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/filter.h>
39 #include <linux/interrupt.h>
40 #include <linux/ptrace.h>
41 #include <linux/ioport.h>
45 #include <linux/icmp.h>
46 #include <linux/icmpv6.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/slab.h>
50 #include <linux/string.h>
51 #include <linux/init.h>
52 #include <linux/timer.h>
53 #include <linux/socket.h>
54 #include <linux/ctype.h>
55 #include <linux/inet.h>
56 #include <linux/bitops.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/phy.h>
76 #include <linux/jiffies.h>
77 #include <linux/preempt.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include <net/netns/generic.h>
81 #include <net/pkt_sched.h>
82 #include <linux/rculist.h>
83 #include <net/flow_dissector.h>
85 #include <net/bonding.h>
86 #include <net/bond_3ad.h>
87 #include <net/bond_alb.h>
88 #if IS_ENABLED(CONFIG_TLS_DEVICE)
91 #include <net/ip6_route.h>
93 #include "bonding_priv.h"
95 /*---------------------------- Module parameters ----------------------------*/
97 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
99 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
100 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
101 static int num_peer_notif = 1;
104 static int downdelay;
105 static int use_carrier = 1;
107 static char *primary;
108 static char *primary_reselect;
109 static char *lacp_rate;
110 static int min_links;
111 static char *ad_select;
112 static char *xmit_hash_policy;
113 static int arp_interval;
114 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
115 static char *arp_validate;
116 static char *arp_all_targets;
117 static char *fail_over_mac;
118 static int all_slaves_active;
119 static struct bond_params bonding_defaults;
120 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
121 static int packets_per_slave = 1;
122 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
124 module_param(max_bonds, int, 0);
125 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
126 module_param(tx_queues, int, 0);
127 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
128 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
129 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
130 "failover event (alias of num_unsol_na)");
131 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
132 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
133 "failover event (alias of num_grat_arp)");
134 module_param(miimon, int, 0);
135 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
136 module_param(updelay, int, 0);
137 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
138 module_param(downdelay, int, 0);
139 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
141 module_param(use_carrier, int, 0);
142 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
143 "0 for off, 1 for on (default)");
144 module_param(mode, charp, 0);
145 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
146 "1 for active-backup, 2 for balance-xor, "
147 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
148 "6 for balance-alb");
149 module_param(primary, charp, 0);
150 MODULE_PARM_DESC(primary, "Primary network device to use");
151 module_param(primary_reselect, charp, 0);
152 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
154 "0 for always (default), "
155 "1 for only if speed of primary is "
157 "2 for only on active slave "
159 module_param(lacp_rate, charp, 0);
160 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
161 "0 for slow, 1 for fast");
162 module_param(ad_select, charp, 0);
163 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
164 "0 for stable (default), 1 for bandwidth, "
166 module_param(min_links, int, 0);
167 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
169 module_param(xmit_hash_policy, charp, 0);
170 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
171 "0 for layer 2 (default), 1 for layer 3+4, "
172 "2 for layer 2+3, 3 for encap layer 2+3, "
173 "4 for encap layer 3+4, 5 for vlan+srcmac");
174 module_param(arp_interval, int, 0);
175 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
176 module_param_array(arp_ip_target, charp, NULL, 0);
177 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
178 module_param(arp_validate, charp, 0);
179 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
180 "0 for none (default), 1 for active, "
181 "2 for backup, 3 for all");
182 module_param(arp_all_targets, charp, 0);
183 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
184 module_param(fail_over_mac, charp, 0);
185 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
186 "the same MAC; 0 for none (default), "
187 "1 for active, 2 for follow");
188 module_param(all_slaves_active, int, 0);
189 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
190 "by setting active flag for all slaves; "
191 "0 for never (default), 1 for always.");
192 module_param(resend_igmp, int, 0);
193 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
195 module_param(packets_per_slave, int, 0);
196 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
197 "mode; 0 for a random slave, 1 packet per "
198 "slave (default), >1 packets per slave.");
199 module_param(lp_interval, uint, 0);
200 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
201 "the bonding driver sends learning packets to "
202 "each slaves peer switch. The default is 1.");
204 /*----------------------------- Global variables ----------------------------*/
206 #ifdef CONFIG_NET_POLL_CONTROLLER
207 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210 unsigned int bond_net_id __read_mostly;
212 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
214 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
215 .offset = offsetof(struct flow_keys, control),
218 .key_id = FLOW_DISSECTOR_KEY_BASIC,
219 .offset = offsetof(struct flow_keys, basic),
222 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
223 .offset = offsetof(struct flow_keys, addrs.v4addrs),
226 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
227 .offset = offsetof(struct flow_keys, addrs.v6addrs),
230 .key_id = FLOW_DISSECTOR_KEY_TIPC,
231 .offset = offsetof(struct flow_keys, addrs.tipckey),
234 .key_id = FLOW_DISSECTOR_KEY_PORTS,
235 .offset = offsetof(struct flow_keys, ports),
238 .key_id = FLOW_DISSECTOR_KEY_ICMP,
239 .offset = offsetof(struct flow_keys, icmp),
242 .key_id = FLOW_DISSECTOR_KEY_VLAN,
243 .offset = offsetof(struct flow_keys, vlan),
246 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
247 .offset = offsetof(struct flow_keys, tags),
250 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
251 .offset = offsetof(struct flow_keys, keyid),
255 static struct flow_dissector flow_keys_bonding __read_mostly;
257 /*-------------------------- Forward declarations ---------------------------*/
259 static int bond_init(struct net_device *bond_dev);
260 static void bond_uninit(struct net_device *bond_dev);
261 static void bond_get_stats(struct net_device *bond_dev,
262 struct rtnl_link_stats64 *stats);
263 static void bond_slave_arr_handler(struct work_struct *work);
264 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
266 static void bond_netdev_notify_work(struct work_struct *work);
268 /*---------------------------- General routines -----------------------------*/
270 const char *bond_mode_name(int mode)
272 static const char *names[] = {
273 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
274 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
275 [BOND_MODE_XOR] = "load balancing (xor)",
276 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
277 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
278 [BOND_MODE_TLB] = "transmit load balancing",
279 [BOND_MODE_ALB] = "adaptive load balancing",
282 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
289 * bond_dev_queue_xmit - Prepare skb for xmit.
291 * @bond: bond device that got this skb for tx.
292 * @skb: hw accel VLAN tagged skb to transmit
293 * @slave_dev: slave that is supposed to xmit this skbuff
295 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
296 struct net_device *slave_dev)
298 skb->dev = slave_dev;
300 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
301 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
302 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
304 if (unlikely(netpoll_tx_running(bond->dev)))
305 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
307 return dev_queue_xmit(skb);
310 static bool bond_sk_check(struct bonding *bond)
312 switch (BOND_MODE(bond)) {
313 case BOND_MODE_8023AD:
315 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
323 static bool bond_xdp_check(struct bonding *bond)
325 switch (BOND_MODE(bond)) {
326 case BOND_MODE_ROUNDROBIN:
327 case BOND_MODE_ACTIVEBACKUP:
329 case BOND_MODE_8023AD:
331 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
332 * payload is not in the packet due to hardware offload.
334 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
342 /*---------------------------------- VLAN -----------------------------------*/
344 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
345 * We don't protect the slave list iteration with a lock because:
346 * a. This operation is performed in IOCTL context,
347 * b. The operation is protected by the RTNL semaphore in the 8021q code,
348 * c. Holding a lock with BH disabled while directly calling a base driver
349 * entry point is generally a BAD idea.
351 * The design of synchronization/protection for this operation in the 8021q
352 * module is good for one or more VLAN devices over a single physical device
353 * and cannot be extended for a teaming solution like bonding, so there is a
354 * potential race condition here where a net device from the vlan group might
355 * be referenced (either by a base driver or the 8021q code) while it is being
356 * removed from the system. However, it turns out we're not making matters
357 * worse, and if it works for regular VLAN usage it will work here too.
361 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
362 * @bond_dev: bonding net device that got called
363 * @proto: network protocol ID
364 * @vid: vlan id being added
366 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
367 __be16 proto, u16 vid)
369 struct bonding *bond = netdev_priv(bond_dev);
370 struct slave *slave, *rollback_slave;
371 struct list_head *iter;
374 bond_for_each_slave(bond, slave, iter) {
375 res = vlan_vid_add(slave->dev, proto, vid);
383 /* unwind to the slave that failed */
384 bond_for_each_slave(bond, rollback_slave, iter) {
385 if (rollback_slave == slave)
388 vlan_vid_del(rollback_slave->dev, proto, vid);
395 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
396 * @bond_dev: bonding net device that got called
397 * @proto: network protocol ID
398 * @vid: vlan id being removed
400 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
401 __be16 proto, u16 vid)
403 struct bonding *bond = netdev_priv(bond_dev);
404 struct list_head *iter;
407 bond_for_each_slave(bond, slave, iter)
408 vlan_vid_del(slave->dev, proto, vid);
410 if (bond_is_lb(bond))
411 bond_alb_clear_vlan(bond, vid);
416 /*---------------------------------- XFRM -----------------------------------*/
418 #ifdef CONFIG_XFRM_OFFLOAD
420 * bond_ipsec_add_sa - program device with a security association
421 * @xs: pointer to transformer state struct
422 * @extack: extack point to fill failure reason
424 static int bond_ipsec_add_sa(struct xfrm_state *xs,
425 struct netlink_ext_ack *extack)
427 struct net_device *bond_dev = xs->xso.dev;
428 struct bond_ipsec *ipsec;
429 struct bonding *bond;
437 bond = netdev_priv(bond_dev);
438 slave = rcu_dereference(bond->curr_active_slave);
444 if (!slave->dev->xfrmdev_ops ||
445 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
446 netif_is_bond_master(slave->dev)) {
447 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
452 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
457 xs->xso.real_dev = slave->dev;
459 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
462 INIT_LIST_HEAD(&ipsec->list);
463 spin_lock_bh(&bond->ipsec_lock);
464 list_add(&ipsec->list, &bond->ipsec_list);
465 spin_unlock_bh(&bond->ipsec_lock);
473 static void bond_ipsec_add_sa_all(struct bonding *bond)
475 struct net_device *bond_dev = bond->dev;
476 struct bond_ipsec *ipsec;
480 slave = rcu_dereference(bond->curr_active_slave);
484 if (!slave->dev->xfrmdev_ops ||
485 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
486 netif_is_bond_master(slave->dev)) {
487 spin_lock_bh(&bond->ipsec_lock);
488 if (!list_empty(&bond->ipsec_list))
489 slave_warn(bond_dev, slave->dev,
490 "%s: no slave xdo_dev_state_add\n",
492 spin_unlock_bh(&bond->ipsec_lock);
496 spin_lock_bh(&bond->ipsec_lock);
497 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
498 ipsec->xs->xso.real_dev = slave->dev;
499 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
500 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
501 ipsec->xs->xso.real_dev = NULL;
504 spin_unlock_bh(&bond->ipsec_lock);
510 * bond_ipsec_del_sa - clear out this specific SA
511 * @xs: pointer to transformer state struct
513 static void bond_ipsec_del_sa(struct xfrm_state *xs)
515 struct net_device *bond_dev = xs->xso.dev;
516 struct bond_ipsec *ipsec;
517 struct bonding *bond;
524 bond = netdev_priv(bond_dev);
525 slave = rcu_dereference(bond->curr_active_slave);
530 if (!xs->xso.real_dev)
533 WARN_ON(xs->xso.real_dev != slave->dev);
535 if (!slave->dev->xfrmdev_ops ||
536 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
537 netif_is_bond_master(slave->dev)) {
538 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
542 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
544 spin_lock_bh(&bond->ipsec_lock);
545 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
546 if (ipsec->xs == xs) {
547 list_del(&ipsec->list);
552 spin_unlock_bh(&bond->ipsec_lock);
556 static void bond_ipsec_del_sa_all(struct bonding *bond)
558 struct net_device *bond_dev = bond->dev;
559 struct bond_ipsec *ipsec;
563 slave = rcu_dereference(bond->curr_active_slave);
569 spin_lock_bh(&bond->ipsec_lock);
570 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
571 if (!ipsec->xs->xso.real_dev)
574 if (!slave->dev->xfrmdev_ops ||
575 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
576 netif_is_bond_master(slave->dev)) {
577 slave_warn(bond_dev, slave->dev,
578 "%s: no slave xdo_dev_state_delete\n",
581 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
583 ipsec->xs->xso.real_dev = NULL;
585 spin_unlock_bh(&bond->ipsec_lock);
590 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
591 * @skb: current data packet
592 * @xs: pointer to transformer state struct
594 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
596 struct net_device *bond_dev = xs->xso.dev;
597 struct net_device *real_dev;
598 struct slave *curr_active;
599 struct bonding *bond;
602 bond = netdev_priv(bond_dev);
604 curr_active = rcu_dereference(bond->curr_active_slave);
605 real_dev = curr_active->dev;
607 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
612 if (!xs->xso.real_dev) {
617 if (!real_dev->xfrmdev_ops ||
618 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
619 netif_is_bond_master(real_dev)) {
624 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
630 static const struct xfrmdev_ops bond_xfrmdev_ops = {
631 .xdo_dev_state_add = bond_ipsec_add_sa,
632 .xdo_dev_state_delete = bond_ipsec_del_sa,
633 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
635 #endif /* CONFIG_XFRM_OFFLOAD */
637 /*------------------------------- Link status -------------------------------*/
639 /* Set the carrier state for the master according to the state of its
640 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
641 * do special 802.3ad magic.
643 * Returns zero if carrier state does not change, nonzero if it does.
645 int bond_set_carrier(struct bonding *bond)
647 struct list_head *iter;
650 if (!bond_has_slaves(bond))
653 if (BOND_MODE(bond) == BOND_MODE_8023AD)
654 return bond_3ad_set_carrier(bond);
656 bond_for_each_slave(bond, slave, iter) {
657 if (slave->link == BOND_LINK_UP) {
658 if (!netif_carrier_ok(bond->dev)) {
659 netif_carrier_on(bond->dev);
667 if (netif_carrier_ok(bond->dev)) {
668 netif_carrier_off(bond->dev);
674 /* Get link speed and duplex from the slave's base driver
675 * using ethtool. If for some reason the call fails or the
676 * values are invalid, set speed and duplex to -1,
677 * and return. Return 1 if speed or duplex settings are
678 * UNKNOWN; 0 otherwise.
680 static int bond_update_speed_duplex(struct slave *slave)
682 struct net_device *slave_dev = slave->dev;
683 struct ethtool_link_ksettings ecmd;
686 slave->speed = SPEED_UNKNOWN;
687 slave->duplex = DUPLEX_UNKNOWN;
689 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
692 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
694 switch (ecmd.base.duplex) {
702 slave->speed = ecmd.base.speed;
703 slave->duplex = ecmd.base.duplex;
708 const char *bond_slave_link_status(s8 link)
724 /* if <dev> supports MII link status reporting, check its link status.
726 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
727 * depending upon the setting of the use_carrier parameter.
729 * Return either BMSR_LSTATUS, meaning that the link is up (or we
730 * can't tell and just pretend it is), or 0, meaning that the link is
733 * If reporting is non-zero, instead of faking link up, return -1 if
734 * both ETHTOOL and MII ioctls fail (meaning the device does not
735 * support them). If use_carrier is set, return whatever it says.
736 * It'd be nice if there was a good way to tell if a driver supports
737 * netif_carrier, but there really isn't.
739 static int bond_check_dev_link(struct bonding *bond,
740 struct net_device *slave_dev, int reporting)
742 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
743 int (*ioctl)(struct net_device *, struct ifreq *, int);
745 struct mii_ioctl_data *mii;
747 if (!reporting && !netif_running(slave_dev))
750 if (bond->params.use_carrier)
751 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
753 /* Try to get link status using Ethtool first. */
754 if (slave_dev->ethtool_ops->get_link)
755 return slave_dev->ethtool_ops->get_link(slave_dev) ?
758 /* Ethtool can't be used, fallback to MII ioctls. */
759 ioctl = slave_ops->ndo_eth_ioctl;
761 /* TODO: set pointer to correct ioctl on a per team member
762 * bases to make this more efficient. that is, once
763 * we determine the correct ioctl, we will always
764 * call it and not the others for that team
768 /* We cannot assume that SIOCGMIIPHY will also read a
769 * register; not all network drivers (e.g., e100)
773 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
774 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
776 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
777 mii->reg_num = MII_BMSR;
778 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
779 return mii->val_out & BMSR_LSTATUS;
783 /* If reporting, report that either there's no ndo_eth_ioctl,
784 * or both SIOCGMIIREG and get_link failed (meaning that we
785 * cannot report link status). If not reporting, pretend
788 return reporting ? -1 : BMSR_LSTATUS;
791 /*----------------------------- Multicast list ------------------------------*/
793 /* Push the promiscuity flag down to appropriate slaves */
794 static int bond_set_promiscuity(struct bonding *bond, int inc)
796 struct list_head *iter;
799 if (bond_uses_primary(bond)) {
800 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
803 err = dev_set_promiscuity(curr_active->dev, inc);
807 bond_for_each_slave(bond, slave, iter) {
808 err = dev_set_promiscuity(slave->dev, inc);
816 /* Push the allmulti flag down to all slaves */
817 static int bond_set_allmulti(struct bonding *bond, int inc)
819 struct list_head *iter;
822 if (bond_uses_primary(bond)) {
823 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
826 err = dev_set_allmulti(curr_active->dev, inc);
830 bond_for_each_slave(bond, slave, iter) {
831 err = dev_set_allmulti(slave->dev, inc);
839 /* Retrieve the list of registered multicast addresses for the bonding
840 * device and retransmit an IGMP JOIN request to the current active
843 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
845 struct bonding *bond = container_of(work, struct bonding,
848 if (!rtnl_trylock()) {
849 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
852 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
854 if (bond->igmp_retrans > 1) {
855 bond->igmp_retrans--;
856 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
861 /* Flush bond's hardware addresses from slave */
862 static void bond_hw_addr_flush(struct net_device *bond_dev,
863 struct net_device *slave_dev)
865 struct bonding *bond = netdev_priv(bond_dev);
867 dev_uc_unsync(slave_dev, bond_dev);
868 dev_mc_unsync(slave_dev, bond_dev);
870 if (BOND_MODE(bond) == BOND_MODE_8023AD)
871 dev_mc_del(slave_dev, lacpdu_mcast_addr);
874 /*--------------------------- Active slave change ---------------------------*/
876 /* Update the hardware address list and promisc/allmulti for the new and
877 * old active slaves (if any). Modes that are not using primary keep all
878 * slaves up date at all times; only the modes that use primary need to call
879 * this function to swap these settings during a failover.
881 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
882 struct slave *old_active)
885 if (bond->dev->flags & IFF_PROMISC)
886 dev_set_promiscuity(old_active->dev, -1);
888 if (bond->dev->flags & IFF_ALLMULTI)
889 dev_set_allmulti(old_active->dev, -1);
891 if (bond->dev->flags & IFF_UP)
892 bond_hw_addr_flush(bond->dev, old_active->dev);
896 /* FIXME: Signal errors upstream. */
897 if (bond->dev->flags & IFF_PROMISC)
898 dev_set_promiscuity(new_active->dev, 1);
900 if (bond->dev->flags & IFF_ALLMULTI)
901 dev_set_allmulti(new_active->dev, 1);
903 if (bond->dev->flags & IFF_UP) {
904 netif_addr_lock_bh(bond->dev);
905 dev_uc_sync(new_active->dev, bond->dev);
906 dev_mc_sync(new_active->dev, bond->dev);
907 netif_addr_unlock_bh(bond->dev);
913 * bond_set_dev_addr - clone slave's address to bond
914 * @bond_dev: bond net device
915 * @slave_dev: slave net device
917 * Should be called with RTNL held.
919 static int bond_set_dev_addr(struct net_device *bond_dev,
920 struct net_device *slave_dev)
924 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
925 bond_dev, slave_dev, slave_dev->addr_len);
926 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
930 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
931 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
932 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
936 static struct slave *bond_get_old_active(struct bonding *bond,
937 struct slave *new_active)
940 struct list_head *iter;
942 bond_for_each_slave(bond, slave, iter) {
943 if (slave == new_active)
946 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
953 /* bond_do_fail_over_mac
955 * Perform special MAC address swapping for fail_over_mac settings
959 static void bond_do_fail_over_mac(struct bonding *bond,
960 struct slave *new_active,
961 struct slave *old_active)
963 u8 tmp_mac[MAX_ADDR_LEN];
964 struct sockaddr_storage ss;
967 switch (bond->params.fail_over_mac) {
968 case BOND_FOM_ACTIVE:
970 rv = bond_set_dev_addr(bond->dev, new_active->dev);
972 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
976 case BOND_FOM_FOLLOW:
977 /* if new_active && old_active, swap them
978 * if just old_active, do nothing (going to no active slave)
979 * if just new_active, set new_active to bond's MAC
985 old_active = bond_get_old_active(bond, new_active);
988 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
989 new_active->dev->addr_len);
990 bond_hw_addr_copy(ss.__data,
991 old_active->dev->dev_addr,
992 old_active->dev->addr_len);
993 ss.ss_family = new_active->dev->type;
995 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
996 bond->dev->addr_len);
997 ss.ss_family = bond->dev->type;
1000 rv = dev_set_mac_address(new_active->dev,
1001 (struct sockaddr *)&ss, NULL);
1003 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1011 bond_hw_addr_copy(ss.__data, tmp_mac,
1012 new_active->dev->addr_len);
1013 ss.ss_family = old_active->dev->type;
1015 rv = dev_set_mac_address(old_active->dev,
1016 (struct sockaddr *)&ss, NULL);
1018 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1023 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1024 bond->params.fail_over_mac);
1031 * bond_choose_primary_or_current - select the primary or high priority slave
1032 * @bond: our bonding struct
1034 * - Check if there is a primary link. If the primary link was set and is up,
1035 * go on and do link reselection.
1037 * - If primary link is not set or down, find the highest priority link.
1038 * If the highest priority link is not current slave, set it as primary
1039 * link and do link reselection.
1041 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1043 struct slave *prim = rtnl_dereference(bond->primary_slave);
1044 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1045 struct slave *slave, *hprio = NULL;
1046 struct list_head *iter;
1048 if (!prim || prim->link != BOND_LINK_UP) {
1049 bond_for_each_slave(bond, slave, iter) {
1050 if (slave->link == BOND_LINK_UP) {
1051 hprio = hprio ?: slave;
1052 if (slave->prio > hprio->prio)
1057 if (hprio && hprio != curr) {
1062 if (!curr || curr->link != BOND_LINK_UP)
1067 if (bond->force_primary) {
1068 bond->force_primary = false;
1073 if (!curr || curr->link != BOND_LINK_UP)
1076 /* At this point, prim and curr are both up */
1077 switch (bond->params.primary_reselect) {
1078 case BOND_PRI_RESELECT_ALWAYS:
1080 case BOND_PRI_RESELECT_BETTER:
1081 if (prim->speed < curr->speed)
1083 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1086 case BOND_PRI_RESELECT_FAILURE:
1089 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1090 bond->params.primary_reselect);
1096 * bond_find_best_slave - select the best available slave to be the active one
1097 * @bond: our bonding struct
1099 static struct slave *bond_find_best_slave(struct bonding *bond)
1101 struct slave *slave, *bestslave = NULL;
1102 struct list_head *iter;
1103 int mintime = bond->params.updelay;
1105 slave = bond_choose_primary_or_current(bond);
1109 bond_for_each_slave(bond, slave, iter) {
1110 if (slave->link == BOND_LINK_UP)
1112 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1113 slave->delay < mintime) {
1114 mintime = slave->delay;
1122 static bool bond_should_notify_peers(struct bonding *bond)
1124 struct slave *slave;
1127 slave = rcu_dereference(bond->curr_active_slave);
1130 if (!slave || !bond->send_peer_notif ||
1131 bond->send_peer_notif %
1132 max(1, bond->params.peer_notif_delay) != 0 ||
1133 !netif_carrier_ok(bond->dev) ||
1134 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1137 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1138 slave ? slave->dev->name : "NULL");
1144 * bond_change_active_slave - change the active slave into the specified one
1145 * @bond: our bonding struct
1146 * @new_active: the new slave to make the active one
1148 * Set the new slave to the bond's settings and unset them on the old
1149 * curr_active_slave.
1150 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1152 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1153 * because it is apparently the best available slave we have, even though its
1154 * updelay hasn't timed out yet.
1156 * Caller must hold RTNL.
1158 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1160 struct slave *old_active;
1164 old_active = rtnl_dereference(bond->curr_active_slave);
1166 if (old_active == new_active)
1169 #ifdef CONFIG_XFRM_OFFLOAD
1170 bond_ipsec_del_sa_all(bond);
1171 #endif /* CONFIG_XFRM_OFFLOAD */
1174 new_active->last_link_up = jiffies;
1176 if (new_active->link == BOND_LINK_BACK) {
1177 if (bond_uses_primary(bond)) {
1178 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1179 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1182 new_active->delay = 0;
1183 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1184 BOND_SLAVE_NOTIFY_NOW);
1186 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1187 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1189 if (bond_is_lb(bond))
1190 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1192 if (bond_uses_primary(bond))
1193 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1197 if (bond_uses_primary(bond))
1198 bond_hw_addr_swap(bond, new_active, old_active);
1200 if (bond_is_lb(bond)) {
1201 bond_alb_handle_active_change(bond, new_active);
1203 bond_set_slave_inactive_flags(old_active,
1204 BOND_SLAVE_NOTIFY_NOW);
1206 bond_set_slave_active_flags(new_active,
1207 BOND_SLAVE_NOTIFY_NOW);
1209 rcu_assign_pointer(bond->curr_active_slave, new_active);
1212 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1214 bond_set_slave_inactive_flags(old_active,
1215 BOND_SLAVE_NOTIFY_NOW);
1218 bool should_notify_peers = false;
1220 bond_set_slave_active_flags(new_active,
1221 BOND_SLAVE_NOTIFY_NOW);
1223 if (bond->params.fail_over_mac)
1224 bond_do_fail_over_mac(bond, new_active,
1227 if (netif_running(bond->dev)) {
1228 bond->send_peer_notif =
1229 bond->params.num_peer_notif *
1230 max(1, bond->params.peer_notif_delay);
1231 should_notify_peers =
1232 bond_should_notify_peers(bond);
1235 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1236 if (should_notify_peers) {
1237 bond->send_peer_notif--;
1238 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1244 #ifdef CONFIG_XFRM_OFFLOAD
1245 bond_ipsec_add_sa_all(bond);
1246 #endif /* CONFIG_XFRM_OFFLOAD */
1248 /* resend IGMP joins since active slave has changed or
1249 * all were sent on curr_active_slave.
1250 * resend only if bond is brought up with the affected
1251 * bonding modes and the retransmission is enabled
1253 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1254 ((bond_uses_primary(bond) && new_active) ||
1255 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1256 bond->igmp_retrans = bond->params.resend_igmp;
1257 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1262 * bond_select_active_slave - select a new active slave, if needed
1263 * @bond: our bonding struct
1265 * This functions should be called when one of the following occurs:
1266 * - The old curr_active_slave has been released or lost its link.
1267 * - The primary_slave has got its link back.
1268 * - A slave has got its link back and there's no old curr_active_slave.
1270 * Caller must hold RTNL.
1272 void bond_select_active_slave(struct bonding *bond)
1274 struct slave *best_slave;
1279 best_slave = bond_find_best_slave(bond);
1280 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1281 bond_change_active_slave(bond, best_slave);
1282 rv = bond_set_carrier(bond);
1286 if (netif_carrier_ok(bond->dev))
1287 netdev_info(bond->dev, "active interface up!\n");
1289 netdev_info(bond->dev, "now running without any active interface!\n");
1293 #ifdef CONFIG_NET_POLL_CONTROLLER
1294 static inline int slave_enable_netpoll(struct slave *slave)
1299 np = kzalloc(sizeof(*np), GFP_KERNEL);
1304 err = __netpoll_setup(np, slave->dev);
1313 static inline void slave_disable_netpoll(struct slave *slave)
1315 struct netpoll *np = slave->np;
1325 static void bond_poll_controller(struct net_device *bond_dev)
1327 struct bonding *bond = netdev_priv(bond_dev);
1328 struct slave *slave = NULL;
1329 struct list_head *iter;
1330 struct ad_info ad_info;
1332 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1333 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1336 bond_for_each_slave_rcu(bond, slave, iter) {
1337 if (!bond_slave_is_up(slave))
1340 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1341 struct aggregator *agg =
1342 SLAVE_AD_INFO(slave)->port.aggregator;
1345 agg->aggregator_identifier != ad_info.aggregator_id)
1349 netpoll_poll_dev(slave->dev);
1353 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1355 struct bonding *bond = netdev_priv(bond_dev);
1356 struct list_head *iter;
1357 struct slave *slave;
1359 bond_for_each_slave(bond, slave, iter)
1360 if (bond_slave_is_up(slave))
1361 slave_disable_netpoll(slave);
1364 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1366 struct bonding *bond = netdev_priv(dev);
1367 struct list_head *iter;
1368 struct slave *slave;
1371 bond_for_each_slave(bond, slave, iter) {
1372 err = slave_enable_netpoll(slave);
1374 bond_netpoll_cleanup(dev);
1381 static inline int slave_enable_netpoll(struct slave *slave)
1385 static inline void slave_disable_netpoll(struct slave *slave)
1388 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1393 /*---------------------------------- IOCTL ----------------------------------*/
1395 static netdev_features_t bond_fix_features(struct net_device *dev,
1396 netdev_features_t features)
1398 struct bonding *bond = netdev_priv(dev);
1399 struct list_head *iter;
1400 netdev_features_t mask;
1401 struct slave *slave;
1405 features &= ~NETIF_F_ONE_FOR_ALL;
1406 features |= NETIF_F_ALL_FOR_ALL;
1408 bond_for_each_slave(bond, slave, iter) {
1409 features = netdev_increment_features(features,
1410 slave->dev->features,
1413 features = netdev_add_tso_features(features, mask);
1418 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1419 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1420 NETIF_F_HIGHDMA | NETIF_F_LRO)
1422 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1423 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1425 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1426 NETIF_F_GSO_SOFTWARE)
1429 static void bond_compute_features(struct bonding *bond)
1431 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1432 IFF_XMIT_DST_RELEASE_PERM;
1433 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1434 netdev_features_t enc_features = BOND_ENC_FEATURES;
1435 #ifdef CONFIG_XFRM_OFFLOAD
1436 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1437 #endif /* CONFIG_XFRM_OFFLOAD */
1438 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1439 struct net_device *bond_dev = bond->dev;
1440 struct list_head *iter;
1441 struct slave *slave;
1442 unsigned short max_hard_header_len = ETH_HLEN;
1443 unsigned int tso_max_size = TSO_MAX_SIZE;
1444 u16 tso_max_segs = TSO_MAX_SEGS;
1446 if (!bond_has_slaves(bond))
1448 vlan_features &= NETIF_F_ALL_FOR_ALL;
1449 mpls_features &= NETIF_F_ALL_FOR_ALL;
1451 bond_for_each_slave(bond, slave, iter) {
1452 vlan_features = netdev_increment_features(vlan_features,
1453 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1455 enc_features = netdev_increment_features(enc_features,
1456 slave->dev->hw_enc_features,
1459 #ifdef CONFIG_XFRM_OFFLOAD
1460 xfrm_features = netdev_increment_features(xfrm_features,
1461 slave->dev->hw_enc_features,
1462 BOND_XFRM_FEATURES);
1463 #endif /* CONFIG_XFRM_OFFLOAD */
1465 mpls_features = netdev_increment_features(mpls_features,
1466 slave->dev->mpls_features,
1467 BOND_MPLS_FEATURES);
1469 dst_release_flag &= slave->dev->priv_flags;
1470 if (slave->dev->hard_header_len > max_hard_header_len)
1471 max_hard_header_len = slave->dev->hard_header_len;
1473 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1474 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1476 bond_dev->hard_header_len = max_hard_header_len;
1479 bond_dev->vlan_features = vlan_features;
1480 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1481 NETIF_F_HW_VLAN_CTAG_TX |
1482 NETIF_F_HW_VLAN_STAG_TX;
1483 #ifdef CONFIG_XFRM_OFFLOAD
1484 bond_dev->hw_enc_features |= xfrm_features;
1485 #endif /* CONFIG_XFRM_OFFLOAD */
1486 bond_dev->mpls_features = mpls_features;
1487 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1488 netif_set_tso_max_size(bond_dev, tso_max_size);
1490 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1491 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1492 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1493 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1495 netdev_change_features(bond_dev);
1498 static void bond_setup_by_slave(struct net_device *bond_dev,
1499 struct net_device *slave_dev)
1501 bond_dev->header_ops = slave_dev->header_ops;
1503 bond_dev->type = slave_dev->type;
1504 bond_dev->hard_header_len = slave_dev->hard_header_len;
1505 bond_dev->needed_headroom = slave_dev->needed_headroom;
1506 bond_dev->addr_len = slave_dev->addr_len;
1508 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1509 slave_dev->addr_len);
1512 /* On bonding slaves other than the currently active slave, suppress
1513 * duplicates except for alb non-mcast/bcast.
1515 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1516 struct slave *slave,
1517 struct bonding *bond)
1519 if (bond_is_slave_inactive(slave)) {
1520 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1521 skb->pkt_type != PACKET_BROADCAST &&
1522 skb->pkt_type != PACKET_MULTICAST)
1529 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1531 struct sk_buff *skb = *pskb;
1532 struct slave *slave;
1533 struct bonding *bond;
1534 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1536 int ret = RX_HANDLER_ANOTHER;
1538 skb = skb_share_check(skb, GFP_ATOMIC);
1540 return RX_HANDLER_CONSUMED;
1544 slave = bond_slave_get_rcu(skb->dev);
1547 recv_probe = READ_ONCE(bond->recv_probe);
1549 ret = recv_probe(skb, bond, slave);
1550 if (ret == RX_HANDLER_CONSUMED) {
1557 * For packets determined by bond_should_deliver_exact_match() call to
1558 * be suppressed we want to make an exception for link-local packets.
1559 * This is necessary for e.g. LLDP daemons to be able to monitor
1560 * inactive slave links without being forced to bind to them
1563 * At the same time, packets that are passed to the bonding master
1564 * (including link-local ones) can have their originating interface
1565 * determined via PACKET_ORIGDEV socket option.
1567 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1568 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1569 return RX_HANDLER_PASS;
1570 return RX_HANDLER_EXACT;
1573 skb->dev = bond->dev;
1575 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1576 netif_is_bridge_port(bond->dev) &&
1577 skb->pkt_type == PACKET_HOST) {
1579 if (unlikely(skb_cow_head(skb,
1580 skb->data - skb_mac_header(skb)))) {
1582 return RX_HANDLER_CONSUMED;
1584 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1585 bond->dev->addr_len);
1591 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1593 switch (BOND_MODE(bond)) {
1594 case BOND_MODE_ROUNDROBIN:
1595 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1596 case BOND_MODE_ACTIVEBACKUP:
1597 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1598 case BOND_MODE_BROADCAST:
1599 return NETDEV_LAG_TX_TYPE_BROADCAST;
1601 case BOND_MODE_8023AD:
1602 return NETDEV_LAG_TX_TYPE_HASH;
1604 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1608 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1609 enum netdev_lag_tx_type type)
1611 if (type != NETDEV_LAG_TX_TYPE_HASH)
1612 return NETDEV_LAG_HASH_NONE;
1614 switch (bond->params.xmit_policy) {
1615 case BOND_XMIT_POLICY_LAYER2:
1616 return NETDEV_LAG_HASH_L2;
1617 case BOND_XMIT_POLICY_LAYER34:
1618 return NETDEV_LAG_HASH_L34;
1619 case BOND_XMIT_POLICY_LAYER23:
1620 return NETDEV_LAG_HASH_L23;
1621 case BOND_XMIT_POLICY_ENCAP23:
1622 return NETDEV_LAG_HASH_E23;
1623 case BOND_XMIT_POLICY_ENCAP34:
1624 return NETDEV_LAG_HASH_E34;
1625 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1626 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1628 return NETDEV_LAG_HASH_UNKNOWN;
1632 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1633 struct netlink_ext_ack *extack)
1635 struct netdev_lag_upper_info lag_upper_info;
1636 enum netdev_lag_tx_type type;
1639 type = bond_lag_tx_type(bond);
1640 lag_upper_info.tx_type = type;
1641 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1643 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1644 &lag_upper_info, extack);
1648 slave->dev->flags |= IFF_SLAVE;
1652 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1654 netdev_upper_dev_unlink(slave->dev, bond->dev);
1655 slave->dev->flags &= ~IFF_SLAVE;
1658 static void slave_kobj_release(struct kobject *kobj)
1660 struct slave *slave = to_slave(kobj);
1661 struct bonding *bond = bond_get_bond_by_slave(slave);
1663 cancel_delayed_work_sync(&slave->notify_work);
1664 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1665 kfree(SLAVE_AD_INFO(slave));
1670 static struct kobj_type slave_ktype = {
1671 .release = slave_kobj_release,
1673 .sysfs_ops = &slave_sysfs_ops,
1677 static int bond_kobj_init(struct slave *slave)
1681 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1682 &(slave->dev->dev.kobj), "bonding_slave");
1684 kobject_put(&slave->kobj);
1689 static struct slave *bond_alloc_slave(struct bonding *bond,
1690 struct net_device *slave_dev)
1692 struct slave *slave = NULL;
1694 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1699 slave->dev = slave_dev;
1700 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1702 if (bond_kobj_init(slave))
1705 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1706 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1708 if (!SLAVE_AD_INFO(slave)) {
1709 kobject_put(&slave->kobj);
1717 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1719 info->bond_mode = BOND_MODE(bond);
1720 info->miimon = bond->params.miimon;
1721 info->num_slaves = bond->slave_cnt;
1724 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1726 strcpy(info->slave_name, slave->dev->name);
1727 info->link = slave->link;
1728 info->state = bond_slave_state(slave);
1729 info->link_failure_count = slave->link_failure_count;
1732 static void bond_netdev_notify_work(struct work_struct *_work)
1734 struct slave *slave = container_of(_work, struct slave,
1737 if (rtnl_trylock()) {
1738 struct netdev_bonding_info binfo;
1740 bond_fill_ifslave(slave, &binfo.slave);
1741 bond_fill_ifbond(slave->bond, &binfo.master);
1742 netdev_bonding_info_change(slave->dev, &binfo);
1745 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1749 void bond_queue_slave_event(struct slave *slave)
1751 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1754 void bond_lower_state_changed(struct slave *slave)
1756 struct netdev_lag_lower_state_info info;
1758 info.link_up = slave->link == BOND_LINK_UP ||
1759 slave->link == BOND_LINK_FAIL;
1760 info.tx_enabled = bond_is_active_slave(slave);
1761 netdev_lower_state_changed(slave->dev, &info);
1764 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1766 NL_SET_ERR_MSG(extack, errmsg); \
1768 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1771 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1773 NL_SET_ERR_MSG(extack, errmsg); \
1775 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1778 /* The bonding driver uses ether_setup() to convert a master bond device
1779 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1780 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1783 static void bond_ether_setup(struct net_device *bond_dev)
1785 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1787 ether_setup(bond_dev);
1788 bond_dev->flags |= IFF_MASTER | flags;
1789 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1792 void bond_xdp_set_features(struct net_device *bond_dev)
1794 struct bonding *bond = netdev_priv(bond_dev);
1795 xdp_features_t val = NETDEV_XDP_ACT_MASK;
1796 struct list_head *iter;
1797 struct slave *slave;
1801 if (!bond_xdp_check(bond)) {
1802 xdp_clear_features_flag(bond_dev);
1806 bond_for_each_slave(bond, slave, iter)
1807 val &= slave->dev->xdp_features;
1809 xdp_set_features_flag(bond_dev, val);
1812 /* enslave device <slave> to bond device <master> */
1813 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1814 struct netlink_ext_ack *extack)
1816 struct bonding *bond = netdev_priv(bond_dev);
1817 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1818 struct slave *new_slave = NULL, *prev_slave;
1819 struct sockaddr_storage ss;
1823 if (slave_dev->flags & IFF_MASTER &&
1824 !netif_is_bond_master(slave_dev)) {
1825 BOND_NL_ERR(bond_dev, extack,
1826 "Device type (master device) cannot be enslaved");
1830 if (!bond->params.use_carrier &&
1831 slave_dev->ethtool_ops->get_link == NULL &&
1832 slave_ops->ndo_eth_ioctl == NULL) {
1833 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1836 /* already in-use? */
1837 if (netdev_is_rx_handler_busy(slave_dev)) {
1838 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1839 "Device is in use and cannot be enslaved");
1843 if (bond_dev == slave_dev) {
1844 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1848 /* vlan challenged mutual exclusion */
1849 /* no need to lock since we're protected by rtnl_lock */
1850 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1851 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1852 if (vlan_uses_dev(bond_dev)) {
1853 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1854 "Can not enslave VLAN challenged device to VLAN enabled bond");
1857 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1860 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1863 if (slave_dev->features & NETIF_F_HW_ESP)
1864 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1866 /* Old ifenslave binaries are no longer supported. These can
1867 * be identified with moderate accuracy by the state of the slave:
1868 * the current ifenslave will set the interface down prior to
1869 * enslaving it; the old ifenslave will not.
1871 if (slave_dev->flags & IFF_UP) {
1872 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1873 "Device can not be enslaved while up");
1877 /* set bonding device ether type by slave - bonding netdevices are
1878 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1879 * there is a need to override some of the type dependent attribs/funcs.
1881 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1882 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1884 if (!bond_has_slaves(bond)) {
1885 if (bond_dev->type != slave_dev->type) {
1886 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1887 bond_dev->type, slave_dev->type);
1889 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1891 res = notifier_to_errno(res);
1893 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1897 /* Flush unicast and multicast addresses */
1898 dev_uc_flush(bond_dev);
1899 dev_mc_flush(bond_dev);
1901 if (slave_dev->type != ARPHRD_ETHER)
1902 bond_setup_by_slave(bond_dev, slave_dev);
1904 bond_ether_setup(bond_dev);
1906 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1909 } else if (bond_dev->type != slave_dev->type) {
1910 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1911 "Device type is different from other slaves");
1915 if (slave_dev->type == ARPHRD_INFINIBAND &&
1916 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1917 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1918 "Only active-backup mode is supported for infiniband slaves");
1920 goto err_undo_flags;
1923 if (!slave_ops->ndo_set_mac_address ||
1924 slave_dev->type == ARPHRD_INFINIBAND) {
1925 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1926 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1927 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1928 if (!bond_has_slaves(bond)) {
1929 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1930 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1932 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1933 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1935 goto err_undo_flags;
1940 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1942 /* If this is the first slave, then we need to set the master's hardware
1943 * address to be the same as the slave's.
1945 if (!bond_has_slaves(bond) &&
1946 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1947 res = bond_set_dev_addr(bond->dev, slave_dev);
1949 goto err_undo_flags;
1952 new_slave = bond_alloc_slave(bond, slave_dev);
1955 goto err_undo_flags;
1958 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1959 * is set via sysfs or module option if desired.
1961 new_slave->queue_id = 0;
1963 /* Save slave's original mtu and then set it to match the bond */
1964 new_slave->original_mtu = slave_dev->mtu;
1965 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1967 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1971 /* Save slave's original ("permanent") mac address for modes
1972 * that need it, and for restoring it upon release, and then
1973 * set it to the master's address
1975 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1976 slave_dev->addr_len);
1978 if (!bond->params.fail_over_mac ||
1979 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1980 /* Set slave to master's mac address. The application already
1981 * set the master's mac address to that of the first slave
1983 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1984 ss.ss_family = slave_dev->type;
1985 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1988 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1989 goto err_restore_mtu;
1993 /* set no_addrconf flag before open to prevent IPv6 addrconf */
1994 slave_dev->priv_flags |= IFF_NO_ADDRCONF;
1996 /* open the slave since the application closed it */
1997 res = dev_open(slave_dev, extack);
1999 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2000 goto err_restore_mac;
2003 slave_dev->priv_flags |= IFF_BONDING;
2004 /* initialize slave stats */
2005 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2007 if (bond_is_lb(bond)) {
2008 /* bond_alb_init_slave() must be called before all other stages since
2009 * it might fail and we do not want to have to undo everything
2011 res = bond_alb_init_slave(bond, new_slave);
2016 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2018 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2022 prev_slave = bond_last_slave(bond);
2024 new_slave->delay = 0;
2025 new_slave->link_failure_count = 0;
2027 if (bond_update_speed_duplex(new_slave) &&
2028 bond_needs_speed_duplex(bond))
2029 new_slave->link = BOND_LINK_DOWN;
2031 new_slave->last_rx = jiffies -
2032 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2033 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2034 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2036 new_slave->last_tx = new_slave->last_rx;
2038 if (bond->params.miimon && !bond->params.use_carrier) {
2039 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2041 if ((link_reporting == -1) && !bond->params.arp_interval) {
2042 /* miimon is set but a bonded network driver
2043 * does not support ETHTOOL/MII and
2044 * arp_interval is not set. Note: if
2045 * use_carrier is enabled, we will never go
2046 * here (because netif_carrier is always
2047 * supported); thus, we don't need to change
2048 * the messages for netif_carrier.
2050 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");
2051 } else if (link_reporting == -1) {
2052 /* unable get link status using mii/ethtool */
2053 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");
2057 /* check for initial state */
2058 new_slave->link = BOND_LINK_NOCHANGE;
2059 if (bond->params.miimon) {
2060 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2061 if (bond->params.updelay) {
2062 bond_set_slave_link_state(new_slave,
2064 BOND_SLAVE_NOTIFY_NOW);
2065 new_slave->delay = bond->params.updelay;
2067 bond_set_slave_link_state(new_slave,
2069 BOND_SLAVE_NOTIFY_NOW);
2072 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2073 BOND_SLAVE_NOTIFY_NOW);
2075 } else if (bond->params.arp_interval) {
2076 bond_set_slave_link_state(new_slave,
2077 (netif_carrier_ok(slave_dev) ?
2078 BOND_LINK_UP : BOND_LINK_DOWN),
2079 BOND_SLAVE_NOTIFY_NOW);
2081 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2082 BOND_SLAVE_NOTIFY_NOW);
2085 if (new_slave->link != BOND_LINK_DOWN)
2086 new_slave->last_link_up = jiffies;
2087 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2088 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2089 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2091 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2092 /* if there is a primary slave, remember it */
2093 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2094 rcu_assign_pointer(bond->primary_slave, new_slave);
2095 bond->force_primary = true;
2099 switch (BOND_MODE(bond)) {
2100 case BOND_MODE_ACTIVEBACKUP:
2101 bond_set_slave_inactive_flags(new_slave,
2102 BOND_SLAVE_NOTIFY_NOW);
2104 case BOND_MODE_8023AD:
2105 /* in 802.3ad mode, the internal mechanism
2106 * will activate the slaves in the selected
2109 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2110 /* if this is the first slave */
2112 SLAVE_AD_INFO(new_slave)->id = 1;
2113 /* Initialize AD with the number of times that the AD timer is called in 1 second
2114 * can be called only after the mac address of the bond is set
2116 bond_3ad_initialize(bond);
2118 SLAVE_AD_INFO(new_slave)->id =
2119 SLAVE_AD_INFO(prev_slave)->id + 1;
2122 bond_3ad_bind_slave(new_slave);
2126 bond_set_active_slave(new_slave);
2127 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2130 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2132 /* always active in trunk mode */
2133 bond_set_active_slave(new_slave);
2135 /* In trunking mode there is little meaning to curr_active_slave
2136 * anyway (it holds no special properties of the bond device),
2137 * so we can change it without calling change_active_interface()
2139 if (!rcu_access_pointer(bond->curr_active_slave) &&
2140 new_slave->link == BOND_LINK_UP)
2141 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2144 } /* switch(bond_mode) */
2146 #ifdef CONFIG_NET_POLL_CONTROLLER
2147 if (bond->dev->npinfo) {
2148 if (slave_enable_netpoll(new_slave)) {
2149 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2156 if (!(bond_dev->features & NETIF_F_LRO))
2157 dev_disable_lro(slave_dev);
2159 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2162 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2166 res = bond_master_upper_dev_link(bond, new_slave, extack);
2168 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2169 goto err_unregister;
2172 bond_lower_state_changed(new_slave);
2174 res = bond_sysfs_slave_add(new_slave);
2176 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2177 goto err_upper_unlink;
2180 /* If the mode uses primary, then the following is handled by
2181 * bond_change_active_slave().
2183 if (!bond_uses_primary(bond)) {
2184 /* set promiscuity level to new slave */
2185 if (bond_dev->flags & IFF_PROMISC) {
2186 res = dev_set_promiscuity(slave_dev, 1);
2191 /* set allmulti level to new slave */
2192 if (bond_dev->flags & IFF_ALLMULTI) {
2193 res = dev_set_allmulti(slave_dev, 1);
2195 if (bond_dev->flags & IFF_PROMISC)
2196 dev_set_promiscuity(slave_dev, -1);
2201 if (bond_dev->flags & IFF_UP) {
2202 netif_addr_lock_bh(bond_dev);
2203 dev_mc_sync_multiple(slave_dev, bond_dev);
2204 dev_uc_sync_multiple(slave_dev, bond_dev);
2205 netif_addr_unlock_bh(bond_dev);
2207 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2208 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2213 bond_compute_features(bond);
2214 bond_set_carrier(bond);
2216 if (bond_uses_primary(bond)) {
2218 bond_select_active_slave(bond);
2219 unblock_netpoll_tx();
2222 if (bond_mode_can_use_xmit_hash(bond))
2223 bond_update_slave_arr(bond, NULL);
2226 if (!slave_dev->netdev_ops->ndo_bpf ||
2227 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2228 if (bond->xdp_prog) {
2229 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2230 "Slave does not support XDP");
2234 } else if (bond->xdp_prog) {
2235 struct netdev_bpf xdp = {
2236 .command = XDP_SETUP_PROG,
2238 .prog = bond->xdp_prog,
2242 if (dev_xdp_prog_count(slave_dev) > 0) {
2243 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2244 "Slave has XDP program loaded, please unload before enslaving");
2249 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2251 /* ndo_bpf() sets extack error message */
2252 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2256 bpf_prog_inc(bond->xdp_prog);
2259 bond_xdp_set_features(bond_dev);
2261 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2262 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2263 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2265 /* enslave is successful */
2266 bond_queue_slave_event(new_slave);
2269 /* Undo stages on error */
2271 bond_sysfs_slave_del(new_slave);
2274 bond_upper_dev_unlink(bond, new_slave);
2277 netdev_rx_handler_unregister(slave_dev);
2280 vlan_vids_del_by_dev(slave_dev, bond_dev);
2281 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2282 RCU_INIT_POINTER(bond->primary_slave, NULL);
2283 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2285 bond_change_active_slave(bond, NULL);
2286 bond_select_active_slave(bond);
2287 unblock_netpoll_tx();
2289 /* either primary_slave or curr_active_slave might've changed */
2291 slave_disable_netpoll(new_slave);
2294 if (!netif_is_bond_master(slave_dev))
2295 slave_dev->priv_flags &= ~IFF_BONDING;
2296 dev_close(slave_dev);
2299 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2300 if (!bond->params.fail_over_mac ||
2301 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2302 /* XXX TODO - fom follow mode needs to change master's
2303 * MAC if this slave's MAC is in use by the bond, or at
2304 * least print a warning.
2306 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2307 new_slave->dev->addr_len);
2308 ss.ss_family = slave_dev->type;
2309 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2313 dev_set_mtu(slave_dev, new_slave->original_mtu);
2316 kobject_put(&new_slave->kobj);
2319 /* Enslave of first slave has failed and we need to fix master's mac */
2320 if (!bond_has_slaves(bond)) {
2321 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2322 slave_dev->dev_addr))
2323 eth_hw_addr_random(bond_dev);
2324 if (bond_dev->type != ARPHRD_ETHER) {
2325 dev_close(bond_dev);
2326 bond_ether_setup(bond_dev);
2333 /* Try to release the slave device <slave> from the bond device <master>
2334 * It is legal to access curr_active_slave without a lock because all the function
2335 * is RTNL-locked. If "all" is true it means that the function is being called
2336 * while destroying a bond interface and all slaves are being released.
2338 * The rules for slave state should be:
2339 * for Active/Backup:
2340 * Active stays on all backups go down
2341 * for Bonded connections:
2342 * The first up interface should be left on and all others downed.
2344 static int __bond_release_one(struct net_device *bond_dev,
2345 struct net_device *slave_dev,
2346 bool all, bool unregister)
2348 struct bonding *bond = netdev_priv(bond_dev);
2349 struct slave *slave, *oldcurrent;
2350 struct sockaddr_storage ss;
2351 int old_flags = bond_dev->flags;
2352 netdev_features_t old_features = bond_dev->features;
2354 /* slave is not a slave or master is not master of this slave */
2355 if (!(slave_dev->flags & IFF_SLAVE) ||
2356 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2357 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2363 slave = bond_get_slave_by_dev(bond, slave_dev);
2365 /* not a slave of this bond */
2366 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2367 unblock_netpoll_tx();
2371 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2373 bond_sysfs_slave_del(slave);
2375 /* recompute stats just before removing the slave */
2376 bond_get_stats(bond->dev, &bond->bond_stats);
2378 if (bond->xdp_prog) {
2379 struct netdev_bpf xdp = {
2380 .command = XDP_SETUP_PROG,
2385 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2386 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2389 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2390 * for this slave anymore.
2392 netdev_rx_handler_unregister(slave_dev);
2394 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2395 bond_3ad_unbind_slave(slave);
2397 bond_upper_dev_unlink(bond, slave);
2399 if (bond_mode_can_use_xmit_hash(bond))
2400 bond_update_slave_arr(bond, slave);
2402 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2403 bond_is_active_slave(slave) ? "active" : "backup");
2405 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2407 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2409 if (!all && (!bond->params.fail_over_mac ||
2410 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2411 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2412 bond_has_slaves(bond))
2413 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",
2414 slave->perm_hwaddr);
2417 if (rtnl_dereference(bond->primary_slave) == slave)
2418 RCU_INIT_POINTER(bond->primary_slave, NULL);
2420 if (oldcurrent == slave)
2421 bond_change_active_slave(bond, NULL);
2423 if (bond_is_lb(bond)) {
2424 /* Must be called only after the slave has been
2425 * detached from the list and the curr_active_slave
2426 * has been cleared (if our_slave == old_current),
2427 * but before a new active slave is selected.
2429 bond_alb_deinit_slave(bond, slave);
2433 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2434 } else if (oldcurrent == slave) {
2435 /* Note that we hold RTNL over this sequence, so there
2436 * is no concern that another slave add/remove event
2439 bond_select_active_slave(bond);
2442 bond_set_carrier(bond);
2443 if (!bond_has_slaves(bond))
2444 eth_hw_addr_random(bond_dev);
2446 unblock_netpoll_tx();
2450 if (!bond_has_slaves(bond)) {
2451 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2452 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2455 bond_compute_features(bond);
2456 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2457 (old_features & NETIF_F_VLAN_CHALLENGED))
2458 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2460 vlan_vids_del_by_dev(slave_dev, bond_dev);
2462 /* If the mode uses primary, then this case was handled above by
2463 * bond_change_active_slave(..., NULL)
2465 if (!bond_uses_primary(bond)) {
2466 /* unset promiscuity level from slave
2467 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2468 * of the IFF_PROMISC flag in the bond_dev, but we need the
2469 * value of that flag before that change, as that was the value
2470 * when this slave was attached, so we cache at the start of the
2471 * function and use it here. Same goes for ALLMULTI below
2473 if (old_flags & IFF_PROMISC)
2474 dev_set_promiscuity(slave_dev, -1);
2476 /* unset allmulti level from slave */
2477 if (old_flags & IFF_ALLMULTI)
2478 dev_set_allmulti(slave_dev, -1);
2480 if (old_flags & IFF_UP)
2481 bond_hw_addr_flush(bond_dev, slave_dev);
2484 slave_disable_netpoll(slave);
2486 /* close slave before restoring its mac address */
2487 dev_close(slave_dev);
2489 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2491 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2492 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2493 /* restore original ("permanent") mac address */
2494 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2495 slave->dev->addr_len);
2496 ss.ss_family = slave_dev->type;
2497 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2501 __dev_set_mtu(slave_dev, slave->original_mtu);
2503 dev_set_mtu(slave_dev, slave->original_mtu);
2505 if (!netif_is_bond_master(slave_dev))
2506 slave_dev->priv_flags &= ~IFF_BONDING;
2508 bond_xdp_set_features(bond_dev);
2509 kobject_put(&slave->kobj);
2514 /* A wrapper used because of ndo_del_link */
2515 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2517 return __bond_release_one(bond_dev, slave_dev, false, false);
2520 /* First release a slave and then destroy the bond if no more slaves are left.
2521 * Must be under rtnl_lock when this function is called.
2523 static int bond_release_and_destroy(struct net_device *bond_dev,
2524 struct net_device *slave_dev)
2526 struct bonding *bond = netdev_priv(bond_dev);
2529 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2530 if (ret == 0 && !bond_has_slaves(bond) &&
2531 bond_dev->reg_state != NETREG_UNREGISTERING) {
2532 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2533 netdev_info(bond_dev, "Destroying bond\n");
2534 bond_remove_proc_entry(bond);
2535 unregister_netdevice(bond_dev);
2540 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2542 struct bonding *bond = netdev_priv(bond_dev);
2544 bond_fill_ifbond(bond, info);
2547 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2549 struct bonding *bond = netdev_priv(bond_dev);
2550 struct list_head *iter;
2551 int i = 0, res = -ENODEV;
2552 struct slave *slave;
2554 bond_for_each_slave(bond, slave, iter) {
2555 if (i++ == (int)info->slave_id) {
2557 bond_fill_ifslave(slave, info);
2565 /*-------------------------------- Monitoring -------------------------------*/
2567 /* called with rcu_read_lock() */
2568 static int bond_miimon_inspect(struct bonding *bond)
2570 bool ignore_updelay = false;
2571 int link_state, commit = 0;
2572 struct list_head *iter;
2573 struct slave *slave;
2575 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2576 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2578 struct bond_up_slave *usable_slaves;
2580 usable_slaves = rcu_dereference(bond->usable_slaves);
2582 if (usable_slaves && usable_slaves->count == 0)
2583 ignore_updelay = true;
2586 bond_for_each_slave_rcu(bond, slave, iter) {
2587 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2589 link_state = bond_check_dev_link(bond, slave->dev, 0);
2591 switch (slave->link) {
2596 bond_propose_link_state(slave, BOND_LINK_FAIL);
2598 slave->delay = bond->params.downdelay;
2600 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2602 BOND_MODE_ACTIVEBACKUP) ?
2603 (bond_is_active_slave(slave) ?
2604 "active " : "backup ") : "",
2605 bond->params.downdelay * bond->params.miimon);
2608 case BOND_LINK_FAIL:
2610 /* recovered before downdelay expired */
2611 bond_propose_link_state(slave, BOND_LINK_UP);
2612 slave->last_link_up = jiffies;
2613 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2614 (bond->params.downdelay - slave->delay) *
2615 bond->params.miimon);
2620 if (slave->delay <= 0) {
2621 bond_propose_link_state(slave, BOND_LINK_DOWN);
2629 case BOND_LINK_DOWN:
2633 bond_propose_link_state(slave, BOND_LINK_BACK);
2635 slave->delay = bond->params.updelay;
2638 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2639 ignore_updelay ? 0 :
2640 bond->params.updelay *
2641 bond->params.miimon);
2644 case BOND_LINK_BACK:
2646 bond_propose_link_state(slave, BOND_LINK_DOWN);
2647 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2648 (bond->params.updelay - slave->delay) *
2649 bond->params.miimon);
2657 if (slave->delay <= 0) {
2658 bond_propose_link_state(slave, BOND_LINK_UP);
2660 ignore_updelay = false;
2672 static void bond_miimon_link_change(struct bonding *bond,
2673 struct slave *slave,
2676 switch (BOND_MODE(bond)) {
2677 case BOND_MODE_8023AD:
2678 bond_3ad_handle_link_change(slave, link);
2682 bond_alb_handle_link_change(bond, slave, link);
2685 bond_update_slave_arr(bond, NULL);
2690 static void bond_miimon_commit(struct bonding *bond)
2692 struct slave *slave, *primary, *active;
2693 bool do_failover = false;
2694 struct list_head *iter;
2698 bond_for_each_slave(bond, slave, iter) {
2699 switch (slave->link_new_state) {
2700 case BOND_LINK_NOCHANGE:
2701 /* For 802.3ad mode, check current slave speed and
2702 * duplex again in case its port was disabled after
2703 * invalid speed/duplex reporting but recovered before
2704 * link monitoring could make a decision on the actual
2707 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2708 slave->link == BOND_LINK_UP)
2709 bond_3ad_adapter_speed_duplex_changed(slave);
2713 if (bond_update_speed_duplex(slave) &&
2714 bond_needs_speed_duplex(bond)) {
2715 slave->link = BOND_LINK_DOWN;
2716 if (net_ratelimit())
2717 slave_warn(bond->dev, slave->dev,
2718 "failed to get link speed/duplex\n");
2721 bond_set_slave_link_state(slave, BOND_LINK_UP,
2722 BOND_SLAVE_NOTIFY_NOW);
2723 slave->last_link_up = jiffies;
2725 primary = rtnl_dereference(bond->primary_slave);
2726 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2727 /* prevent it from being the active one */
2728 bond_set_backup_slave(slave);
2729 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2730 /* make it immediately active */
2731 bond_set_active_slave(slave);
2734 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2735 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2736 slave->duplex ? "full" : "half");
2738 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2740 active = rtnl_dereference(bond->curr_active_slave);
2741 if (!active || slave == primary || slave->prio > active->prio)
2746 case BOND_LINK_DOWN:
2747 if (slave->link_failure_count < UINT_MAX)
2748 slave->link_failure_count++;
2750 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2751 BOND_SLAVE_NOTIFY_NOW);
2753 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2754 BOND_MODE(bond) == BOND_MODE_8023AD)
2755 bond_set_slave_inactive_flags(slave,
2756 BOND_SLAVE_NOTIFY_NOW);
2758 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2760 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2762 if (slave == rcu_access_pointer(bond->curr_active_slave))
2768 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2769 slave->link_new_state);
2770 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2778 bond_select_active_slave(bond);
2779 unblock_netpoll_tx();
2782 bond_set_carrier(bond);
2787 * Really a wrapper that splits the mii monitor into two phases: an
2788 * inspection, then (if inspection indicates something needs to be done)
2789 * an acquisition of appropriate locks followed by a commit phase to
2790 * implement whatever link state changes are indicated.
2792 static void bond_mii_monitor(struct work_struct *work)
2794 struct bonding *bond = container_of(work, struct bonding,
2796 bool should_notify_peers = false;
2798 unsigned long delay;
2799 struct slave *slave;
2800 struct list_head *iter;
2802 delay = msecs_to_jiffies(bond->params.miimon);
2804 if (!bond_has_slaves(bond))
2808 should_notify_peers = bond_should_notify_peers(bond);
2809 commit = !!bond_miimon_inspect(bond);
2810 if (bond->send_peer_notif) {
2812 if (rtnl_trylock()) {
2813 bond->send_peer_notif--;
2821 /* Race avoidance with bond_close cancel of workqueue */
2822 if (!rtnl_trylock()) {
2824 should_notify_peers = false;
2828 bond_for_each_slave(bond, slave, iter) {
2829 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2831 bond_miimon_commit(bond);
2833 rtnl_unlock(); /* might sleep, hold no other locks */
2837 if (bond->params.miimon)
2838 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2840 if (should_notify_peers) {
2841 if (!rtnl_trylock())
2843 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2848 static int bond_upper_dev_walk(struct net_device *upper,
2849 struct netdev_nested_priv *priv)
2851 __be32 ip = *(__be32 *)priv->data;
2853 return ip == bond_confirm_addr(upper, 0, ip);
2856 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2858 struct netdev_nested_priv priv = {
2859 .data = (void *)&ip,
2863 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2867 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2874 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2875 struct sk_buff *skb)
2877 struct net_device *bond_dev = slave->bond->dev;
2878 struct net_device *slave_dev = slave->dev;
2879 struct bond_vlan_tag *outer_tag = tags;
2881 if (!tags || tags->vlan_proto == VLAN_N_VID)
2886 /* Go through all the tags backwards and add them to the packet */
2887 while (tags->vlan_proto != VLAN_N_VID) {
2888 if (!tags->vlan_id) {
2893 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2894 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2895 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2898 net_err_ratelimited("failed to insert inner VLAN tag\n");
2904 /* Set the outer tag */
2905 if (outer_tag->vlan_id) {
2906 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2907 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2908 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2909 outer_tag->vlan_id);
2915 /* We go to the (large) trouble of VLAN tagging ARP frames because
2916 * switches in VLAN mode (especially if ports are configured as
2917 * "native" to a VLAN) might not pass non-tagged frames.
2919 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2920 __be32 src_ip, struct bond_vlan_tag *tags)
2922 struct net_device *bond_dev = slave->bond->dev;
2923 struct net_device *slave_dev = slave->dev;
2924 struct sk_buff *skb;
2926 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2927 arp_op, &dest_ip, &src_ip);
2929 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2930 NULL, slave_dev->dev_addr, NULL);
2933 net_err_ratelimited("ARP packet allocation failed\n");
2937 if (bond_handle_vlan(slave, tags, skb)) {
2938 slave_update_last_tx(slave);
2945 /* Validate the device path between the @start_dev and the @end_dev.
2946 * The path is valid if the @end_dev is reachable through device
2948 * When the path is validated, collect any vlan information in the
2951 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2952 struct net_device *end_dev,
2955 struct bond_vlan_tag *tags;
2956 struct net_device *upper;
2957 struct list_head *iter;
2959 if (start_dev == end_dev) {
2960 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2962 return ERR_PTR(-ENOMEM);
2963 tags[level].vlan_proto = VLAN_N_VID;
2967 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2968 tags = bond_verify_device_path(upper, end_dev, level + 1);
2969 if (IS_ERR_OR_NULL(tags)) {
2974 if (is_vlan_dev(upper)) {
2975 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2976 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2985 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2988 struct bond_vlan_tag *tags;
2989 __be32 *targets = bond->params.arp_targets, addr;
2992 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2993 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2994 __func__, &targets[i]);
2997 /* Find out through which dev should the packet go */
2998 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
3001 /* there's no route to target - try to send arp
3002 * probe to generate any traffic (arp_validate=0)
3004 if (bond->params.arp_validate)
3005 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3008 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3013 /* bond device itself */
3014 if (rt->dst.dev == bond->dev)
3018 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3021 if (!IS_ERR_OR_NULL(tags))
3024 /* Not our device - skip */
3025 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3026 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3032 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3034 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3039 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3043 if (!sip || !bond_has_this_ip(bond, tip)) {
3044 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3045 __func__, &sip, &tip);
3049 i = bond_get_targets_ip(bond->params.arp_targets, sip);
3051 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3055 slave->last_rx = jiffies;
3056 slave->target_last_arp_rx[i] = jiffies;
3059 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3060 struct slave *slave)
3062 struct arphdr *arp = (struct arphdr *)skb->data;
3063 struct slave *curr_active_slave, *curr_arp_slave;
3064 unsigned char *arp_ptr;
3068 alen = arp_hdr_len(bond->dev);
3070 if (alen > skb_headlen(skb)) {
3071 arp = kmalloc(alen, GFP_ATOMIC);
3074 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3078 if (arp->ar_hln != bond->dev->addr_len ||
3079 skb->pkt_type == PACKET_OTHERHOST ||
3080 skb->pkt_type == PACKET_LOOPBACK ||
3081 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3082 arp->ar_pro != htons(ETH_P_IP) ||
3086 arp_ptr = (unsigned char *)(arp + 1);
3087 arp_ptr += bond->dev->addr_len;
3088 memcpy(&sip, arp_ptr, 4);
3089 arp_ptr += 4 + bond->dev->addr_len;
3090 memcpy(&tip, arp_ptr, 4);
3092 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3093 __func__, slave->dev->name, bond_slave_state(slave),
3094 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3097 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3098 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3100 /* We 'trust' the received ARP enough to validate it if:
3102 * (a) the slave receiving the ARP is active (which includes the
3103 * current ARP slave, if any), or
3105 * (b) the receiving slave isn't active, but there is a currently
3106 * active slave and it received valid arp reply(s) after it became
3107 * the currently active slave, or
3109 * (c) there is an ARP slave that sent an ARP during the prior ARP
3110 * interval, and we receive an ARP reply on any slave. We accept
3111 * these because switch FDB update delays may deliver the ARP
3112 * reply to a slave other than the sender of the ARP request.
3114 * Note: for (b), backup slaves are receiving the broadcast ARP
3115 * request, not a reply. This request passes from the sending
3116 * slave through the L2 switch(es) to the receiving slave. Since
3117 * this is checking the request, sip/tip are swapped for
3120 * This is done to avoid endless looping when we can't reach the
3121 * arp_ip_target and fool ourselves with our own arp requests.
3123 if (bond_is_active_slave(slave))
3124 bond_validate_arp(bond, slave, sip, tip);
3125 else if (curr_active_slave &&
3126 time_after(slave_last_rx(bond, curr_active_slave),
3127 curr_active_slave->last_link_up))
3128 bond_validate_arp(bond, slave, tip, sip);
3129 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3130 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3131 bond_validate_arp(bond, slave, sip, tip);
3134 if (arp != (struct arphdr *)skb->data)
3136 return RX_HANDLER_ANOTHER;
3139 #if IS_ENABLED(CONFIG_IPV6)
3140 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3141 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3143 struct net_device *bond_dev = slave->bond->dev;
3144 struct net_device *slave_dev = slave->dev;
3145 struct in6_addr mcaddr;
3146 struct sk_buff *skb;
3148 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3151 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3153 net_err_ratelimited("NS packet allocation failed\n");
3157 addrconf_addr_solict_mult(daddr, &mcaddr);
3158 if (bond_handle_vlan(slave, tags, skb)) {
3159 slave_update_last_tx(slave);
3160 ndisc_send_skb(skb, &mcaddr, saddr);
3164 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3166 struct in6_addr *targets = bond->params.ns_targets;
3167 struct bond_vlan_tag *tags;
3168 struct dst_entry *dst;
3169 struct in6_addr saddr;
3173 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3174 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3175 __func__, &targets[i]);
3178 /* Find out through which dev should the packet go */
3179 memset(&fl6, 0, sizeof(struct flowi6));
3180 fl6.daddr = targets[i];
3181 fl6.flowi6_oif = bond->dev->ifindex;
3183 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3186 /* there's no route to target - try to send arp
3187 * probe to generate any traffic (arp_validate=0)
3189 if (bond->params.arp_validate)
3190 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3193 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3197 /* bond device itself */
3198 if (dst->dev == bond->dev)
3202 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3205 if (!IS_ERR_OR_NULL(tags))
3208 /* Not our device - skip */
3209 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3210 &targets[i], dst->dev ? dst->dev->name : "NULL");
3216 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3217 bond_ns_send(slave, &targets[i], &saddr, tags);
3219 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3226 static int bond_confirm_addr6(struct net_device *dev,
3227 struct netdev_nested_priv *priv)
3229 struct in6_addr *addr = (struct in6_addr *)priv->data;
3231 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3234 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3236 struct netdev_nested_priv priv = {
3241 if (bond_confirm_addr6(bond->dev, &priv))
3245 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3252 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3253 struct in6_addr *saddr, struct in6_addr *daddr)
3258 * 1. Source address is unspecified address.
3259 * 2. Dest address is neither all-nodes multicast address nor
3260 * exist on bond interface.
3262 if (ipv6_addr_any(saddr) ||
3263 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3264 !bond_has_this_ip6(bond, daddr))) {
3265 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3266 __func__, saddr, daddr);
3270 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3272 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3276 slave->last_rx = jiffies;
3277 slave->target_last_arp_rx[i] = jiffies;
3280 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3281 struct slave *slave)
3283 struct slave *curr_active_slave, *curr_arp_slave;
3284 struct in6_addr *saddr, *daddr;
3287 struct icmp6hdr icmp6;
3288 } *combined, _combined;
3290 if (skb->pkt_type == PACKET_OTHERHOST ||
3291 skb->pkt_type == PACKET_LOOPBACK)
3294 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3295 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3296 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3297 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3300 saddr = &combined->ip6.saddr;
3301 daddr = &combined->ip6.daddr;
3303 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3304 __func__, slave->dev->name, bond_slave_state(slave),
3305 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3308 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3309 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3311 /* We 'trust' the received ARP enough to validate it if:
3312 * see bond_arp_rcv().
3314 if (bond_is_active_slave(slave))
3315 bond_validate_na(bond, slave, saddr, daddr);
3316 else if (curr_active_slave &&
3317 time_after(slave_last_rx(bond, curr_active_slave),
3318 curr_active_slave->last_link_up))
3319 bond_validate_na(bond, slave, daddr, saddr);
3320 else if (curr_arp_slave &&
3321 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3322 bond_validate_na(bond, slave, saddr, daddr);
3325 return RX_HANDLER_ANOTHER;
3329 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3330 struct slave *slave)
3332 #if IS_ENABLED(CONFIG_IPV6)
3333 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3335 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3337 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3338 __func__, skb->dev->name);
3340 /* Use arp validate logic for both ARP and NS */
3341 if (!slave_do_arp_validate(bond, slave)) {
3342 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3343 #if IS_ENABLED(CONFIG_IPV6)
3344 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3346 !slave_do_arp_validate_only(bond))
3347 slave->last_rx = jiffies;
3348 return RX_HANDLER_ANOTHER;
3349 } else if (is_arp) {
3350 return bond_arp_rcv(skb, bond, slave);
3351 #if IS_ENABLED(CONFIG_IPV6)
3352 } else if (is_ipv6) {
3353 return bond_na_rcv(skb, bond, slave);
3356 return RX_HANDLER_ANOTHER;
3360 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3362 bond_arp_send_all(bond, slave);
3363 #if IS_ENABLED(CONFIG_IPV6)
3364 bond_ns_send_all(bond, slave);
3368 /* function to verify if we're in the arp_interval timeslice, returns true if
3369 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3370 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3372 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3375 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3377 return time_in_range(jiffies,
3378 last_act - delta_in_ticks,
3379 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3382 /* This function is called regularly to monitor each slave's link
3383 * ensuring that traffic is being sent and received when arp monitoring
3384 * is used in load-balancing mode. if the adapter has been dormant, then an
3385 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3386 * arp monitoring in active backup mode.
3388 static void bond_loadbalance_arp_mon(struct bonding *bond)
3390 struct slave *slave, *oldcurrent;
3391 struct list_head *iter;
3392 int do_failover = 0, slave_state_changed = 0;
3394 if (!bond_has_slaves(bond))
3399 oldcurrent = rcu_dereference(bond->curr_active_slave);
3400 /* see if any of the previous devices are up now (i.e. they have
3401 * xmt and rcv traffic). the curr_active_slave does not come into
3402 * the picture unless it is null. also, slave->last_link_up is not
3403 * needed here because we send an arp on each slave and give a slave
3404 * as long as it needs to get the tx/rx within the delta.
3405 * TODO: what about up/down delay in arp mode? it wasn't here before
3408 bond_for_each_slave_rcu(bond, slave, iter) {
3409 unsigned long last_tx = slave_last_tx(slave);
3411 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3413 if (slave->link != BOND_LINK_UP) {
3414 if (bond_time_in_interval(bond, last_tx, 1) &&
3415 bond_time_in_interval(bond, slave->last_rx, 1)) {
3417 bond_propose_link_state(slave, BOND_LINK_UP);
3418 slave_state_changed = 1;
3420 /* primary_slave has no meaning in round-robin
3421 * mode. the window of a slave being up and
3422 * curr_active_slave being null after enslaving
3426 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3429 slave_info(bond->dev, slave->dev, "interface is now up\n");
3433 /* slave->link == BOND_LINK_UP */
3435 /* not all switches will respond to an arp request
3436 * when the source ip is 0, so don't take the link down
3437 * if we don't know our ip yet
3439 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3440 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3442 bond_propose_link_state(slave, BOND_LINK_DOWN);
3443 slave_state_changed = 1;
3445 if (slave->link_failure_count < UINT_MAX)
3446 slave->link_failure_count++;
3448 slave_info(bond->dev, slave->dev, "interface is now down\n");
3450 if (slave == oldcurrent)
3455 /* note: if switch is in round-robin mode, all links
3456 * must tx arp to ensure all links rx an arp - otherwise
3457 * links may oscillate or not come up at all; if switch is
3458 * in something like xor mode, there is nothing we can
3459 * do - all replies will be rx'ed on same link causing slaves
3460 * to be unstable during low/no traffic periods
3462 if (bond_slave_is_up(slave))
3463 bond_send_validate(bond, slave);
3468 if (do_failover || slave_state_changed) {
3469 if (!rtnl_trylock())
3472 bond_for_each_slave(bond, slave, iter) {
3473 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3474 slave->link = slave->link_new_state;
3477 if (slave_state_changed) {
3478 bond_slave_state_change(bond);
3479 if (BOND_MODE(bond) == BOND_MODE_XOR)
3480 bond_update_slave_arr(bond, NULL);
3484 bond_select_active_slave(bond);
3485 unblock_netpoll_tx();
3491 if (bond->params.arp_interval)
3492 queue_delayed_work(bond->wq, &bond->arp_work,
3493 msecs_to_jiffies(bond->params.arp_interval));
3496 /* Called to inspect slaves for active-backup mode ARP monitor link state
3497 * changes. Sets proposed link state in slaves to specify what action
3498 * should take place for the slave. Returns 0 if no changes are found, >0
3499 * if changes to link states must be committed.
3501 * Called with rcu_read_lock held.
3503 static int bond_ab_arp_inspect(struct bonding *bond)
3505 unsigned long last_tx, last_rx;
3506 struct list_head *iter;
3507 struct slave *slave;
3510 bond_for_each_slave_rcu(bond, slave, iter) {
3511 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3512 last_rx = slave_last_rx(bond, slave);
3514 if (slave->link != BOND_LINK_UP) {
3515 if (bond_time_in_interval(bond, last_rx, 1)) {
3516 bond_propose_link_state(slave, BOND_LINK_UP);
3518 } else if (slave->link == BOND_LINK_BACK) {
3519 bond_propose_link_state(slave, BOND_LINK_FAIL);
3525 /* Give slaves 2*delta after being enslaved or made
3526 * active. This avoids bouncing, as the last receive
3527 * times need a full ARP monitor cycle to be updated.
3529 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3532 /* Backup slave is down if:
3533 * - No current_arp_slave AND
3534 * - more than (missed_max+1)*delta since last receive AND
3535 * - the bond has an IP address
3537 * Note: a non-null current_arp_slave indicates
3538 * the curr_active_slave went down and we are
3539 * searching for a new one; under this condition
3540 * we only take the curr_active_slave down - this
3541 * gives each slave a chance to tx/rx traffic
3542 * before being taken out
3544 if (!bond_is_active_slave(slave) &&
3545 !rcu_access_pointer(bond->current_arp_slave) &&
3546 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3547 bond_propose_link_state(slave, BOND_LINK_DOWN);
3551 /* Active slave is down if:
3552 * - more than missed_max*delta since transmitting OR
3553 * - (more than missed_max*delta since receive AND
3554 * the bond has an IP address)
3556 last_tx = slave_last_tx(slave);
3557 if (bond_is_active_slave(slave) &&
3558 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3559 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3560 bond_propose_link_state(slave, BOND_LINK_DOWN);
3568 /* Called to commit link state changes noted by inspection step of
3569 * active-backup mode ARP monitor.
3571 * Called with RTNL hold.
3573 static void bond_ab_arp_commit(struct bonding *bond)
3575 bool do_failover = false;
3576 struct list_head *iter;
3577 unsigned long last_tx;
3578 struct slave *slave;
3580 bond_for_each_slave(bond, slave, iter) {
3581 switch (slave->link_new_state) {
3582 case BOND_LINK_NOCHANGE:
3586 last_tx = slave_last_tx(slave);
3587 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3588 (!rtnl_dereference(bond->curr_active_slave) &&
3589 bond_time_in_interval(bond, last_tx, 1))) {
3590 struct slave *current_arp_slave;
3592 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3593 bond_set_slave_link_state(slave, BOND_LINK_UP,
3594 BOND_SLAVE_NOTIFY_NOW);
3595 if (current_arp_slave) {
3596 bond_set_slave_inactive_flags(
3598 BOND_SLAVE_NOTIFY_NOW);
3599 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3602 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3604 if (!rtnl_dereference(bond->curr_active_slave) ||
3605 slave == rtnl_dereference(bond->primary_slave) ||
3606 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3613 case BOND_LINK_DOWN:
3614 if (slave->link_failure_count < UINT_MAX)
3615 slave->link_failure_count++;
3617 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3618 BOND_SLAVE_NOTIFY_NOW);
3619 bond_set_slave_inactive_flags(slave,
3620 BOND_SLAVE_NOTIFY_NOW);
3622 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3624 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3625 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3631 case BOND_LINK_FAIL:
3632 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3633 BOND_SLAVE_NOTIFY_NOW);
3634 bond_set_slave_inactive_flags(slave,
3635 BOND_SLAVE_NOTIFY_NOW);
3637 /* A slave has just been enslaved and has become
3638 * the current active slave.
3640 if (rtnl_dereference(bond->curr_active_slave))
3641 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3645 slave_err(bond->dev, slave->dev,
3646 "impossible: link_new_state %d on slave\n",
3647 slave->link_new_state);
3654 bond_select_active_slave(bond);
3655 unblock_netpoll_tx();
3658 bond_set_carrier(bond);
3661 /* Send ARP probes for active-backup mode ARP monitor.
3663 * Called with rcu_read_lock held.
3665 static bool bond_ab_arp_probe(struct bonding *bond)
3667 struct slave *slave, *before = NULL, *new_slave = NULL,
3668 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3669 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3670 struct list_head *iter;
3672 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3674 if (curr_arp_slave && curr_active_slave)
3675 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3676 curr_arp_slave->dev->name,
3677 curr_active_slave->dev->name);
3679 if (curr_active_slave) {
3680 bond_send_validate(bond, curr_active_slave);
3681 return should_notify_rtnl;
3684 /* if we don't have a curr_active_slave, search for the next available
3685 * backup slave from the current_arp_slave and make it the candidate
3686 * for becoming the curr_active_slave
3689 if (!curr_arp_slave) {
3690 curr_arp_slave = bond_first_slave_rcu(bond);
3691 if (!curr_arp_slave)
3692 return should_notify_rtnl;
3695 bond_for_each_slave_rcu(bond, slave, iter) {
3696 if (!found && !before && bond_slave_is_up(slave))
3699 if (found && !new_slave && bond_slave_is_up(slave))
3701 /* if the link state is up at this point, we
3702 * mark it down - this can happen if we have
3703 * simultaneous link failures and
3704 * reselect_active_interface doesn't make this
3705 * one the current slave so it is still marked
3706 * up when it is actually down
3708 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3709 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3710 BOND_SLAVE_NOTIFY_LATER);
3711 if (slave->link_failure_count < UINT_MAX)
3712 slave->link_failure_count++;
3714 bond_set_slave_inactive_flags(slave,
3715 BOND_SLAVE_NOTIFY_LATER);
3717 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3719 if (slave == curr_arp_slave)
3723 if (!new_slave && before)
3729 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3730 BOND_SLAVE_NOTIFY_LATER);
3731 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3732 bond_send_validate(bond, new_slave);
3733 new_slave->last_link_up = jiffies;
3734 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3737 bond_for_each_slave_rcu(bond, slave, iter) {
3738 if (slave->should_notify || slave->should_notify_link) {
3739 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3743 return should_notify_rtnl;
3746 static void bond_activebackup_arp_mon(struct bonding *bond)
3748 bool should_notify_peers = false;
3749 bool should_notify_rtnl = false;
3752 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3754 if (!bond_has_slaves(bond))
3759 should_notify_peers = bond_should_notify_peers(bond);
3761 if (bond_ab_arp_inspect(bond)) {
3764 /* Race avoidance with bond_close flush of workqueue */
3765 if (!rtnl_trylock()) {
3767 should_notify_peers = false;
3771 bond_ab_arp_commit(bond);
3777 should_notify_rtnl = bond_ab_arp_probe(bond);
3781 if (bond->params.arp_interval)
3782 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3784 if (should_notify_peers || should_notify_rtnl) {
3785 if (!rtnl_trylock())
3788 if (should_notify_peers) {
3789 bond->send_peer_notif--;
3790 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3793 if (should_notify_rtnl) {
3794 bond_slave_state_notify(bond);
3795 bond_slave_link_notify(bond);
3802 static void bond_arp_monitor(struct work_struct *work)
3804 struct bonding *bond = container_of(work, struct bonding,
3807 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3808 bond_activebackup_arp_mon(bond);
3810 bond_loadbalance_arp_mon(bond);
3813 /*-------------------------- netdev event handling --------------------------*/
3815 /* Change device name */
3816 static int bond_event_changename(struct bonding *bond)
3818 bond_remove_proc_entry(bond);
3819 bond_create_proc_entry(bond);
3821 bond_debug_reregister(bond);
3826 static int bond_master_netdev_event(unsigned long event,
3827 struct net_device *bond_dev)
3829 struct bonding *event_bond = netdev_priv(bond_dev);
3831 netdev_dbg(bond_dev, "%s called\n", __func__);
3834 case NETDEV_CHANGENAME:
3835 return bond_event_changename(event_bond);
3836 case NETDEV_UNREGISTER:
3837 bond_remove_proc_entry(event_bond);
3838 #ifdef CONFIG_XFRM_OFFLOAD
3839 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3840 #endif /* CONFIG_XFRM_OFFLOAD */
3842 case NETDEV_REGISTER:
3843 bond_create_proc_entry(event_bond);
3852 static int bond_slave_netdev_event(unsigned long event,
3853 struct net_device *slave_dev)
3855 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3856 struct bonding *bond;
3857 struct net_device *bond_dev;
3859 /* A netdev event can be generated while enslaving a device
3860 * before netdev_rx_handler_register is called in which case
3861 * slave will be NULL
3864 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3868 bond_dev = slave->bond->dev;
3870 primary = rtnl_dereference(bond->primary_slave);
3872 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3875 case NETDEV_UNREGISTER:
3876 if (bond_dev->type != ARPHRD_ETHER)
3877 bond_release_and_destroy(bond_dev, slave_dev);
3879 __bond_release_one(bond_dev, slave_dev, false, true);
3883 /* For 802.3ad mode only:
3884 * Getting invalid Speed/Duplex values here will put slave
3885 * in weird state. Mark it as link-fail if the link was
3886 * previously up or link-down if it hasn't yet come up, and
3887 * let link-monitoring (miimon) set it right when correct
3888 * speeds/duplex are available.
3890 if (bond_update_speed_duplex(slave) &&
3891 BOND_MODE(bond) == BOND_MODE_8023AD) {
3892 if (slave->last_link_up)
3893 slave->link = BOND_LINK_FAIL;
3895 slave->link = BOND_LINK_DOWN;
3898 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3899 bond_3ad_adapter_speed_duplex_changed(slave);
3902 /* Refresh slave-array if applicable!
3903 * If the setup does not use miimon or arpmon (mode-specific!),
3904 * then these events will not cause the slave-array to be
3905 * refreshed. This will cause xmit to use a slave that is not
3906 * usable. Avoid such situation by refeshing the array at these
3907 * events. If these (miimon/arpmon) parameters are configured
3908 * then array gets refreshed twice and that should be fine!
3910 if (bond_mode_can_use_xmit_hash(bond))
3911 bond_update_slave_arr(bond, NULL);
3913 case NETDEV_CHANGEMTU:
3914 /* TODO: Should slaves be allowed to
3915 * independently alter their MTU? For
3916 * an active-backup bond, slaves need
3917 * not be the same type of device, so
3918 * MTUs may vary. For other modes,
3919 * slaves arguably should have the
3920 * same MTUs. To do this, we'd need to
3921 * take over the slave's change_mtu
3922 * function for the duration of their
3926 case NETDEV_CHANGENAME:
3927 /* we don't care if we don't have primary set */
3928 if (!bond_uses_primary(bond) ||
3929 !bond->params.primary[0])
3932 if (slave == primary) {
3933 /* slave's name changed - he's no longer primary */
3934 RCU_INIT_POINTER(bond->primary_slave, NULL);
3935 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3936 /* we have a new primary slave */
3937 rcu_assign_pointer(bond->primary_slave, slave);
3938 } else { /* we didn't change primary - exit */
3942 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3943 primary ? slave_dev->name : "none");
3946 bond_select_active_slave(bond);
3947 unblock_netpoll_tx();
3949 case NETDEV_FEAT_CHANGE:
3950 bond_compute_features(bond);
3952 case NETDEV_RESEND_IGMP:
3953 /* Propagate to master device */
3954 call_netdevice_notifiers(event, slave->bond->dev);
3956 case NETDEV_XDP_FEAT_CHANGE:
3957 bond_xdp_set_features(bond_dev);
3966 /* bond_netdev_event: handle netdev notifier chain events.
3968 * This function receives events for the netdev chain. The caller (an
3969 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3970 * locks for us to safely manipulate the slave devices (RTNL lock,
3973 static int bond_netdev_event(struct notifier_block *this,
3974 unsigned long event, void *ptr)
3976 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3978 netdev_dbg(event_dev, "%s received %s\n",
3979 __func__, netdev_cmd_to_name(event));
3981 if (!(event_dev->priv_flags & IFF_BONDING))
3984 if (event_dev->flags & IFF_MASTER) {
3987 ret = bond_master_netdev_event(event, event_dev);
3988 if (ret != NOTIFY_DONE)
3992 if (event_dev->flags & IFF_SLAVE)
3993 return bond_slave_netdev_event(event, event_dev);
3998 static struct notifier_block bond_netdev_notifier = {
3999 .notifier_call = bond_netdev_event,
4002 /*---------------------------- Hashing Policies -----------------------------*/
4004 /* Helper to access data in a packet, with or without a backing skb.
4005 * If skb is given the data is linearized if necessary via pskb_may_pull.
4007 static inline const void *bond_pull_data(struct sk_buff *skb,
4008 const void *data, int hlen, int n)
4010 if (likely(n <= hlen))
4012 else if (skb && likely(pskb_may_pull(skb, n)))
4018 /* L2 hash helper */
4019 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4023 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4027 ep = (struct ethhdr *)(data + mhoff);
4028 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4031 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4032 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4034 const struct ipv6hdr *iph6;
4035 const struct iphdr *iph;
4037 if (l2_proto == htons(ETH_P_IP)) {
4038 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4042 iph = (const struct iphdr *)(data + *nhoff);
4043 iph_to_flow_copy_v4addrs(fk, iph);
4044 *nhoff += iph->ihl << 2;
4045 if (!ip_is_fragment(iph))
4046 *ip_proto = iph->protocol;
4047 } else if (l2_proto == htons(ETH_P_IPV6)) {
4048 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4052 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4053 iph_to_flow_copy_v6addrs(fk, iph6);
4054 *nhoff += sizeof(*iph6);
4055 *ip_proto = iph6->nexthdr;
4060 if (l34 && *ip_proto >= 0)
4061 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4066 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4068 u32 srcmac_vendor = 0, srcmac_dev = 0;
4069 struct ethhdr *mac_hdr;
4073 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4076 mac_hdr = (struct ethhdr *)(data + mhoff);
4078 for (i = 0; i < 3; i++)
4079 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4081 for (i = 3; i < ETH_ALEN; i++)
4082 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4084 if (skb && skb_vlan_tag_present(skb))
4085 vlan = skb_vlan_tag_get(skb);
4087 return vlan ^ srcmac_vendor ^ srcmac_dev;
4090 /* Extract the appropriate headers based on bond's xmit policy */
4091 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4092 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4094 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4097 switch (bond->params.xmit_policy) {
4098 case BOND_XMIT_POLICY_ENCAP23:
4099 case BOND_XMIT_POLICY_ENCAP34:
4100 memset(fk, 0, sizeof(*fk));
4101 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4102 fk, data, l2_proto, nhoff, hlen, 0);
4107 fk->ports.ports = 0;
4108 memset(&fk->icmp, 0, sizeof(fk->icmp));
4109 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4112 /* ICMP error packets contains at least 8 bytes of the header
4113 * of the packet which generated the error. Use this information
4114 * to correlate ICMP error packets within the same flow which
4115 * generated the error.
4117 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4118 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4119 if (ip_proto == IPPROTO_ICMP) {
4120 if (!icmp_is_err(fk->icmp.type))
4123 nhoff += sizeof(struct icmphdr);
4124 } else if (ip_proto == IPPROTO_ICMPV6) {
4125 if (!icmpv6_is_err(fk->icmp.type))
4128 nhoff += sizeof(struct icmp6hdr);
4130 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4136 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4138 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4139 (__force u32)flow_get_u32_src(flow);
4140 hash ^= (hash >> 16);
4141 hash ^= (hash >> 8);
4143 /* discard lowest hash bit to deal with the common even ports pattern */
4144 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4145 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4151 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4152 * the data as required, but this function can be used without it if the data is
4153 * known to be linear (e.g. with xdp_buff).
4155 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4156 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4158 struct flow_keys flow;
4161 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4162 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4164 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4165 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4166 return bond_eth_hash(skb, data, mhoff, hlen);
4168 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4169 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4170 hash = bond_eth_hash(skb, data, mhoff, hlen);
4173 memcpy(&hash, &flow.icmp, sizeof(hash));
4175 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4178 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4182 * bond_xmit_hash - generate a hash value based on the xmit policy
4183 * @bond: bonding device
4184 * @skb: buffer to use for headers
4186 * This function will extract the necessary headers from the skb buffer and use
4187 * them to generate a hash based on the xmit_policy set in the bonding device
4189 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4191 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4195 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4196 skb_mac_offset(skb), skb_network_offset(skb),
4201 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4202 * @bond: bonding device
4203 * @xdp: buffer to use for headers
4205 * The XDP variant of bond_xmit_hash.
4207 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4211 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4214 eth = (struct ethhdr *)xdp->data;
4216 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4217 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4220 /*-------------------------- Device entry points ----------------------------*/
4222 void bond_work_init_all(struct bonding *bond)
4224 INIT_DELAYED_WORK(&bond->mcast_work,
4225 bond_resend_igmp_join_requests_delayed);
4226 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4227 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4228 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4229 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4230 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4233 static void bond_work_cancel_all(struct bonding *bond)
4235 cancel_delayed_work_sync(&bond->mii_work);
4236 cancel_delayed_work_sync(&bond->arp_work);
4237 cancel_delayed_work_sync(&bond->alb_work);
4238 cancel_delayed_work_sync(&bond->ad_work);
4239 cancel_delayed_work_sync(&bond->mcast_work);
4240 cancel_delayed_work_sync(&bond->slave_arr_work);
4243 static int bond_open(struct net_device *bond_dev)
4245 struct bonding *bond = netdev_priv(bond_dev);
4246 struct list_head *iter;
4247 struct slave *slave;
4249 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4250 bond->rr_tx_counter = alloc_percpu(u32);
4251 if (!bond->rr_tx_counter)
4255 /* reset slave->backup and slave->inactive */
4256 if (bond_has_slaves(bond)) {
4257 bond_for_each_slave(bond, slave, iter) {
4258 if (bond_uses_primary(bond) &&
4259 slave != rcu_access_pointer(bond->curr_active_slave)) {
4260 bond_set_slave_inactive_flags(slave,
4261 BOND_SLAVE_NOTIFY_NOW);
4262 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4263 bond_set_slave_active_flags(slave,
4264 BOND_SLAVE_NOTIFY_NOW);
4269 if (bond_is_lb(bond)) {
4270 /* bond_alb_initialize must be called before the timer
4273 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4275 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4276 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4279 if (bond->params.miimon) /* link check interval, in milliseconds. */
4280 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4282 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4283 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4284 bond->recv_probe = bond_rcv_validate;
4287 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4288 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4289 /* register to receive LACPDUs */
4290 bond->recv_probe = bond_3ad_lacpdu_recv;
4291 bond_3ad_initiate_agg_selection(bond, 1);
4293 bond_for_each_slave(bond, slave, iter)
4294 dev_mc_add(slave->dev, lacpdu_mcast_addr);
4297 if (bond_mode_can_use_xmit_hash(bond))
4298 bond_update_slave_arr(bond, NULL);
4303 static int bond_close(struct net_device *bond_dev)
4305 struct bonding *bond = netdev_priv(bond_dev);
4306 struct slave *slave;
4308 bond_work_cancel_all(bond);
4309 bond->send_peer_notif = 0;
4310 if (bond_is_lb(bond))
4311 bond_alb_deinitialize(bond);
4312 bond->recv_probe = NULL;
4314 if (bond_uses_primary(bond)) {
4316 slave = rcu_dereference(bond->curr_active_slave);
4318 bond_hw_addr_flush(bond_dev, slave->dev);
4321 struct list_head *iter;
4323 bond_for_each_slave(bond, slave, iter)
4324 bond_hw_addr_flush(bond_dev, slave->dev);
4330 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4331 * that some drivers can provide 32bit values only.
4333 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4334 const struct rtnl_link_stats64 *_new,
4335 const struct rtnl_link_stats64 *_old)
4337 const u64 *new = (const u64 *)_new;
4338 const u64 *old = (const u64 *)_old;
4339 u64 *res = (u64 *)_res;
4342 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4345 s64 delta = nv - ov;
4347 /* detects if this particular field is 32bit only */
4348 if (((nv | ov) >> 32) == 0)
4349 delta = (s64)(s32)((u32)nv - (u32)ov);
4351 /* filter anomalies, some drivers reset their stats
4352 * at down/up events.
4359 #ifdef CONFIG_LOCKDEP
4360 static int bond_get_lowest_level_rcu(struct net_device *dev)
4362 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4363 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4364 int cur = 0, max = 0;
4367 iter = &dev->adj_list.lower;
4372 ldev = netdev_next_lower_dev_rcu(now, &iter);
4377 niter = &ldev->adj_list.lower;
4378 dev_stack[cur] = now;
4379 iter_stack[cur++] = iter;
4388 next = dev_stack[--cur];
4389 niter = iter_stack[cur];
4400 static void bond_get_stats(struct net_device *bond_dev,
4401 struct rtnl_link_stats64 *stats)
4403 struct bonding *bond = netdev_priv(bond_dev);
4404 struct rtnl_link_stats64 temp;
4405 struct list_head *iter;
4406 struct slave *slave;
4411 #ifdef CONFIG_LOCKDEP
4412 nest_level = bond_get_lowest_level_rcu(bond_dev);
4415 spin_lock_nested(&bond->stats_lock, nest_level);
4416 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4418 bond_for_each_slave_rcu(bond, slave, iter) {
4419 const struct rtnl_link_stats64 *new =
4420 dev_get_stats(slave->dev, &temp);
4422 bond_fold_stats(stats, new, &slave->slave_stats);
4424 /* save off the slave stats for the next run */
4425 memcpy(&slave->slave_stats, new, sizeof(*new));
4428 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4429 spin_unlock(&bond->stats_lock);
4433 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4435 struct bonding *bond = netdev_priv(bond_dev);
4436 struct mii_ioctl_data *mii = NULL;
4437 const struct net_device_ops *ops;
4438 struct net_device *real_dev;
4439 struct hwtstamp_config cfg;
4443 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4454 /* We do this again just in case we were called by SIOCGMIIREG
4455 * instead of SIOCGMIIPHY.
4461 if (mii->reg_num == 1) {
4463 if (netif_carrier_ok(bond->dev))
4464 mii->val_out = BMSR_LSTATUS;
4469 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4472 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4477 real_dev = bond_option_active_slave_get_rcu(bond);
4481 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4482 ifrr.ifr_ifru = ifr->ifr_ifru;
4484 ops = real_dev->netdev_ops;
4485 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4486 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4490 ifr->ifr_ifru = ifrr.ifr_ifru;
4491 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4494 /* Set the BOND_PHC_INDEX flag to notify user space */
4495 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4497 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4508 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4510 struct bonding *bond = netdev_priv(bond_dev);
4511 struct net_device *slave_dev = NULL;
4512 struct ifbond k_binfo;
4513 struct ifbond __user *u_binfo = NULL;
4514 struct ifslave k_sinfo;
4515 struct ifslave __user *u_sinfo = NULL;
4516 struct bond_opt_value newval;
4520 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4523 case SIOCBONDINFOQUERY:
4524 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4526 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4529 bond_info_query(bond_dev, &k_binfo);
4530 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4534 case SIOCBONDSLAVEINFOQUERY:
4535 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4537 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4540 res = bond_slave_info_query(bond_dev, &k_sinfo);
4542 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4550 net = dev_net(bond_dev);
4552 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4555 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4557 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4563 case SIOCBONDENSLAVE:
4564 res = bond_enslave(bond_dev, slave_dev, NULL);
4566 case SIOCBONDRELEASE:
4567 res = bond_release(bond_dev, slave_dev);
4569 case SIOCBONDSETHWADDR:
4570 res = bond_set_dev_addr(bond_dev, slave_dev);
4572 case SIOCBONDCHANGEACTIVE:
4573 bond_opt_initstr(&newval, slave_dev->name);
4574 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4584 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4585 void __user *data, int cmd)
4587 struct ifreq ifrdata = { .ifr_data = data };
4590 case BOND_INFO_QUERY_OLD:
4591 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4592 case BOND_SLAVE_INFO_QUERY_OLD:
4593 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4594 case BOND_ENSLAVE_OLD:
4595 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4596 case BOND_RELEASE_OLD:
4597 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4598 case BOND_SETHWADDR_OLD:
4599 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4600 case BOND_CHANGE_ACTIVE_OLD:
4601 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4607 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4609 struct bonding *bond = netdev_priv(bond_dev);
4611 if (change & IFF_PROMISC)
4612 bond_set_promiscuity(bond,
4613 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4615 if (change & IFF_ALLMULTI)
4616 bond_set_allmulti(bond,
4617 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4620 static void bond_set_rx_mode(struct net_device *bond_dev)
4622 struct bonding *bond = netdev_priv(bond_dev);
4623 struct list_head *iter;
4624 struct slave *slave;
4627 if (bond_uses_primary(bond)) {
4628 slave = rcu_dereference(bond->curr_active_slave);
4630 dev_uc_sync(slave->dev, bond_dev);
4631 dev_mc_sync(slave->dev, bond_dev);
4634 bond_for_each_slave_rcu(bond, slave, iter) {
4635 dev_uc_sync_multiple(slave->dev, bond_dev);
4636 dev_mc_sync_multiple(slave->dev, bond_dev);
4642 static int bond_neigh_init(struct neighbour *n)
4644 struct bonding *bond = netdev_priv(n->dev);
4645 const struct net_device_ops *slave_ops;
4646 struct neigh_parms parms;
4647 struct slave *slave;
4651 slave = bond_first_slave_rcu(bond);
4654 slave_ops = slave->dev->netdev_ops;
4655 if (!slave_ops->ndo_neigh_setup)
4658 /* TODO: find another way [1] to implement this.
4659 * Passing a zeroed structure is fragile,
4660 * but at least we do not pass garbage.
4662 * [1] One way would be that ndo_neigh_setup() never touch
4663 * struct neigh_parms, but propagate the new neigh_setup()
4664 * back to ___neigh_create() / neigh_parms_alloc()
4666 memset(&parms, 0, sizeof(parms));
4667 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4672 if (parms.neigh_setup)
4673 ret = parms.neigh_setup(n);
4679 /* The bonding ndo_neigh_setup is called at init time beofre any
4680 * slave exists. So we must declare proxy setup function which will
4681 * be used at run time to resolve the actual slave neigh param setup.
4683 * It's also called by master devices (such as vlans) to setup their
4684 * underlying devices. In that case - do nothing, we're already set up from
4687 static int bond_neigh_setup(struct net_device *dev,
4688 struct neigh_parms *parms)
4690 /* modify only our neigh_parms */
4691 if (parms->dev == dev)
4692 parms->neigh_setup = bond_neigh_init;
4697 /* Change the MTU of all of a master's slaves to match the master */
4698 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4700 struct bonding *bond = netdev_priv(bond_dev);
4701 struct slave *slave, *rollback_slave;
4702 struct list_head *iter;
4705 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4707 bond_for_each_slave(bond, slave, iter) {
4708 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4709 slave, slave->dev->netdev_ops->ndo_change_mtu);
4711 res = dev_set_mtu(slave->dev, new_mtu);
4714 /* If we failed to set the slave's mtu to the new value
4715 * we must abort the operation even in ACTIVE_BACKUP
4716 * mode, because if we allow the backup slaves to have
4717 * different mtu values than the active slave we'll
4718 * need to change their mtu when doing a failover. That
4719 * means changing their mtu from timer context, which
4720 * is probably not a good idea.
4722 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4728 bond_dev->mtu = new_mtu;
4733 /* unwind from head to the slave that failed */
4734 bond_for_each_slave(bond, rollback_slave, iter) {
4737 if (rollback_slave == slave)
4740 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4742 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4749 /* Change HW address
4751 * Note that many devices must be down to change the HW address, and
4752 * downing the master releases all slaves. We can make bonds full of
4753 * bonding devices to test this, however.
4755 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4757 struct bonding *bond = netdev_priv(bond_dev);
4758 struct slave *slave, *rollback_slave;
4759 struct sockaddr_storage *ss = addr, tmp_ss;
4760 struct list_head *iter;
4763 if (BOND_MODE(bond) == BOND_MODE_ALB)
4764 return bond_alb_set_mac_address(bond_dev, addr);
4767 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4769 /* If fail_over_mac is enabled, do nothing and return success.
4770 * Returning an error causes ifenslave to fail.
4772 if (bond->params.fail_over_mac &&
4773 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4776 if (!is_valid_ether_addr(ss->__data))
4777 return -EADDRNOTAVAIL;
4779 bond_for_each_slave(bond, slave, iter) {
4780 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4782 res = dev_set_mac_address(slave->dev, addr, NULL);
4784 /* TODO: consider downing the slave
4786 * User should expect communications
4787 * breakage anyway until ARP finish
4790 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4797 dev_addr_set(bond_dev, ss->__data);
4801 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4802 tmp_ss.ss_family = bond_dev->type;
4804 /* unwind from head to the slave that failed */
4805 bond_for_each_slave(bond, rollback_slave, iter) {
4808 if (rollback_slave == slave)
4811 tmp_res = dev_set_mac_address(rollback_slave->dev,
4812 (struct sockaddr *)&tmp_ss, NULL);
4814 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4823 * bond_get_slave_by_id - get xmit slave with slave_id
4824 * @bond: bonding device that is transmitting
4825 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4827 * This function tries to get slave with slave_id but in case
4828 * it fails, it tries to find the first available slave for transmission.
4830 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4833 struct list_head *iter;
4834 struct slave *slave;
4837 /* Here we start from the slave with slave_id */
4838 bond_for_each_slave_rcu(bond, slave, iter) {
4840 if (bond_slave_can_tx(slave))
4845 /* Here we start from the first slave up to slave_id */
4847 bond_for_each_slave_rcu(bond, slave, iter) {
4850 if (bond_slave_can_tx(slave))
4853 /* no slave that can tx has been found */
4858 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4859 * @bond: bonding device to use
4861 * Based on the value of the bonding device's packets_per_slave parameter
4862 * this function generates a slave id, which is usually used as the next
4863 * slave to transmit through.
4865 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4868 struct reciprocal_value reciprocal_packets_per_slave;
4869 int packets_per_slave = bond->params.packets_per_slave;
4871 switch (packets_per_slave) {
4873 slave_id = get_random_u32();
4876 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4879 reciprocal_packets_per_slave =
4880 bond->params.reciprocal_packets_per_slave;
4881 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4882 slave_id = reciprocal_divide(slave_id,
4883 reciprocal_packets_per_slave);
4890 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4891 struct sk_buff *skb)
4893 struct slave *slave;
4897 /* Start with the curr_active_slave that joined the bond as the
4898 * default for sending IGMP traffic. For failover purposes one
4899 * needs to maintain some consistency for the interface that will
4900 * send the join/membership reports. The curr_active_slave found
4901 * will send all of this type of traffic.
4903 if (skb->protocol == htons(ETH_P_IP)) {
4904 int noff = skb_network_offset(skb);
4907 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4911 if (iph->protocol == IPPROTO_IGMP) {
4912 slave = rcu_dereference(bond->curr_active_slave);
4915 return bond_get_slave_by_id(bond, 0);
4920 slave_cnt = READ_ONCE(bond->slave_cnt);
4921 if (likely(slave_cnt)) {
4922 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4923 return bond_get_slave_by_id(bond, slave_id);
4928 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4929 struct xdp_buff *xdp)
4931 struct slave *slave;
4934 const struct ethhdr *eth;
4935 void *data = xdp->data;
4937 if (data + sizeof(struct ethhdr) > xdp->data_end)
4940 eth = (struct ethhdr *)data;
4941 data += sizeof(struct ethhdr);
4943 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4944 if (eth->h_proto == htons(ETH_P_IP)) {
4945 const struct iphdr *iph;
4947 if (data + sizeof(struct iphdr) > xdp->data_end)
4950 iph = (struct iphdr *)data;
4952 if (iph->protocol == IPPROTO_IGMP) {
4953 slave = rcu_dereference(bond->curr_active_slave);
4956 return bond_get_slave_by_id(bond, 0);
4961 slave_cnt = READ_ONCE(bond->slave_cnt);
4962 if (likely(slave_cnt)) {
4963 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4964 return bond_get_slave_by_id(bond, slave_id);
4969 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4970 struct net_device *bond_dev)
4972 struct bonding *bond = netdev_priv(bond_dev);
4973 struct slave *slave;
4975 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4977 return bond_dev_queue_xmit(bond, skb, slave->dev);
4979 return bond_tx_drop(bond_dev, skb);
4982 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4984 return rcu_dereference(bond->curr_active_slave);
4987 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4988 * the bond has a usable interface.
4990 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4991 struct net_device *bond_dev)
4993 struct bonding *bond = netdev_priv(bond_dev);
4994 struct slave *slave;
4996 slave = bond_xmit_activebackup_slave_get(bond);
4998 return bond_dev_queue_xmit(bond, skb, slave->dev);
5000 return bond_tx_drop(bond_dev, skb);
5003 /* Use this to update slave_array when (a) it's not appropriate to update
5004 * slave_array right away (note that update_slave_array() may sleep)
5005 * and / or (b) RTNL is not held.
5007 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
5009 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
5012 /* Slave array work handler. Holds only RTNL */
5013 static void bond_slave_arr_handler(struct work_struct *work)
5015 struct bonding *bond = container_of(work, struct bonding,
5016 slave_arr_work.work);
5019 if (!rtnl_trylock())
5022 ret = bond_update_slave_arr(bond, NULL);
5025 pr_warn_ratelimited("Failed to update slave array from WT\n");
5031 bond_slave_arr_work_rearm(bond, 1);
5034 static void bond_skip_slave(struct bond_up_slave *slaves,
5035 struct slave *skipslave)
5039 /* Rare situation where caller has asked to skip a specific
5040 * slave but allocation failed (most likely!). BTW this is
5041 * only possible when the call is initiated from
5042 * __bond_release_one(). In this situation; overwrite the
5043 * skipslave entry in the array with the last entry from the
5044 * array to avoid a situation where the xmit path may choose
5045 * this to-be-skipped slave to send a packet out.
5047 for (idx = 0; slaves && idx < slaves->count; idx++) {
5048 if (skipslave == slaves->arr[idx]) {
5050 slaves->arr[slaves->count - 1];
5057 static void bond_set_slave_arr(struct bonding *bond,
5058 struct bond_up_slave *usable_slaves,
5059 struct bond_up_slave *all_slaves)
5061 struct bond_up_slave *usable, *all;
5063 usable = rtnl_dereference(bond->usable_slaves);
5064 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5065 kfree_rcu(usable, rcu);
5067 all = rtnl_dereference(bond->all_slaves);
5068 rcu_assign_pointer(bond->all_slaves, all_slaves);
5069 kfree_rcu(all, rcu);
5072 static void bond_reset_slave_arr(struct bonding *bond)
5074 struct bond_up_slave *usable, *all;
5076 usable = rtnl_dereference(bond->usable_slaves);
5078 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5079 kfree_rcu(usable, rcu);
5082 all = rtnl_dereference(bond->all_slaves);
5084 RCU_INIT_POINTER(bond->all_slaves, NULL);
5085 kfree_rcu(all, rcu);
5089 /* Build the usable slaves array in control path for modes that use xmit-hash
5090 * to determine the slave interface -
5091 * (a) BOND_MODE_8023AD
5093 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5095 * The caller is expected to hold RTNL only and NO other lock!
5097 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5099 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5100 struct slave *slave;
5101 struct list_head *iter;
5107 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5108 bond->slave_cnt), GFP_KERNEL);
5109 all_slaves = kzalloc(struct_size(all_slaves, arr,
5110 bond->slave_cnt), GFP_KERNEL);
5111 if (!usable_slaves || !all_slaves) {
5115 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5116 struct ad_info ad_info;
5118 spin_lock_bh(&bond->mode_lock);
5119 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5120 spin_unlock_bh(&bond->mode_lock);
5121 pr_debug("bond_3ad_get_active_agg_info failed\n");
5122 /* No active aggragator means it's not safe to use
5123 * the previous array.
5125 bond_reset_slave_arr(bond);
5128 spin_unlock_bh(&bond->mode_lock);
5129 agg_id = ad_info.aggregator_id;
5131 bond_for_each_slave(bond, slave, iter) {
5132 if (skipslave == slave)
5135 all_slaves->arr[all_slaves->count++] = slave;
5136 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5137 struct aggregator *agg;
5139 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5140 if (!agg || agg->aggregator_identifier != agg_id)
5143 if (!bond_slave_can_tx(slave))
5146 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5147 usable_slaves->count);
5149 usable_slaves->arr[usable_slaves->count++] = slave;
5152 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5155 if (ret != 0 && skipslave) {
5156 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5158 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5161 kfree_rcu(all_slaves, rcu);
5162 kfree_rcu(usable_slaves, rcu);
5167 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5168 struct sk_buff *skb,
5169 struct bond_up_slave *slaves)
5171 struct slave *slave;
5175 hash = bond_xmit_hash(bond, skb);
5176 count = slaves ? READ_ONCE(slaves->count) : 0;
5177 if (unlikely(!count))
5180 slave = slaves->arr[hash % count];
5184 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5185 struct xdp_buff *xdp)
5187 struct bond_up_slave *slaves;
5191 hash = bond_xmit_hash_xdp(bond, xdp);
5192 slaves = rcu_dereference(bond->usable_slaves);
5193 count = slaves ? READ_ONCE(slaves->count) : 0;
5194 if (unlikely(!count))
5197 return slaves->arr[hash % count];
5200 /* Use this Xmit function for 3AD as well as XOR modes. The current
5201 * usable slave array is formed in the control path. The xmit function
5202 * just calculates hash and sends the packet out.
5204 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5205 struct net_device *dev)
5207 struct bonding *bond = netdev_priv(dev);
5208 struct bond_up_slave *slaves;
5209 struct slave *slave;
5211 slaves = rcu_dereference(bond->usable_slaves);
5212 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5214 return bond_dev_queue_xmit(bond, skb, slave->dev);
5216 return bond_tx_drop(dev, skb);
5219 /* in broadcast mode, we send everything to all usable interfaces. */
5220 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5221 struct net_device *bond_dev)
5223 struct bonding *bond = netdev_priv(bond_dev);
5224 struct slave *slave = NULL;
5225 struct list_head *iter;
5226 bool xmit_suc = false;
5227 bool skb_used = false;
5229 bond_for_each_slave_rcu(bond, slave, iter) {
5230 struct sk_buff *skb2;
5232 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5235 if (bond_is_last_slave(bond, slave)) {
5239 skb2 = skb_clone(skb, GFP_ATOMIC);
5241 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5242 bond_dev->name, __func__);
5247 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5252 dev_kfree_skb_any(skb);
5255 return NETDEV_TX_OK;
5257 dev_core_stats_tx_dropped_inc(bond_dev);
5258 return NET_XMIT_DROP;
5261 /*------------------------- Device initialization ---------------------------*/
5263 /* Lookup the slave that corresponds to a qid */
5264 static inline int bond_slave_override(struct bonding *bond,
5265 struct sk_buff *skb)
5267 struct slave *slave = NULL;
5268 struct list_head *iter;
5270 if (!skb_rx_queue_recorded(skb))
5273 /* Find out if any slaves have the same mapping as this skb. */
5274 bond_for_each_slave_rcu(bond, slave, iter) {
5275 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5276 if (bond_slave_is_up(slave) &&
5277 slave->link == BOND_LINK_UP) {
5278 bond_dev_queue_xmit(bond, skb, slave->dev);
5281 /* If the slave isn't UP, use default transmit policy. */
5290 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5291 struct net_device *sb_dev)
5293 /* This helper function exists to help dev_pick_tx get the correct
5294 * destination queue. Using a helper function skips a call to
5295 * skb_tx_hash and will put the skbs in the queue we expect on their
5296 * way down to the bonding driver.
5298 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5300 /* Save the original txq to restore before passing to the driver */
5301 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5303 if (unlikely(txq >= dev->real_num_tx_queues)) {
5305 txq -= dev->real_num_tx_queues;
5306 } while (txq >= dev->real_num_tx_queues);
5311 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5312 struct sk_buff *skb,
5315 struct bonding *bond = netdev_priv(master_dev);
5316 struct bond_up_slave *slaves;
5317 struct slave *slave = NULL;
5319 switch (BOND_MODE(bond)) {
5320 case BOND_MODE_ROUNDROBIN:
5321 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5323 case BOND_MODE_ACTIVEBACKUP:
5324 slave = bond_xmit_activebackup_slave_get(bond);
5326 case BOND_MODE_8023AD:
5329 slaves = rcu_dereference(bond->all_slaves);
5331 slaves = rcu_dereference(bond->usable_slaves);
5332 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5334 case BOND_MODE_BROADCAST:
5337 slave = bond_xmit_alb_slave_get(bond, skb);
5340 slave = bond_xmit_tlb_slave_get(bond, skb);
5343 /* Should never happen, mode already checked */
5344 WARN_ONCE(true, "Unknown bonding mode");
5353 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5355 switch (sk->sk_family) {
5356 #if IS_ENABLED(CONFIG_IPV6)
5358 if (ipv6_only_sock(sk) ||
5359 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5360 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5361 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5362 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5367 default: /* AF_INET */
5368 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5369 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5370 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5374 flow->ports.src = inet_sk(sk)->inet_sport;
5375 flow->ports.dst = inet_sk(sk)->inet_dport;
5379 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5380 * @sk: socket to use for headers
5382 * This function will extract the necessary field from the socket and use
5383 * them to generate a hash based on the LAYER34 xmit_policy.
5384 * Assumes that sk is a TCP or UDP socket.
5386 static u32 bond_sk_hash_l34(struct sock *sk)
5388 struct flow_keys flow;
5391 bond_sk_to_flow(sk, &flow);
5394 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5396 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5399 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5402 struct bond_up_slave *slaves;
5403 struct slave *slave;
5407 slaves = rcu_dereference(bond->usable_slaves);
5408 count = slaves ? READ_ONCE(slaves->count) : 0;
5409 if (unlikely(!count))
5412 hash = bond_sk_hash_l34(sk);
5413 slave = slaves->arr[hash % count];
5418 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5421 struct bonding *bond = netdev_priv(dev);
5422 struct net_device *lower = NULL;
5425 if (bond_sk_check(bond))
5426 lower = __bond_sk_get_lower_dev(bond, sk);
5432 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5433 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5434 struct net_device *dev)
5436 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5438 /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded
5439 * was true, if tls_device_down is running in parallel, but it's OK,
5440 * because bond_get_slave_by_dev has a NULL check.
5442 if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5443 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5444 return bond_tx_drop(dev, skb);
5448 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5450 struct bonding *bond = netdev_priv(dev);
5452 if (bond_should_override_tx_queue(bond) &&
5453 !bond_slave_override(bond, skb))
5454 return NETDEV_TX_OK;
5456 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5457 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5458 return bond_tls_device_xmit(bond, skb, dev);
5461 switch (BOND_MODE(bond)) {
5462 case BOND_MODE_ROUNDROBIN:
5463 return bond_xmit_roundrobin(skb, dev);
5464 case BOND_MODE_ACTIVEBACKUP:
5465 return bond_xmit_activebackup(skb, dev);
5466 case BOND_MODE_8023AD:
5468 return bond_3ad_xor_xmit(skb, dev);
5469 case BOND_MODE_BROADCAST:
5470 return bond_xmit_broadcast(skb, dev);
5472 return bond_alb_xmit(skb, dev);
5474 return bond_tlb_xmit(skb, dev);
5476 /* Should never happen, mode already checked */
5477 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5479 return bond_tx_drop(dev, skb);
5483 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5485 struct bonding *bond = netdev_priv(dev);
5486 netdev_tx_t ret = NETDEV_TX_OK;
5488 /* If we risk deadlock from transmitting this in the
5489 * netpoll path, tell netpoll to queue the frame for later tx
5491 if (unlikely(is_netpoll_tx_blocked(dev)))
5492 return NETDEV_TX_BUSY;
5495 if (bond_has_slaves(bond))
5496 ret = __bond_start_xmit(skb, dev);
5498 ret = bond_tx_drop(dev, skb);
5504 static struct net_device *
5505 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5507 struct bonding *bond = netdev_priv(bond_dev);
5508 struct slave *slave;
5510 /* Caller needs to hold rcu_read_lock() */
5512 switch (BOND_MODE(bond)) {
5513 case BOND_MODE_ROUNDROBIN:
5514 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5517 case BOND_MODE_ACTIVEBACKUP:
5518 slave = bond_xmit_activebackup_slave_get(bond);
5521 case BOND_MODE_8023AD:
5523 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5527 /* Should never happen. Mode guarded by bond_xdp_check() */
5528 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5539 static int bond_xdp_xmit(struct net_device *bond_dev,
5540 int n, struct xdp_frame **frames, u32 flags)
5542 int nxmit, err = -ENXIO;
5546 for (nxmit = 0; nxmit < n; nxmit++) {
5547 struct xdp_frame *frame = frames[nxmit];
5548 struct xdp_frame *frames1[] = {frame};
5549 struct net_device *slave_dev;
5550 struct xdp_buff xdp;
5552 xdp_convert_frame_to_buff(frame, &xdp);
5554 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5560 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5567 /* If error happened on the first frame then we can pass the error up, otherwise
5568 * report the number of frames that were xmitted.
5571 return (nxmit == 0 ? err : nxmit);
5576 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5577 struct netlink_ext_ack *extack)
5579 struct bonding *bond = netdev_priv(dev);
5580 struct list_head *iter;
5581 struct slave *slave, *rollback_slave;
5582 struct bpf_prog *old_prog;
5583 struct netdev_bpf xdp = {
5584 .command = XDP_SETUP_PROG,
5593 if (!bond_xdp_check(bond))
5596 old_prog = bond->xdp_prog;
5597 bond->xdp_prog = prog;
5599 bond_for_each_slave(bond, slave, iter) {
5600 struct net_device *slave_dev = slave->dev;
5602 if (!slave_dev->netdev_ops->ndo_bpf ||
5603 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5604 SLAVE_NL_ERR(dev, slave_dev, extack,
5605 "Slave device does not support XDP");
5610 if (dev_xdp_prog_count(slave_dev) > 0) {
5611 SLAVE_NL_ERR(dev, slave_dev, extack,
5612 "Slave has XDP program loaded, please unload before enslaving");
5617 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5619 /* ndo_bpf() sets extack error message */
5620 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5628 static_branch_inc(&bpf_master_redirect_enabled_key);
5629 } else if (old_prog) {
5630 bpf_prog_put(old_prog);
5631 static_branch_dec(&bpf_master_redirect_enabled_key);
5637 /* unwind the program changes */
5638 bond->xdp_prog = old_prog;
5639 xdp.prog = old_prog;
5640 xdp.extack = NULL; /* do not overwrite original error */
5642 bond_for_each_slave(bond, rollback_slave, iter) {
5643 struct net_device *slave_dev = rollback_slave->dev;
5646 if (slave == rollback_slave)
5649 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5651 slave_err(dev, slave_dev,
5652 "Error %d when unwinding XDP program change\n", err_unwind);
5654 bpf_prog_inc(xdp.prog);
5659 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5661 switch (xdp->command) {
5662 case XDP_SETUP_PROG:
5663 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5669 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5671 if (speed == 0 || speed == SPEED_UNKNOWN)
5672 speed = slave->speed;
5674 speed = min(speed, slave->speed);
5679 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5680 struct ethtool_link_ksettings *cmd)
5682 struct bonding *bond = netdev_priv(bond_dev);
5683 struct list_head *iter;
5684 struct slave *slave;
5687 cmd->base.duplex = DUPLEX_UNKNOWN;
5688 cmd->base.port = PORT_OTHER;
5690 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5691 * do not need to check mode. Though link speed might not represent
5692 * the true receive or transmit bandwidth (not all modes are symmetric)
5693 * this is an accurate maximum.
5695 bond_for_each_slave(bond, slave, iter) {
5696 if (bond_slave_can_tx(slave)) {
5697 if (slave->speed != SPEED_UNKNOWN) {
5698 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5699 speed = bond_mode_bcast_speed(slave,
5702 speed += slave->speed;
5704 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5705 slave->duplex != DUPLEX_UNKNOWN)
5706 cmd->base.duplex = slave->duplex;
5709 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5714 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5715 struct ethtool_drvinfo *drvinfo)
5717 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5718 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5722 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5723 struct ethtool_ts_info *info)
5725 struct bonding *bond = netdev_priv(bond_dev);
5726 struct ethtool_ts_info ts_info;
5727 const struct ethtool_ops *ops;
5728 struct net_device *real_dev;
5729 bool sw_tx_support = false;
5730 struct phy_device *phydev;
5731 struct list_head *iter;
5732 struct slave *slave;
5736 real_dev = bond_option_active_slave_get_rcu(bond);
5741 ops = real_dev->ethtool_ops;
5742 phydev = real_dev->phydev;
5744 if (phy_has_tsinfo(phydev)) {
5745 ret = phy_ts_info(phydev, info);
5747 } else if (ops->get_ts_info) {
5748 ret = ops->get_ts_info(real_dev, info);
5752 /* Check if all slaves support software tx timestamping */
5754 bond_for_each_slave_rcu(bond, slave, iter) {
5756 ops = slave->dev->ethtool_ops;
5757 phydev = slave->dev->phydev;
5759 if (phy_has_tsinfo(phydev))
5760 ret = phy_ts_info(phydev, &ts_info);
5761 else if (ops->get_ts_info)
5762 ret = ops->get_ts_info(slave->dev, &ts_info);
5764 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5765 sw_tx_support = true;
5769 sw_tx_support = false;
5776 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5777 SOF_TIMESTAMPING_SOFTWARE;
5779 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5781 info->phc_index = -1;
5788 static const struct ethtool_ops bond_ethtool_ops = {
5789 .get_drvinfo = bond_ethtool_get_drvinfo,
5790 .get_link = ethtool_op_get_link,
5791 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5792 .get_ts_info = bond_ethtool_get_ts_info,
5795 static const struct net_device_ops bond_netdev_ops = {
5796 .ndo_init = bond_init,
5797 .ndo_uninit = bond_uninit,
5798 .ndo_open = bond_open,
5799 .ndo_stop = bond_close,
5800 .ndo_start_xmit = bond_start_xmit,
5801 .ndo_select_queue = bond_select_queue,
5802 .ndo_get_stats64 = bond_get_stats,
5803 .ndo_eth_ioctl = bond_eth_ioctl,
5804 .ndo_siocbond = bond_do_ioctl,
5805 .ndo_siocdevprivate = bond_siocdevprivate,
5806 .ndo_change_rx_flags = bond_change_rx_flags,
5807 .ndo_set_rx_mode = bond_set_rx_mode,
5808 .ndo_change_mtu = bond_change_mtu,
5809 .ndo_set_mac_address = bond_set_mac_address,
5810 .ndo_neigh_setup = bond_neigh_setup,
5811 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5812 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5813 #ifdef CONFIG_NET_POLL_CONTROLLER
5814 .ndo_netpoll_setup = bond_netpoll_setup,
5815 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5816 .ndo_poll_controller = bond_poll_controller,
5818 .ndo_add_slave = bond_enslave,
5819 .ndo_del_slave = bond_release,
5820 .ndo_fix_features = bond_fix_features,
5821 .ndo_features_check = passthru_features_check,
5822 .ndo_get_xmit_slave = bond_xmit_get_slave,
5823 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5824 .ndo_bpf = bond_xdp,
5825 .ndo_xdp_xmit = bond_xdp_xmit,
5826 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5829 static const struct device_type bond_type = {
5833 static void bond_destructor(struct net_device *bond_dev)
5835 struct bonding *bond = netdev_priv(bond_dev);
5838 destroy_workqueue(bond->wq);
5840 if (bond->rr_tx_counter)
5841 free_percpu(bond->rr_tx_counter);
5844 void bond_setup(struct net_device *bond_dev)
5846 struct bonding *bond = netdev_priv(bond_dev);
5848 spin_lock_init(&bond->mode_lock);
5849 bond->params = bonding_defaults;
5851 /* Initialize pointers */
5852 bond->dev = bond_dev;
5854 /* Initialize the device entry points */
5855 ether_setup(bond_dev);
5856 bond_dev->max_mtu = ETH_MAX_MTU;
5857 bond_dev->netdev_ops = &bond_netdev_ops;
5858 bond_dev->ethtool_ops = &bond_ethtool_ops;
5860 bond_dev->needs_free_netdev = true;
5861 bond_dev->priv_destructor = bond_destructor;
5863 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5865 /* Initialize the device options */
5866 bond_dev->flags |= IFF_MASTER;
5867 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5868 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5870 #ifdef CONFIG_XFRM_OFFLOAD
5871 /* set up xfrm device ops (only supported in active-backup right now) */
5872 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5873 INIT_LIST_HEAD(&bond->ipsec_list);
5874 spin_lock_init(&bond->ipsec_lock);
5875 #endif /* CONFIG_XFRM_OFFLOAD */
5877 /* don't acquire bond device's netif_tx_lock when transmitting */
5878 bond_dev->features |= NETIF_F_LLTX;
5880 /* By default, we declare the bond to be fully
5881 * VLAN hardware accelerated capable. Special
5882 * care is taken in the various xmit functions
5883 * when there are slaves that are not hw accel
5887 /* Don't allow bond devices to change network namespaces. */
5888 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5890 bond_dev->hw_features = BOND_VLAN_FEATURES |
5891 NETIF_F_HW_VLAN_CTAG_RX |
5892 NETIF_F_HW_VLAN_CTAG_FILTER;
5894 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5895 bond_dev->features |= bond_dev->hw_features;
5896 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5897 #ifdef CONFIG_XFRM_OFFLOAD
5898 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5899 /* Only enable XFRM features if this is an active-backup config */
5900 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5901 bond_dev->features |= BOND_XFRM_FEATURES;
5902 #endif /* CONFIG_XFRM_OFFLOAD */
5904 if (bond_xdp_check(bond))
5905 bond_dev->xdp_features = NETDEV_XDP_ACT_MASK;
5908 /* Destroy a bonding device.
5909 * Must be under rtnl_lock when this function is called.
5911 static void bond_uninit(struct net_device *bond_dev)
5913 struct bonding *bond = netdev_priv(bond_dev);
5914 struct bond_up_slave *usable, *all;
5915 struct list_head *iter;
5916 struct slave *slave;
5918 bond_netpoll_cleanup(bond_dev);
5920 /* Release the bonded slaves */
5921 bond_for_each_slave(bond, slave, iter)
5922 __bond_release_one(bond_dev, slave->dev, true, true);
5923 netdev_info(bond_dev, "Released all slaves\n");
5925 usable = rtnl_dereference(bond->usable_slaves);
5927 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5928 kfree_rcu(usable, rcu);
5931 all = rtnl_dereference(bond->all_slaves);
5933 RCU_INIT_POINTER(bond->all_slaves, NULL);
5934 kfree_rcu(all, rcu);
5937 list_del(&bond->bond_list);
5939 bond_debug_unregister(bond);
5942 /*------------------------- Module initialization ---------------------------*/
5944 static int bond_check_params(struct bond_params *params)
5946 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5947 struct bond_opt_value newval;
5948 const struct bond_opt_value *valptr;
5949 int arp_all_targets_value = 0;
5950 u16 ad_actor_sys_prio = 0;
5951 u16 ad_user_port_key = 0;
5952 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5954 int bond_mode = BOND_MODE_ROUNDROBIN;
5955 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5959 /* Convert string parameters. */
5961 bond_opt_initstr(&newval, mode);
5962 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5964 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5967 bond_mode = valptr->value;
5970 if (xmit_hash_policy) {
5971 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5972 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5973 bond_mode == BOND_MODE_BROADCAST) {
5974 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5975 bond_mode_name(bond_mode));
5977 bond_opt_initstr(&newval, xmit_hash_policy);
5978 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5981 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5985 xmit_hashtype = valptr->value;
5990 if (bond_mode != BOND_MODE_8023AD) {
5991 pr_info("lacp_rate param is irrelevant in mode %s\n",
5992 bond_mode_name(bond_mode));
5994 bond_opt_initstr(&newval, lacp_rate);
5995 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5998 pr_err("Error: Invalid lacp rate \"%s\"\n",
6002 lacp_fast = valptr->value;
6007 bond_opt_initstr(&newval, ad_select);
6008 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6011 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6014 params->ad_select = valptr->value;
6015 if (bond_mode != BOND_MODE_8023AD)
6016 pr_warn("ad_select param only affects 802.3ad mode\n");
6018 params->ad_select = BOND_AD_STABLE;
6021 if (max_bonds < 0) {
6022 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6023 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6024 max_bonds = BOND_DEFAULT_MAX_BONDS;
6028 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6034 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6039 if (downdelay < 0) {
6040 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6041 downdelay, INT_MAX);
6045 if ((use_carrier != 0) && (use_carrier != 1)) {
6046 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6051 if (num_peer_notif < 0 || num_peer_notif > 255) {
6052 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6057 /* reset values for 802.3ad/TLB/ALB */
6058 if (!bond_mode_uses_arp(bond_mode)) {
6060 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");
6061 pr_warn("Forcing miimon to 100msec\n");
6062 miimon = BOND_DEFAULT_MIIMON;
6066 if (tx_queues < 1 || tx_queues > 255) {
6067 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6068 tx_queues, BOND_DEFAULT_TX_QUEUES);
6069 tx_queues = BOND_DEFAULT_TX_QUEUES;
6072 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6073 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6075 all_slaves_active = 0;
6078 if (resend_igmp < 0 || resend_igmp > 255) {
6079 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6080 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6081 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6084 bond_opt_initval(&newval, packets_per_slave);
6085 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6086 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6087 packets_per_slave, USHRT_MAX);
6088 packets_per_slave = 1;
6091 if (bond_mode == BOND_MODE_ALB) {
6092 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",
6097 if (updelay || downdelay) {
6098 /* just warn the user the up/down delay will have
6099 * no effect since miimon is zero...
6101 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",
6102 updelay, downdelay);
6105 /* don't allow arp monitoring */
6107 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6108 miimon, arp_interval);
6112 if ((updelay % miimon) != 0) {
6113 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6114 updelay, miimon, (updelay / miimon) * miimon);
6119 if ((downdelay % miimon) != 0) {
6120 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6122 (downdelay / miimon) * miimon);
6125 downdelay /= miimon;
6128 if (arp_interval < 0) {
6129 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6130 arp_interval, INT_MAX);
6134 for (arp_ip_count = 0, i = 0;
6135 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6138 /* not a complete check, but good enough to catch mistakes */
6139 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6140 !bond_is_ip_target_ok(ip)) {
6141 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6145 if (bond_get_targets_ip(arp_target, ip) == -1)
6146 arp_target[arp_ip_count++] = ip;
6148 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6153 if (arp_interval && !arp_ip_count) {
6154 /* don't allow arping if no arp_ip_target given... */
6155 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6161 if (!arp_interval) {
6162 pr_err("arp_validate requires arp_interval\n");
6166 bond_opt_initstr(&newval, arp_validate);
6167 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6170 pr_err("Error: invalid arp_validate \"%s\"\n",
6174 arp_validate_value = valptr->value;
6176 arp_validate_value = 0;
6179 if (arp_all_targets) {
6180 bond_opt_initstr(&newval, arp_all_targets);
6181 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6184 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6186 arp_all_targets_value = 0;
6188 arp_all_targets_value = valptr->value;
6193 pr_info("MII link monitoring set to %d ms\n", miimon);
6194 } else if (arp_interval) {
6195 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6196 arp_validate_value);
6197 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6198 arp_interval, valptr->string, arp_ip_count);
6200 for (i = 0; i < arp_ip_count; i++)
6201 pr_cont(" %s", arp_ip_target[i]);
6205 } else if (max_bonds) {
6206 /* miimon and arp_interval not set, we need one so things
6207 * work as expected, see bonding.txt for details
6209 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");
6212 if (primary && !bond_mode_uses_primary(bond_mode)) {
6213 /* currently, using a primary only makes sense
6214 * in active backup, TLB or ALB modes
6216 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6217 primary, bond_mode_name(bond_mode));
6221 if (primary && primary_reselect) {
6222 bond_opt_initstr(&newval, primary_reselect);
6223 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6226 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6230 primary_reselect_value = valptr->value;
6232 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6235 if (fail_over_mac) {
6236 bond_opt_initstr(&newval, fail_over_mac);
6237 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6240 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6244 fail_over_mac_value = valptr->value;
6245 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6246 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6248 fail_over_mac_value = BOND_FOM_NONE;
6251 bond_opt_initstr(&newval, "default");
6252 valptr = bond_opt_parse(
6253 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6256 pr_err("Error: No ad_actor_sys_prio default value");
6259 ad_actor_sys_prio = valptr->value;
6261 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6264 pr_err("Error: No ad_user_port_key default value");
6267 ad_user_port_key = valptr->value;
6269 bond_opt_initstr(&newval, "default");
6270 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6272 pr_err("Error: No tlb_dynamic_lb default value");
6275 tlb_dynamic_lb = valptr->value;
6277 if (lp_interval == 0) {
6278 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6279 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6280 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6283 /* fill params struct with the proper values */
6284 params->mode = bond_mode;
6285 params->xmit_policy = xmit_hashtype;
6286 params->miimon = miimon;
6287 params->num_peer_notif = num_peer_notif;
6288 params->arp_interval = arp_interval;
6289 params->arp_validate = arp_validate_value;
6290 params->arp_all_targets = arp_all_targets_value;
6291 params->missed_max = 2;
6292 params->updelay = updelay;
6293 params->downdelay = downdelay;
6294 params->peer_notif_delay = 0;
6295 params->use_carrier = use_carrier;
6296 params->lacp_active = 1;
6297 params->lacp_fast = lacp_fast;
6298 params->primary[0] = 0;
6299 params->primary_reselect = primary_reselect_value;
6300 params->fail_over_mac = fail_over_mac_value;
6301 params->tx_queues = tx_queues;
6302 params->all_slaves_active = all_slaves_active;
6303 params->resend_igmp = resend_igmp;
6304 params->min_links = min_links;
6305 params->lp_interval = lp_interval;
6306 params->packets_per_slave = packets_per_slave;
6307 params->tlb_dynamic_lb = tlb_dynamic_lb;
6308 params->ad_actor_sys_prio = ad_actor_sys_prio;
6309 eth_zero_addr(params->ad_actor_system);
6310 params->ad_user_port_key = ad_user_port_key;
6311 if (packets_per_slave > 0) {
6312 params->reciprocal_packets_per_slave =
6313 reciprocal_value(packets_per_slave);
6315 /* reciprocal_packets_per_slave is unused if
6316 * packets_per_slave is 0 or 1, just initialize it
6318 params->reciprocal_packets_per_slave =
6319 (struct reciprocal_value) { 0 };
6323 strscpy_pad(params->primary, primary, sizeof(params->primary));
6325 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6326 #if IS_ENABLED(CONFIG_IPV6)
6327 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6333 /* Called from registration process */
6334 static int bond_init(struct net_device *bond_dev)
6336 struct bonding *bond = netdev_priv(bond_dev);
6337 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6339 netdev_dbg(bond_dev, "Begin bond_init\n");
6341 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6345 spin_lock_init(&bond->stats_lock);
6346 netdev_lockdep_set_classes(bond_dev);
6348 list_add_tail(&bond->bond_list, &bn->dev_list);
6350 bond_prepare_sysfs_group(bond);
6352 bond_debug_register(bond);
6354 /* Ensure valid dev_addr */
6355 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6356 bond_dev->addr_assign_type == NET_ADDR_PERM)
6357 eth_hw_addr_random(bond_dev);
6362 unsigned int bond_get_num_tx_queues(void)
6367 /* Create a new bond based on the specified name and bonding parameters.
6368 * If name is NULL, obtain a suitable "bond%d" name for us.
6369 * Caller must NOT hold rtnl_lock; we need to release it here before we
6370 * set up our sysfs entries.
6372 int bond_create(struct net *net, const char *name)
6374 struct net_device *bond_dev;
6375 struct bonding *bond;
6380 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6381 name ? name : "bond%d", NET_NAME_UNKNOWN,
6382 bond_setup, tx_queues);
6386 bond = netdev_priv(bond_dev);
6387 dev_net_set(bond_dev, net);
6388 bond_dev->rtnl_link_ops = &bond_link_ops;
6390 res = register_netdevice(bond_dev);
6392 free_netdev(bond_dev);
6396 netif_carrier_off(bond_dev);
6398 bond_work_init_all(bond);
6405 static int __net_init bond_net_init(struct net *net)
6407 struct bond_net *bn = net_generic(net, bond_net_id);
6410 INIT_LIST_HEAD(&bn->dev_list);
6412 bond_create_proc_dir(bn);
6413 bond_create_sysfs(bn);
6418 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6420 struct bond_net *bn;
6424 list_for_each_entry(net, net_list, exit_list) {
6425 bn = net_generic(net, bond_net_id);
6426 bond_destroy_sysfs(bn);
6429 /* Kill off any bonds created after unregistering bond rtnl ops */
6431 list_for_each_entry(net, net_list, exit_list) {
6432 struct bonding *bond, *tmp_bond;
6434 bn = net_generic(net, bond_net_id);
6435 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6436 unregister_netdevice_queue(bond->dev, &list);
6438 unregister_netdevice_many(&list);
6441 list_for_each_entry(net, net_list, exit_list) {
6442 bn = net_generic(net, bond_net_id);
6443 bond_destroy_proc_dir(bn);
6447 static struct pernet_operations bond_net_ops = {
6448 .init = bond_net_init,
6449 .exit_batch = bond_net_exit_batch,
6451 .size = sizeof(struct bond_net),
6454 static int __init bonding_init(void)
6459 res = bond_check_params(&bonding_defaults);
6463 res = register_pernet_subsys(&bond_net_ops);
6467 res = bond_netlink_init();
6471 bond_create_debugfs();
6473 for (i = 0; i < max_bonds; i++) {
6474 res = bond_create(&init_net, NULL);
6479 skb_flow_dissector_init(&flow_keys_bonding,
6480 flow_keys_bonding_keys,
6481 ARRAY_SIZE(flow_keys_bonding_keys));
6483 register_netdevice_notifier(&bond_netdev_notifier);
6487 bond_destroy_debugfs();
6488 bond_netlink_fini();
6490 unregister_pernet_subsys(&bond_net_ops);
6495 static void __exit bonding_exit(void)
6497 unregister_netdevice_notifier(&bond_netdev_notifier);
6499 bond_destroy_debugfs();
6501 bond_netlink_fini();
6502 unregister_pernet_subsys(&bond_net_ops);
6504 #ifdef CONFIG_NET_POLL_CONTROLLER
6505 /* Make sure we don't have an imbalance on our netpoll blocking */
6506 WARN_ON(atomic_read(&netpoll_block_tx));
6510 module_init(bonding_init);
6511 module_exit(bonding_exit);
6512 MODULE_LICENSE("GPL");
6513 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6514 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");