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 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
423 static int bond_ipsec_add_sa(struct xfrm_state *xs)
425 struct net_device *bond_dev = xs->xso.dev;
426 struct bond_ipsec *ipsec;
427 struct bonding *bond;
435 bond = netdev_priv(bond_dev);
436 slave = rcu_dereference(bond->curr_active_slave);
442 if (!slave->dev->xfrmdev_ops ||
443 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
444 netif_is_bond_master(slave->dev)) {
445 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
450 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
455 xs->xso.real_dev = slave->dev;
457 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
460 INIT_LIST_HEAD(&ipsec->list);
461 spin_lock_bh(&bond->ipsec_lock);
462 list_add(&ipsec->list, &bond->ipsec_list);
463 spin_unlock_bh(&bond->ipsec_lock);
471 static void bond_ipsec_add_sa_all(struct bonding *bond)
473 struct net_device *bond_dev = bond->dev;
474 struct bond_ipsec *ipsec;
478 slave = rcu_dereference(bond->curr_active_slave);
482 if (!slave->dev->xfrmdev_ops ||
483 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
484 netif_is_bond_master(slave->dev)) {
485 spin_lock_bh(&bond->ipsec_lock);
486 if (!list_empty(&bond->ipsec_list))
487 slave_warn(bond_dev, slave->dev,
488 "%s: no slave xdo_dev_state_add\n",
490 spin_unlock_bh(&bond->ipsec_lock);
494 spin_lock_bh(&bond->ipsec_lock);
495 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
496 ipsec->xs->xso.real_dev = slave->dev;
497 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
498 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
499 ipsec->xs->xso.real_dev = NULL;
502 spin_unlock_bh(&bond->ipsec_lock);
508 * bond_ipsec_del_sa - clear out this specific SA
509 * @xs: pointer to transformer state struct
511 static void bond_ipsec_del_sa(struct xfrm_state *xs)
513 struct net_device *bond_dev = xs->xso.dev;
514 struct bond_ipsec *ipsec;
515 struct bonding *bond;
522 bond = netdev_priv(bond_dev);
523 slave = rcu_dereference(bond->curr_active_slave);
528 if (!xs->xso.real_dev)
531 WARN_ON(xs->xso.real_dev != slave->dev);
533 if (!slave->dev->xfrmdev_ops ||
534 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
535 netif_is_bond_master(slave->dev)) {
536 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
540 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
542 spin_lock_bh(&bond->ipsec_lock);
543 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
544 if (ipsec->xs == xs) {
545 list_del(&ipsec->list);
550 spin_unlock_bh(&bond->ipsec_lock);
554 static void bond_ipsec_del_sa_all(struct bonding *bond)
556 struct net_device *bond_dev = bond->dev;
557 struct bond_ipsec *ipsec;
561 slave = rcu_dereference(bond->curr_active_slave);
567 spin_lock_bh(&bond->ipsec_lock);
568 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
569 if (!ipsec->xs->xso.real_dev)
572 if (!slave->dev->xfrmdev_ops ||
573 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
574 netif_is_bond_master(slave->dev)) {
575 slave_warn(bond_dev, slave->dev,
576 "%s: no slave xdo_dev_state_delete\n",
579 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
581 ipsec->xs->xso.real_dev = NULL;
583 spin_unlock_bh(&bond->ipsec_lock);
588 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
589 * @skb: current data packet
590 * @xs: pointer to transformer state struct
592 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
594 struct net_device *bond_dev = xs->xso.dev;
595 struct net_device *real_dev;
596 struct slave *curr_active;
597 struct bonding *bond;
600 bond = netdev_priv(bond_dev);
602 curr_active = rcu_dereference(bond->curr_active_slave);
603 real_dev = curr_active->dev;
605 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
610 if (!xs->xso.real_dev) {
615 if (!real_dev->xfrmdev_ops ||
616 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
617 netif_is_bond_master(real_dev)) {
622 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
628 static const struct xfrmdev_ops bond_xfrmdev_ops = {
629 .xdo_dev_state_add = bond_ipsec_add_sa,
630 .xdo_dev_state_delete = bond_ipsec_del_sa,
631 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
633 #endif /* CONFIG_XFRM_OFFLOAD */
635 /*------------------------------- Link status -------------------------------*/
637 /* Set the carrier state for the master according to the state of its
638 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
639 * do special 802.3ad magic.
641 * Returns zero if carrier state does not change, nonzero if it does.
643 int bond_set_carrier(struct bonding *bond)
645 struct list_head *iter;
648 if (!bond_has_slaves(bond))
651 if (BOND_MODE(bond) == BOND_MODE_8023AD)
652 return bond_3ad_set_carrier(bond);
654 bond_for_each_slave(bond, slave, iter) {
655 if (slave->link == BOND_LINK_UP) {
656 if (!netif_carrier_ok(bond->dev)) {
657 netif_carrier_on(bond->dev);
665 if (netif_carrier_ok(bond->dev)) {
666 netif_carrier_off(bond->dev);
672 /* Get link speed and duplex from the slave's base driver
673 * using ethtool. If for some reason the call fails or the
674 * values are invalid, set speed and duplex to -1,
675 * and return. Return 1 if speed or duplex settings are
676 * UNKNOWN; 0 otherwise.
678 static int bond_update_speed_duplex(struct slave *slave)
680 struct net_device *slave_dev = slave->dev;
681 struct ethtool_link_ksettings ecmd;
684 slave->speed = SPEED_UNKNOWN;
685 slave->duplex = DUPLEX_UNKNOWN;
687 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
690 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
692 switch (ecmd.base.duplex) {
700 slave->speed = ecmd.base.speed;
701 slave->duplex = ecmd.base.duplex;
706 const char *bond_slave_link_status(s8 link)
722 /* if <dev> supports MII link status reporting, check its link status.
724 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
725 * depending upon the setting of the use_carrier parameter.
727 * Return either BMSR_LSTATUS, meaning that the link is up (or we
728 * can't tell and just pretend it is), or 0, meaning that the link is
731 * If reporting is non-zero, instead of faking link up, return -1 if
732 * both ETHTOOL and MII ioctls fail (meaning the device does not
733 * support them). If use_carrier is set, return whatever it says.
734 * It'd be nice if there was a good way to tell if a driver supports
735 * netif_carrier, but there really isn't.
737 static int bond_check_dev_link(struct bonding *bond,
738 struct net_device *slave_dev, int reporting)
740 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
741 int (*ioctl)(struct net_device *, struct ifreq *, int);
743 struct mii_ioctl_data *mii;
745 if (!reporting && !netif_running(slave_dev))
748 if (bond->params.use_carrier)
749 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
751 /* Try to get link status using Ethtool first. */
752 if (slave_dev->ethtool_ops->get_link)
753 return slave_dev->ethtool_ops->get_link(slave_dev) ?
756 /* Ethtool can't be used, fallback to MII ioctls. */
757 ioctl = slave_ops->ndo_eth_ioctl;
759 /* TODO: set pointer to correct ioctl on a per team member
760 * bases to make this more efficient. that is, once
761 * we determine the correct ioctl, we will always
762 * call it and not the others for that team
766 /* We cannot assume that SIOCGMIIPHY will also read a
767 * register; not all network drivers (e.g., e100)
771 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
772 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
774 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
775 mii->reg_num = MII_BMSR;
776 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
777 return mii->val_out & BMSR_LSTATUS;
781 /* If reporting, report that either there's no ndo_eth_ioctl,
782 * or both SIOCGMIIREG and get_link failed (meaning that we
783 * cannot report link status). If not reporting, pretend
786 return reporting ? -1 : BMSR_LSTATUS;
789 /*----------------------------- Multicast list ------------------------------*/
791 /* Push the promiscuity flag down to appropriate slaves */
792 static int bond_set_promiscuity(struct bonding *bond, int inc)
794 struct list_head *iter;
797 if (bond_uses_primary(bond)) {
798 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
801 err = dev_set_promiscuity(curr_active->dev, inc);
805 bond_for_each_slave(bond, slave, iter) {
806 err = dev_set_promiscuity(slave->dev, inc);
814 /* Push the allmulti flag down to all slaves */
815 static int bond_set_allmulti(struct bonding *bond, int inc)
817 struct list_head *iter;
820 if (bond_uses_primary(bond)) {
821 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
824 err = dev_set_allmulti(curr_active->dev, inc);
828 bond_for_each_slave(bond, slave, iter) {
829 err = dev_set_allmulti(slave->dev, inc);
837 /* Retrieve the list of registered multicast addresses for the bonding
838 * device and retransmit an IGMP JOIN request to the current active
841 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
843 struct bonding *bond = container_of(work, struct bonding,
846 if (!rtnl_trylock()) {
847 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
850 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
852 if (bond->igmp_retrans > 1) {
853 bond->igmp_retrans--;
854 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
859 /* Flush bond's hardware addresses from slave */
860 static void bond_hw_addr_flush(struct net_device *bond_dev,
861 struct net_device *slave_dev)
863 struct bonding *bond = netdev_priv(bond_dev);
865 dev_uc_unsync(slave_dev, bond_dev);
866 dev_mc_unsync(slave_dev, bond_dev);
868 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
869 /* del lacpdu mc addr from mc list */
870 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
872 dev_mc_del(slave_dev, lacpdu_multicast);
876 /*--------------------------- Active slave change ---------------------------*/
878 /* Update the hardware address list and promisc/allmulti for the new and
879 * old active slaves (if any). Modes that are not using primary keep all
880 * slaves up date at all times; only the modes that use primary need to call
881 * this function to swap these settings during a failover.
883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
884 struct slave *old_active)
887 if (bond->dev->flags & IFF_PROMISC)
888 dev_set_promiscuity(old_active->dev, -1);
890 if (bond->dev->flags & IFF_ALLMULTI)
891 dev_set_allmulti(old_active->dev, -1);
893 bond_hw_addr_flush(bond->dev, old_active->dev);
897 /* FIXME: Signal errors upstream. */
898 if (bond->dev->flags & IFF_PROMISC)
899 dev_set_promiscuity(new_active->dev, 1);
901 if (bond->dev->flags & IFF_ALLMULTI)
902 dev_set_allmulti(new_active->dev, 1);
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);
912 * bond_set_dev_addr - clone slave's address to bond
913 * @bond_dev: bond net device
914 * @slave_dev: slave net device
916 * Should be called with RTNL held.
918 static int bond_set_dev_addr(struct net_device *bond_dev,
919 struct net_device *slave_dev)
923 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
924 bond_dev, slave_dev, slave_dev->addr_len);
925 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
929 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
930 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
931 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
935 static struct slave *bond_get_old_active(struct bonding *bond,
936 struct slave *new_active)
939 struct list_head *iter;
941 bond_for_each_slave(bond, slave, iter) {
942 if (slave == new_active)
945 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
952 /* bond_do_fail_over_mac
954 * Perform special MAC address swapping for fail_over_mac settings
958 static void bond_do_fail_over_mac(struct bonding *bond,
959 struct slave *new_active,
960 struct slave *old_active)
962 u8 tmp_mac[MAX_ADDR_LEN];
963 struct sockaddr_storage ss;
966 switch (bond->params.fail_over_mac) {
967 case BOND_FOM_ACTIVE:
969 rv = bond_set_dev_addr(bond->dev, new_active->dev);
971 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
975 case BOND_FOM_FOLLOW:
976 /* if new_active && old_active, swap them
977 * if just old_active, do nothing (going to no active slave)
978 * if just new_active, set new_active to bond's MAC
984 old_active = bond_get_old_active(bond, new_active);
987 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
988 new_active->dev->addr_len);
989 bond_hw_addr_copy(ss.__data,
990 old_active->dev->dev_addr,
991 old_active->dev->addr_len);
992 ss.ss_family = new_active->dev->type;
994 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
995 bond->dev->addr_len);
996 ss.ss_family = bond->dev->type;
999 rv = dev_set_mac_address(new_active->dev,
1000 (struct sockaddr *)&ss, NULL);
1002 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1010 bond_hw_addr_copy(ss.__data, tmp_mac,
1011 new_active->dev->addr_len);
1012 ss.ss_family = old_active->dev->type;
1014 rv = dev_set_mac_address(old_active->dev,
1015 (struct sockaddr *)&ss, NULL);
1017 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1022 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1023 bond->params.fail_over_mac);
1030 * bond_choose_primary_or_current - select the primary or high priority slave
1031 * @bond: our bonding struct
1033 * - Check if there is a primary link. If the primary link was set and is up,
1034 * go on and do link reselection.
1036 * - If primary link is not set or down, find the highest priority link.
1037 * If the highest priority link is not current slave, set it as primary
1038 * link and do link reselection.
1040 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1042 struct slave *prim = rtnl_dereference(bond->primary_slave);
1043 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1044 struct slave *slave, *hprio = NULL;
1045 struct list_head *iter;
1047 if (!prim || prim->link != BOND_LINK_UP) {
1048 bond_for_each_slave(bond, slave, iter) {
1049 if (slave->link == BOND_LINK_UP) {
1050 hprio = hprio ?: slave;
1051 if (slave->prio > hprio->prio)
1056 if (hprio && hprio != curr) {
1061 if (!curr || curr->link != BOND_LINK_UP)
1066 if (bond->force_primary) {
1067 bond->force_primary = false;
1072 if (!curr || curr->link != BOND_LINK_UP)
1075 /* At this point, prim and curr are both up */
1076 switch (bond->params.primary_reselect) {
1077 case BOND_PRI_RESELECT_ALWAYS:
1079 case BOND_PRI_RESELECT_BETTER:
1080 if (prim->speed < curr->speed)
1082 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1085 case BOND_PRI_RESELECT_FAILURE:
1088 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1089 bond->params.primary_reselect);
1095 * bond_find_best_slave - select the best available slave to be the active one
1096 * @bond: our bonding struct
1098 static struct slave *bond_find_best_slave(struct bonding *bond)
1100 struct slave *slave, *bestslave = NULL;
1101 struct list_head *iter;
1102 int mintime = bond->params.updelay;
1104 slave = bond_choose_primary_or_current(bond);
1108 bond_for_each_slave(bond, slave, iter) {
1109 if (slave->link == BOND_LINK_UP)
1111 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1112 slave->delay < mintime) {
1113 mintime = slave->delay;
1121 static bool bond_should_notify_peers(struct bonding *bond)
1123 struct slave *slave;
1126 slave = rcu_dereference(bond->curr_active_slave);
1129 if (!slave || !bond->send_peer_notif ||
1130 bond->send_peer_notif %
1131 max(1, bond->params.peer_notif_delay) != 0 ||
1132 !netif_carrier_ok(bond->dev) ||
1133 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1136 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1137 slave ? slave->dev->name : "NULL");
1143 * bond_change_active_slave - change the active slave into the specified one
1144 * @bond: our bonding struct
1145 * @new_active: the new slave to make the active one
1147 * Set the new slave to the bond's settings and unset them on the old
1148 * curr_active_slave.
1149 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1151 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1152 * because it is apparently the best available slave we have, even though its
1153 * updelay hasn't timed out yet.
1155 * Caller must hold RTNL.
1157 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1159 struct slave *old_active;
1163 old_active = rtnl_dereference(bond->curr_active_slave);
1165 if (old_active == new_active)
1168 #ifdef CONFIG_XFRM_OFFLOAD
1169 bond_ipsec_del_sa_all(bond);
1170 #endif /* CONFIG_XFRM_OFFLOAD */
1173 new_active->last_link_up = jiffies;
1175 if (new_active->link == BOND_LINK_BACK) {
1176 if (bond_uses_primary(bond)) {
1177 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1178 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1181 new_active->delay = 0;
1182 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1183 BOND_SLAVE_NOTIFY_NOW);
1185 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1186 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1188 if (bond_is_lb(bond))
1189 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1191 if (bond_uses_primary(bond))
1192 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1196 if (bond_uses_primary(bond))
1197 bond_hw_addr_swap(bond, new_active, old_active);
1199 if (bond_is_lb(bond)) {
1200 bond_alb_handle_active_change(bond, new_active);
1202 bond_set_slave_inactive_flags(old_active,
1203 BOND_SLAVE_NOTIFY_NOW);
1205 bond_set_slave_active_flags(new_active,
1206 BOND_SLAVE_NOTIFY_NOW);
1208 rcu_assign_pointer(bond->curr_active_slave, new_active);
1211 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1213 bond_set_slave_inactive_flags(old_active,
1214 BOND_SLAVE_NOTIFY_NOW);
1217 bool should_notify_peers = false;
1219 bond_set_slave_active_flags(new_active,
1220 BOND_SLAVE_NOTIFY_NOW);
1222 if (bond->params.fail_over_mac)
1223 bond_do_fail_over_mac(bond, new_active,
1226 if (netif_running(bond->dev)) {
1227 bond->send_peer_notif =
1228 bond->params.num_peer_notif *
1229 max(1, bond->params.peer_notif_delay);
1230 should_notify_peers =
1231 bond_should_notify_peers(bond);
1234 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1235 if (should_notify_peers) {
1236 bond->send_peer_notif--;
1237 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1243 #ifdef CONFIG_XFRM_OFFLOAD
1244 bond_ipsec_add_sa_all(bond);
1245 #endif /* CONFIG_XFRM_OFFLOAD */
1247 /* resend IGMP joins since active slave has changed or
1248 * all were sent on curr_active_slave.
1249 * resend only if bond is brought up with the affected
1250 * bonding modes and the retransmission is enabled
1252 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1253 ((bond_uses_primary(bond) && new_active) ||
1254 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1255 bond->igmp_retrans = bond->params.resend_igmp;
1256 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1261 * bond_select_active_slave - select a new active slave, if needed
1262 * @bond: our bonding struct
1264 * This functions should be called when one of the following occurs:
1265 * - The old curr_active_slave has been released or lost its link.
1266 * - The primary_slave has got its link back.
1267 * - A slave has got its link back and there's no old curr_active_slave.
1269 * Caller must hold RTNL.
1271 void bond_select_active_slave(struct bonding *bond)
1273 struct slave *best_slave;
1278 best_slave = bond_find_best_slave(bond);
1279 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1280 bond_change_active_slave(bond, best_slave);
1281 rv = bond_set_carrier(bond);
1285 if (netif_carrier_ok(bond->dev))
1286 netdev_info(bond->dev, "active interface up!\n");
1288 netdev_info(bond->dev, "now running without any active interface!\n");
1292 #ifdef CONFIG_NET_POLL_CONTROLLER
1293 static inline int slave_enable_netpoll(struct slave *slave)
1298 np = kzalloc(sizeof(*np), GFP_KERNEL);
1303 err = __netpoll_setup(np, slave->dev);
1312 static inline void slave_disable_netpoll(struct slave *slave)
1314 struct netpoll *np = slave->np;
1324 static void bond_poll_controller(struct net_device *bond_dev)
1326 struct bonding *bond = netdev_priv(bond_dev);
1327 struct slave *slave = NULL;
1328 struct list_head *iter;
1329 struct ad_info ad_info;
1331 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1332 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1335 bond_for_each_slave_rcu(bond, slave, iter) {
1336 if (!bond_slave_is_up(slave))
1339 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1340 struct aggregator *agg =
1341 SLAVE_AD_INFO(slave)->port.aggregator;
1344 agg->aggregator_identifier != ad_info.aggregator_id)
1348 netpoll_poll_dev(slave->dev);
1352 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1354 struct bonding *bond = netdev_priv(bond_dev);
1355 struct list_head *iter;
1356 struct slave *slave;
1358 bond_for_each_slave(bond, slave, iter)
1359 if (bond_slave_is_up(slave))
1360 slave_disable_netpoll(slave);
1363 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1365 struct bonding *bond = netdev_priv(dev);
1366 struct list_head *iter;
1367 struct slave *slave;
1370 bond_for_each_slave(bond, slave, iter) {
1371 err = slave_enable_netpoll(slave);
1373 bond_netpoll_cleanup(dev);
1380 static inline int slave_enable_netpoll(struct slave *slave)
1384 static inline void slave_disable_netpoll(struct slave *slave)
1387 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1392 /*---------------------------------- IOCTL ----------------------------------*/
1394 static netdev_features_t bond_fix_features(struct net_device *dev,
1395 netdev_features_t features)
1397 struct bonding *bond = netdev_priv(dev);
1398 struct list_head *iter;
1399 netdev_features_t mask;
1400 struct slave *slave;
1402 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1403 if (bond_sk_check(bond))
1404 features |= BOND_TLS_FEATURES;
1406 features &= ~BOND_TLS_FEATURES;
1411 features &= ~NETIF_F_ONE_FOR_ALL;
1412 features |= NETIF_F_ALL_FOR_ALL;
1414 bond_for_each_slave(bond, slave, iter) {
1415 features = netdev_increment_features(features,
1416 slave->dev->features,
1419 features = netdev_add_tso_features(features, mask);
1424 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1425 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1426 NETIF_F_HIGHDMA | NETIF_F_LRO)
1428 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1429 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1431 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1432 NETIF_F_GSO_SOFTWARE)
1435 static void bond_compute_features(struct bonding *bond)
1437 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1438 IFF_XMIT_DST_RELEASE_PERM;
1439 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1440 netdev_features_t enc_features = BOND_ENC_FEATURES;
1441 #ifdef CONFIG_XFRM_OFFLOAD
1442 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1443 #endif /* CONFIG_XFRM_OFFLOAD */
1444 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1445 struct net_device *bond_dev = bond->dev;
1446 struct list_head *iter;
1447 struct slave *slave;
1448 unsigned short max_hard_header_len = ETH_HLEN;
1449 unsigned int tso_max_size = TSO_MAX_SIZE;
1450 u16 tso_max_segs = TSO_MAX_SEGS;
1452 if (!bond_has_slaves(bond))
1454 vlan_features &= NETIF_F_ALL_FOR_ALL;
1455 mpls_features &= NETIF_F_ALL_FOR_ALL;
1457 bond_for_each_slave(bond, slave, iter) {
1458 vlan_features = netdev_increment_features(vlan_features,
1459 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1461 enc_features = netdev_increment_features(enc_features,
1462 slave->dev->hw_enc_features,
1465 #ifdef CONFIG_XFRM_OFFLOAD
1466 xfrm_features = netdev_increment_features(xfrm_features,
1467 slave->dev->hw_enc_features,
1468 BOND_XFRM_FEATURES);
1469 #endif /* CONFIG_XFRM_OFFLOAD */
1471 mpls_features = netdev_increment_features(mpls_features,
1472 slave->dev->mpls_features,
1473 BOND_MPLS_FEATURES);
1475 dst_release_flag &= slave->dev->priv_flags;
1476 if (slave->dev->hard_header_len > max_hard_header_len)
1477 max_hard_header_len = slave->dev->hard_header_len;
1479 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1480 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1482 bond_dev->hard_header_len = max_hard_header_len;
1485 bond_dev->vlan_features = vlan_features;
1486 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1487 NETIF_F_HW_VLAN_CTAG_TX |
1488 NETIF_F_HW_VLAN_STAG_TX;
1489 #ifdef CONFIG_XFRM_OFFLOAD
1490 bond_dev->hw_enc_features |= xfrm_features;
1491 #endif /* CONFIG_XFRM_OFFLOAD */
1492 bond_dev->mpls_features = mpls_features;
1493 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1494 netif_set_tso_max_size(bond_dev, tso_max_size);
1496 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1497 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1498 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1499 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1501 netdev_change_features(bond_dev);
1504 static void bond_setup_by_slave(struct net_device *bond_dev,
1505 struct net_device *slave_dev)
1507 bond_dev->header_ops = slave_dev->header_ops;
1509 bond_dev->type = slave_dev->type;
1510 bond_dev->hard_header_len = slave_dev->hard_header_len;
1511 bond_dev->needed_headroom = slave_dev->needed_headroom;
1512 bond_dev->addr_len = slave_dev->addr_len;
1514 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1515 slave_dev->addr_len);
1518 /* On bonding slaves other than the currently active slave, suppress
1519 * duplicates except for alb non-mcast/bcast.
1521 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1522 struct slave *slave,
1523 struct bonding *bond)
1525 if (bond_is_slave_inactive(slave)) {
1526 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1527 skb->pkt_type != PACKET_BROADCAST &&
1528 skb->pkt_type != PACKET_MULTICAST)
1535 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1537 struct sk_buff *skb = *pskb;
1538 struct slave *slave;
1539 struct bonding *bond;
1540 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1542 int ret = RX_HANDLER_ANOTHER;
1544 skb = skb_share_check(skb, GFP_ATOMIC);
1546 return RX_HANDLER_CONSUMED;
1550 slave = bond_slave_get_rcu(skb->dev);
1553 recv_probe = READ_ONCE(bond->recv_probe);
1555 ret = recv_probe(skb, bond, slave);
1556 if (ret == RX_HANDLER_CONSUMED) {
1563 * For packets determined by bond_should_deliver_exact_match() call to
1564 * be suppressed we want to make an exception for link-local packets.
1565 * This is necessary for e.g. LLDP daemons to be able to monitor
1566 * inactive slave links without being forced to bind to them
1569 * At the same time, packets that are passed to the bonding master
1570 * (including link-local ones) can have their originating interface
1571 * determined via PACKET_ORIGDEV socket option.
1573 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1574 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1575 return RX_HANDLER_PASS;
1576 return RX_HANDLER_EXACT;
1579 skb->dev = bond->dev;
1581 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1582 netif_is_bridge_port(bond->dev) &&
1583 skb->pkt_type == PACKET_HOST) {
1585 if (unlikely(skb_cow_head(skb,
1586 skb->data - skb_mac_header(skb)))) {
1588 return RX_HANDLER_CONSUMED;
1590 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1591 bond->dev->addr_len);
1597 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1599 switch (BOND_MODE(bond)) {
1600 case BOND_MODE_ROUNDROBIN:
1601 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1602 case BOND_MODE_ACTIVEBACKUP:
1603 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1604 case BOND_MODE_BROADCAST:
1605 return NETDEV_LAG_TX_TYPE_BROADCAST;
1607 case BOND_MODE_8023AD:
1608 return NETDEV_LAG_TX_TYPE_HASH;
1610 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1614 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1615 enum netdev_lag_tx_type type)
1617 if (type != NETDEV_LAG_TX_TYPE_HASH)
1618 return NETDEV_LAG_HASH_NONE;
1620 switch (bond->params.xmit_policy) {
1621 case BOND_XMIT_POLICY_LAYER2:
1622 return NETDEV_LAG_HASH_L2;
1623 case BOND_XMIT_POLICY_LAYER34:
1624 return NETDEV_LAG_HASH_L34;
1625 case BOND_XMIT_POLICY_LAYER23:
1626 return NETDEV_LAG_HASH_L23;
1627 case BOND_XMIT_POLICY_ENCAP23:
1628 return NETDEV_LAG_HASH_E23;
1629 case BOND_XMIT_POLICY_ENCAP34:
1630 return NETDEV_LAG_HASH_E34;
1631 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1632 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1634 return NETDEV_LAG_HASH_UNKNOWN;
1638 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1639 struct netlink_ext_ack *extack)
1641 struct netdev_lag_upper_info lag_upper_info;
1642 enum netdev_lag_tx_type type;
1644 type = bond_lag_tx_type(bond);
1645 lag_upper_info.tx_type = type;
1646 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1648 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1649 &lag_upper_info, extack);
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 /* enslave device <slave> to bond device <master> */
1779 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1780 struct netlink_ext_ack *extack)
1782 struct bonding *bond = netdev_priv(bond_dev);
1783 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1784 struct slave *new_slave = NULL, *prev_slave;
1785 struct sockaddr_storage ss;
1789 if (slave_dev->flags & IFF_MASTER &&
1790 !netif_is_bond_master(slave_dev)) {
1791 BOND_NL_ERR(bond_dev, extack,
1792 "Device type (master device) cannot be enslaved");
1796 if (!bond->params.use_carrier &&
1797 slave_dev->ethtool_ops->get_link == NULL &&
1798 slave_ops->ndo_eth_ioctl == NULL) {
1799 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1802 /* already in-use? */
1803 if (netdev_is_rx_handler_busy(slave_dev)) {
1804 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1805 "Device is in use and cannot be enslaved");
1809 if (bond_dev == slave_dev) {
1810 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1814 /* vlan challenged mutual exclusion */
1815 /* no need to lock since we're protected by rtnl_lock */
1816 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1817 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1818 if (vlan_uses_dev(bond_dev)) {
1819 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1820 "Can not enslave VLAN challenged device to VLAN enabled bond");
1823 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1826 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1829 if (slave_dev->features & NETIF_F_HW_ESP)
1830 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1832 /* Old ifenslave binaries are no longer supported. These can
1833 * be identified with moderate accuracy by the state of the slave:
1834 * the current ifenslave will set the interface down prior to
1835 * enslaving it; the old ifenslave will not.
1837 if (slave_dev->flags & IFF_UP) {
1838 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1839 "Device can not be enslaved while up");
1843 /* set bonding device ether type by slave - bonding netdevices are
1844 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1845 * there is a need to override some of the type dependent attribs/funcs.
1847 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1848 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1850 if (!bond_has_slaves(bond)) {
1851 if (bond_dev->type != slave_dev->type) {
1852 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1853 bond_dev->type, slave_dev->type);
1855 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1857 res = notifier_to_errno(res);
1859 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1863 /* Flush unicast and multicast addresses */
1864 dev_uc_flush(bond_dev);
1865 dev_mc_flush(bond_dev);
1867 if (slave_dev->type != ARPHRD_ETHER)
1868 bond_setup_by_slave(bond_dev, slave_dev);
1870 ether_setup(bond_dev);
1871 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1874 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1877 } else if (bond_dev->type != slave_dev->type) {
1878 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1879 "Device type is different from other slaves");
1883 if (slave_dev->type == ARPHRD_INFINIBAND &&
1884 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1885 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1886 "Only active-backup mode is supported for infiniband slaves");
1888 goto err_undo_flags;
1891 if (!slave_ops->ndo_set_mac_address ||
1892 slave_dev->type == ARPHRD_INFINIBAND) {
1893 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1894 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1895 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1896 if (!bond_has_slaves(bond)) {
1897 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1898 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1900 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1901 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1903 goto err_undo_flags;
1908 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1910 /* If this is the first slave, then we need to set the master's hardware
1911 * address to be the same as the slave's.
1913 if (!bond_has_slaves(bond) &&
1914 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1915 res = bond_set_dev_addr(bond->dev, slave_dev);
1917 goto err_undo_flags;
1920 new_slave = bond_alloc_slave(bond, slave_dev);
1923 goto err_undo_flags;
1926 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1927 * is set via sysfs or module option if desired.
1929 new_slave->queue_id = 0;
1931 /* Save slave's original mtu and then set it to match the bond */
1932 new_slave->original_mtu = slave_dev->mtu;
1933 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1935 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1939 /* Save slave's original ("permanent") mac address for modes
1940 * that need it, and for restoring it upon release, and then
1941 * set it to the master's address
1943 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1944 slave_dev->addr_len);
1946 if (!bond->params.fail_over_mac ||
1947 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1948 /* Set slave to master's mac address. The application already
1949 * set the master's mac address to that of the first slave
1951 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1952 ss.ss_family = slave_dev->type;
1953 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1956 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1957 goto err_restore_mtu;
1961 /* set slave flag before open to prevent IPv6 addrconf */
1962 slave_dev->flags |= IFF_SLAVE;
1964 /* open the slave since the application closed it */
1965 res = dev_open(slave_dev, extack);
1967 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1968 goto err_restore_mac;
1971 slave_dev->priv_flags |= IFF_BONDING;
1972 /* initialize slave stats */
1973 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1975 if (bond_is_lb(bond)) {
1976 /* bond_alb_init_slave() must be called before all other stages since
1977 * it might fail and we do not want to have to undo everything
1979 res = bond_alb_init_slave(bond, new_slave);
1984 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1986 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1990 prev_slave = bond_last_slave(bond);
1992 new_slave->delay = 0;
1993 new_slave->link_failure_count = 0;
1995 if (bond_update_speed_duplex(new_slave) &&
1996 bond_needs_speed_duplex(bond))
1997 new_slave->link = BOND_LINK_DOWN;
1999 new_slave->last_rx = jiffies -
2000 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2001 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2002 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2004 new_slave->last_tx = new_slave->last_rx;
2006 if (bond->params.miimon && !bond->params.use_carrier) {
2007 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2009 if ((link_reporting == -1) && !bond->params.arp_interval) {
2010 /* miimon is set but a bonded network driver
2011 * does not support ETHTOOL/MII and
2012 * arp_interval is not set. Note: if
2013 * use_carrier is enabled, we will never go
2014 * here (because netif_carrier is always
2015 * supported); thus, we don't need to change
2016 * the messages for netif_carrier.
2018 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");
2019 } else if (link_reporting == -1) {
2020 /* unable get link status using mii/ethtool */
2021 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");
2025 /* check for initial state */
2026 new_slave->link = BOND_LINK_NOCHANGE;
2027 if (bond->params.miimon) {
2028 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2029 if (bond->params.updelay) {
2030 bond_set_slave_link_state(new_slave,
2032 BOND_SLAVE_NOTIFY_NOW);
2033 new_slave->delay = bond->params.updelay;
2035 bond_set_slave_link_state(new_slave,
2037 BOND_SLAVE_NOTIFY_NOW);
2040 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2041 BOND_SLAVE_NOTIFY_NOW);
2043 } else if (bond->params.arp_interval) {
2044 bond_set_slave_link_state(new_slave,
2045 (netif_carrier_ok(slave_dev) ?
2046 BOND_LINK_UP : BOND_LINK_DOWN),
2047 BOND_SLAVE_NOTIFY_NOW);
2049 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2050 BOND_SLAVE_NOTIFY_NOW);
2053 if (new_slave->link != BOND_LINK_DOWN)
2054 new_slave->last_link_up = jiffies;
2055 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2056 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2057 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2059 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2060 /* if there is a primary slave, remember it */
2061 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2062 rcu_assign_pointer(bond->primary_slave, new_slave);
2063 bond->force_primary = true;
2067 switch (BOND_MODE(bond)) {
2068 case BOND_MODE_ACTIVEBACKUP:
2069 bond_set_slave_inactive_flags(new_slave,
2070 BOND_SLAVE_NOTIFY_NOW);
2072 case BOND_MODE_8023AD:
2073 /* in 802.3ad mode, the internal mechanism
2074 * will activate the slaves in the selected
2077 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2078 /* if this is the first slave */
2080 SLAVE_AD_INFO(new_slave)->id = 1;
2081 /* Initialize AD with the number of times that the AD timer is called in 1 second
2082 * can be called only after the mac address of the bond is set
2084 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2086 SLAVE_AD_INFO(new_slave)->id =
2087 SLAVE_AD_INFO(prev_slave)->id + 1;
2090 bond_3ad_bind_slave(new_slave);
2094 bond_set_active_slave(new_slave);
2095 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2098 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2100 /* always active in trunk mode */
2101 bond_set_active_slave(new_slave);
2103 /* In trunking mode there is little meaning to curr_active_slave
2104 * anyway (it holds no special properties of the bond device),
2105 * so we can change it without calling change_active_interface()
2107 if (!rcu_access_pointer(bond->curr_active_slave) &&
2108 new_slave->link == BOND_LINK_UP)
2109 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2112 } /* switch(bond_mode) */
2114 #ifdef CONFIG_NET_POLL_CONTROLLER
2115 if (bond->dev->npinfo) {
2116 if (slave_enable_netpoll(new_slave)) {
2117 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2124 if (!(bond_dev->features & NETIF_F_LRO))
2125 dev_disable_lro(slave_dev);
2127 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2130 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2134 res = bond_master_upper_dev_link(bond, new_slave, extack);
2136 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2137 goto err_unregister;
2140 bond_lower_state_changed(new_slave);
2142 res = bond_sysfs_slave_add(new_slave);
2144 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2145 goto err_upper_unlink;
2148 /* If the mode uses primary, then the following is handled by
2149 * bond_change_active_slave().
2151 if (!bond_uses_primary(bond)) {
2152 /* set promiscuity level to new slave */
2153 if (bond_dev->flags & IFF_PROMISC) {
2154 res = dev_set_promiscuity(slave_dev, 1);
2159 /* set allmulti level to new slave */
2160 if (bond_dev->flags & IFF_ALLMULTI) {
2161 res = dev_set_allmulti(slave_dev, 1);
2163 if (bond_dev->flags & IFF_PROMISC)
2164 dev_set_promiscuity(slave_dev, -1);
2169 netif_addr_lock_bh(bond_dev);
2170 dev_mc_sync_multiple(slave_dev, bond_dev);
2171 dev_uc_sync_multiple(slave_dev, bond_dev);
2172 netif_addr_unlock_bh(bond_dev);
2174 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2175 /* add lacpdu mc addr to mc list */
2176 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2178 dev_mc_add(slave_dev, lacpdu_multicast);
2183 bond_compute_features(bond);
2184 bond_set_carrier(bond);
2186 if (bond_uses_primary(bond)) {
2188 bond_select_active_slave(bond);
2189 unblock_netpoll_tx();
2192 if (bond_mode_can_use_xmit_hash(bond))
2193 bond_update_slave_arr(bond, NULL);
2196 if (!slave_dev->netdev_ops->ndo_bpf ||
2197 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2198 if (bond->xdp_prog) {
2199 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2200 "Slave does not support XDP");
2204 } else if (bond->xdp_prog) {
2205 struct netdev_bpf xdp = {
2206 .command = XDP_SETUP_PROG,
2208 .prog = bond->xdp_prog,
2212 if (dev_xdp_prog_count(slave_dev) > 0) {
2213 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2214 "Slave has XDP program loaded, please unload before enslaving");
2219 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2221 /* ndo_bpf() sets extack error message */
2222 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2226 bpf_prog_inc(bond->xdp_prog);
2229 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2230 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2231 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2233 /* enslave is successful */
2234 bond_queue_slave_event(new_slave);
2237 /* Undo stages on error */
2239 bond_sysfs_slave_del(new_slave);
2242 bond_upper_dev_unlink(bond, new_slave);
2245 netdev_rx_handler_unregister(slave_dev);
2248 vlan_vids_del_by_dev(slave_dev, bond_dev);
2249 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2250 RCU_INIT_POINTER(bond->primary_slave, NULL);
2251 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2253 bond_change_active_slave(bond, NULL);
2254 bond_select_active_slave(bond);
2255 unblock_netpoll_tx();
2257 /* either primary_slave or curr_active_slave might've changed */
2259 slave_disable_netpoll(new_slave);
2262 if (!netif_is_bond_master(slave_dev))
2263 slave_dev->priv_flags &= ~IFF_BONDING;
2264 dev_close(slave_dev);
2267 slave_dev->flags &= ~IFF_SLAVE;
2268 if (!bond->params.fail_over_mac ||
2269 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2270 /* XXX TODO - fom follow mode needs to change master's
2271 * MAC if this slave's MAC is in use by the bond, or at
2272 * least print a warning.
2274 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2275 new_slave->dev->addr_len);
2276 ss.ss_family = slave_dev->type;
2277 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2281 dev_set_mtu(slave_dev, new_slave->original_mtu);
2284 kobject_put(&new_slave->kobj);
2287 /* Enslave of first slave has failed and we need to fix master's mac */
2288 if (!bond_has_slaves(bond)) {
2289 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2290 slave_dev->dev_addr))
2291 eth_hw_addr_random(bond_dev);
2292 if (bond_dev->type != ARPHRD_ETHER) {
2293 dev_close(bond_dev);
2294 ether_setup(bond_dev);
2295 bond_dev->flags |= IFF_MASTER;
2296 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2303 /* Try to release the slave device <slave> from the bond device <master>
2304 * It is legal to access curr_active_slave without a lock because all the function
2305 * is RTNL-locked. If "all" is true it means that the function is being called
2306 * while destroying a bond interface and all slaves are being released.
2308 * The rules for slave state should be:
2309 * for Active/Backup:
2310 * Active stays on all backups go down
2311 * for Bonded connections:
2312 * The first up interface should be left on and all others downed.
2314 static int __bond_release_one(struct net_device *bond_dev,
2315 struct net_device *slave_dev,
2316 bool all, bool unregister)
2318 struct bonding *bond = netdev_priv(bond_dev);
2319 struct slave *slave, *oldcurrent;
2320 struct sockaddr_storage ss;
2321 int old_flags = bond_dev->flags;
2322 netdev_features_t old_features = bond_dev->features;
2324 /* slave is not a slave or master is not master of this slave */
2325 if (!(slave_dev->flags & IFF_SLAVE) ||
2326 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2327 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2333 slave = bond_get_slave_by_dev(bond, slave_dev);
2335 /* not a slave of this bond */
2336 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2337 unblock_netpoll_tx();
2341 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2343 bond_sysfs_slave_del(slave);
2345 /* recompute stats just before removing the slave */
2346 bond_get_stats(bond->dev, &bond->bond_stats);
2348 if (bond->xdp_prog) {
2349 struct netdev_bpf xdp = {
2350 .command = XDP_SETUP_PROG,
2355 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2356 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2359 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2360 * for this slave anymore.
2362 netdev_rx_handler_unregister(slave_dev);
2364 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2365 bond_3ad_unbind_slave(slave);
2367 bond_upper_dev_unlink(bond, slave);
2369 if (bond_mode_can_use_xmit_hash(bond))
2370 bond_update_slave_arr(bond, slave);
2372 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2373 bond_is_active_slave(slave) ? "active" : "backup");
2375 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2377 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2379 if (!all && (!bond->params.fail_over_mac ||
2380 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2381 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2382 bond_has_slaves(bond))
2383 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",
2384 slave->perm_hwaddr);
2387 if (rtnl_dereference(bond->primary_slave) == slave)
2388 RCU_INIT_POINTER(bond->primary_slave, NULL);
2390 if (oldcurrent == slave)
2391 bond_change_active_slave(bond, NULL);
2393 if (bond_is_lb(bond)) {
2394 /* Must be called only after the slave has been
2395 * detached from the list and the curr_active_slave
2396 * has been cleared (if our_slave == old_current),
2397 * but before a new active slave is selected.
2399 bond_alb_deinit_slave(bond, slave);
2403 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2404 } else if (oldcurrent == slave) {
2405 /* Note that we hold RTNL over this sequence, so there
2406 * is no concern that another slave add/remove event
2409 bond_select_active_slave(bond);
2412 bond_set_carrier(bond);
2413 if (!bond_has_slaves(bond))
2414 eth_hw_addr_random(bond_dev);
2416 unblock_netpoll_tx();
2420 if (!bond_has_slaves(bond)) {
2421 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2422 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2425 bond_compute_features(bond);
2426 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2427 (old_features & NETIF_F_VLAN_CHALLENGED))
2428 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2430 vlan_vids_del_by_dev(slave_dev, bond_dev);
2432 /* If the mode uses primary, then this case was handled above by
2433 * bond_change_active_slave(..., NULL)
2435 if (!bond_uses_primary(bond)) {
2436 /* unset promiscuity level from slave
2437 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2438 * of the IFF_PROMISC flag in the bond_dev, but we need the
2439 * value of that flag before that change, as that was the value
2440 * when this slave was attached, so we cache at the start of the
2441 * function and use it here. Same goes for ALLMULTI below
2443 if (old_flags & IFF_PROMISC)
2444 dev_set_promiscuity(slave_dev, -1);
2446 /* unset allmulti level from slave */
2447 if (old_flags & IFF_ALLMULTI)
2448 dev_set_allmulti(slave_dev, -1);
2450 bond_hw_addr_flush(bond_dev, slave_dev);
2453 slave_disable_netpoll(slave);
2455 /* close slave before restoring its mac address */
2456 dev_close(slave_dev);
2458 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2459 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2460 /* restore original ("permanent") mac address */
2461 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2462 slave->dev->addr_len);
2463 ss.ss_family = slave_dev->type;
2464 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2468 __dev_set_mtu(slave_dev, slave->original_mtu);
2470 dev_set_mtu(slave_dev, slave->original_mtu);
2472 if (!netif_is_bond_master(slave_dev))
2473 slave_dev->priv_flags &= ~IFF_BONDING;
2475 kobject_put(&slave->kobj);
2480 /* A wrapper used because of ndo_del_link */
2481 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2483 return __bond_release_one(bond_dev, slave_dev, false, false);
2486 /* First release a slave and then destroy the bond if no more slaves are left.
2487 * Must be under rtnl_lock when this function is called.
2489 static int bond_release_and_destroy(struct net_device *bond_dev,
2490 struct net_device *slave_dev)
2492 struct bonding *bond = netdev_priv(bond_dev);
2495 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2496 if (ret == 0 && !bond_has_slaves(bond) &&
2497 bond_dev->reg_state != NETREG_UNREGISTERING) {
2498 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2499 netdev_info(bond_dev, "Destroying bond\n");
2500 bond_remove_proc_entry(bond);
2501 unregister_netdevice(bond_dev);
2506 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2508 struct bonding *bond = netdev_priv(bond_dev);
2510 bond_fill_ifbond(bond, info);
2513 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2515 struct bonding *bond = netdev_priv(bond_dev);
2516 struct list_head *iter;
2517 int i = 0, res = -ENODEV;
2518 struct slave *slave;
2520 bond_for_each_slave(bond, slave, iter) {
2521 if (i++ == (int)info->slave_id) {
2523 bond_fill_ifslave(slave, info);
2531 /*-------------------------------- Monitoring -------------------------------*/
2533 /* called with rcu_read_lock() */
2534 static int bond_miimon_inspect(struct bonding *bond)
2536 int link_state, commit = 0;
2537 struct list_head *iter;
2538 struct slave *slave;
2539 bool ignore_updelay;
2541 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2543 bond_for_each_slave_rcu(bond, slave, iter) {
2544 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2546 link_state = bond_check_dev_link(bond, slave->dev, 0);
2548 switch (slave->link) {
2553 bond_propose_link_state(slave, BOND_LINK_FAIL);
2555 slave->delay = bond->params.downdelay;
2557 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2559 BOND_MODE_ACTIVEBACKUP) ?
2560 (bond_is_active_slave(slave) ?
2561 "active " : "backup ") : "",
2562 bond->params.downdelay * bond->params.miimon);
2565 case BOND_LINK_FAIL:
2567 /* recovered before downdelay expired */
2568 bond_propose_link_state(slave, BOND_LINK_UP);
2569 slave->last_link_up = jiffies;
2570 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2571 (bond->params.downdelay - slave->delay) *
2572 bond->params.miimon);
2577 if (slave->delay <= 0) {
2578 bond_propose_link_state(slave, BOND_LINK_DOWN);
2586 case BOND_LINK_DOWN:
2590 bond_propose_link_state(slave, BOND_LINK_BACK);
2592 slave->delay = bond->params.updelay;
2595 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2596 ignore_updelay ? 0 :
2597 bond->params.updelay *
2598 bond->params.miimon);
2601 case BOND_LINK_BACK:
2603 bond_propose_link_state(slave, BOND_LINK_DOWN);
2604 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2605 (bond->params.updelay - slave->delay) *
2606 bond->params.miimon);
2614 if (slave->delay <= 0) {
2615 bond_propose_link_state(slave, BOND_LINK_UP);
2617 ignore_updelay = false;
2629 static void bond_miimon_link_change(struct bonding *bond,
2630 struct slave *slave,
2633 switch (BOND_MODE(bond)) {
2634 case BOND_MODE_8023AD:
2635 bond_3ad_handle_link_change(slave, link);
2639 bond_alb_handle_link_change(bond, slave, link);
2642 bond_update_slave_arr(bond, NULL);
2647 static void bond_miimon_commit(struct bonding *bond)
2649 struct list_head *iter;
2650 struct slave *slave, *primary;
2652 bond_for_each_slave(bond, slave, iter) {
2653 switch (slave->link_new_state) {
2654 case BOND_LINK_NOCHANGE:
2655 /* For 802.3ad mode, check current slave speed and
2656 * duplex again in case its port was disabled after
2657 * invalid speed/duplex reporting but recovered before
2658 * link monitoring could make a decision on the actual
2661 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2662 slave->link == BOND_LINK_UP)
2663 bond_3ad_adapter_speed_duplex_changed(slave);
2667 if (bond_update_speed_duplex(slave) &&
2668 bond_needs_speed_duplex(bond)) {
2669 slave->link = BOND_LINK_DOWN;
2670 if (net_ratelimit())
2671 slave_warn(bond->dev, slave->dev,
2672 "failed to get link speed/duplex\n");
2675 bond_set_slave_link_state(slave, BOND_LINK_UP,
2676 BOND_SLAVE_NOTIFY_NOW);
2677 slave->last_link_up = jiffies;
2679 primary = rtnl_dereference(bond->primary_slave);
2680 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2681 /* prevent it from being the active one */
2682 bond_set_backup_slave(slave);
2683 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2684 /* make it immediately active */
2685 bond_set_active_slave(slave);
2688 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2689 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2690 slave->duplex ? "full" : "half");
2692 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2694 if (!bond->curr_active_slave || slave == primary)
2699 case BOND_LINK_DOWN:
2700 if (slave->link_failure_count < UINT_MAX)
2701 slave->link_failure_count++;
2703 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2704 BOND_SLAVE_NOTIFY_NOW);
2706 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2707 BOND_MODE(bond) == BOND_MODE_8023AD)
2708 bond_set_slave_inactive_flags(slave,
2709 BOND_SLAVE_NOTIFY_NOW);
2711 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2713 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2715 if (slave == rcu_access_pointer(bond->curr_active_slave))
2721 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2722 slave->link_new_state);
2723 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2730 bond_select_active_slave(bond);
2731 unblock_netpoll_tx();
2734 bond_set_carrier(bond);
2739 * Really a wrapper that splits the mii monitor into two phases: an
2740 * inspection, then (if inspection indicates something needs to be done)
2741 * an acquisition of appropriate locks followed by a commit phase to
2742 * implement whatever link state changes are indicated.
2744 static void bond_mii_monitor(struct work_struct *work)
2746 struct bonding *bond = container_of(work, struct bonding,
2748 bool should_notify_peers = false;
2750 unsigned long delay;
2751 struct slave *slave;
2752 struct list_head *iter;
2754 delay = msecs_to_jiffies(bond->params.miimon);
2756 if (!bond_has_slaves(bond))
2760 should_notify_peers = bond_should_notify_peers(bond);
2761 commit = !!bond_miimon_inspect(bond);
2762 if (bond->send_peer_notif) {
2764 if (rtnl_trylock()) {
2765 bond->send_peer_notif--;
2773 /* Race avoidance with bond_close cancel of workqueue */
2774 if (!rtnl_trylock()) {
2776 should_notify_peers = false;
2780 bond_for_each_slave(bond, slave, iter) {
2781 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2783 bond_miimon_commit(bond);
2785 rtnl_unlock(); /* might sleep, hold no other locks */
2789 if (bond->params.miimon)
2790 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2792 if (should_notify_peers) {
2793 if (!rtnl_trylock())
2795 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2800 static int bond_upper_dev_walk(struct net_device *upper,
2801 struct netdev_nested_priv *priv)
2803 __be32 ip = *(__be32 *)priv->data;
2805 return ip == bond_confirm_addr(upper, 0, ip);
2808 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2810 struct netdev_nested_priv priv = {
2811 .data = (void *)&ip,
2815 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2819 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2826 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2827 struct sk_buff *skb)
2829 struct net_device *bond_dev = slave->bond->dev;
2830 struct net_device *slave_dev = slave->dev;
2831 struct bond_vlan_tag *outer_tag = tags;
2833 if (!tags || tags->vlan_proto == VLAN_N_VID)
2838 /* Go through all the tags backwards and add them to the packet */
2839 while (tags->vlan_proto != VLAN_N_VID) {
2840 if (!tags->vlan_id) {
2845 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2846 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2847 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2850 net_err_ratelimited("failed to insert inner VLAN tag\n");
2856 /* Set the outer tag */
2857 if (outer_tag->vlan_id) {
2858 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2859 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2860 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2861 outer_tag->vlan_id);
2867 /* We go to the (large) trouble of VLAN tagging ARP frames because
2868 * switches in VLAN mode (especially if ports are configured as
2869 * "native" to a VLAN) might not pass non-tagged frames.
2871 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2872 __be32 src_ip, struct bond_vlan_tag *tags)
2874 struct net_device *bond_dev = slave->bond->dev;
2875 struct net_device *slave_dev = slave->dev;
2876 struct sk_buff *skb;
2878 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2879 arp_op, &dest_ip, &src_ip);
2881 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2882 NULL, slave_dev->dev_addr, NULL);
2885 net_err_ratelimited("ARP packet allocation failed\n");
2889 if (bond_handle_vlan(slave, tags, skb)) {
2890 slave_update_last_tx(slave);
2897 /* Validate the device path between the @start_dev and the @end_dev.
2898 * The path is valid if the @end_dev is reachable through device
2900 * When the path is validated, collect any vlan information in the
2903 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2904 struct net_device *end_dev,
2907 struct bond_vlan_tag *tags;
2908 struct net_device *upper;
2909 struct list_head *iter;
2911 if (start_dev == end_dev) {
2912 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2914 return ERR_PTR(-ENOMEM);
2915 tags[level].vlan_proto = VLAN_N_VID;
2919 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2920 tags = bond_verify_device_path(upper, end_dev, level + 1);
2921 if (IS_ERR_OR_NULL(tags)) {
2926 if (is_vlan_dev(upper)) {
2927 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2928 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2937 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2940 struct bond_vlan_tag *tags;
2941 __be32 *targets = bond->params.arp_targets, addr;
2944 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2945 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2946 __func__, &targets[i]);
2949 /* Find out through which dev should the packet go */
2950 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2953 /* there's no route to target - try to send arp
2954 * probe to generate any traffic (arp_validate=0)
2956 if (bond->params.arp_validate)
2957 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2960 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2965 /* bond device itself */
2966 if (rt->dst.dev == bond->dev)
2970 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2973 if (!IS_ERR_OR_NULL(tags))
2976 /* Not our device - skip */
2977 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2978 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2984 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2986 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2991 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2995 if (!sip || !bond_has_this_ip(bond, tip)) {
2996 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2997 __func__, &sip, &tip);
3001 i = bond_get_targets_ip(bond->params.arp_targets, sip);
3003 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3007 slave->last_rx = jiffies;
3008 slave->target_last_arp_rx[i] = jiffies;
3011 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3012 struct slave *slave)
3014 struct arphdr *arp = (struct arphdr *)skb->data;
3015 struct slave *curr_active_slave, *curr_arp_slave;
3016 unsigned char *arp_ptr;
3020 alen = arp_hdr_len(bond->dev);
3022 if (alen > skb_headlen(skb)) {
3023 arp = kmalloc(alen, GFP_ATOMIC);
3026 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3030 if (arp->ar_hln != bond->dev->addr_len ||
3031 skb->pkt_type == PACKET_OTHERHOST ||
3032 skb->pkt_type == PACKET_LOOPBACK ||
3033 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3034 arp->ar_pro != htons(ETH_P_IP) ||
3038 arp_ptr = (unsigned char *)(arp + 1);
3039 arp_ptr += bond->dev->addr_len;
3040 memcpy(&sip, arp_ptr, 4);
3041 arp_ptr += 4 + bond->dev->addr_len;
3042 memcpy(&tip, arp_ptr, 4);
3044 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3045 __func__, slave->dev->name, bond_slave_state(slave),
3046 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3049 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3050 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3052 /* We 'trust' the received ARP enough to validate it if:
3054 * (a) the slave receiving the ARP is active (which includes the
3055 * current ARP slave, if any), or
3057 * (b) the receiving slave isn't active, but there is a currently
3058 * active slave and it received valid arp reply(s) after it became
3059 * the currently active slave, or
3061 * (c) there is an ARP slave that sent an ARP during the prior ARP
3062 * interval, and we receive an ARP reply on any slave. We accept
3063 * these because switch FDB update delays may deliver the ARP
3064 * reply to a slave other than the sender of the ARP request.
3066 * Note: for (b), backup slaves are receiving the broadcast ARP
3067 * request, not a reply. This request passes from the sending
3068 * slave through the L2 switch(es) to the receiving slave. Since
3069 * this is checking the request, sip/tip are swapped for
3072 * This is done to avoid endless looping when we can't reach the
3073 * arp_ip_target and fool ourselves with our own arp requests.
3075 if (bond_is_active_slave(slave))
3076 bond_validate_arp(bond, slave, sip, tip);
3077 else if (curr_active_slave &&
3078 time_after(slave_last_rx(bond, curr_active_slave),
3079 curr_active_slave->last_link_up))
3080 bond_validate_arp(bond, slave, tip, sip);
3081 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3082 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3083 bond_validate_arp(bond, slave, sip, tip);
3086 if (arp != (struct arphdr *)skb->data)
3088 return RX_HANDLER_ANOTHER;
3091 #if IS_ENABLED(CONFIG_IPV6)
3092 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3093 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3095 struct net_device *bond_dev = slave->bond->dev;
3096 struct net_device *slave_dev = slave->dev;
3097 struct in6_addr mcaddr;
3098 struct sk_buff *skb;
3100 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3103 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3105 net_err_ratelimited("NS packet allocation failed\n");
3109 addrconf_addr_solict_mult(daddr, &mcaddr);
3110 if (bond_handle_vlan(slave, tags, skb)) {
3111 slave_update_last_tx(slave);
3112 ndisc_send_skb(skb, &mcaddr, saddr);
3116 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3118 struct in6_addr *targets = bond->params.ns_targets;
3119 struct bond_vlan_tag *tags;
3120 struct dst_entry *dst;
3121 struct in6_addr saddr;
3125 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3126 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3127 __func__, &targets[i]);
3130 /* Find out through which dev should the packet go */
3131 memset(&fl6, 0, sizeof(struct flowi6));
3132 fl6.daddr = targets[i];
3133 fl6.flowi6_oif = bond->dev->ifindex;
3135 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3138 /* there's no route to target - try to send arp
3139 * probe to generate any traffic (arp_validate=0)
3141 if (bond->params.arp_validate)
3142 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3145 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3149 /* bond device itself */
3150 if (dst->dev == bond->dev)
3154 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3157 if (!IS_ERR_OR_NULL(tags))
3160 /* Not our device - skip */
3161 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3162 &targets[i], dst->dev ? dst->dev->name : "NULL");
3168 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3169 bond_ns_send(slave, &targets[i], &saddr, tags);
3175 static int bond_confirm_addr6(struct net_device *dev,
3176 struct netdev_nested_priv *priv)
3178 struct in6_addr *addr = (struct in6_addr *)priv->data;
3180 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3183 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3185 struct netdev_nested_priv priv = {
3190 if (bond_confirm_addr6(bond->dev, &priv))
3194 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3201 static void bond_validate_ns(struct bonding *bond, struct slave *slave,
3202 struct in6_addr *saddr, struct in6_addr *daddr)
3206 if (ipv6_addr_any(saddr) || !bond_has_this_ip6(bond, daddr)) {
3207 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3208 __func__, saddr, daddr);
3212 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3214 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3218 slave->last_rx = jiffies;
3219 slave->target_last_arp_rx[i] = jiffies;
3222 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3223 struct slave *slave)
3225 struct slave *curr_active_slave, *curr_arp_slave;
3226 struct icmp6hdr *hdr = icmp6_hdr(skb);
3227 struct in6_addr *saddr, *daddr;
3229 if (skb->pkt_type == PACKET_OTHERHOST ||
3230 skb->pkt_type == PACKET_LOOPBACK ||
3231 hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
3234 saddr = &ipv6_hdr(skb)->saddr;
3235 daddr = &ipv6_hdr(skb)->daddr;
3237 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3238 __func__, slave->dev->name, bond_slave_state(slave),
3239 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3242 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3243 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3245 /* We 'trust' the received ARP enough to validate it if:
3246 * see bond_arp_rcv().
3248 if (bond_is_active_slave(slave))
3249 bond_validate_ns(bond, slave, saddr, daddr);
3250 else if (curr_active_slave &&
3251 time_after(slave_last_rx(bond, curr_active_slave),
3252 curr_active_slave->last_link_up))
3253 bond_validate_ns(bond, slave, saddr, daddr);
3254 else if (curr_arp_slave &&
3255 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3256 bond_validate_ns(bond, slave, saddr, daddr);
3259 return RX_HANDLER_ANOTHER;
3263 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3264 struct slave *slave)
3266 #if IS_ENABLED(CONFIG_IPV6)
3267 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3269 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3271 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3272 __func__, skb->dev->name);
3274 /* Use arp validate logic for both ARP and NS */
3275 if (!slave_do_arp_validate(bond, slave)) {
3276 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3277 #if IS_ENABLED(CONFIG_IPV6)
3278 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3280 !slave_do_arp_validate_only(bond))
3281 slave->last_rx = jiffies;
3282 return RX_HANDLER_ANOTHER;
3283 } else if (is_arp) {
3284 return bond_arp_rcv(skb, bond, slave);
3285 #if IS_ENABLED(CONFIG_IPV6)
3286 } else if (is_ipv6) {
3287 return bond_na_rcv(skb, bond, slave);
3290 return RX_HANDLER_ANOTHER;
3294 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3296 bond_arp_send_all(bond, slave);
3297 #if IS_ENABLED(CONFIG_IPV6)
3298 bond_ns_send_all(bond, slave);
3302 /* function to verify if we're in the arp_interval timeslice, returns true if
3303 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3304 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3306 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3309 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3311 return time_in_range(jiffies,
3312 last_act - delta_in_ticks,
3313 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3316 /* This function is called regularly to monitor each slave's link
3317 * ensuring that traffic is being sent and received when arp monitoring
3318 * is used in load-balancing mode. if the adapter has been dormant, then an
3319 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3320 * arp monitoring in active backup mode.
3322 static void bond_loadbalance_arp_mon(struct bonding *bond)
3324 struct slave *slave, *oldcurrent;
3325 struct list_head *iter;
3326 int do_failover = 0, slave_state_changed = 0;
3328 if (!bond_has_slaves(bond))
3333 oldcurrent = rcu_dereference(bond->curr_active_slave);
3334 /* see if any of the previous devices are up now (i.e. they have
3335 * xmt and rcv traffic). the curr_active_slave does not come into
3336 * the picture unless it is null. also, slave->last_link_up is not
3337 * needed here because we send an arp on each slave and give a slave
3338 * as long as it needs to get the tx/rx within the delta.
3339 * TODO: what about up/down delay in arp mode? it wasn't here before
3342 bond_for_each_slave_rcu(bond, slave, iter) {
3343 unsigned long last_tx = slave_last_tx(slave);
3345 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3347 if (slave->link != BOND_LINK_UP) {
3348 if (bond_time_in_interval(bond, last_tx, 1) &&
3349 bond_time_in_interval(bond, slave->last_rx, 1)) {
3351 bond_propose_link_state(slave, BOND_LINK_UP);
3352 slave_state_changed = 1;
3354 /* primary_slave has no meaning in round-robin
3355 * mode. the window of a slave being up and
3356 * curr_active_slave being null after enslaving
3360 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3363 slave_info(bond->dev, slave->dev, "interface is now up\n");
3367 /* slave->link == BOND_LINK_UP */
3369 /* not all switches will respond to an arp request
3370 * when the source ip is 0, so don't take the link down
3371 * if we don't know our ip yet
3373 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3374 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3376 bond_propose_link_state(slave, BOND_LINK_DOWN);
3377 slave_state_changed = 1;
3379 if (slave->link_failure_count < UINT_MAX)
3380 slave->link_failure_count++;
3382 slave_info(bond->dev, slave->dev, "interface is now down\n");
3384 if (slave == oldcurrent)
3389 /* note: if switch is in round-robin mode, all links
3390 * must tx arp to ensure all links rx an arp - otherwise
3391 * links may oscillate or not come up at all; if switch is
3392 * in something like xor mode, there is nothing we can
3393 * do - all replies will be rx'ed on same link causing slaves
3394 * to be unstable during low/no traffic periods
3396 if (bond_slave_is_up(slave))
3397 bond_send_validate(bond, slave);
3402 if (do_failover || slave_state_changed) {
3403 if (!rtnl_trylock())
3406 bond_for_each_slave(bond, slave, iter) {
3407 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3408 slave->link = slave->link_new_state;
3411 if (slave_state_changed) {
3412 bond_slave_state_change(bond);
3413 if (BOND_MODE(bond) == BOND_MODE_XOR)
3414 bond_update_slave_arr(bond, NULL);
3418 bond_select_active_slave(bond);
3419 unblock_netpoll_tx();
3425 if (bond->params.arp_interval)
3426 queue_delayed_work(bond->wq, &bond->arp_work,
3427 msecs_to_jiffies(bond->params.arp_interval));
3430 /* Called to inspect slaves for active-backup mode ARP monitor link state
3431 * changes. Sets proposed link state in slaves to specify what action
3432 * should take place for the slave. Returns 0 if no changes are found, >0
3433 * if changes to link states must be committed.
3435 * Called with rcu_read_lock held.
3437 static int bond_ab_arp_inspect(struct bonding *bond)
3439 unsigned long last_tx, last_rx;
3440 struct list_head *iter;
3441 struct slave *slave;
3444 bond_for_each_slave_rcu(bond, slave, iter) {
3445 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3446 last_rx = slave_last_rx(bond, slave);
3448 if (slave->link != BOND_LINK_UP) {
3449 if (bond_time_in_interval(bond, last_rx, 1)) {
3450 bond_propose_link_state(slave, BOND_LINK_UP);
3452 } else if (slave->link == BOND_LINK_BACK) {
3453 bond_propose_link_state(slave, BOND_LINK_FAIL);
3459 /* Give slaves 2*delta after being enslaved or made
3460 * active. This avoids bouncing, as the last receive
3461 * times need a full ARP monitor cycle to be updated.
3463 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3466 /* Backup slave is down if:
3467 * - No current_arp_slave AND
3468 * - more than (missed_max+1)*delta since last receive AND
3469 * - the bond has an IP address
3471 * Note: a non-null current_arp_slave indicates
3472 * the curr_active_slave went down and we are
3473 * searching for a new one; under this condition
3474 * we only take the curr_active_slave down - this
3475 * gives each slave a chance to tx/rx traffic
3476 * before being taken out
3478 if (!bond_is_active_slave(slave) &&
3479 !rcu_access_pointer(bond->current_arp_slave) &&
3480 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3481 bond_propose_link_state(slave, BOND_LINK_DOWN);
3485 /* Active slave is down if:
3486 * - more than missed_max*delta since transmitting OR
3487 * - (more than missed_max*delta since receive AND
3488 * the bond has an IP address)
3490 last_tx = slave_last_tx(slave);
3491 if (bond_is_active_slave(slave) &&
3492 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3493 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3494 bond_propose_link_state(slave, BOND_LINK_DOWN);
3502 /* Called to commit link state changes noted by inspection step of
3503 * active-backup mode ARP monitor.
3505 * Called with RTNL hold.
3507 static void bond_ab_arp_commit(struct bonding *bond)
3509 struct list_head *iter;
3510 unsigned long last_tx;
3511 struct slave *slave;
3513 bond_for_each_slave(bond, slave, iter) {
3514 switch (slave->link_new_state) {
3515 case BOND_LINK_NOCHANGE:
3519 last_tx = slave_last_tx(slave);
3520 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3521 (!rtnl_dereference(bond->curr_active_slave) &&
3522 bond_time_in_interval(bond, last_tx, 1))) {
3523 struct slave *current_arp_slave;
3525 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3526 bond_set_slave_link_state(slave, BOND_LINK_UP,
3527 BOND_SLAVE_NOTIFY_NOW);
3528 if (current_arp_slave) {
3529 bond_set_slave_inactive_flags(
3531 BOND_SLAVE_NOTIFY_NOW);
3532 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3535 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3537 if (!rtnl_dereference(bond->curr_active_slave) ||
3538 slave == rtnl_dereference(bond->primary_slave))
3545 case BOND_LINK_DOWN:
3546 if (slave->link_failure_count < UINT_MAX)
3547 slave->link_failure_count++;
3549 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3550 BOND_SLAVE_NOTIFY_NOW);
3551 bond_set_slave_inactive_flags(slave,
3552 BOND_SLAVE_NOTIFY_NOW);
3554 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3556 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3557 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3563 case BOND_LINK_FAIL:
3564 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3565 BOND_SLAVE_NOTIFY_NOW);
3566 bond_set_slave_inactive_flags(slave,
3567 BOND_SLAVE_NOTIFY_NOW);
3569 /* A slave has just been enslaved and has become
3570 * the current active slave.
3572 if (rtnl_dereference(bond->curr_active_slave))
3573 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3577 slave_err(bond->dev, slave->dev,
3578 "impossible: link_new_state %d on slave\n",
3579 slave->link_new_state);
3585 bond_select_active_slave(bond);
3586 unblock_netpoll_tx();
3589 bond_set_carrier(bond);
3592 /* Send ARP probes for active-backup mode ARP monitor.
3594 * Called with rcu_read_lock held.
3596 static bool bond_ab_arp_probe(struct bonding *bond)
3598 struct slave *slave, *before = NULL, *new_slave = NULL,
3599 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3600 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3601 struct list_head *iter;
3603 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3605 if (curr_arp_slave && curr_active_slave)
3606 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3607 curr_arp_slave->dev->name,
3608 curr_active_slave->dev->name);
3610 if (curr_active_slave) {
3611 bond_send_validate(bond, curr_active_slave);
3612 return should_notify_rtnl;
3615 /* if we don't have a curr_active_slave, search for the next available
3616 * backup slave from the current_arp_slave and make it the candidate
3617 * for becoming the curr_active_slave
3620 if (!curr_arp_slave) {
3621 curr_arp_slave = bond_first_slave_rcu(bond);
3622 if (!curr_arp_slave)
3623 return should_notify_rtnl;
3626 bond_for_each_slave_rcu(bond, slave, iter) {
3627 if (!found && !before && bond_slave_is_up(slave))
3630 if (found && !new_slave && bond_slave_is_up(slave))
3632 /* if the link state is up at this point, we
3633 * mark it down - this can happen if we have
3634 * simultaneous link failures and
3635 * reselect_active_interface doesn't make this
3636 * one the current slave so it is still marked
3637 * up when it is actually down
3639 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3640 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3641 BOND_SLAVE_NOTIFY_LATER);
3642 if (slave->link_failure_count < UINT_MAX)
3643 slave->link_failure_count++;
3645 bond_set_slave_inactive_flags(slave,
3646 BOND_SLAVE_NOTIFY_LATER);
3648 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3650 if (slave == curr_arp_slave)
3654 if (!new_slave && before)
3660 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3661 BOND_SLAVE_NOTIFY_LATER);
3662 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3663 bond_send_validate(bond, new_slave);
3664 new_slave->last_link_up = jiffies;
3665 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3668 bond_for_each_slave_rcu(bond, slave, iter) {
3669 if (slave->should_notify || slave->should_notify_link) {
3670 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3674 return should_notify_rtnl;
3677 static void bond_activebackup_arp_mon(struct bonding *bond)
3679 bool should_notify_peers = false;
3680 bool should_notify_rtnl = false;
3683 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3685 if (!bond_has_slaves(bond))
3690 should_notify_peers = bond_should_notify_peers(bond);
3692 if (bond_ab_arp_inspect(bond)) {
3695 /* Race avoidance with bond_close flush of workqueue */
3696 if (!rtnl_trylock()) {
3698 should_notify_peers = false;
3702 bond_ab_arp_commit(bond);
3708 should_notify_rtnl = bond_ab_arp_probe(bond);
3712 if (bond->params.arp_interval)
3713 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3715 if (should_notify_peers || should_notify_rtnl) {
3716 if (!rtnl_trylock())
3719 if (should_notify_peers) {
3720 bond->send_peer_notif--;
3721 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3724 if (should_notify_rtnl) {
3725 bond_slave_state_notify(bond);
3726 bond_slave_link_notify(bond);
3733 static void bond_arp_monitor(struct work_struct *work)
3735 struct bonding *bond = container_of(work, struct bonding,
3738 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3739 bond_activebackup_arp_mon(bond);
3741 bond_loadbalance_arp_mon(bond);
3744 /*-------------------------- netdev event handling --------------------------*/
3746 /* Change device name */
3747 static int bond_event_changename(struct bonding *bond)
3749 bond_remove_proc_entry(bond);
3750 bond_create_proc_entry(bond);
3752 bond_debug_reregister(bond);
3757 static int bond_master_netdev_event(unsigned long event,
3758 struct net_device *bond_dev)
3760 struct bonding *event_bond = netdev_priv(bond_dev);
3762 netdev_dbg(bond_dev, "%s called\n", __func__);
3765 case NETDEV_CHANGENAME:
3766 return bond_event_changename(event_bond);
3767 case NETDEV_UNREGISTER:
3768 bond_remove_proc_entry(event_bond);
3769 #ifdef CONFIG_XFRM_OFFLOAD
3770 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3771 #endif /* CONFIG_XFRM_OFFLOAD */
3773 case NETDEV_REGISTER:
3774 bond_create_proc_entry(event_bond);
3783 static int bond_slave_netdev_event(unsigned long event,
3784 struct net_device *slave_dev)
3786 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3787 struct bonding *bond;
3788 struct net_device *bond_dev;
3790 /* A netdev event can be generated while enslaving a device
3791 * before netdev_rx_handler_register is called in which case
3792 * slave will be NULL
3795 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3799 bond_dev = slave->bond->dev;
3801 primary = rtnl_dereference(bond->primary_slave);
3803 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3806 case NETDEV_UNREGISTER:
3807 if (bond_dev->type != ARPHRD_ETHER)
3808 bond_release_and_destroy(bond_dev, slave_dev);
3810 __bond_release_one(bond_dev, slave_dev, false, true);
3814 /* For 802.3ad mode only:
3815 * Getting invalid Speed/Duplex values here will put slave
3816 * in weird state. Mark it as link-fail if the link was
3817 * previously up or link-down if it hasn't yet come up, and
3818 * let link-monitoring (miimon) set it right when correct
3819 * speeds/duplex are available.
3821 if (bond_update_speed_duplex(slave) &&
3822 BOND_MODE(bond) == BOND_MODE_8023AD) {
3823 if (slave->last_link_up)
3824 slave->link = BOND_LINK_FAIL;
3826 slave->link = BOND_LINK_DOWN;
3829 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3830 bond_3ad_adapter_speed_duplex_changed(slave);
3833 /* Refresh slave-array if applicable!
3834 * If the setup does not use miimon or arpmon (mode-specific!),
3835 * then these events will not cause the slave-array to be
3836 * refreshed. This will cause xmit to use a slave that is not
3837 * usable. Avoid such situation by refeshing the array at these
3838 * events. If these (miimon/arpmon) parameters are configured
3839 * then array gets refreshed twice and that should be fine!
3841 if (bond_mode_can_use_xmit_hash(bond))
3842 bond_update_slave_arr(bond, NULL);
3844 case NETDEV_CHANGEMTU:
3845 /* TODO: Should slaves be allowed to
3846 * independently alter their MTU? For
3847 * an active-backup bond, slaves need
3848 * not be the same type of device, so
3849 * MTUs may vary. For other modes,
3850 * slaves arguably should have the
3851 * same MTUs. To do this, we'd need to
3852 * take over the slave's change_mtu
3853 * function for the duration of their
3857 case NETDEV_CHANGENAME:
3858 /* we don't care if we don't have primary set */
3859 if (!bond_uses_primary(bond) ||
3860 !bond->params.primary[0])
3863 if (slave == primary) {
3864 /* slave's name changed - he's no longer primary */
3865 RCU_INIT_POINTER(bond->primary_slave, NULL);
3866 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3867 /* we have a new primary slave */
3868 rcu_assign_pointer(bond->primary_slave, slave);
3869 } else { /* we didn't change primary - exit */
3873 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3874 primary ? slave_dev->name : "none");
3877 bond_select_active_slave(bond);
3878 unblock_netpoll_tx();
3880 case NETDEV_FEAT_CHANGE:
3881 bond_compute_features(bond);
3883 case NETDEV_RESEND_IGMP:
3884 /* Propagate to master device */
3885 call_netdevice_notifiers(event, slave->bond->dev);
3894 /* bond_netdev_event: handle netdev notifier chain events.
3896 * This function receives events for the netdev chain. The caller (an
3897 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3898 * locks for us to safely manipulate the slave devices (RTNL lock,
3901 static int bond_netdev_event(struct notifier_block *this,
3902 unsigned long event, void *ptr)
3904 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3906 netdev_dbg(event_dev, "%s received %s\n",
3907 __func__, netdev_cmd_to_name(event));
3909 if (!(event_dev->priv_flags & IFF_BONDING))
3912 if (event_dev->flags & IFF_MASTER) {
3915 ret = bond_master_netdev_event(event, event_dev);
3916 if (ret != NOTIFY_DONE)
3920 if (event_dev->flags & IFF_SLAVE)
3921 return bond_slave_netdev_event(event, event_dev);
3926 static struct notifier_block bond_netdev_notifier = {
3927 .notifier_call = bond_netdev_event,
3930 /*---------------------------- Hashing Policies -----------------------------*/
3932 /* Helper to access data in a packet, with or without a backing skb.
3933 * If skb is given the data is linearized if necessary via pskb_may_pull.
3935 static inline const void *bond_pull_data(struct sk_buff *skb,
3936 const void *data, int hlen, int n)
3938 if (likely(n <= hlen))
3940 else if (skb && likely(pskb_may_pull(skb, n)))
3946 /* L2 hash helper */
3947 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3951 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3955 ep = (struct ethhdr *)(data + mhoff);
3956 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
3959 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
3960 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
3962 const struct ipv6hdr *iph6;
3963 const struct iphdr *iph;
3965 if (l2_proto == htons(ETH_P_IP)) {
3966 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
3970 iph = (const struct iphdr *)(data + *nhoff);
3971 iph_to_flow_copy_v4addrs(fk, iph);
3972 *nhoff += iph->ihl << 2;
3973 if (!ip_is_fragment(iph))
3974 *ip_proto = iph->protocol;
3975 } else if (l2_proto == htons(ETH_P_IPV6)) {
3976 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
3980 iph6 = (const struct ipv6hdr *)(data + *nhoff);
3981 iph_to_flow_copy_v6addrs(fk, iph6);
3982 *nhoff += sizeof(*iph6);
3983 *ip_proto = iph6->nexthdr;
3988 if (l34 && *ip_proto >= 0)
3989 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
3994 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3996 u32 srcmac_vendor = 0, srcmac_dev = 0;
3997 struct ethhdr *mac_hdr;
4001 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4004 mac_hdr = (struct ethhdr *)(data + mhoff);
4006 for (i = 0; i < 3; i++)
4007 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4009 for (i = 3; i < ETH_ALEN; i++)
4010 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4012 if (skb && skb_vlan_tag_present(skb))
4013 vlan = skb_vlan_tag_get(skb);
4015 return vlan ^ srcmac_vendor ^ srcmac_dev;
4018 /* Extract the appropriate headers based on bond's xmit policy */
4019 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4020 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4022 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4025 switch (bond->params.xmit_policy) {
4026 case BOND_XMIT_POLICY_ENCAP23:
4027 case BOND_XMIT_POLICY_ENCAP34:
4028 memset(fk, 0, sizeof(*fk));
4029 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4030 fk, data, l2_proto, nhoff, hlen, 0);
4035 fk->ports.ports = 0;
4036 memset(&fk->icmp, 0, sizeof(fk->icmp));
4037 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4040 /* ICMP error packets contains at least 8 bytes of the header
4041 * of the packet which generated the error. Use this information
4042 * to correlate ICMP error packets within the same flow which
4043 * generated the error.
4045 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4046 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4047 if (ip_proto == IPPROTO_ICMP) {
4048 if (!icmp_is_err(fk->icmp.type))
4051 nhoff += sizeof(struct icmphdr);
4052 } else if (ip_proto == IPPROTO_ICMPV6) {
4053 if (!icmpv6_is_err(fk->icmp.type))
4056 nhoff += sizeof(struct icmp6hdr);
4058 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4064 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4066 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4067 (__force u32)flow_get_u32_src(flow);
4068 hash ^= (hash >> 16);
4069 hash ^= (hash >> 8);
4071 /* discard lowest hash bit to deal with the common even ports pattern */
4072 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4073 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4079 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4080 * the data as required, but this function can be used without it if the data is
4081 * known to be linear (e.g. with xdp_buff).
4083 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4084 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4086 struct flow_keys flow;
4089 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4090 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4092 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4093 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4094 return bond_eth_hash(skb, data, mhoff, hlen);
4096 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4097 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4098 hash = bond_eth_hash(skb, data, mhoff, hlen);
4101 memcpy(&hash, &flow.icmp, sizeof(hash));
4103 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4106 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4110 * bond_xmit_hash - generate a hash value based on the xmit policy
4111 * @bond: bonding device
4112 * @skb: buffer to use for headers
4114 * This function will extract the necessary headers from the skb buffer and use
4115 * them to generate a hash based on the xmit_policy set in the bonding device
4117 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4119 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4123 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4124 skb_mac_offset(skb), skb_network_offset(skb),
4129 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4130 * @bond: bonding device
4131 * @xdp: buffer to use for headers
4133 * The XDP variant of bond_xmit_hash.
4135 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4139 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4142 eth = (struct ethhdr *)xdp->data;
4144 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4145 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4148 /*-------------------------- Device entry points ----------------------------*/
4150 void bond_work_init_all(struct bonding *bond)
4152 INIT_DELAYED_WORK(&bond->mcast_work,
4153 bond_resend_igmp_join_requests_delayed);
4154 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4155 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4156 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4157 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4158 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4161 static void bond_work_cancel_all(struct bonding *bond)
4163 cancel_delayed_work_sync(&bond->mii_work);
4164 cancel_delayed_work_sync(&bond->arp_work);
4165 cancel_delayed_work_sync(&bond->alb_work);
4166 cancel_delayed_work_sync(&bond->ad_work);
4167 cancel_delayed_work_sync(&bond->mcast_work);
4168 cancel_delayed_work_sync(&bond->slave_arr_work);
4171 static int bond_open(struct net_device *bond_dev)
4173 struct bonding *bond = netdev_priv(bond_dev);
4174 struct list_head *iter;
4175 struct slave *slave;
4177 /* reset slave->backup and slave->inactive */
4178 if (bond_has_slaves(bond)) {
4179 bond_for_each_slave(bond, slave, iter) {
4180 if (bond_uses_primary(bond) &&
4181 slave != rcu_access_pointer(bond->curr_active_slave)) {
4182 bond_set_slave_inactive_flags(slave,
4183 BOND_SLAVE_NOTIFY_NOW);
4184 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4185 bond_set_slave_active_flags(slave,
4186 BOND_SLAVE_NOTIFY_NOW);
4191 if (bond_is_lb(bond)) {
4192 /* bond_alb_initialize must be called before the timer
4195 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4197 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4198 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4201 if (bond->params.miimon) /* link check interval, in milliseconds. */
4202 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4204 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4205 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4206 bond->recv_probe = bond_rcv_validate;
4209 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4210 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4211 /* register to receive LACPDUs */
4212 bond->recv_probe = bond_3ad_lacpdu_recv;
4213 bond_3ad_initiate_agg_selection(bond, 1);
4216 if (bond_mode_can_use_xmit_hash(bond))
4217 bond_update_slave_arr(bond, NULL);
4222 static int bond_close(struct net_device *bond_dev)
4224 struct bonding *bond = netdev_priv(bond_dev);
4226 bond_work_cancel_all(bond);
4227 bond->send_peer_notif = 0;
4228 if (bond_is_lb(bond))
4229 bond_alb_deinitialize(bond);
4230 bond->recv_probe = NULL;
4235 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4236 * that some drivers can provide 32bit values only.
4238 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4239 const struct rtnl_link_stats64 *_new,
4240 const struct rtnl_link_stats64 *_old)
4242 const u64 *new = (const u64 *)_new;
4243 const u64 *old = (const u64 *)_old;
4244 u64 *res = (u64 *)_res;
4247 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4250 s64 delta = nv - ov;
4252 /* detects if this particular field is 32bit only */
4253 if (((nv | ov) >> 32) == 0)
4254 delta = (s64)(s32)((u32)nv - (u32)ov);
4256 /* filter anomalies, some drivers reset their stats
4257 * at down/up events.
4264 #ifdef CONFIG_LOCKDEP
4265 static int bond_get_lowest_level_rcu(struct net_device *dev)
4267 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4268 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4269 int cur = 0, max = 0;
4272 iter = &dev->adj_list.lower;
4277 ldev = netdev_next_lower_dev_rcu(now, &iter);
4282 niter = &ldev->adj_list.lower;
4283 dev_stack[cur] = now;
4284 iter_stack[cur++] = iter;
4293 next = dev_stack[--cur];
4294 niter = iter_stack[cur];
4305 static void bond_get_stats(struct net_device *bond_dev,
4306 struct rtnl_link_stats64 *stats)
4308 struct bonding *bond = netdev_priv(bond_dev);
4309 struct rtnl_link_stats64 temp;
4310 struct list_head *iter;
4311 struct slave *slave;
4316 #ifdef CONFIG_LOCKDEP
4317 nest_level = bond_get_lowest_level_rcu(bond_dev);
4320 spin_lock_nested(&bond->stats_lock, nest_level);
4321 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4323 bond_for_each_slave_rcu(bond, slave, iter) {
4324 const struct rtnl_link_stats64 *new =
4325 dev_get_stats(slave->dev, &temp);
4327 bond_fold_stats(stats, new, &slave->slave_stats);
4329 /* save off the slave stats for the next run */
4330 memcpy(&slave->slave_stats, new, sizeof(*new));
4333 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4334 spin_unlock(&bond->stats_lock);
4338 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4340 struct bonding *bond = netdev_priv(bond_dev);
4341 struct mii_ioctl_data *mii = NULL;
4342 const struct net_device_ops *ops;
4343 struct net_device *real_dev;
4344 struct hwtstamp_config cfg;
4348 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4359 /* We do this again just in case we were called by SIOCGMIIREG
4360 * instead of SIOCGMIIPHY.
4366 if (mii->reg_num == 1) {
4368 if (netif_carrier_ok(bond->dev))
4369 mii->val_out = BMSR_LSTATUS;
4374 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4377 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4382 real_dev = bond_option_active_slave_get_rcu(bond);
4386 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4387 ifrr.ifr_ifru = ifr->ifr_ifru;
4389 ops = real_dev->netdev_ops;
4390 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4391 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4395 ifr->ifr_ifru = ifrr.ifr_ifru;
4396 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4399 /* Set the BOND_PHC_INDEX flag to notify user space */
4400 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4402 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4413 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4415 struct bonding *bond = netdev_priv(bond_dev);
4416 struct net_device *slave_dev = NULL;
4417 struct ifbond k_binfo;
4418 struct ifbond __user *u_binfo = NULL;
4419 struct ifslave k_sinfo;
4420 struct ifslave __user *u_sinfo = NULL;
4421 struct bond_opt_value newval;
4425 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4428 case SIOCBONDINFOQUERY:
4429 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4431 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4434 bond_info_query(bond_dev, &k_binfo);
4435 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4439 case SIOCBONDSLAVEINFOQUERY:
4440 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4442 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4445 res = bond_slave_info_query(bond_dev, &k_sinfo);
4447 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4455 net = dev_net(bond_dev);
4457 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4460 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4462 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4468 case SIOCBONDENSLAVE:
4469 res = bond_enslave(bond_dev, slave_dev, NULL);
4471 case SIOCBONDRELEASE:
4472 res = bond_release(bond_dev, slave_dev);
4474 case SIOCBONDSETHWADDR:
4475 res = bond_set_dev_addr(bond_dev, slave_dev);
4477 case SIOCBONDCHANGEACTIVE:
4478 bond_opt_initstr(&newval, slave_dev->name);
4479 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4489 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4490 void __user *data, int cmd)
4492 struct ifreq ifrdata = { .ifr_data = data };
4495 case BOND_INFO_QUERY_OLD:
4496 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4497 case BOND_SLAVE_INFO_QUERY_OLD:
4498 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4499 case BOND_ENSLAVE_OLD:
4500 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4501 case BOND_RELEASE_OLD:
4502 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4503 case BOND_SETHWADDR_OLD:
4504 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4505 case BOND_CHANGE_ACTIVE_OLD:
4506 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4512 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4514 struct bonding *bond = netdev_priv(bond_dev);
4516 if (change & IFF_PROMISC)
4517 bond_set_promiscuity(bond,
4518 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4520 if (change & IFF_ALLMULTI)
4521 bond_set_allmulti(bond,
4522 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4525 static void bond_set_rx_mode(struct net_device *bond_dev)
4527 struct bonding *bond = netdev_priv(bond_dev);
4528 struct list_head *iter;
4529 struct slave *slave;
4532 if (bond_uses_primary(bond)) {
4533 slave = rcu_dereference(bond->curr_active_slave);
4535 dev_uc_sync(slave->dev, bond_dev);
4536 dev_mc_sync(slave->dev, bond_dev);
4539 bond_for_each_slave_rcu(bond, slave, iter) {
4540 dev_uc_sync_multiple(slave->dev, bond_dev);
4541 dev_mc_sync_multiple(slave->dev, bond_dev);
4547 static int bond_neigh_init(struct neighbour *n)
4549 struct bonding *bond = netdev_priv(n->dev);
4550 const struct net_device_ops *slave_ops;
4551 struct neigh_parms parms;
4552 struct slave *slave;
4556 slave = bond_first_slave_rcu(bond);
4559 slave_ops = slave->dev->netdev_ops;
4560 if (!slave_ops->ndo_neigh_setup)
4563 /* TODO: find another way [1] to implement this.
4564 * Passing a zeroed structure is fragile,
4565 * but at least we do not pass garbage.
4567 * [1] One way would be that ndo_neigh_setup() never touch
4568 * struct neigh_parms, but propagate the new neigh_setup()
4569 * back to ___neigh_create() / neigh_parms_alloc()
4571 memset(&parms, 0, sizeof(parms));
4572 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4577 if (parms.neigh_setup)
4578 ret = parms.neigh_setup(n);
4584 /* The bonding ndo_neigh_setup is called at init time beofre any
4585 * slave exists. So we must declare proxy setup function which will
4586 * be used at run time to resolve the actual slave neigh param setup.
4588 * It's also called by master devices (such as vlans) to setup their
4589 * underlying devices. In that case - do nothing, we're already set up from
4592 static int bond_neigh_setup(struct net_device *dev,
4593 struct neigh_parms *parms)
4595 /* modify only our neigh_parms */
4596 if (parms->dev == dev)
4597 parms->neigh_setup = bond_neigh_init;
4602 /* Change the MTU of all of a master's slaves to match the master */
4603 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4605 struct bonding *bond = netdev_priv(bond_dev);
4606 struct slave *slave, *rollback_slave;
4607 struct list_head *iter;
4610 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4612 bond_for_each_slave(bond, slave, iter) {
4613 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4614 slave, slave->dev->netdev_ops->ndo_change_mtu);
4616 res = dev_set_mtu(slave->dev, new_mtu);
4619 /* If we failed to set the slave's mtu to the new value
4620 * we must abort the operation even in ACTIVE_BACKUP
4621 * mode, because if we allow the backup slaves to have
4622 * different mtu values than the active slave we'll
4623 * need to change their mtu when doing a failover. That
4624 * means changing their mtu from timer context, which
4625 * is probably not a good idea.
4627 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4633 bond_dev->mtu = new_mtu;
4638 /* unwind from head to the slave that failed */
4639 bond_for_each_slave(bond, rollback_slave, iter) {
4642 if (rollback_slave == slave)
4645 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4647 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4654 /* Change HW address
4656 * Note that many devices must be down to change the HW address, and
4657 * downing the master releases all slaves. We can make bonds full of
4658 * bonding devices to test this, however.
4660 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4662 struct bonding *bond = netdev_priv(bond_dev);
4663 struct slave *slave, *rollback_slave;
4664 struct sockaddr_storage *ss = addr, tmp_ss;
4665 struct list_head *iter;
4668 if (BOND_MODE(bond) == BOND_MODE_ALB)
4669 return bond_alb_set_mac_address(bond_dev, addr);
4672 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4674 /* If fail_over_mac is enabled, do nothing and return success.
4675 * Returning an error causes ifenslave to fail.
4677 if (bond->params.fail_over_mac &&
4678 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4681 if (!is_valid_ether_addr(ss->__data))
4682 return -EADDRNOTAVAIL;
4684 bond_for_each_slave(bond, slave, iter) {
4685 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4687 res = dev_set_mac_address(slave->dev, addr, NULL);
4689 /* TODO: consider downing the slave
4691 * User should expect communications
4692 * breakage anyway until ARP finish
4695 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4702 dev_addr_set(bond_dev, ss->__data);
4706 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4707 tmp_ss.ss_family = bond_dev->type;
4709 /* unwind from head to the slave that failed */
4710 bond_for_each_slave(bond, rollback_slave, iter) {
4713 if (rollback_slave == slave)
4716 tmp_res = dev_set_mac_address(rollback_slave->dev,
4717 (struct sockaddr *)&tmp_ss, NULL);
4719 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4728 * bond_get_slave_by_id - get xmit slave with slave_id
4729 * @bond: bonding device that is transmitting
4730 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4732 * This function tries to get slave with slave_id but in case
4733 * it fails, it tries to find the first available slave for transmission.
4735 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4738 struct list_head *iter;
4739 struct slave *slave;
4742 /* Here we start from the slave with slave_id */
4743 bond_for_each_slave_rcu(bond, slave, iter) {
4745 if (bond_slave_can_tx(slave))
4750 /* Here we start from the first slave up to slave_id */
4752 bond_for_each_slave_rcu(bond, slave, iter) {
4755 if (bond_slave_can_tx(slave))
4758 /* no slave that can tx has been found */
4763 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4764 * @bond: bonding device to use
4766 * Based on the value of the bonding device's packets_per_slave parameter
4767 * this function generates a slave id, which is usually used as the next
4768 * slave to transmit through.
4770 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4773 struct reciprocal_value reciprocal_packets_per_slave;
4774 int packets_per_slave = bond->params.packets_per_slave;
4776 switch (packets_per_slave) {
4778 slave_id = prandom_u32();
4781 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4784 reciprocal_packets_per_slave =
4785 bond->params.reciprocal_packets_per_slave;
4786 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4787 slave_id = reciprocal_divide(slave_id,
4788 reciprocal_packets_per_slave);
4795 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4796 struct sk_buff *skb)
4798 struct slave *slave;
4802 /* Start with the curr_active_slave that joined the bond as the
4803 * default for sending IGMP traffic. For failover purposes one
4804 * needs to maintain some consistency for the interface that will
4805 * send the join/membership reports. The curr_active_slave found
4806 * will send all of this type of traffic.
4808 if (skb->protocol == htons(ETH_P_IP)) {
4809 int noff = skb_network_offset(skb);
4812 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4816 if (iph->protocol == IPPROTO_IGMP) {
4817 slave = rcu_dereference(bond->curr_active_slave);
4820 return bond_get_slave_by_id(bond, 0);
4825 slave_cnt = READ_ONCE(bond->slave_cnt);
4826 if (likely(slave_cnt)) {
4827 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4828 return bond_get_slave_by_id(bond, slave_id);
4833 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4834 struct xdp_buff *xdp)
4836 struct slave *slave;
4839 const struct ethhdr *eth;
4840 void *data = xdp->data;
4842 if (data + sizeof(struct ethhdr) > xdp->data_end)
4845 eth = (struct ethhdr *)data;
4846 data += sizeof(struct ethhdr);
4848 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4849 if (eth->h_proto == htons(ETH_P_IP)) {
4850 const struct iphdr *iph;
4852 if (data + sizeof(struct iphdr) > xdp->data_end)
4855 iph = (struct iphdr *)data;
4857 if (iph->protocol == IPPROTO_IGMP) {
4858 slave = rcu_dereference(bond->curr_active_slave);
4861 return bond_get_slave_by_id(bond, 0);
4866 slave_cnt = READ_ONCE(bond->slave_cnt);
4867 if (likely(slave_cnt)) {
4868 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4869 return bond_get_slave_by_id(bond, slave_id);
4874 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4875 struct net_device *bond_dev)
4877 struct bonding *bond = netdev_priv(bond_dev);
4878 struct slave *slave;
4880 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4882 return bond_dev_queue_xmit(bond, skb, slave->dev);
4884 return bond_tx_drop(bond_dev, skb);
4887 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4889 return rcu_dereference(bond->curr_active_slave);
4892 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4893 * the bond has a usable interface.
4895 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4896 struct net_device *bond_dev)
4898 struct bonding *bond = netdev_priv(bond_dev);
4899 struct slave *slave;
4901 slave = bond_xmit_activebackup_slave_get(bond);
4903 return bond_dev_queue_xmit(bond, skb, slave->dev);
4905 return bond_tx_drop(bond_dev, skb);
4908 /* Use this to update slave_array when (a) it's not appropriate to update
4909 * slave_array right away (note that update_slave_array() may sleep)
4910 * and / or (b) RTNL is not held.
4912 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4914 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4917 /* Slave array work handler. Holds only RTNL */
4918 static void bond_slave_arr_handler(struct work_struct *work)
4920 struct bonding *bond = container_of(work, struct bonding,
4921 slave_arr_work.work);
4924 if (!rtnl_trylock())
4927 ret = bond_update_slave_arr(bond, NULL);
4930 pr_warn_ratelimited("Failed to update slave array from WT\n");
4936 bond_slave_arr_work_rearm(bond, 1);
4939 static void bond_skip_slave(struct bond_up_slave *slaves,
4940 struct slave *skipslave)
4944 /* Rare situation where caller has asked to skip a specific
4945 * slave but allocation failed (most likely!). BTW this is
4946 * only possible when the call is initiated from
4947 * __bond_release_one(). In this situation; overwrite the
4948 * skipslave entry in the array with the last entry from the
4949 * array to avoid a situation where the xmit path may choose
4950 * this to-be-skipped slave to send a packet out.
4952 for (idx = 0; slaves && idx < slaves->count; idx++) {
4953 if (skipslave == slaves->arr[idx]) {
4955 slaves->arr[slaves->count - 1];
4962 static void bond_set_slave_arr(struct bonding *bond,
4963 struct bond_up_slave *usable_slaves,
4964 struct bond_up_slave *all_slaves)
4966 struct bond_up_slave *usable, *all;
4968 usable = rtnl_dereference(bond->usable_slaves);
4969 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4970 kfree_rcu(usable, rcu);
4972 all = rtnl_dereference(bond->all_slaves);
4973 rcu_assign_pointer(bond->all_slaves, all_slaves);
4974 kfree_rcu(all, rcu);
4977 static void bond_reset_slave_arr(struct bonding *bond)
4979 struct bond_up_slave *usable, *all;
4981 usable = rtnl_dereference(bond->usable_slaves);
4983 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4984 kfree_rcu(usable, rcu);
4987 all = rtnl_dereference(bond->all_slaves);
4989 RCU_INIT_POINTER(bond->all_slaves, NULL);
4990 kfree_rcu(all, rcu);
4994 /* Build the usable slaves array in control path for modes that use xmit-hash
4995 * to determine the slave interface -
4996 * (a) BOND_MODE_8023AD
4998 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5000 * The caller is expected to hold RTNL only and NO other lock!
5002 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5004 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5005 struct slave *slave;
5006 struct list_head *iter;
5012 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5013 bond->slave_cnt), GFP_KERNEL);
5014 all_slaves = kzalloc(struct_size(all_slaves, arr,
5015 bond->slave_cnt), GFP_KERNEL);
5016 if (!usable_slaves || !all_slaves) {
5020 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5021 struct ad_info ad_info;
5023 spin_lock_bh(&bond->mode_lock);
5024 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5025 spin_unlock_bh(&bond->mode_lock);
5026 pr_debug("bond_3ad_get_active_agg_info failed\n");
5027 /* No active aggragator means it's not safe to use
5028 * the previous array.
5030 bond_reset_slave_arr(bond);
5033 spin_unlock_bh(&bond->mode_lock);
5034 agg_id = ad_info.aggregator_id;
5036 bond_for_each_slave(bond, slave, iter) {
5037 if (skipslave == slave)
5040 all_slaves->arr[all_slaves->count++] = slave;
5041 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5042 struct aggregator *agg;
5044 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5045 if (!agg || agg->aggregator_identifier != agg_id)
5048 if (!bond_slave_can_tx(slave))
5051 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5052 usable_slaves->count);
5054 usable_slaves->arr[usable_slaves->count++] = slave;
5057 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5060 if (ret != 0 && skipslave) {
5061 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5063 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5066 kfree_rcu(all_slaves, rcu);
5067 kfree_rcu(usable_slaves, rcu);
5072 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5073 struct sk_buff *skb,
5074 struct bond_up_slave *slaves)
5076 struct slave *slave;
5080 hash = bond_xmit_hash(bond, skb);
5081 count = slaves ? READ_ONCE(slaves->count) : 0;
5082 if (unlikely(!count))
5085 slave = slaves->arr[hash % count];
5089 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5090 struct xdp_buff *xdp)
5092 struct bond_up_slave *slaves;
5096 hash = bond_xmit_hash_xdp(bond, xdp);
5097 slaves = rcu_dereference(bond->usable_slaves);
5098 count = slaves ? READ_ONCE(slaves->count) : 0;
5099 if (unlikely(!count))
5102 return slaves->arr[hash % count];
5105 /* Use this Xmit function for 3AD as well as XOR modes. The current
5106 * usable slave array is formed in the control path. The xmit function
5107 * just calculates hash and sends the packet out.
5109 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5110 struct net_device *dev)
5112 struct bonding *bond = netdev_priv(dev);
5113 struct bond_up_slave *slaves;
5114 struct slave *slave;
5116 slaves = rcu_dereference(bond->usable_slaves);
5117 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5119 return bond_dev_queue_xmit(bond, skb, slave->dev);
5121 return bond_tx_drop(dev, skb);
5124 /* in broadcast mode, we send everything to all usable interfaces. */
5125 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5126 struct net_device *bond_dev)
5128 struct bonding *bond = netdev_priv(bond_dev);
5129 struct slave *slave = NULL;
5130 struct list_head *iter;
5131 bool xmit_suc = false;
5132 bool skb_used = false;
5134 bond_for_each_slave_rcu(bond, slave, iter) {
5135 struct sk_buff *skb2;
5137 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5140 if (bond_is_last_slave(bond, slave)) {
5144 skb2 = skb_clone(skb, GFP_ATOMIC);
5146 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5147 bond_dev->name, __func__);
5152 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5157 dev_kfree_skb_any(skb);
5160 return NETDEV_TX_OK;
5162 dev_core_stats_tx_dropped_inc(bond_dev);
5163 return NET_XMIT_DROP;
5166 /*------------------------- Device initialization ---------------------------*/
5168 /* Lookup the slave that corresponds to a qid */
5169 static inline int bond_slave_override(struct bonding *bond,
5170 struct sk_buff *skb)
5172 struct slave *slave = NULL;
5173 struct list_head *iter;
5175 if (!skb_rx_queue_recorded(skb))
5178 /* Find out if any slaves have the same mapping as this skb. */
5179 bond_for_each_slave_rcu(bond, slave, iter) {
5180 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5181 if (bond_slave_is_up(slave) &&
5182 slave->link == BOND_LINK_UP) {
5183 bond_dev_queue_xmit(bond, skb, slave->dev);
5186 /* If the slave isn't UP, use default transmit policy. */
5195 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5196 struct net_device *sb_dev)
5198 /* This helper function exists to help dev_pick_tx get the correct
5199 * destination queue. Using a helper function skips a call to
5200 * skb_tx_hash and will put the skbs in the queue we expect on their
5201 * way down to the bonding driver.
5203 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5205 /* Save the original txq to restore before passing to the driver */
5206 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5208 if (unlikely(txq >= dev->real_num_tx_queues)) {
5210 txq -= dev->real_num_tx_queues;
5211 } while (txq >= dev->real_num_tx_queues);
5216 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5217 struct sk_buff *skb,
5220 struct bonding *bond = netdev_priv(master_dev);
5221 struct bond_up_slave *slaves;
5222 struct slave *slave = NULL;
5224 switch (BOND_MODE(bond)) {
5225 case BOND_MODE_ROUNDROBIN:
5226 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5228 case BOND_MODE_ACTIVEBACKUP:
5229 slave = bond_xmit_activebackup_slave_get(bond);
5231 case BOND_MODE_8023AD:
5234 slaves = rcu_dereference(bond->all_slaves);
5236 slaves = rcu_dereference(bond->usable_slaves);
5237 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5239 case BOND_MODE_BROADCAST:
5242 slave = bond_xmit_alb_slave_get(bond, skb);
5245 slave = bond_xmit_tlb_slave_get(bond, skb);
5248 /* Should never happen, mode already checked */
5249 WARN_ONCE(true, "Unknown bonding mode");
5258 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5260 switch (sk->sk_family) {
5261 #if IS_ENABLED(CONFIG_IPV6)
5263 if (ipv6_only_sock(sk) ||
5264 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5265 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5266 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5267 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5272 default: /* AF_INET */
5273 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5274 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5275 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5279 flow->ports.src = inet_sk(sk)->inet_sport;
5280 flow->ports.dst = inet_sk(sk)->inet_dport;
5284 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5285 * @sk: socket to use for headers
5287 * This function will extract the necessary field from the socket and use
5288 * them to generate a hash based on the LAYER34 xmit_policy.
5289 * Assumes that sk is a TCP or UDP socket.
5291 static u32 bond_sk_hash_l34(struct sock *sk)
5293 struct flow_keys flow;
5296 bond_sk_to_flow(sk, &flow);
5299 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5301 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5304 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5307 struct bond_up_slave *slaves;
5308 struct slave *slave;
5312 slaves = rcu_dereference(bond->usable_slaves);
5313 count = slaves ? READ_ONCE(slaves->count) : 0;
5314 if (unlikely(!count))
5317 hash = bond_sk_hash_l34(sk);
5318 slave = slaves->arr[hash % count];
5323 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5326 struct bonding *bond = netdev_priv(dev);
5327 struct net_device *lower = NULL;
5330 if (bond_sk_check(bond))
5331 lower = __bond_sk_get_lower_dev(bond, sk);
5337 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5338 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5339 struct net_device *dev)
5341 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
5342 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
5343 return bond_tx_drop(dev, skb);
5347 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5349 struct bonding *bond = netdev_priv(dev);
5351 if (bond_should_override_tx_queue(bond) &&
5352 !bond_slave_override(bond, skb))
5353 return NETDEV_TX_OK;
5355 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5356 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5357 return bond_tls_device_xmit(bond, skb, dev);
5360 switch (BOND_MODE(bond)) {
5361 case BOND_MODE_ROUNDROBIN:
5362 return bond_xmit_roundrobin(skb, dev);
5363 case BOND_MODE_ACTIVEBACKUP:
5364 return bond_xmit_activebackup(skb, dev);
5365 case BOND_MODE_8023AD:
5367 return bond_3ad_xor_xmit(skb, dev);
5368 case BOND_MODE_BROADCAST:
5369 return bond_xmit_broadcast(skb, dev);
5371 return bond_alb_xmit(skb, dev);
5373 return bond_tlb_xmit(skb, dev);
5375 /* Should never happen, mode already checked */
5376 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5378 return bond_tx_drop(dev, skb);
5382 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5384 struct bonding *bond = netdev_priv(dev);
5385 netdev_tx_t ret = NETDEV_TX_OK;
5387 /* If we risk deadlock from transmitting this in the
5388 * netpoll path, tell netpoll to queue the frame for later tx
5390 if (unlikely(is_netpoll_tx_blocked(dev)))
5391 return NETDEV_TX_BUSY;
5394 if (bond_has_slaves(bond))
5395 ret = __bond_start_xmit(skb, dev);
5397 ret = bond_tx_drop(dev, skb);
5403 static struct net_device *
5404 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5406 struct bonding *bond = netdev_priv(bond_dev);
5407 struct slave *slave;
5409 /* Caller needs to hold rcu_read_lock() */
5411 switch (BOND_MODE(bond)) {
5412 case BOND_MODE_ROUNDROBIN:
5413 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5416 case BOND_MODE_ACTIVEBACKUP:
5417 slave = bond_xmit_activebackup_slave_get(bond);
5420 case BOND_MODE_8023AD:
5422 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5426 /* Should never happen. Mode guarded by bond_xdp_check() */
5427 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5438 static int bond_xdp_xmit(struct net_device *bond_dev,
5439 int n, struct xdp_frame **frames, u32 flags)
5441 int nxmit, err = -ENXIO;
5445 for (nxmit = 0; nxmit < n; nxmit++) {
5446 struct xdp_frame *frame = frames[nxmit];
5447 struct xdp_frame *frames1[] = {frame};
5448 struct net_device *slave_dev;
5449 struct xdp_buff xdp;
5451 xdp_convert_frame_to_buff(frame, &xdp);
5453 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5459 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5466 /* If error happened on the first frame then we can pass the error up, otherwise
5467 * report the number of frames that were xmitted.
5470 return (nxmit == 0 ? err : nxmit);
5475 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5476 struct netlink_ext_ack *extack)
5478 struct bonding *bond = netdev_priv(dev);
5479 struct list_head *iter;
5480 struct slave *slave, *rollback_slave;
5481 struct bpf_prog *old_prog;
5482 struct netdev_bpf xdp = {
5483 .command = XDP_SETUP_PROG,
5492 if (!bond_xdp_check(bond))
5495 old_prog = bond->xdp_prog;
5496 bond->xdp_prog = prog;
5498 bond_for_each_slave(bond, slave, iter) {
5499 struct net_device *slave_dev = slave->dev;
5501 if (!slave_dev->netdev_ops->ndo_bpf ||
5502 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5503 SLAVE_NL_ERR(dev, slave_dev, extack,
5504 "Slave device does not support XDP");
5509 if (dev_xdp_prog_count(slave_dev) > 0) {
5510 SLAVE_NL_ERR(dev, slave_dev, extack,
5511 "Slave has XDP program loaded, please unload before enslaving");
5516 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5518 /* ndo_bpf() sets extack error message */
5519 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5527 static_branch_inc(&bpf_master_redirect_enabled_key);
5528 } else if (old_prog) {
5529 bpf_prog_put(old_prog);
5530 static_branch_dec(&bpf_master_redirect_enabled_key);
5536 /* unwind the program changes */
5537 bond->xdp_prog = old_prog;
5538 xdp.prog = old_prog;
5539 xdp.extack = NULL; /* do not overwrite original error */
5541 bond_for_each_slave(bond, rollback_slave, iter) {
5542 struct net_device *slave_dev = rollback_slave->dev;
5545 if (slave == rollback_slave)
5548 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5550 slave_err(dev, slave_dev,
5551 "Error %d when unwinding XDP program change\n", err_unwind);
5553 bpf_prog_inc(xdp.prog);
5558 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5560 switch (xdp->command) {
5561 case XDP_SETUP_PROG:
5562 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5568 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5570 if (speed == 0 || speed == SPEED_UNKNOWN)
5571 speed = slave->speed;
5573 speed = min(speed, slave->speed);
5578 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5579 struct ethtool_link_ksettings *cmd)
5581 struct bonding *bond = netdev_priv(bond_dev);
5582 struct list_head *iter;
5583 struct slave *slave;
5586 cmd->base.duplex = DUPLEX_UNKNOWN;
5587 cmd->base.port = PORT_OTHER;
5589 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5590 * do not need to check mode. Though link speed might not represent
5591 * the true receive or transmit bandwidth (not all modes are symmetric)
5592 * this is an accurate maximum.
5594 bond_for_each_slave(bond, slave, iter) {
5595 if (bond_slave_can_tx(slave)) {
5596 if (slave->speed != SPEED_UNKNOWN) {
5597 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5598 speed = bond_mode_bcast_speed(slave,
5601 speed += slave->speed;
5603 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5604 slave->duplex != DUPLEX_UNKNOWN)
5605 cmd->base.duplex = slave->duplex;
5608 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5613 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5614 struct ethtool_drvinfo *drvinfo)
5616 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5617 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5621 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5622 struct ethtool_ts_info *info)
5624 struct bonding *bond = netdev_priv(bond_dev);
5625 const struct ethtool_ops *ops;
5626 struct net_device *real_dev;
5627 struct phy_device *phydev;
5631 real_dev = bond_option_active_slave_get_rcu(bond);
5636 ops = real_dev->ethtool_ops;
5637 phydev = real_dev->phydev;
5639 if (phy_has_tsinfo(phydev)) {
5640 ret = phy_ts_info(phydev, info);
5642 } else if (ops->get_ts_info) {
5643 ret = ops->get_ts_info(real_dev, info);
5648 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5649 SOF_TIMESTAMPING_SOFTWARE;
5650 info->phc_index = -1;
5657 static const struct ethtool_ops bond_ethtool_ops = {
5658 .get_drvinfo = bond_ethtool_get_drvinfo,
5659 .get_link = ethtool_op_get_link,
5660 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5661 .get_ts_info = bond_ethtool_get_ts_info,
5664 static const struct net_device_ops bond_netdev_ops = {
5665 .ndo_init = bond_init,
5666 .ndo_uninit = bond_uninit,
5667 .ndo_open = bond_open,
5668 .ndo_stop = bond_close,
5669 .ndo_start_xmit = bond_start_xmit,
5670 .ndo_select_queue = bond_select_queue,
5671 .ndo_get_stats64 = bond_get_stats,
5672 .ndo_eth_ioctl = bond_eth_ioctl,
5673 .ndo_siocbond = bond_do_ioctl,
5674 .ndo_siocdevprivate = bond_siocdevprivate,
5675 .ndo_change_rx_flags = bond_change_rx_flags,
5676 .ndo_set_rx_mode = bond_set_rx_mode,
5677 .ndo_change_mtu = bond_change_mtu,
5678 .ndo_set_mac_address = bond_set_mac_address,
5679 .ndo_neigh_setup = bond_neigh_setup,
5680 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5681 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5682 #ifdef CONFIG_NET_POLL_CONTROLLER
5683 .ndo_netpoll_setup = bond_netpoll_setup,
5684 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5685 .ndo_poll_controller = bond_poll_controller,
5687 .ndo_add_slave = bond_enslave,
5688 .ndo_del_slave = bond_release,
5689 .ndo_fix_features = bond_fix_features,
5690 .ndo_features_check = passthru_features_check,
5691 .ndo_get_xmit_slave = bond_xmit_get_slave,
5692 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5693 .ndo_bpf = bond_xdp,
5694 .ndo_xdp_xmit = bond_xdp_xmit,
5695 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5698 static const struct device_type bond_type = {
5702 static void bond_destructor(struct net_device *bond_dev)
5704 struct bonding *bond = netdev_priv(bond_dev);
5707 destroy_workqueue(bond->wq);
5709 if (bond->rr_tx_counter)
5710 free_percpu(bond->rr_tx_counter);
5713 void bond_setup(struct net_device *bond_dev)
5715 struct bonding *bond = netdev_priv(bond_dev);
5717 spin_lock_init(&bond->mode_lock);
5718 bond->params = bonding_defaults;
5720 /* Initialize pointers */
5721 bond->dev = bond_dev;
5723 /* Initialize the device entry points */
5724 ether_setup(bond_dev);
5725 bond_dev->max_mtu = ETH_MAX_MTU;
5726 bond_dev->netdev_ops = &bond_netdev_ops;
5727 bond_dev->ethtool_ops = &bond_ethtool_ops;
5729 bond_dev->needs_free_netdev = true;
5730 bond_dev->priv_destructor = bond_destructor;
5732 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5734 /* Initialize the device options */
5735 bond_dev->flags |= IFF_MASTER;
5736 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5737 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5739 #ifdef CONFIG_XFRM_OFFLOAD
5740 /* set up xfrm device ops (only supported in active-backup right now) */
5741 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5742 INIT_LIST_HEAD(&bond->ipsec_list);
5743 spin_lock_init(&bond->ipsec_lock);
5744 #endif /* CONFIG_XFRM_OFFLOAD */
5746 /* don't acquire bond device's netif_tx_lock when transmitting */
5747 bond_dev->features |= NETIF_F_LLTX;
5749 /* By default, we declare the bond to be fully
5750 * VLAN hardware accelerated capable. Special
5751 * care is taken in the various xmit functions
5752 * when there are slaves that are not hw accel
5756 /* Don't allow bond devices to change network namespaces. */
5757 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5759 bond_dev->hw_features = BOND_VLAN_FEATURES |
5760 NETIF_F_HW_VLAN_CTAG_RX |
5761 NETIF_F_HW_VLAN_CTAG_FILTER;
5763 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5764 bond_dev->features |= bond_dev->hw_features;
5765 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5766 #ifdef CONFIG_XFRM_OFFLOAD
5767 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5768 /* Only enable XFRM features if this is an active-backup config */
5769 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5770 bond_dev->features |= BOND_XFRM_FEATURES;
5771 #endif /* CONFIG_XFRM_OFFLOAD */
5772 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5773 if (bond_sk_check(bond))
5774 bond_dev->features |= BOND_TLS_FEATURES;
5778 /* Destroy a bonding device.
5779 * Must be under rtnl_lock when this function is called.
5781 static void bond_uninit(struct net_device *bond_dev)
5783 struct bonding *bond = netdev_priv(bond_dev);
5784 struct bond_up_slave *usable, *all;
5785 struct list_head *iter;
5786 struct slave *slave;
5788 bond_netpoll_cleanup(bond_dev);
5790 /* Release the bonded slaves */
5791 bond_for_each_slave(bond, slave, iter)
5792 __bond_release_one(bond_dev, slave->dev, true, true);
5793 netdev_info(bond_dev, "Released all slaves\n");
5795 usable = rtnl_dereference(bond->usable_slaves);
5797 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5798 kfree_rcu(usable, rcu);
5801 all = rtnl_dereference(bond->all_slaves);
5803 RCU_INIT_POINTER(bond->all_slaves, NULL);
5804 kfree_rcu(all, rcu);
5807 list_del(&bond->bond_list);
5809 bond_debug_unregister(bond);
5812 /*------------------------- Module initialization ---------------------------*/
5814 static int bond_check_params(struct bond_params *params)
5816 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5817 struct bond_opt_value newval;
5818 const struct bond_opt_value *valptr;
5819 int arp_all_targets_value = 0;
5820 u16 ad_actor_sys_prio = 0;
5821 u16 ad_user_port_key = 0;
5822 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5824 int bond_mode = BOND_MODE_ROUNDROBIN;
5825 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5829 /* Convert string parameters. */
5831 bond_opt_initstr(&newval, mode);
5832 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5834 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5837 bond_mode = valptr->value;
5840 if (xmit_hash_policy) {
5841 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5842 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5843 bond_mode == BOND_MODE_BROADCAST) {
5844 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5845 bond_mode_name(bond_mode));
5847 bond_opt_initstr(&newval, xmit_hash_policy);
5848 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5851 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5855 xmit_hashtype = valptr->value;
5860 if (bond_mode != BOND_MODE_8023AD) {
5861 pr_info("lacp_rate param is irrelevant in mode %s\n",
5862 bond_mode_name(bond_mode));
5864 bond_opt_initstr(&newval, lacp_rate);
5865 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5868 pr_err("Error: Invalid lacp rate \"%s\"\n",
5872 lacp_fast = valptr->value;
5877 bond_opt_initstr(&newval, ad_select);
5878 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5881 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5884 params->ad_select = valptr->value;
5885 if (bond_mode != BOND_MODE_8023AD)
5886 pr_warn("ad_select param only affects 802.3ad mode\n");
5888 params->ad_select = BOND_AD_STABLE;
5891 if (max_bonds < 0) {
5892 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5893 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5894 max_bonds = BOND_DEFAULT_MAX_BONDS;
5898 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5904 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5909 if (downdelay < 0) {
5910 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5911 downdelay, INT_MAX);
5915 if ((use_carrier != 0) && (use_carrier != 1)) {
5916 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5921 if (num_peer_notif < 0 || num_peer_notif > 255) {
5922 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5927 /* reset values for 802.3ad/TLB/ALB */
5928 if (!bond_mode_uses_arp(bond_mode)) {
5930 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");
5931 pr_warn("Forcing miimon to 100msec\n");
5932 miimon = BOND_DEFAULT_MIIMON;
5936 if (tx_queues < 1 || tx_queues > 255) {
5937 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5938 tx_queues, BOND_DEFAULT_TX_QUEUES);
5939 tx_queues = BOND_DEFAULT_TX_QUEUES;
5942 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5943 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5945 all_slaves_active = 0;
5948 if (resend_igmp < 0 || resend_igmp > 255) {
5949 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5950 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5951 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5954 bond_opt_initval(&newval, packets_per_slave);
5955 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5956 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5957 packets_per_slave, USHRT_MAX);
5958 packets_per_slave = 1;
5961 if (bond_mode == BOND_MODE_ALB) {
5962 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",
5967 if (updelay || downdelay) {
5968 /* just warn the user the up/down delay will have
5969 * no effect since miimon is zero...
5971 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",
5972 updelay, downdelay);
5975 /* don't allow arp monitoring */
5977 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5978 miimon, arp_interval);
5982 if ((updelay % miimon) != 0) {
5983 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5984 updelay, miimon, (updelay / miimon) * miimon);
5989 if ((downdelay % miimon) != 0) {
5990 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5992 (downdelay / miimon) * miimon);
5995 downdelay /= miimon;
5998 if (arp_interval < 0) {
5999 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6000 arp_interval, INT_MAX);
6004 for (arp_ip_count = 0, i = 0;
6005 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6008 /* not a complete check, but good enough to catch mistakes */
6009 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6010 !bond_is_ip_target_ok(ip)) {
6011 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6015 if (bond_get_targets_ip(arp_target, ip) == -1)
6016 arp_target[arp_ip_count++] = ip;
6018 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6023 if (arp_interval && !arp_ip_count) {
6024 /* don't allow arping if no arp_ip_target given... */
6025 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6031 if (!arp_interval) {
6032 pr_err("arp_validate requires arp_interval\n");
6036 bond_opt_initstr(&newval, arp_validate);
6037 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6040 pr_err("Error: invalid arp_validate \"%s\"\n",
6044 arp_validate_value = valptr->value;
6046 arp_validate_value = 0;
6049 if (arp_all_targets) {
6050 bond_opt_initstr(&newval, arp_all_targets);
6051 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6054 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6056 arp_all_targets_value = 0;
6058 arp_all_targets_value = valptr->value;
6063 pr_info("MII link monitoring set to %d ms\n", miimon);
6064 } else if (arp_interval) {
6065 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6066 arp_validate_value);
6067 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6068 arp_interval, valptr->string, arp_ip_count);
6070 for (i = 0; i < arp_ip_count; i++)
6071 pr_cont(" %s", arp_ip_target[i]);
6075 } else if (max_bonds) {
6076 /* miimon and arp_interval not set, we need one so things
6077 * work as expected, see bonding.txt for details
6079 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");
6082 if (primary && !bond_mode_uses_primary(bond_mode)) {
6083 /* currently, using a primary only makes sense
6084 * in active backup, TLB or ALB modes
6086 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6087 primary, bond_mode_name(bond_mode));
6091 if (primary && primary_reselect) {
6092 bond_opt_initstr(&newval, primary_reselect);
6093 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6096 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6100 primary_reselect_value = valptr->value;
6102 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6105 if (fail_over_mac) {
6106 bond_opt_initstr(&newval, fail_over_mac);
6107 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6110 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6114 fail_over_mac_value = valptr->value;
6115 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6116 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6118 fail_over_mac_value = BOND_FOM_NONE;
6121 bond_opt_initstr(&newval, "default");
6122 valptr = bond_opt_parse(
6123 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6126 pr_err("Error: No ad_actor_sys_prio default value");
6129 ad_actor_sys_prio = valptr->value;
6131 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6134 pr_err("Error: No ad_user_port_key default value");
6137 ad_user_port_key = valptr->value;
6139 bond_opt_initstr(&newval, "default");
6140 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6142 pr_err("Error: No tlb_dynamic_lb default value");
6145 tlb_dynamic_lb = valptr->value;
6147 if (lp_interval == 0) {
6148 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6149 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6150 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6153 /* fill params struct with the proper values */
6154 params->mode = bond_mode;
6155 params->xmit_policy = xmit_hashtype;
6156 params->miimon = miimon;
6157 params->num_peer_notif = num_peer_notif;
6158 params->arp_interval = arp_interval;
6159 params->arp_validate = arp_validate_value;
6160 params->arp_all_targets = arp_all_targets_value;
6161 params->missed_max = 2;
6162 params->updelay = updelay;
6163 params->downdelay = downdelay;
6164 params->peer_notif_delay = 0;
6165 params->use_carrier = use_carrier;
6166 params->lacp_active = 1;
6167 params->lacp_fast = lacp_fast;
6168 params->primary[0] = 0;
6169 params->primary_reselect = primary_reselect_value;
6170 params->fail_over_mac = fail_over_mac_value;
6171 params->tx_queues = tx_queues;
6172 params->all_slaves_active = all_slaves_active;
6173 params->resend_igmp = resend_igmp;
6174 params->min_links = min_links;
6175 params->lp_interval = lp_interval;
6176 params->packets_per_slave = packets_per_slave;
6177 params->tlb_dynamic_lb = tlb_dynamic_lb;
6178 params->ad_actor_sys_prio = ad_actor_sys_prio;
6179 eth_zero_addr(params->ad_actor_system);
6180 params->ad_user_port_key = ad_user_port_key;
6181 if (packets_per_slave > 0) {
6182 params->reciprocal_packets_per_slave =
6183 reciprocal_value(packets_per_slave);
6185 /* reciprocal_packets_per_slave is unused if
6186 * packets_per_slave is 0 or 1, just initialize it
6188 params->reciprocal_packets_per_slave =
6189 (struct reciprocal_value) { 0 };
6193 strscpy_pad(params->primary, primary, sizeof(params->primary));
6195 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6196 #if IS_ENABLED(CONFIG_IPV6)
6197 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6203 /* Called from registration process */
6204 static int bond_init(struct net_device *bond_dev)
6206 struct bonding *bond = netdev_priv(bond_dev);
6207 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6209 netdev_dbg(bond_dev, "Begin bond_init\n");
6211 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6215 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
6216 bond->rr_tx_counter = alloc_percpu(u32);
6217 if (!bond->rr_tx_counter) {
6218 destroy_workqueue(bond->wq);
6224 spin_lock_init(&bond->stats_lock);
6225 netdev_lockdep_set_classes(bond_dev);
6227 list_add_tail(&bond->bond_list, &bn->dev_list);
6229 bond_prepare_sysfs_group(bond);
6231 bond_debug_register(bond);
6233 /* Ensure valid dev_addr */
6234 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6235 bond_dev->addr_assign_type == NET_ADDR_PERM)
6236 eth_hw_addr_random(bond_dev);
6241 unsigned int bond_get_num_tx_queues(void)
6246 /* Create a new bond based on the specified name and bonding parameters.
6247 * If name is NULL, obtain a suitable "bond%d" name for us.
6248 * Caller must NOT hold rtnl_lock; we need to release it here before we
6249 * set up our sysfs entries.
6251 int bond_create(struct net *net, const char *name)
6253 struct net_device *bond_dev;
6254 struct bonding *bond;
6259 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6260 name ? name : "bond%d", NET_NAME_UNKNOWN,
6261 bond_setup, tx_queues);
6265 bond = netdev_priv(bond_dev);
6266 dev_net_set(bond_dev, net);
6267 bond_dev->rtnl_link_ops = &bond_link_ops;
6269 res = register_netdevice(bond_dev);
6271 free_netdev(bond_dev);
6275 netif_carrier_off(bond_dev);
6277 bond_work_init_all(bond);
6284 static int __net_init bond_net_init(struct net *net)
6286 struct bond_net *bn = net_generic(net, bond_net_id);
6289 INIT_LIST_HEAD(&bn->dev_list);
6291 bond_create_proc_dir(bn);
6292 bond_create_sysfs(bn);
6297 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6299 struct bond_net *bn;
6303 list_for_each_entry(net, net_list, exit_list) {
6304 bn = net_generic(net, bond_net_id);
6305 bond_destroy_sysfs(bn);
6308 /* Kill off any bonds created after unregistering bond rtnl ops */
6310 list_for_each_entry(net, net_list, exit_list) {
6311 struct bonding *bond, *tmp_bond;
6313 bn = net_generic(net, bond_net_id);
6314 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6315 unregister_netdevice_queue(bond->dev, &list);
6317 unregister_netdevice_many(&list);
6320 list_for_each_entry(net, net_list, exit_list) {
6321 bn = net_generic(net, bond_net_id);
6322 bond_destroy_proc_dir(bn);
6326 static struct pernet_operations bond_net_ops = {
6327 .init = bond_net_init,
6328 .exit_batch = bond_net_exit_batch,
6330 .size = sizeof(struct bond_net),
6333 static int __init bonding_init(void)
6338 res = bond_check_params(&bonding_defaults);
6342 res = register_pernet_subsys(&bond_net_ops);
6346 res = bond_netlink_init();
6350 bond_create_debugfs();
6352 for (i = 0; i < max_bonds; i++) {
6353 res = bond_create(&init_net, NULL);
6358 skb_flow_dissector_init(&flow_keys_bonding,
6359 flow_keys_bonding_keys,
6360 ARRAY_SIZE(flow_keys_bonding_keys));
6362 register_netdevice_notifier(&bond_netdev_notifier);
6366 bond_destroy_debugfs();
6367 bond_netlink_fini();
6369 unregister_pernet_subsys(&bond_net_ops);
6374 static void __exit bonding_exit(void)
6376 unregister_netdevice_notifier(&bond_netdev_notifier);
6378 bond_destroy_debugfs();
6380 bond_netlink_fini();
6381 unregister_pernet_subsys(&bond_net_ops);
6383 #ifdef CONFIG_NET_POLL_CONTROLLER
6384 /* Make sure we don't have an imbalance on our netpoll blocking */
6385 WARN_ON(atomic_read(&netpoll_block_tx));
6389 module_init(bonding_init);
6390 module_exit(bonding_exit);
6391 MODULE_LICENSE("GPL");
6392 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6393 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");