> 16);
3731 hash ^= (hash >> 8);
3732 /* discard lowest hash bit to deal with the common even ports pattern */
3737 * bond_xmit_hash - generate a hash value based on the xmit policy
3738 * @bond: bonding device
3739 * @skb: buffer to use for headers
3741 * This function will extract the necessary headers from the skb buffer and use
3742 * them to generate a hash based on the xmit_policy set in the bonding device
3744 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3746 struct flow_keys flow;
3749 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3753 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
3754 return bond_vlan_srcmac_hash(skb);
3756 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3757 !bond_flow_dissect(bond, skb, &flow))
3758 return bond_eth_hash(skb);
3760 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3761 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3762 hash = bond_eth_hash(skb);
3765 memcpy(&hash, &flow.icmp, sizeof(hash));
3767 memcpy(&hash, &flow.ports.ports, sizeof(hash));
3770 return bond_ip_hash(hash, &flow);
3773 /*-------------------------- Device entry points ----------------------------*/
3775 void bond_work_init_all(struct bonding *bond)
3777 INIT_DELAYED_WORK(&bond->mcast_work,
3778 bond_resend_igmp_join_requests_delayed);
3779 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3780 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3781 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3782 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3783 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3786 static void bond_work_cancel_all(struct bonding *bond)
3788 cancel_delayed_work_sync(&bond->mii_work);
3789 cancel_delayed_work_sync(&bond->arp_work);
3790 cancel_delayed_work_sync(&bond->alb_work);
3791 cancel_delayed_work_sync(&bond->ad_work);
3792 cancel_delayed_work_sync(&bond->mcast_work);
3793 cancel_delayed_work_sync(&bond->slave_arr_work);
3796 static int bond_open(struct net_device *bond_dev)
3798 struct bonding *bond = netdev_priv(bond_dev);
3799 struct list_head *iter;
3800 struct slave *slave;
3802 /* reset slave->backup and slave->inactive */
3803 if (bond_has_slaves(bond)) {
3804 bond_for_each_slave(bond, slave, iter) {
3805 if (bond_uses_primary(bond) &&
3806 slave != rcu_access_pointer(bond->curr_active_slave)) {
3807 bond_set_slave_inactive_flags(slave,
3808 BOND_SLAVE_NOTIFY_NOW);
3809 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3810 bond_set_slave_active_flags(slave,
3811 BOND_SLAVE_NOTIFY_NOW);
3816 if (bond_is_lb(bond)) {
3817 /* bond_alb_initialize must be called before the timer
3820 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3822 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3823 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3826 if (bond->params.miimon) /* link check interval, in milliseconds. */
3827 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3829 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3830 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3831 bond->recv_probe = bond_arp_rcv;
3834 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3835 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3836 /* register to receive LACPDUs */
3837 bond->recv_probe = bond_3ad_lacpdu_recv;
3838 bond_3ad_initiate_agg_selection(bond, 1);
3841 if (bond_mode_can_use_xmit_hash(bond))
3842 bond_update_slave_arr(bond, NULL);
3847 static int bond_close(struct net_device *bond_dev)
3849 struct bonding *bond = netdev_priv(bond_dev);
3851 bond_work_cancel_all(bond);
3852 bond->send_peer_notif = 0;
3853 if (bond_is_lb(bond))
3854 bond_alb_deinitialize(bond);
3855 bond->recv_probe = NULL;
3860 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3861 * that some drivers can provide 32bit values only.
3863 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3864 const struct rtnl_link_stats64 *_new,
3865 const struct rtnl_link_stats64 *_old)
3867 const u64 *new = (const u64 *)_new;
3868 const u64 *old = (const u64 *)_old;
3869 u64 *res = (u64 *)_res;
3872 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3875 s64 delta = nv - ov;
3877 /* detects if this particular field is 32bit only */
3878 if (((nv | ov) >> 32) == 0)
3879 delta = (s64)(s32)((u32)nv - (u32)ov);
3881 /* filter anomalies, some drivers reset their stats
3882 * at down/up events.
3889 #ifdef CONFIG_LOCKDEP
3890 static int bond_get_lowest_level_rcu(struct net_device *dev)
3892 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3893 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3894 int cur = 0, max = 0;
3897 iter = &dev->adj_list.lower;
3902 ldev = netdev_next_lower_dev_rcu(now, &iter);
3907 niter = &ldev->adj_list.lower;
3908 dev_stack[cur] = now;
3909 iter_stack[cur++] = iter;
3918 next = dev_stack[--cur];
3919 niter = iter_stack[cur];
3930 static void bond_get_stats(struct net_device *bond_dev,
3931 struct rtnl_link_stats64 *stats)
3933 struct bonding *bond = netdev_priv(bond_dev);
3934 struct rtnl_link_stats64 temp;
3935 struct list_head *iter;
3936 struct slave *slave;
3941 #ifdef CONFIG_LOCKDEP
3942 nest_level = bond_get_lowest_level_rcu(bond_dev);
3945 spin_lock_nested(&bond->stats_lock, nest_level);
3946 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3948 bond_for_each_slave_rcu(bond, slave, iter) {
3949 const struct rtnl_link_stats64 *new =
3950 dev_get_stats(slave->dev, &temp);
3952 bond_fold_stats(stats, new, &slave->slave_stats);
3954 /* save off the slave stats for the next run */
3955 memcpy(&slave->slave_stats, new, sizeof(*new));
3958 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3959 spin_unlock(&bond->stats_lock);
3963 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3965 struct bonding *bond = netdev_priv(bond_dev);
3966 struct net_device *slave_dev = NULL;
3967 struct ifbond k_binfo;
3968 struct ifbond __user *u_binfo = NULL;
3969 struct ifslave k_sinfo;
3970 struct ifslave __user *u_sinfo = NULL;
3971 struct mii_ioctl_data *mii = NULL;
3972 struct bond_opt_value newval;
3976 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3987 /* We do this again just in case we were called by SIOCGMIIREG
3988 * instead of SIOCGMIIPHY.
3994 if (mii->reg_num == 1) {
3996 if (netif_carrier_ok(bond->dev))
3997 mii->val_out = BMSR_LSTATUS;
4001 case BOND_INFO_QUERY_OLD:
4002 case SIOCBONDINFOQUERY:
4003 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4005 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4008 bond_info_query(bond_dev, &k_binfo);
4009 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4013 case BOND_SLAVE_INFO_QUERY_OLD:
4014 case SIOCBONDSLAVEINFOQUERY:
4015 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4017 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4020 res = bond_slave_info_query(bond_dev, &k_sinfo);
4022 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4030 net = dev_net(bond_dev);
4032 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4035 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4037 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4043 case BOND_ENSLAVE_OLD:
4044 case SIOCBONDENSLAVE:
4045 res = bond_enslave(bond_dev, slave_dev, NULL);
4047 case BOND_RELEASE_OLD:
4048 case SIOCBONDRELEASE:
4049 res = bond_release(bond_dev, slave_dev);
4051 case BOND_SETHWADDR_OLD:
4052 case SIOCBONDSETHWADDR:
4053 res = bond_set_dev_addr(bond_dev, slave_dev);
4055 case BOND_CHANGE_ACTIVE_OLD:
4056 case SIOCBONDCHANGEACTIVE:
4057 bond_opt_initstr(&newval, slave_dev->name);
4058 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4068 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4070 struct bonding *bond = netdev_priv(bond_dev);
4072 if (change & IFF_PROMISC)
4073 bond_set_promiscuity(bond,
4074 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4076 if (change & IFF_ALLMULTI)
4077 bond_set_allmulti(bond,
4078 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4081 static void bond_set_rx_mode(struct net_device *bond_dev)
4083 struct bonding *bond = netdev_priv(bond_dev);
4084 struct list_head *iter;
4085 struct slave *slave;
4088 if (bond_uses_primary(bond)) {
4089 slave = rcu_dereference(bond->curr_active_slave);
4091 dev_uc_sync(slave->dev, bond_dev);
4092 dev_mc_sync(slave->dev, bond_dev);
4095 bond_for_each_slave_rcu(bond, slave, iter) {
4096 dev_uc_sync_multiple(slave->dev, bond_dev);
4097 dev_mc_sync_multiple(slave->dev, bond_dev);
4103 static int bond_neigh_init(struct neighbour *n)
4105 struct bonding *bond = netdev_priv(n->dev);
4106 const struct net_device_ops *slave_ops;
4107 struct neigh_parms parms;
4108 struct slave *slave;
4112 slave = bond_first_slave_rcu(bond);
4115 slave_ops = slave->dev->netdev_ops;
4116 if (!slave_ops->ndo_neigh_setup)
4119 /* TODO: find another way [1] to implement this.
4120 * Passing a zeroed structure is fragile,
4121 * but at least we do not pass garbage.
4123 * [1] One way would be that ndo_neigh_setup() never touch
4124 * struct neigh_parms, but propagate the new neigh_setup()
4125 * back to ___neigh_create() / neigh_parms_alloc()
4127 memset(&parms, 0, sizeof(parms));
4128 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4133 if (parms.neigh_setup)
4134 ret = parms.neigh_setup(n);
4140 /* The bonding ndo_neigh_setup is called at init time beofre any
4141 * slave exists. So we must declare proxy setup function which will
4142 * be used at run time to resolve the actual slave neigh param setup.
4144 * It's also called by master devices (such as vlans) to setup their
4145 * underlying devices. In that case - do nothing, we're already set up from
4148 static int bond_neigh_setup(struct net_device *dev,
4149 struct neigh_parms *parms)
4151 /* modify only our neigh_parms */
4152 if (parms->dev == dev)
4153 parms->neigh_setup = bond_neigh_init;
4158 /* Change the MTU of all of a master's slaves to match the master */
4159 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4161 struct bonding *bond = netdev_priv(bond_dev);
4162 struct slave *slave, *rollback_slave;
4163 struct list_head *iter;
4166 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4168 bond_for_each_slave(bond, slave, iter) {
4169 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4170 slave, slave->dev->netdev_ops->ndo_change_mtu);
4172 res = dev_set_mtu(slave->dev, new_mtu);
4175 /* If we failed to set the slave's mtu to the new value
4176 * we must abort the operation even in ACTIVE_BACKUP
4177 * mode, because if we allow the backup slaves to have
4178 * different mtu values than the active slave we'll
4179 * need to change their mtu when doing a failover. That
4180 * means changing their mtu from timer context, which
4181 * is probably not a good idea.
4183 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4189 bond_dev->mtu = new_mtu;
4194 /* unwind from head to the slave that failed */
4195 bond_for_each_slave(bond, rollback_slave, iter) {
4198 if (rollback_slave == slave)
4201 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4203 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4210 /* Change HW address
4212 * Note that many devices must be down to change the HW address, and
4213 * downing the master releases all slaves. We can make bonds full of
4214 * bonding devices to test this, however.
4216 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4218 struct bonding *bond = netdev_priv(bond_dev);
4219 struct slave *slave, *rollback_slave;
4220 struct sockaddr_storage *ss = addr, tmp_ss;
4221 struct list_head *iter;
4224 if (BOND_MODE(bond) == BOND_MODE_ALB)
4225 return bond_alb_set_mac_address(bond_dev, addr);
4228 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4230 /* If fail_over_mac is enabled, do nothing and return success.
4231 * Returning an error causes ifenslave to fail.
4233 if (bond->params.fail_over_mac &&
4234 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4237 if (!is_valid_ether_addr(ss->__data))
4238 return -EADDRNOTAVAIL;
4240 bond_for_each_slave(bond, slave, iter) {
4241 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4243 res = dev_set_mac_address(slave->dev, addr, NULL);
4245 /* TODO: consider downing the slave
4247 * User should expect communications
4248 * breakage anyway until ARP finish
4251 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4258 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4262 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4263 tmp_ss.ss_family = bond_dev->type;
4265 /* unwind from head to the slave that failed */
4266 bond_for_each_slave(bond, rollback_slave, iter) {
4269 if (rollback_slave == slave)
4272 tmp_res = dev_set_mac_address(rollback_slave->dev,
4273 (struct sockaddr *)&tmp_ss, NULL);
4275 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4284 * bond_get_slave_by_id - get xmit slave with slave_id
4285 * @bond: bonding device that is transmitting
4286 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4288 * This function tries to get slave with slave_id but in case
4289 * it fails, it tries to find the first available slave for transmission.
4291 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4294 struct list_head *iter;
4295 struct slave *slave;
4298 /* Here we start from the slave with slave_id */
4299 bond_for_each_slave_rcu(bond, slave, iter) {
4301 if (bond_slave_can_tx(slave))
4306 /* Here we start from the first slave up to slave_id */
4308 bond_for_each_slave_rcu(bond, slave, iter) {
4311 if (bond_slave_can_tx(slave))
4314 /* no slave that can tx has been found */
4319 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4320 * @bond: bonding device to use
4322 * Based on the value of the bonding device's packets_per_slave parameter
4323 * this function generates a slave id, which is usually used as the next
4324 * slave to transmit through.
4326 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4329 struct reciprocal_value reciprocal_packets_per_slave;
4330 int packets_per_slave = bond->params.packets_per_slave;
4332 switch (packets_per_slave) {
4334 slave_id = prandom_u32();
4337 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4340 reciprocal_packets_per_slave =
4341 bond->params.reciprocal_packets_per_slave;
4342 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4343 slave_id = reciprocal_divide(slave_id,
4344 reciprocal_packets_per_slave);
4351 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4352 struct sk_buff *skb)
4354 struct slave *slave;
4358 /* Start with the curr_active_slave that joined the bond as the
4359 * default for sending IGMP traffic. For failover purposes one
4360 * needs to maintain some consistency for the interface that will
4361 * send the join/membership reports. The curr_active_slave found
4362 * will send all of this type of traffic.
4364 if (skb->protocol == htons(ETH_P_IP)) {
4365 int noff = skb_network_offset(skb);
4368 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4372 if (iph->protocol == IPPROTO_IGMP) {
4373 slave = rcu_dereference(bond->curr_active_slave);
4376 return bond_get_slave_by_id(bond, 0);
4381 slave_cnt = READ_ONCE(bond->slave_cnt);
4382 if (likely(slave_cnt)) {
4383 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4384 return bond_get_slave_by_id(bond, slave_id);
4389 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4390 struct net_device *bond_dev)
4392 struct bonding *bond = netdev_priv(bond_dev);
4393 struct slave *slave;
4395 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4397 return bond_dev_queue_xmit(bond, skb, slave->dev);
4399 return bond_tx_drop(bond_dev, skb);
4402 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4403 struct sk_buff *skb)
4405 return rcu_dereference(bond->curr_active_slave);
4408 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4409 * the bond has a usable interface.
4411 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4412 struct net_device *bond_dev)
4414 struct bonding *bond = netdev_priv(bond_dev);
4415 struct slave *slave;
4417 slave = bond_xmit_activebackup_slave_get(bond, skb);
4419 return bond_dev_queue_xmit(bond, skb, slave->dev);
4421 return bond_tx_drop(bond_dev, skb);
4424 /* Use this to update slave_array when (a) it's not appropriate to update
4425 * slave_array right away (note that update_slave_array() may sleep)
4426 * and / or (b) RTNL is not held.
4428 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4430 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4433 /* Slave array work handler. Holds only RTNL */
4434 static void bond_slave_arr_handler(struct work_struct *work)
4436 struct bonding *bond = container_of(work, struct bonding,
4437 slave_arr_work.work);
4440 if (!rtnl_trylock())
4443 ret = bond_update_slave_arr(bond, NULL);
4446 pr_warn_ratelimited("Failed to update slave array from WT\n");
4452 bond_slave_arr_work_rearm(bond, 1);
4455 static void bond_skip_slave(struct bond_up_slave *slaves,
4456 struct slave *skipslave)
4460 /* Rare situation where caller has asked to skip a specific
4461 * slave but allocation failed (most likely!). BTW this is
4462 * only possible when the call is initiated from
4463 * __bond_release_one(). In this situation; overwrite the
4464 * skipslave entry in the array with the last entry from the
4465 * array to avoid a situation where the xmit path may choose
4466 * this to-be-skipped slave to send a packet out.
4468 for (idx = 0; slaves && idx < slaves->count; idx++) {
4469 if (skipslave == slaves->arr[idx]) {
4471 slaves->arr[slaves->count - 1];
4478 static void bond_set_slave_arr(struct bonding *bond,
4479 struct bond_up_slave *usable_slaves,
4480 struct bond_up_slave *all_slaves)
4482 struct bond_up_slave *usable, *all;
4484 usable = rtnl_dereference(bond->usable_slaves);
4485 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4486 kfree_rcu(usable, rcu);
4488 all = rtnl_dereference(bond->all_slaves);
4489 rcu_assign_pointer(bond->all_slaves, all_slaves);
4490 kfree_rcu(all, rcu);
4493 static void bond_reset_slave_arr(struct bonding *bond)
4495 struct bond_up_slave *usable, *all;
4497 usable = rtnl_dereference(bond->usable_slaves);
4499 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4500 kfree_rcu(usable, rcu);
4503 all = rtnl_dereference(bond->all_slaves);
4505 RCU_INIT_POINTER(bond->all_slaves, NULL);
4506 kfree_rcu(all, rcu);
4510 /* Build the usable slaves array in control path for modes that use xmit-hash
4511 * to determine the slave interface -
4512 * (a) BOND_MODE_8023AD
4514 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4516 * The caller is expected to hold RTNL only and NO other lock!
4518 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4520 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4521 struct slave *slave;
4522 struct list_head *iter;
4528 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4529 bond->slave_cnt), GFP_KERNEL);
4530 all_slaves = kzalloc(struct_size(all_slaves, arr,
4531 bond->slave_cnt), GFP_KERNEL);
4532 if (!usable_slaves || !all_slaves) {
4536 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4537 struct ad_info ad_info;
4539 spin_lock_bh(&bond->mode_lock);
4540 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4541 spin_unlock_bh(&bond->mode_lock);
4542 pr_debug("bond_3ad_get_active_agg_info failed\n");
4543 /* No active aggragator means it's not safe to use
4544 * the previous array.
4546 bond_reset_slave_arr(bond);
4549 spin_unlock_bh(&bond->mode_lock);
4550 agg_id = ad_info.aggregator_id;
4552 bond_for_each_slave(bond, slave, iter) {
4553 if (skipslave == slave)
4556 all_slaves->arr[all_slaves->count++] = slave;
4557 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4558 struct aggregator *agg;
4560 agg = SLAVE_AD_INFO(slave)->port.aggregator;
4561 if (!agg || agg->aggregator_identifier != agg_id)
4564 if (!bond_slave_can_tx(slave))
4567 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4568 usable_slaves->count);
4570 usable_slaves->arr[usable_slaves->count++] = slave;
4573 bond_set_slave_arr(bond, usable_slaves, all_slaves);
4576 if (ret != 0 && skipslave) {
4577 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4579 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4582 kfree_rcu(all_slaves, rcu);
4583 kfree_rcu(usable_slaves, rcu);
4588 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4589 struct sk_buff *skb,
4590 struct bond_up_slave *slaves)
4592 struct slave *slave;
4596 hash = bond_xmit_hash(bond, skb);
4597 count = slaves ? READ_ONCE(slaves->count) : 0;
4598 if (unlikely(!count))
4601 slave = slaves->arr[hash % count];
4605 /* Use this Xmit function for 3AD as well as XOR modes. The current
4606 * usable slave array is formed in the control path. The xmit function
4607 * just calculates hash and sends the packet out.
4609 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4610 struct net_device *dev)
4612 struct bonding *bond = netdev_priv(dev);
4613 struct bond_up_slave *slaves;
4614 struct slave *slave;
4616 slaves = rcu_dereference(bond->usable_slaves);
4617 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4619 return bond_dev_queue_xmit(bond, skb, slave->dev);
4621 return bond_tx_drop(dev, skb);
4624 /* in broadcast mode, we send everything to all usable interfaces. */
4625 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4626 struct net_device *bond_dev)
4628 struct bonding *bond = netdev_priv(bond_dev);
4629 struct slave *slave = NULL;
4630 struct list_head *iter;
4632 bond_for_each_slave_rcu(bond, slave, iter) {
4633 if (bond_is_last_slave(bond, slave))
4635 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4636 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4639 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4640 bond_dev->name, __func__);
4643 bond_dev_queue_xmit(bond, skb2, slave->dev);
4646 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4647 return bond_dev_queue_xmit(bond, skb, slave->dev);
4649 return bond_tx_drop(bond_dev, skb);
4652 /*------------------------- Device initialization ---------------------------*/
4654 /* Lookup the slave that corresponds to a qid */
4655 static inline int bond_slave_override(struct bonding *bond,
4656 struct sk_buff *skb)
4658 struct slave *slave = NULL;
4659 struct list_head *iter;
4661 if (!skb_rx_queue_recorded(skb))
4664 /* Find out if any slaves have the same mapping as this skb. */
4665 bond_for_each_slave_rcu(bond, slave, iter) {
4666 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4667 if (bond_slave_is_up(slave) &&
4668 slave->link == BOND_LINK_UP) {
4669 bond_dev_queue_xmit(bond, skb, slave->dev);
4672 /* If the slave isn't UP, use default transmit policy. */
4681 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4682 struct net_device *sb_dev)
4684 /* This helper function exists to help dev_pick_tx get the correct
4685 * destination queue. Using a helper function skips a call to
4686 * skb_tx_hash and will put the skbs in the queue we expect on their
4687 * way down to the bonding driver.
4689 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4691 /* Save the original txq to restore before passing to the driver */
4692 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4694 if (unlikely(txq >= dev->real_num_tx_queues)) {
4696 txq -= dev->real_num_tx_queues;
4697 } while (txq >= dev->real_num_tx_queues);
4702 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4703 struct sk_buff *skb,
4706 struct bonding *bond = netdev_priv(master_dev);
4707 struct bond_up_slave *slaves;
4708 struct slave *slave = NULL;
4710 switch (BOND_MODE(bond)) {
4711 case BOND_MODE_ROUNDROBIN:
4712 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4714 case BOND_MODE_ACTIVEBACKUP:
4715 slave = bond_xmit_activebackup_slave_get(bond, skb);
4717 case BOND_MODE_8023AD:
4720 slaves = rcu_dereference(bond->all_slaves);
4722 slaves = rcu_dereference(bond->usable_slaves);
4723 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4725 case BOND_MODE_BROADCAST:
4728 slave = bond_xmit_alb_slave_get(bond, skb);
4731 slave = bond_xmit_tlb_slave_get(bond, skb);
4734 /* Should never happen, mode already checked */
4735 WARN_ONCE(true, "Unknown bonding mode");
4744 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
4746 switch (sk->sk_family) {
4747 #if IS_ENABLED(CONFIG_IPV6)
4749 if (sk->sk_ipv6only ||
4750 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
4751 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
4752 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
4753 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
4758 default: /* AF_INET */
4759 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
4760 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
4761 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
4765 flow->ports.src = inet_sk(sk)->inet_sport;
4766 flow->ports.dst = inet_sk(sk)->inet_dport;
4770 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
4771 * @sk: socket to use for headers
4773 * This function will extract the necessary field from the socket and use
4774 * them to generate a hash based on the LAYER34 xmit_policy.
4775 * Assumes that sk is a TCP or UDP socket.
4777 static u32 bond_sk_hash_l34(struct sock *sk)
4779 struct flow_keys flow;
4782 bond_sk_to_flow(sk, &flow);
4785 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4787 return bond_ip_hash(hash, &flow);
4790 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
4793 struct bond_up_slave *slaves;
4794 struct slave *slave;
4798 slaves = rcu_dereference(bond->usable_slaves);
4799 count = slaves ? READ_ONCE(slaves->count) : 0;
4800 if (unlikely(!count))
4803 hash = bond_sk_hash_l34(sk);
4804 slave = slaves->arr[hash % count];
4809 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
4812 struct bonding *bond = netdev_priv(dev);
4813 struct net_device *lower = NULL;
4816 if (bond_sk_check(bond))
4817 lower = __bond_sk_get_lower_dev(bond, sk);
4823 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4824 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
4825 struct net_device *dev)
4827 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
4828 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
4829 return bond_tx_drop(dev, skb);
4833 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4835 struct bonding *bond = netdev_priv(dev);
4837 if (bond_should_override_tx_queue(bond) &&
4838 !bond_slave_override(bond, skb))
4839 return NETDEV_TX_OK;
4841 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4842 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
4843 return bond_tls_device_xmit(bond, skb, dev);
4846 switch (BOND_MODE(bond)) {
4847 case BOND_MODE_ROUNDROBIN:
4848 return bond_xmit_roundrobin(skb, dev);
4849 case BOND_MODE_ACTIVEBACKUP:
4850 return bond_xmit_activebackup(skb, dev);
4851 case BOND_MODE_8023AD:
4853 return bond_3ad_xor_xmit(skb, dev);
4854 case BOND_MODE_BROADCAST:
4855 return bond_xmit_broadcast(skb, dev);
4857 return bond_alb_xmit(skb, dev);
4859 return bond_tlb_xmit(skb, dev);
4861 /* Should never happen, mode already checked */
4862 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4864 return bond_tx_drop(dev, skb);
4868 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4870 struct bonding *bond = netdev_priv(dev);
4871 netdev_tx_t ret = NETDEV_TX_OK;
4873 /* If we risk deadlock from transmitting this in the
4874 * netpoll path, tell netpoll to queue the frame for later tx
4876 if (unlikely(is_netpoll_tx_blocked(dev)))
4877 return NETDEV_TX_BUSY;
4880 if (bond_has_slaves(bond))
4881 ret = __bond_start_xmit(skb, dev);
4883 ret = bond_tx_drop(dev, skb);
4889 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4891 if (speed == 0 || speed == SPEED_UNKNOWN)
4892 speed = slave->speed;
4894 speed = min(speed, slave->speed);
4899 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4900 struct ethtool_link_ksettings *cmd)
4902 struct bonding *bond = netdev_priv(bond_dev);
4903 struct list_head *iter;
4904 struct slave *slave;
4907 cmd->base.duplex = DUPLEX_UNKNOWN;
4908 cmd->base.port = PORT_OTHER;
4910 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4911 * do not need to check mode. Though link speed might not represent
4912 * the true receive or transmit bandwidth (not all modes are symmetric)
4913 * this is an accurate maximum.
4915 bond_for_each_slave(bond, slave, iter) {
4916 if (bond_slave_can_tx(slave)) {
4917 if (slave->speed != SPEED_UNKNOWN) {
4918 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4919 speed = bond_mode_bcast_speed(slave,
4922 speed += slave->speed;
4924 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4925 slave->duplex != DUPLEX_UNKNOWN)
4926 cmd->base.duplex = slave->duplex;
4929 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4934 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4935 struct ethtool_drvinfo *drvinfo)
4937 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4938 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4942 static const struct ethtool_ops bond_ethtool_ops = {
4943 .get_drvinfo = bond_ethtool_get_drvinfo,
4944 .get_link = ethtool_op_get_link,
4945 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4948 static const struct net_device_ops bond_netdev_ops = {
4949 .ndo_init = bond_init,
4950 .ndo_uninit = bond_uninit,
4951 .ndo_open = bond_open,
4952 .ndo_stop = bond_close,
4953 .ndo_start_xmit = bond_start_xmit,
4954 .ndo_select_queue = bond_select_queue,
4955 .ndo_get_stats64 = bond_get_stats,
4956 .ndo_do_ioctl = bond_do_ioctl,
4957 .ndo_change_rx_flags = bond_change_rx_flags,
4958 .ndo_set_rx_mode = bond_set_rx_mode,
4959 .ndo_change_mtu = bond_change_mtu,
4960 .ndo_set_mac_address = bond_set_mac_address,
4961 .ndo_neigh_setup = bond_neigh_setup,
4962 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4963 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4964 #ifdef CONFIG_NET_POLL_CONTROLLER
4965 .ndo_netpoll_setup = bond_netpoll_setup,
4966 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4967 .ndo_poll_controller = bond_poll_controller,
4969 .ndo_add_slave = bond_enslave,
4970 .ndo_del_slave = bond_release,
4971 .ndo_fix_features = bond_fix_features,
4972 .ndo_features_check = passthru_features_check,
4973 .ndo_get_xmit_slave = bond_xmit_get_slave,
4974 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
4977 static const struct device_type bond_type = {
4981 static void bond_destructor(struct net_device *bond_dev)
4983 struct bonding *bond = netdev_priv(bond_dev);
4986 destroy_workqueue(bond->wq);
4988 if (bond->rr_tx_counter)
4989 free_percpu(bond->rr_tx_counter);
4992 void bond_setup(struct net_device *bond_dev)
4994 struct bonding *bond = netdev_priv(bond_dev);
4996 spin_lock_init(&bond->mode_lock);
4997 bond->params = bonding_defaults;
4999 /* Initialize pointers */
5000 bond->dev = bond_dev;
5002 /* Initialize the device entry points */
5003 ether_setup(bond_dev);
5004 bond_dev->max_mtu = ETH_MAX_MTU;
5005 bond_dev->netdev_ops = &bond_netdev_ops;
5006 bond_dev->ethtool_ops = &bond_ethtool_ops;
5008 bond_dev->needs_free_netdev = true;
5009 bond_dev->priv_destructor = bond_destructor;
5011 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5013 /* Initialize the device options */
5014 bond_dev->flags |= IFF_MASTER;
5015 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5016 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5018 #ifdef CONFIG_XFRM_OFFLOAD
5019 /* set up xfrm device ops (only supported in active-backup right now) */
5020 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5021 INIT_LIST_HEAD(&bond->ipsec_list);
5022 spin_lock_init(&bond->ipsec_lock);
5023 #endif /* CONFIG_XFRM_OFFLOAD */
5025 /* don't acquire bond device's netif_tx_lock when transmitting */
5026 bond_dev->features |= NETIF_F_LLTX;
5028 /* By default, we declare the bond to be fully
5029 * VLAN hardware accelerated capable. Special
5030 * care is taken in the various xmit functions
5031 * when there are slaves that are not hw accel
5035 /* Don't allow bond devices to change network namespaces. */
5036 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5038 bond_dev->hw_features = BOND_VLAN_FEATURES |
5039 NETIF_F_HW_VLAN_CTAG_RX |
5040 NETIF_F_HW_VLAN_CTAG_FILTER;
5042 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5043 bond_dev->features |= bond_dev->hw_features;
5044 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5045 #ifdef CONFIG_XFRM_OFFLOAD
5046 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5047 /* Only enable XFRM features if this is an active-backup config */
5048 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5049 bond_dev->features |= BOND_XFRM_FEATURES;
5050 #endif /* CONFIG_XFRM_OFFLOAD */
5051 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5052 if (bond_sk_check(bond))
5053 bond_dev->features |= BOND_TLS_FEATURES;
5057 /* Destroy a bonding device.
5058 * Must be under rtnl_lock when this function is called.
5060 static void bond_uninit(struct net_device *bond_dev)
5062 struct bonding *bond = netdev_priv(bond_dev);
5063 struct bond_up_slave *usable, *all;
5064 struct list_head *iter;
5065 struct slave *slave;
5067 bond_netpoll_cleanup(bond_dev);
5069 /* Release the bonded slaves */
5070 bond_for_each_slave(bond, slave, iter)
5071 __bond_release_one(bond_dev, slave->dev, true, true);
5072 netdev_info(bond_dev, "Released all slaves\n");
5074 usable = rtnl_dereference(bond->usable_slaves);
5076 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5077 kfree_rcu(usable, rcu);
5080 all = rtnl_dereference(bond->all_slaves);
5082 RCU_INIT_POINTER(bond->all_slaves, NULL);
5083 kfree_rcu(all, rcu);
5086 list_del(&bond->bond_list);
5088 bond_debug_unregister(bond);
5091 /*------------------------- Module initialization ---------------------------*/
5093 static int bond_check_params(struct bond_params *params)
5095 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5096 struct bond_opt_value newval;
5097 const struct bond_opt_value *valptr;
5098 int arp_all_targets_value = 0;
5099 u16 ad_actor_sys_prio = 0;
5100 u16 ad_user_port_key = 0;
5101 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5103 int bond_mode = BOND_MODE_ROUNDROBIN;
5104 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5108 /* Convert string parameters. */
5110 bond_opt_initstr(&newval, mode);
5111 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5113 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5116 bond_mode = valptr->value;
5119 if (xmit_hash_policy) {
5120 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5121 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5122 bond_mode == BOND_MODE_BROADCAST) {
5123 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5124 bond_mode_name(bond_mode));
5126 bond_opt_initstr(&newval, xmit_hash_policy);
5127 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5130 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5134 xmit_hashtype = valptr->value;
5139 if (bond_mode != BOND_MODE_8023AD) {
5140 pr_info("lacp_rate param is irrelevant in mode %s\n",
5141 bond_mode_name(bond_mode));
5143 bond_opt_initstr(&newval, lacp_rate);
5144 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5147 pr_err("Error: Invalid lacp rate \"%s\"\n",
5151 lacp_fast = valptr->value;
5156 bond_opt_initstr(&newval, ad_select);
5157 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5160 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5163 params->ad_select = valptr->value;
5164 if (bond_mode != BOND_MODE_8023AD)
5165 pr_warn("ad_select param only affects 802.3ad mode\n");
5167 params->ad_select = BOND_AD_STABLE;
5170 if (max_bonds < 0) {
5171 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5172 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5173 max_bonds = BOND_DEFAULT_MAX_BONDS;
5177 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5183 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5188 if (downdelay < 0) {
5189 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5190 downdelay, INT_MAX);
5194 if ((use_carrier != 0) && (use_carrier != 1)) {
5195 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5200 if (num_peer_notif < 0 || num_peer_notif > 255) {
5201 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5206 /* reset values for 802.3ad/TLB/ALB */
5207 if (!bond_mode_uses_arp(bond_mode)) {
5209 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");
5210 pr_warn("Forcing miimon to 100msec\n");
5211 miimon = BOND_DEFAULT_MIIMON;
5215 if (tx_queues < 1 || tx_queues > 255) {
5216 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5217 tx_queues, BOND_DEFAULT_TX_QUEUES);
5218 tx_queues = BOND_DEFAULT_TX_QUEUES;
5221 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5222 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5224 all_slaves_active = 0;
5227 if (resend_igmp < 0 || resend_igmp > 255) {
5228 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5229 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5230 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5233 bond_opt_initval(&newval, packets_per_slave);
5234 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5235 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5236 packets_per_slave, USHRT_MAX);
5237 packets_per_slave = 1;
5240 if (bond_mode == BOND_MODE_ALB) {
5241 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",
5246 if (updelay || downdelay) {
5247 /* just warn the user the up/down delay will have
5248 * no effect since miimon is zero...
5250 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",
5251 updelay, downdelay);
5254 /* don't allow arp monitoring */
5256 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5257 miimon, arp_interval);
5261 if ((updelay % miimon) != 0) {
5262 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5263 updelay, miimon, (updelay / miimon) * miimon);
5268 if ((downdelay % miimon) != 0) {
5269 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5271 (downdelay / miimon) * miimon);
5274 downdelay /= miimon;
5277 if (arp_interval < 0) {
5278 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5279 arp_interval, INT_MAX);
5283 for (arp_ip_count = 0, i = 0;
5284 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5287 /* not a complete check, but good enough to catch mistakes */
5288 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5289 !bond_is_ip_target_ok(ip)) {
5290 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5294 if (bond_get_targets_ip(arp_target, ip) == -1)
5295 arp_target[arp_ip_count++] = ip;
5297 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5302 if (arp_interval && !arp_ip_count) {
5303 /* don't allow arping if no arp_ip_target given... */
5304 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5310 if (!arp_interval) {
5311 pr_err("arp_validate requires arp_interval\n");
5315 bond_opt_initstr(&newval, arp_validate);
5316 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5319 pr_err("Error: invalid arp_validate \"%s\"\n",
5323 arp_validate_value = valptr->value;
5325 arp_validate_value = 0;
5328 if (arp_all_targets) {
5329 bond_opt_initstr(&newval, arp_all_targets);
5330 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5333 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5335 arp_all_targets_value = 0;
5337 arp_all_targets_value = valptr->value;
5342 pr_info("MII link monitoring set to %d ms\n", miimon);
5343 } else if (arp_interval) {
5344 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5345 arp_validate_value);
5346 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5347 arp_interval, valptr->string, arp_ip_count);
5349 for (i = 0; i < arp_ip_count; i++)
5350 pr_cont(" %s", arp_ip_target[i]);
5354 } else if (max_bonds) {
5355 /* miimon and arp_interval not set, we need one so things
5356 * work as expected, see bonding.txt for details
5358 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");
5361 if (primary && !bond_mode_uses_primary(bond_mode)) {
5362 /* currently, using a primary only makes sense
5363 * in active backup, TLB or ALB modes
5365 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5366 primary, bond_mode_name(bond_mode));
5370 if (primary && primary_reselect) {
5371 bond_opt_initstr(&newval, primary_reselect);
5372 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5375 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5379 primary_reselect_value = valptr->value;
5381 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5384 if (fail_over_mac) {
5385 bond_opt_initstr(&newval, fail_over_mac);
5386 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5389 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5393 fail_over_mac_value = valptr->value;
5394 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5395 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5397 fail_over_mac_value = BOND_FOM_NONE;
5400 bond_opt_initstr(&newval, "default");
5401 valptr = bond_opt_parse(
5402 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5405 pr_err("Error: No ad_actor_sys_prio default value");
5408 ad_actor_sys_prio = valptr->value;
5410 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5413 pr_err("Error: No ad_user_port_key default value");
5416 ad_user_port_key = valptr->value;
5418 bond_opt_initstr(&newval, "default");
5419 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5421 pr_err("Error: No tlb_dynamic_lb default value");
5424 tlb_dynamic_lb = valptr->value;
5426 if (lp_interval == 0) {
5427 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5428 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5429 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5432 /* fill params struct with the proper values */
5433 params->mode = bond_mode;
5434 params->xmit_policy = xmit_hashtype;
5435 params->miimon = miimon;
5436 params->num_peer_notif = num_peer_notif;
5437 params->arp_interval = arp_interval;
5438 params->arp_validate = arp_validate_value;
5439 params->arp_all_targets = arp_all_targets_value;
5440 params->updelay = updelay;
5441 params->downdelay = downdelay;
5442 params->peer_notif_delay = 0;
5443 params->use_carrier = use_carrier;
5444 params->lacp_fast = lacp_fast;
5445 params->primary[0] = 0;
5446 params->primary_reselect = primary_reselect_value;
5447 params->fail_over_mac = fail_over_mac_value;
5448 params->tx_queues = tx_queues;
5449 params->all_slaves_active = all_slaves_active;
5450 params->resend_igmp = resend_igmp;
5451 params->min_links = min_links;
5452 params->lp_interval = lp_interval;
5453 params->packets_per_slave = packets_per_slave;
5454 params->tlb_dynamic_lb = tlb_dynamic_lb;
5455 params->ad_actor_sys_prio = ad_actor_sys_prio;
5456 eth_zero_addr(params->ad_actor_system);
5457 params->ad_user_port_key = ad_user_port_key;
5458 if (packets_per_slave > 0) {
5459 params->reciprocal_packets_per_slave =
5460 reciprocal_value(packets_per_slave);
5462 /* reciprocal_packets_per_slave is unused if
5463 * packets_per_slave is 0 or 1, just initialize it
5465 params->reciprocal_packets_per_slave =
5466 (struct reciprocal_value) { 0 };
5470 strscpy_pad(params->primary, primary, sizeof(params->primary));
5472 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5477 /* Called from registration process */
5478 static int bond_init(struct net_device *bond_dev)
5480 struct bonding *bond = netdev_priv(bond_dev);
5481 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5483 netdev_dbg(bond_dev, "Begin bond_init\n");
5485 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5489 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
5490 bond->rr_tx_counter = alloc_percpu(u32);
5491 if (!bond->rr_tx_counter) {
5492 destroy_workqueue(bond->wq);
5498 spin_lock_init(&bond->stats_lock);
5499 netdev_lockdep_set_classes(bond_dev);
5501 list_add_tail(&bond->bond_list, &bn->dev_list);
5503 bond_prepare_sysfs_group(bond);
5505 bond_debug_register(bond);
5507 /* Ensure valid dev_addr */
5508 if (is_zero_ether_addr(bond_dev->dev_addr) &&
5509 bond_dev->addr_assign_type == NET_ADDR_PERM)
5510 eth_hw_addr_random(bond_dev);
5515 unsigned int bond_get_num_tx_queues(void)
5520 /* Create a new bond based on the specified name and bonding parameters.
5521 * If name is NULL, obtain a suitable "bond%d" name for us.
5522 * Caller must NOT hold rtnl_lock; we need to release it here before we
5523 * set up our sysfs entries.
5525 int bond_create(struct net *net, const char *name)
5527 struct net_device *bond_dev;
5528 struct bonding *bond;
5529 struct alb_bond_info *bond_info;
5534 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5535 name ? name : "bond%d", NET_NAME_UNKNOWN,
5536 bond_setup, tx_queues);
5538 pr_err("%s: eek! can't alloc netdev!\n", name);
5544 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5545 * It is set to 0 by default which is wrong.
5547 bond = netdev_priv(bond_dev);
5548 bond_info = &(BOND_ALB_INFO(bond));
5549 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5551 dev_net_set(bond_dev, net);
5552 bond_dev->rtnl_link_ops = &bond_link_ops;
5554 res = register_netdevice(bond_dev);
5556 free_netdev(bond_dev);
5562 netif_carrier_off(bond_dev);
5564 bond_work_init_all(bond);
5570 static int __net_init bond_net_init(struct net *net)
5572 struct bond_net *bn = net_generic(net, bond_net_id);
5575 INIT_LIST_HEAD(&bn->dev_list);
5577 bond_create_proc_dir(bn);
5578 bond_create_sysfs(bn);
5583 static void __net_exit bond_net_exit(struct net *net)
5585 struct bond_net *bn = net_generic(net, bond_net_id);
5586 struct bonding *bond, *tmp_bond;
5589 bond_destroy_sysfs(bn);
5591 /* Kill off any bonds created after unregistering bond rtnl ops */
5593 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5594 unregister_netdevice_queue(bond->dev, &list);
5595 unregister_netdevice_many(&list);
5598 bond_destroy_proc_dir(bn);
5601 static struct pernet_operations bond_net_ops = {
5602 .init = bond_net_init,
5603 .exit = bond_net_exit,
5605 .size = sizeof(struct bond_net),
5608 static int __init bonding_init(void)
5613 res = bond_check_params(&bonding_defaults);
5617 res = register_pernet_subsys(&bond_net_ops);
5621 res = bond_netlink_init();
5625 bond_create_debugfs();
5627 for (i = 0; i < max_bonds; i++) {
5628 res = bond_create(&init_net, NULL);
5633 skb_flow_dissector_init(&flow_keys_bonding,
5634 flow_keys_bonding_keys,
5635 ARRAY_SIZE(flow_keys_bonding_keys));
5637 register_netdevice_notifier(&bond_netdev_notifier);
5641 bond_destroy_debugfs();
5642 bond_netlink_fini();
5644 unregister_pernet_subsys(&bond_net_ops);
5649 static void __exit bonding_exit(void)
5651 unregister_netdevice_notifier(&bond_netdev_notifier);
5653 bond_destroy_debugfs();
5655 bond_netlink_fini();
5656 unregister_pernet_subsys(&bond_net_ops);
5658 #ifdef CONFIG_NET_POLL_CONTROLLER
5659 /* Make sure we don't have an imbalance on our netpoll blocking */
5660 WARN_ON(atomic_read(&netpoll_block_tx));
5664 module_init(bonding_init);
5665 module_exit(bonding_exit);
5666 MODULE_LICENSE("GPL");
5667 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5668 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");