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 dev_mc_del(slave_dev, lacpdu_mcast_addr);
872 /*--------------------------- Active slave change ---------------------------*/
874 /* Update the hardware address list and promisc/allmulti for the new and
875 * old active slaves (if any). Modes that are not using primary keep all
876 * slaves up date at all times; only the modes that use primary need to call
877 * this function to swap these settings during a failover.
879 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
880 struct slave *old_active)
883 if (bond->dev->flags & IFF_PROMISC)
884 dev_set_promiscuity(old_active->dev, -1);
886 if (bond->dev->flags & IFF_ALLMULTI)
887 dev_set_allmulti(old_active->dev, -1);
889 if (bond->dev->flags & IFF_UP)
890 bond_hw_addr_flush(bond->dev, old_active->dev);
894 /* FIXME: Signal errors upstream. */
895 if (bond->dev->flags & IFF_PROMISC)
896 dev_set_promiscuity(new_active->dev, 1);
898 if (bond->dev->flags & IFF_ALLMULTI)
899 dev_set_allmulti(new_active->dev, 1);
901 if (bond->dev->flags & IFF_UP) {
902 netif_addr_lock_bh(bond->dev);
903 dev_uc_sync(new_active->dev, bond->dev);
904 dev_mc_sync(new_active->dev, bond->dev);
905 netif_addr_unlock_bh(bond->dev);
911 * bond_set_dev_addr - clone slave's address to bond
912 * @bond_dev: bond net device
913 * @slave_dev: slave net device
915 * Should be called with RTNL held.
917 static int bond_set_dev_addr(struct net_device *bond_dev,
918 struct net_device *slave_dev)
922 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
923 bond_dev, slave_dev, slave_dev->addr_len);
924 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
928 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
929 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
930 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
934 static struct slave *bond_get_old_active(struct bonding *bond,
935 struct slave *new_active)
938 struct list_head *iter;
940 bond_for_each_slave(bond, slave, iter) {
941 if (slave == new_active)
944 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
951 /* bond_do_fail_over_mac
953 * Perform special MAC address swapping for fail_over_mac settings
957 static void bond_do_fail_over_mac(struct bonding *bond,
958 struct slave *new_active,
959 struct slave *old_active)
961 u8 tmp_mac[MAX_ADDR_LEN];
962 struct sockaddr_storage ss;
965 switch (bond->params.fail_over_mac) {
966 case BOND_FOM_ACTIVE:
968 rv = bond_set_dev_addr(bond->dev, new_active->dev);
970 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
974 case BOND_FOM_FOLLOW:
975 /* if new_active && old_active, swap them
976 * if just old_active, do nothing (going to no active slave)
977 * if just new_active, set new_active to bond's MAC
983 old_active = bond_get_old_active(bond, new_active);
986 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
987 new_active->dev->addr_len);
988 bond_hw_addr_copy(ss.__data,
989 old_active->dev->dev_addr,
990 old_active->dev->addr_len);
991 ss.ss_family = new_active->dev->type;
993 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
994 bond->dev->addr_len);
995 ss.ss_family = bond->dev->type;
998 rv = dev_set_mac_address(new_active->dev,
999 (struct sockaddr *)&ss, NULL);
1001 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1009 bond_hw_addr_copy(ss.__data, tmp_mac,
1010 new_active->dev->addr_len);
1011 ss.ss_family = old_active->dev->type;
1013 rv = dev_set_mac_address(old_active->dev,
1014 (struct sockaddr *)&ss, NULL);
1016 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1021 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1022 bond->params.fail_over_mac);
1029 * bond_choose_primary_or_current - select the primary or high priority slave
1030 * @bond: our bonding struct
1032 * - Check if there is a primary link. If the primary link was set and is up,
1033 * go on and do link reselection.
1035 * - If primary link is not set or down, find the highest priority link.
1036 * If the highest priority link is not current slave, set it as primary
1037 * link and do link reselection.
1039 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1041 struct slave *prim = rtnl_dereference(bond->primary_slave);
1042 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1043 struct slave *slave, *hprio = NULL;
1044 struct list_head *iter;
1046 if (!prim || prim->link != BOND_LINK_UP) {
1047 bond_for_each_slave(bond, slave, iter) {
1048 if (slave->link == BOND_LINK_UP) {
1049 hprio = hprio ?: slave;
1050 if (slave->prio > hprio->prio)
1055 if (hprio && hprio != curr) {
1060 if (!curr || curr->link != BOND_LINK_UP)
1065 if (bond->force_primary) {
1066 bond->force_primary = false;
1071 if (!curr || curr->link != BOND_LINK_UP)
1074 /* At this point, prim and curr are both up */
1075 switch (bond->params.primary_reselect) {
1076 case BOND_PRI_RESELECT_ALWAYS:
1078 case BOND_PRI_RESELECT_BETTER:
1079 if (prim->speed < curr->speed)
1081 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1084 case BOND_PRI_RESELECT_FAILURE:
1087 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1088 bond->params.primary_reselect);
1094 * bond_find_best_slave - select the best available slave to be the active one
1095 * @bond: our bonding struct
1097 static struct slave *bond_find_best_slave(struct bonding *bond)
1099 struct slave *slave, *bestslave = NULL;
1100 struct list_head *iter;
1101 int mintime = bond->params.updelay;
1103 slave = bond_choose_primary_or_current(bond);
1107 bond_for_each_slave(bond, slave, iter) {
1108 if (slave->link == BOND_LINK_UP)
1110 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1111 slave->delay < mintime) {
1112 mintime = slave->delay;
1120 static bool bond_should_notify_peers(struct bonding *bond)
1122 struct slave *slave;
1125 slave = rcu_dereference(bond->curr_active_slave);
1128 if (!slave || !bond->send_peer_notif ||
1129 bond->send_peer_notif %
1130 max(1, bond->params.peer_notif_delay) != 0 ||
1131 !netif_carrier_ok(bond->dev) ||
1132 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1135 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1136 slave ? slave->dev->name : "NULL");
1142 * bond_change_active_slave - change the active slave into the specified one
1143 * @bond: our bonding struct
1144 * @new_active: the new slave to make the active one
1146 * Set the new slave to the bond's settings and unset them on the old
1147 * curr_active_slave.
1148 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1150 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1151 * because it is apparently the best available slave we have, even though its
1152 * updelay hasn't timed out yet.
1154 * Caller must hold RTNL.
1156 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1158 struct slave *old_active;
1162 old_active = rtnl_dereference(bond->curr_active_slave);
1164 if (old_active == new_active)
1167 #ifdef CONFIG_XFRM_OFFLOAD
1168 bond_ipsec_del_sa_all(bond);
1169 #endif /* CONFIG_XFRM_OFFLOAD */
1172 new_active->last_link_up = jiffies;
1174 if (new_active->link == BOND_LINK_BACK) {
1175 if (bond_uses_primary(bond)) {
1176 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1177 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1180 new_active->delay = 0;
1181 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1182 BOND_SLAVE_NOTIFY_NOW);
1184 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1185 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1187 if (bond_is_lb(bond))
1188 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1190 if (bond_uses_primary(bond))
1191 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1195 if (bond_uses_primary(bond))
1196 bond_hw_addr_swap(bond, new_active, old_active);
1198 if (bond_is_lb(bond)) {
1199 bond_alb_handle_active_change(bond, new_active);
1201 bond_set_slave_inactive_flags(old_active,
1202 BOND_SLAVE_NOTIFY_NOW);
1204 bond_set_slave_active_flags(new_active,
1205 BOND_SLAVE_NOTIFY_NOW);
1207 rcu_assign_pointer(bond->curr_active_slave, new_active);
1210 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1212 bond_set_slave_inactive_flags(old_active,
1213 BOND_SLAVE_NOTIFY_NOW);
1216 bool should_notify_peers = false;
1218 bond_set_slave_active_flags(new_active,
1219 BOND_SLAVE_NOTIFY_NOW);
1221 if (bond->params.fail_over_mac)
1222 bond_do_fail_over_mac(bond, new_active,
1225 if (netif_running(bond->dev)) {
1226 bond->send_peer_notif =
1227 bond->params.num_peer_notif *
1228 max(1, bond->params.peer_notif_delay);
1229 should_notify_peers =
1230 bond_should_notify_peers(bond);
1233 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1234 if (should_notify_peers) {
1235 bond->send_peer_notif--;
1236 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1242 #ifdef CONFIG_XFRM_OFFLOAD
1243 bond_ipsec_add_sa_all(bond);
1244 #endif /* CONFIG_XFRM_OFFLOAD */
1246 /* resend IGMP joins since active slave has changed or
1247 * all were sent on curr_active_slave.
1248 * resend only if bond is brought up with the affected
1249 * bonding modes and the retransmission is enabled
1251 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1252 ((bond_uses_primary(bond) && new_active) ||
1253 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1254 bond->igmp_retrans = bond->params.resend_igmp;
1255 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1260 * bond_select_active_slave - select a new active slave, if needed
1261 * @bond: our bonding struct
1263 * This functions should be called when one of the following occurs:
1264 * - The old curr_active_slave has been released or lost its link.
1265 * - The primary_slave has got its link back.
1266 * - A slave has got its link back and there's no old curr_active_slave.
1268 * Caller must hold RTNL.
1270 void bond_select_active_slave(struct bonding *bond)
1272 struct slave *best_slave;
1277 best_slave = bond_find_best_slave(bond);
1278 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1279 bond_change_active_slave(bond, best_slave);
1280 rv = bond_set_carrier(bond);
1284 if (netif_carrier_ok(bond->dev))
1285 netdev_info(bond->dev, "active interface up!\n");
1287 netdev_info(bond->dev, "now running without any active interface!\n");
1291 #ifdef CONFIG_NET_POLL_CONTROLLER
1292 static inline int slave_enable_netpoll(struct slave *slave)
1297 np = kzalloc(sizeof(*np), GFP_KERNEL);
1302 err = __netpoll_setup(np, slave->dev);
1311 static inline void slave_disable_netpoll(struct slave *slave)
1313 struct netpoll *np = slave->np;
1323 static void bond_poll_controller(struct net_device *bond_dev)
1325 struct bonding *bond = netdev_priv(bond_dev);
1326 struct slave *slave = NULL;
1327 struct list_head *iter;
1328 struct ad_info ad_info;
1330 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1331 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1334 bond_for_each_slave_rcu(bond, slave, iter) {
1335 if (!bond_slave_is_up(slave))
1338 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1339 struct aggregator *agg =
1340 SLAVE_AD_INFO(slave)->port.aggregator;
1343 agg->aggregator_identifier != ad_info.aggregator_id)
1347 netpoll_poll_dev(slave->dev);
1351 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1353 struct bonding *bond = netdev_priv(bond_dev);
1354 struct list_head *iter;
1355 struct slave *slave;
1357 bond_for_each_slave(bond, slave, iter)
1358 if (bond_slave_is_up(slave))
1359 slave_disable_netpoll(slave);
1362 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1364 struct bonding *bond = netdev_priv(dev);
1365 struct list_head *iter;
1366 struct slave *slave;
1369 bond_for_each_slave(bond, slave, iter) {
1370 err = slave_enable_netpoll(slave);
1372 bond_netpoll_cleanup(dev);
1379 static inline int slave_enable_netpoll(struct slave *slave)
1383 static inline void slave_disable_netpoll(struct slave *slave)
1386 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1391 /*---------------------------------- IOCTL ----------------------------------*/
1393 static netdev_features_t bond_fix_features(struct net_device *dev,
1394 netdev_features_t features)
1396 struct bonding *bond = netdev_priv(dev);
1397 struct list_head *iter;
1398 netdev_features_t mask;
1399 struct slave *slave;
1401 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1402 if (bond_sk_check(bond))
1403 features |= BOND_TLS_FEATURES;
1405 features &= ~BOND_TLS_FEATURES;
1410 features &= ~NETIF_F_ONE_FOR_ALL;
1411 features |= NETIF_F_ALL_FOR_ALL;
1413 bond_for_each_slave(bond, slave, iter) {
1414 features = netdev_increment_features(features,
1415 slave->dev->features,
1418 features = netdev_add_tso_features(features, mask);
1423 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1424 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1425 NETIF_F_HIGHDMA | NETIF_F_LRO)
1427 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1428 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1430 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1431 NETIF_F_GSO_SOFTWARE)
1434 static void bond_compute_features(struct bonding *bond)
1436 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1437 IFF_XMIT_DST_RELEASE_PERM;
1438 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1439 netdev_features_t enc_features = BOND_ENC_FEATURES;
1440 #ifdef CONFIG_XFRM_OFFLOAD
1441 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1442 #endif /* CONFIG_XFRM_OFFLOAD */
1443 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1444 struct net_device *bond_dev = bond->dev;
1445 struct list_head *iter;
1446 struct slave *slave;
1447 unsigned short max_hard_header_len = ETH_HLEN;
1448 unsigned int tso_max_size = TSO_MAX_SIZE;
1449 u16 tso_max_segs = TSO_MAX_SEGS;
1451 if (!bond_has_slaves(bond))
1453 vlan_features &= NETIF_F_ALL_FOR_ALL;
1454 mpls_features &= NETIF_F_ALL_FOR_ALL;
1456 bond_for_each_slave(bond, slave, iter) {
1457 vlan_features = netdev_increment_features(vlan_features,
1458 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1460 enc_features = netdev_increment_features(enc_features,
1461 slave->dev->hw_enc_features,
1464 #ifdef CONFIG_XFRM_OFFLOAD
1465 xfrm_features = netdev_increment_features(xfrm_features,
1466 slave->dev->hw_enc_features,
1467 BOND_XFRM_FEATURES);
1468 #endif /* CONFIG_XFRM_OFFLOAD */
1470 mpls_features = netdev_increment_features(mpls_features,
1471 slave->dev->mpls_features,
1472 BOND_MPLS_FEATURES);
1474 dst_release_flag &= slave->dev->priv_flags;
1475 if (slave->dev->hard_header_len > max_hard_header_len)
1476 max_hard_header_len = slave->dev->hard_header_len;
1478 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1479 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1481 bond_dev->hard_header_len = max_hard_header_len;
1484 bond_dev->vlan_features = vlan_features;
1485 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1486 NETIF_F_HW_VLAN_CTAG_TX |
1487 NETIF_F_HW_VLAN_STAG_TX;
1488 #ifdef CONFIG_XFRM_OFFLOAD
1489 bond_dev->hw_enc_features |= xfrm_features;
1490 #endif /* CONFIG_XFRM_OFFLOAD */
1491 bond_dev->mpls_features = mpls_features;
1492 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1493 netif_set_tso_max_size(bond_dev, tso_max_size);
1495 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1496 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1497 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1498 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1500 netdev_change_features(bond_dev);
1503 static void bond_setup_by_slave(struct net_device *bond_dev,
1504 struct net_device *slave_dev)
1506 bond_dev->header_ops = slave_dev->header_ops;
1508 bond_dev->type = slave_dev->type;
1509 bond_dev->hard_header_len = slave_dev->hard_header_len;
1510 bond_dev->needed_headroom = slave_dev->needed_headroom;
1511 bond_dev->addr_len = slave_dev->addr_len;
1513 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1514 slave_dev->addr_len);
1517 /* On bonding slaves other than the currently active slave, suppress
1518 * duplicates except for alb non-mcast/bcast.
1520 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1521 struct slave *slave,
1522 struct bonding *bond)
1524 if (bond_is_slave_inactive(slave)) {
1525 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1526 skb->pkt_type != PACKET_BROADCAST &&
1527 skb->pkt_type != PACKET_MULTICAST)
1534 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1536 struct sk_buff *skb = *pskb;
1537 struct slave *slave;
1538 struct bonding *bond;
1539 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1541 int ret = RX_HANDLER_ANOTHER;
1543 skb = skb_share_check(skb, GFP_ATOMIC);
1545 return RX_HANDLER_CONSUMED;
1549 slave = bond_slave_get_rcu(skb->dev);
1552 recv_probe = READ_ONCE(bond->recv_probe);
1554 ret = recv_probe(skb, bond, slave);
1555 if (ret == RX_HANDLER_CONSUMED) {
1562 * For packets determined by bond_should_deliver_exact_match() call to
1563 * be suppressed we want to make an exception for link-local packets.
1564 * This is necessary for e.g. LLDP daemons to be able to monitor
1565 * inactive slave links without being forced to bind to them
1568 * At the same time, packets that are passed to the bonding master
1569 * (including link-local ones) can have their originating interface
1570 * determined via PACKET_ORIGDEV socket option.
1572 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1573 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1574 return RX_HANDLER_PASS;
1575 return RX_HANDLER_EXACT;
1578 skb->dev = bond->dev;
1580 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1581 netif_is_bridge_port(bond->dev) &&
1582 skb->pkt_type == PACKET_HOST) {
1584 if (unlikely(skb_cow_head(skb,
1585 skb->data - skb_mac_header(skb)))) {
1587 return RX_HANDLER_CONSUMED;
1589 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1590 bond->dev->addr_len);
1596 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1598 switch (BOND_MODE(bond)) {
1599 case BOND_MODE_ROUNDROBIN:
1600 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1601 case BOND_MODE_ACTIVEBACKUP:
1602 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1603 case BOND_MODE_BROADCAST:
1604 return NETDEV_LAG_TX_TYPE_BROADCAST;
1606 case BOND_MODE_8023AD:
1607 return NETDEV_LAG_TX_TYPE_HASH;
1609 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1613 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1614 enum netdev_lag_tx_type type)
1616 if (type != NETDEV_LAG_TX_TYPE_HASH)
1617 return NETDEV_LAG_HASH_NONE;
1619 switch (bond->params.xmit_policy) {
1620 case BOND_XMIT_POLICY_LAYER2:
1621 return NETDEV_LAG_HASH_L2;
1622 case BOND_XMIT_POLICY_LAYER34:
1623 return NETDEV_LAG_HASH_L34;
1624 case BOND_XMIT_POLICY_LAYER23:
1625 return NETDEV_LAG_HASH_L23;
1626 case BOND_XMIT_POLICY_ENCAP23:
1627 return NETDEV_LAG_HASH_E23;
1628 case BOND_XMIT_POLICY_ENCAP34:
1629 return NETDEV_LAG_HASH_E34;
1630 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1631 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1633 return NETDEV_LAG_HASH_UNKNOWN;
1637 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1638 struct netlink_ext_ack *extack)
1640 struct netdev_lag_upper_info lag_upper_info;
1641 enum netdev_lag_tx_type type;
1643 type = bond_lag_tx_type(bond);
1644 lag_upper_info.tx_type = type;
1645 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1647 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1648 &lag_upper_info, extack);
1651 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1653 netdev_upper_dev_unlink(slave->dev, bond->dev);
1654 slave->dev->flags &= ~IFF_SLAVE;
1657 static void slave_kobj_release(struct kobject *kobj)
1659 struct slave *slave = to_slave(kobj);
1660 struct bonding *bond = bond_get_bond_by_slave(slave);
1662 cancel_delayed_work_sync(&slave->notify_work);
1663 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1664 kfree(SLAVE_AD_INFO(slave));
1669 static struct kobj_type slave_ktype = {
1670 .release = slave_kobj_release,
1672 .sysfs_ops = &slave_sysfs_ops,
1676 static int bond_kobj_init(struct slave *slave)
1680 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1681 &(slave->dev->dev.kobj), "bonding_slave");
1683 kobject_put(&slave->kobj);
1688 static struct slave *bond_alloc_slave(struct bonding *bond,
1689 struct net_device *slave_dev)
1691 struct slave *slave = NULL;
1693 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1698 slave->dev = slave_dev;
1699 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1701 if (bond_kobj_init(slave))
1704 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1705 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1707 if (!SLAVE_AD_INFO(slave)) {
1708 kobject_put(&slave->kobj);
1716 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1718 info->bond_mode = BOND_MODE(bond);
1719 info->miimon = bond->params.miimon;
1720 info->num_slaves = bond->slave_cnt;
1723 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1725 strcpy(info->slave_name, slave->dev->name);
1726 info->link = slave->link;
1727 info->state = bond_slave_state(slave);
1728 info->link_failure_count = slave->link_failure_count;
1731 static void bond_netdev_notify_work(struct work_struct *_work)
1733 struct slave *slave = container_of(_work, struct slave,
1736 if (rtnl_trylock()) {
1737 struct netdev_bonding_info binfo;
1739 bond_fill_ifslave(slave, &binfo.slave);
1740 bond_fill_ifbond(slave->bond, &binfo.master);
1741 netdev_bonding_info_change(slave->dev, &binfo);
1744 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1748 void bond_queue_slave_event(struct slave *slave)
1750 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1753 void bond_lower_state_changed(struct slave *slave)
1755 struct netdev_lag_lower_state_info info;
1757 info.link_up = slave->link == BOND_LINK_UP ||
1758 slave->link == BOND_LINK_FAIL;
1759 info.tx_enabled = bond_is_active_slave(slave);
1760 netdev_lower_state_changed(slave->dev, &info);
1763 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1765 NL_SET_ERR_MSG(extack, errmsg); \
1767 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1770 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1772 NL_SET_ERR_MSG(extack, errmsg); \
1774 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1777 /* enslave device <slave> to bond device <master> */
1778 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1779 struct netlink_ext_ack *extack)
1781 struct bonding *bond = netdev_priv(bond_dev);
1782 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1783 struct slave *new_slave = NULL, *prev_slave;
1784 struct sockaddr_storage ss;
1788 if (slave_dev->flags & IFF_MASTER &&
1789 !netif_is_bond_master(slave_dev)) {
1790 BOND_NL_ERR(bond_dev, extack,
1791 "Device type (master device) cannot be enslaved");
1795 if (!bond->params.use_carrier &&
1796 slave_dev->ethtool_ops->get_link == NULL &&
1797 slave_ops->ndo_eth_ioctl == NULL) {
1798 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1801 /* already in-use? */
1802 if (netdev_is_rx_handler_busy(slave_dev)) {
1803 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1804 "Device is in use and cannot be enslaved");
1808 if (bond_dev == slave_dev) {
1809 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1813 /* vlan challenged mutual exclusion */
1814 /* no need to lock since we're protected by rtnl_lock */
1815 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1816 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1817 if (vlan_uses_dev(bond_dev)) {
1818 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1819 "Can not enslave VLAN challenged device to VLAN enabled bond");
1822 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1825 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1828 if (slave_dev->features & NETIF_F_HW_ESP)
1829 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1831 /* Old ifenslave binaries are no longer supported. These can
1832 * be identified with moderate accuracy by the state of the slave:
1833 * the current ifenslave will set the interface down prior to
1834 * enslaving it; the old ifenslave will not.
1836 if (slave_dev->flags & IFF_UP) {
1837 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1838 "Device can not be enslaved while up");
1842 /* set bonding device ether type by slave - bonding netdevices are
1843 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1844 * there is a need to override some of the type dependent attribs/funcs.
1846 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1847 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1849 if (!bond_has_slaves(bond)) {
1850 if (bond_dev->type != slave_dev->type) {
1851 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1852 bond_dev->type, slave_dev->type);
1854 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1856 res = notifier_to_errno(res);
1858 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1862 /* Flush unicast and multicast addresses */
1863 dev_uc_flush(bond_dev);
1864 dev_mc_flush(bond_dev);
1866 if (slave_dev->type != ARPHRD_ETHER)
1867 bond_setup_by_slave(bond_dev, slave_dev);
1869 ether_setup(bond_dev);
1870 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1873 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1876 } else if (bond_dev->type != slave_dev->type) {
1877 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1878 "Device type is different from other slaves");
1882 if (slave_dev->type == ARPHRD_INFINIBAND &&
1883 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1884 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1885 "Only active-backup mode is supported for infiniband slaves");
1887 goto err_undo_flags;
1890 if (!slave_ops->ndo_set_mac_address ||
1891 slave_dev->type == ARPHRD_INFINIBAND) {
1892 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1893 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1894 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1895 if (!bond_has_slaves(bond)) {
1896 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1897 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1899 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1900 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1902 goto err_undo_flags;
1907 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1909 /* If this is the first slave, then we need to set the master's hardware
1910 * address to be the same as the slave's.
1912 if (!bond_has_slaves(bond) &&
1913 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1914 res = bond_set_dev_addr(bond->dev, slave_dev);
1916 goto err_undo_flags;
1919 new_slave = bond_alloc_slave(bond, slave_dev);
1922 goto err_undo_flags;
1925 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1926 * is set via sysfs or module option if desired.
1928 new_slave->queue_id = 0;
1930 /* Save slave's original mtu and then set it to match the bond */
1931 new_slave->original_mtu = slave_dev->mtu;
1932 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1934 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1938 /* Save slave's original ("permanent") mac address for modes
1939 * that need it, and for restoring it upon release, and then
1940 * set it to the master's address
1942 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1943 slave_dev->addr_len);
1945 if (!bond->params.fail_over_mac ||
1946 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1947 /* Set slave to master's mac address. The application already
1948 * set the master's mac address to that of the first slave
1950 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1951 ss.ss_family = slave_dev->type;
1952 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1955 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1956 goto err_restore_mtu;
1960 /* set slave flag before open to prevent IPv6 addrconf */
1961 slave_dev->flags |= IFF_SLAVE;
1963 /* open the slave since the application closed it */
1964 res = dev_open(slave_dev, extack);
1966 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1967 goto err_restore_mac;
1970 slave_dev->priv_flags |= IFF_BONDING;
1971 /* initialize slave stats */
1972 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1974 if (bond_is_lb(bond)) {
1975 /* bond_alb_init_slave() must be called before all other stages since
1976 * it might fail and we do not want to have to undo everything
1978 res = bond_alb_init_slave(bond, new_slave);
1983 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1985 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1989 prev_slave = bond_last_slave(bond);
1991 new_slave->delay = 0;
1992 new_slave->link_failure_count = 0;
1994 if (bond_update_speed_duplex(new_slave) &&
1995 bond_needs_speed_duplex(bond))
1996 new_slave->link = BOND_LINK_DOWN;
1998 new_slave->last_rx = jiffies -
1999 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2000 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2001 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2003 new_slave->last_tx = new_slave->last_rx;
2005 if (bond->params.miimon && !bond->params.use_carrier) {
2006 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2008 if ((link_reporting == -1) && !bond->params.arp_interval) {
2009 /* miimon is set but a bonded network driver
2010 * does not support ETHTOOL/MII and
2011 * arp_interval is not set. Note: if
2012 * use_carrier is enabled, we will never go
2013 * here (because netif_carrier is always
2014 * supported); thus, we don't need to change
2015 * the messages for netif_carrier.
2017 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");
2018 } else if (link_reporting == -1) {
2019 /* unable get link status using mii/ethtool */
2020 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");
2024 /* check for initial state */
2025 new_slave->link = BOND_LINK_NOCHANGE;
2026 if (bond->params.miimon) {
2027 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2028 if (bond->params.updelay) {
2029 bond_set_slave_link_state(new_slave,
2031 BOND_SLAVE_NOTIFY_NOW);
2032 new_slave->delay = bond->params.updelay;
2034 bond_set_slave_link_state(new_slave,
2036 BOND_SLAVE_NOTIFY_NOW);
2039 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2040 BOND_SLAVE_NOTIFY_NOW);
2042 } else if (bond->params.arp_interval) {
2043 bond_set_slave_link_state(new_slave,
2044 (netif_carrier_ok(slave_dev) ?
2045 BOND_LINK_UP : BOND_LINK_DOWN),
2046 BOND_SLAVE_NOTIFY_NOW);
2048 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2049 BOND_SLAVE_NOTIFY_NOW);
2052 if (new_slave->link != BOND_LINK_DOWN)
2053 new_slave->last_link_up = jiffies;
2054 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2055 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2056 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2058 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2059 /* if there is a primary slave, remember it */
2060 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2061 rcu_assign_pointer(bond->primary_slave, new_slave);
2062 bond->force_primary = true;
2066 switch (BOND_MODE(bond)) {
2067 case BOND_MODE_ACTIVEBACKUP:
2068 bond_set_slave_inactive_flags(new_slave,
2069 BOND_SLAVE_NOTIFY_NOW);
2071 case BOND_MODE_8023AD:
2072 /* in 802.3ad mode, the internal mechanism
2073 * will activate the slaves in the selected
2076 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2077 /* if this is the first slave */
2079 SLAVE_AD_INFO(new_slave)->id = 1;
2080 /* Initialize AD with the number of times that the AD timer is called in 1 second
2081 * can be called only after the mac address of the bond is set
2083 bond_3ad_initialize(bond);
2085 SLAVE_AD_INFO(new_slave)->id =
2086 SLAVE_AD_INFO(prev_slave)->id + 1;
2089 bond_3ad_bind_slave(new_slave);
2093 bond_set_active_slave(new_slave);
2094 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2097 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2099 /* always active in trunk mode */
2100 bond_set_active_slave(new_slave);
2102 /* In trunking mode there is little meaning to curr_active_slave
2103 * anyway (it holds no special properties of the bond device),
2104 * so we can change it without calling change_active_interface()
2106 if (!rcu_access_pointer(bond->curr_active_slave) &&
2107 new_slave->link == BOND_LINK_UP)
2108 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2111 } /* switch(bond_mode) */
2113 #ifdef CONFIG_NET_POLL_CONTROLLER
2114 if (bond->dev->npinfo) {
2115 if (slave_enable_netpoll(new_slave)) {
2116 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2123 if (!(bond_dev->features & NETIF_F_LRO))
2124 dev_disable_lro(slave_dev);
2126 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2129 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2133 res = bond_master_upper_dev_link(bond, new_slave, extack);
2135 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2136 goto err_unregister;
2139 bond_lower_state_changed(new_slave);
2141 res = bond_sysfs_slave_add(new_slave);
2143 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2144 goto err_upper_unlink;
2147 /* If the mode uses primary, then the following is handled by
2148 * bond_change_active_slave().
2150 if (!bond_uses_primary(bond)) {
2151 /* set promiscuity level to new slave */
2152 if (bond_dev->flags & IFF_PROMISC) {
2153 res = dev_set_promiscuity(slave_dev, 1);
2158 /* set allmulti level to new slave */
2159 if (bond_dev->flags & IFF_ALLMULTI) {
2160 res = dev_set_allmulti(slave_dev, 1);
2162 if (bond_dev->flags & IFF_PROMISC)
2163 dev_set_promiscuity(slave_dev, -1);
2168 if (bond_dev->flags & IFF_UP) {
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 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2180 bond_compute_features(bond);
2181 bond_set_carrier(bond);
2183 if (bond_uses_primary(bond)) {
2185 bond_select_active_slave(bond);
2186 unblock_netpoll_tx();
2189 if (bond_mode_can_use_xmit_hash(bond))
2190 bond_update_slave_arr(bond, NULL);
2193 if (!slave_dev->netdev_ops->ndo_bpf ||
2194 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2195 if (bond->xdp_prog) {
2196 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2197 "Slave does not support XDP");
2201 } else if (bond->xdp_prog) {
2202 struct netdev_bpf xdp = {
2203 .command = XDP_SETUP_PROG,
2205 .prog = bond->xdp_prog,
2209 if (dev_xdp_prog_count(slave_dev) > 0) {
2210 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2211 "Slave has XDP program loaded, please unload before enslaving");
2216 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2218 /* ndo_bpf() sets extack error message */
2219 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2223 bpf_prog_inc(bond->xdp_prog);
2226 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2227 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2228 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2230 /* enslave is successful */
2231 bond_queue_slave_event(new_slave);
2234 /* Undo stages on error */
2236 bond_sysfs_slave_del(new_slave);
2239 bond_upper_dev_unlink(bond, new_slave);
2242 netdev_rx_handler_unregister(slave_dev);
2245 vlan_vids_del_by_dev(slave_dev, bond_dev);
2246 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2247 RCU_INIT_POINTER(bond->primary_slave, NULL);
2248 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2250 bond_change_active_slave(bond, NULL);
2251 bond_select_active_slave(bond);
2252 unblock_netpoll_tx();
2254 /* either primary_slave or curr_active_slave might've changed */
2256 slave_disable_netpoll(new_slave);
2259 if (!netif_is_bond_master(slave_dev))
2260 slave_dev->priv_flags &= ~IFF_BONDING;
2261 dev_close(slave_dev);
2264 slave_dev->flags &= ~IFF_SLAVE;
2265 if (!bond->params.fail_over_mac ||
2266 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2267 /* XXX TODO - fom follow mode needs to change master's
2268 * MAC if this slave's MAC is in use by the bond, or at
2269 * least print a warning.
2271 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2272 new_slave->dev->addr_len);
2273 ss.ss_family = slave_dev->type;
2274 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2278 dev_set_mtu(slave_dev, new_slave->original_mtu);
2281 kobject_put(&new_slave->kobj);
2284 /* Enslave of first slave has failed and we need to fix master's mac */
2285 if (!bond_has_slaves(bond)) {
2286 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2287 slave_dev->dev_addr))
2288 eth_hw_addr_random(bond_dev);
2289 if (bond_dev->type != ARPHRD_ETHER) {
2290 dev_close(bond_dev);
2291 ether_setup(bond_dev);
2292 bond_dev->flags |= IFF_MASTER;
2293 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2300 /* Try to release the slave device <slave> from the bond device <master>
2301 * It is legal to access curr_active_slave without a lock because all the function
2302 * is RTNL-locked. If "all" is true it means that the function is being called
2303 * while destroying a bond interface and all slaves are being released.
2305 * The rules for slave state should be:
2306 * for Active/Backup:
2307 * Active stays on all backups go down
2308 * for Bonded connections:
2309 * The first up interface should be left on and all others downed.
2311 static int __bond_release_one(struct net_device *bond_dev,
2312 struct net_device *slave_dev,
2313 bool all, bool unregister)
2315 struct bonding *bond = netdev_priv(bond_dev);
2316 struct slave *slave, *oldcurrent;
2317 struct sockaddr_storage ss;
2318 int old_flags = bond_dev->flags;
2319 netdev_features_t old_features = bond_dev->features;
2321 /* slave is not a slave or master is not master of this slave */
2322 if (!(slave_dev->flags & IFF_SLAVE) ||
2323 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2324 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2330 slave = bond_get_slave_by_dev(bond, slave_dev);
2332 /* not a slave of this bond */
2333 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2334 unblock_netpoll_tx();
2338 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2340 bond_sysfs_slave_del(slave);
2342 /* recompute stats just before removing the slave */
2343 bond_get_stats(bond->dev, &bond->bond_stats);
2345 if (bond->xdp_prog) {
2346 struct netdev_bpf xdp = {
2347 .command = XDP_SETUP_PROG,
2352 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2353 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2356 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2357 * for this slave anymore.
2359 netdev_rx_handler_unregister(slave_dev);
2361 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2362 bond_3ad_unbind_slave(slave);
2364 bond_upper_dev_unlink(bond, slave);
2366 if (bond_mode_can_use_xmit_hash(bond))
2367 bond_update_slave_arr(bond, slave);
2369 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2370 bond_is_active_slave(slave) ? "active" : "backup");
2372 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2374 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2376 if (!all && (!bond->params.fail_over_mac ||
2377 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2378 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2379 bond_has_slaves(bond))
2380 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",
2381 slave->perm_hwaddr);
2384 if (rtnl_dereference(bond->primary_slave) == slave)
2385 RCU_INIT_POINTER(bond->primary_slave, NULL);
2387 if (oldcurrent == slave)
2388 bond_change_active_slave(bond, NULL);
2390 if (bond_is_lb(bond)) {
2391 /* Must be called only after the slave has been
2392 * detached from the list and the curr_active_slave
2393 * has been cleared (if our_slave == old_current),
2394 * but before a new active slave is selected.
2396 bond_alb_deinit_slave(bond, slave);
2400 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2401 } else if (oldcurrent == slave) {
2402 /* Note that we hold RTNL over this sequence, so there
2403 * is no concern that another slave add/remove event
2406 bond_select_active_slave(bond);
2409 bond_set_carrier(bond);
2410 if (!bond_has_slaves(bond))
2411 eth_hw_addr_random(bond_dev);
2413 unblock_netpoll_tx();
2417 if (!bond_has_slaves(bond)) {
2418 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2419 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2422 bond_compute_features(bond);
2423 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2424 (old_features & NETIF_F_VLAN_CHALLENGED))
2425 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2427 vlan_vids_del_by_dev(slave_dev, bond_dev);
2429 /* If the mode uses primary, then this case was handled above by
2430 * bond_change_active_slave(..., NULL)
2432 if (!bond_uses_primary(bond)) {
2433 /* unset promiscuity level from slave
2434 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2435 * of the IFF_PROMISC flag in the bond_dev, but we need the
2436 * value of that flag before that change, as that was the value
2437 * when this slave was attached, so we cache at the start of the
2438 * function and use it here. Same goes for ALLMULTI below
2440 if (old_flags & IFF_PROMISC)
2441 dev_set_promiscuity(slave_dev, -1);
2443 /* unset allmulti level from slave */
2444 if (old_flags & IFF_ALLMULTI)
2445 dev_set_allmulti(slave_dev, -1);
2447 if (old_flags & IFF_UP)
2448 bond_hw_addr_flush(bond_dev, slave_dev);
2451 slave_disable_netpoll(slave);
2453 /* close slave before restoring its mac address */
2454 dev_close(slave_dev);
2456 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2457 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2458 /* restore original ("permanent") mac address */
2459 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2460 slave->dev->addr_len);
2461 ss.ss_family = slave_dev->type;
2462 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2466 __dev_set_mtu(slave_dev, slave->original_mtu);
2468 dev_set_mtu(slave_dev, slave->original_mtu);
2470 if (!netif_is_bond_master(slave_dev))
2471 slave_dev->priv_flags &= ~IFF_BONDING;
2473 kobject_put(&slave->kobj);
2478 /* A wrapper used because of ndo_del_link */
2479 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2481 return __bond_release_one(bond_dev, slave_dev, false, false);
2484 /* First release a slave and then destroy the bond if no more slaves are left.
2485 * Must be under rtnl_lock when this function is called.
2487 static int bond_release_and_destroy(struct net_device *bond_dev,
2488 struct net_device *slave_dev)
2490 struct bonding *bond = netdev_priv(bond_dev);
2493 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2494 if (ret == 0 && !bond_has_slaves(bond) &&
2495 bond_dev->reg_state != NETREG_UNREGISTERING) {
2496 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2497 netdev_info(bond_dev, "Destroying bond\n");
2498 bond_remove_proc_entry(bond);
2499 unregister_netdevice(bond_dev);
2504 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2506 struct bonding *bond = netdev_priv(bond_dev);
2508 bond_fill_ifbond(bond, info);
2511 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2513 struct bonding *bond = netdev_priv(bond_dev);
2514 struct list_head *iter;
2515 int i = 0, res = -ENODEV;
2516 struct slave *slave;
2518 bond_for_each_slave(bond, slave, iter) {
2519 if (i++ == (int)info->slave_id) {
2521 bond_fill_ifslave(slave, info);
2529 /*-------------------------------- Monitoring -------------------------------*/
2531 /* called with rcu_read_lock() */
2532 static int bond_miimon_inspect(struct bonding *bond)
2534 int link_state, commit = 0;
2535 struct list_head *iter;
2536 struct slave *slave;
2537 bool ignore_updelay;
2539 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2541 bond_for_each_slave_rcu(bond, slave, iter) {
2542 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2544 link_state = bond_check_dev_link(bond, slave->dev, 0);
2546 switch (slave->link) {
2551 bond_propose_link_state(slave, BOND_LINK_FAIL);
2553 slave->delay = bond->params.downdelay;
2555 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2557 BOND_MODE_ACTIVEBACKUP) ?
2558 (bond_is_active_slave(slave) ?
2559 "active " : "backup ") : "",
2560 bond->params.downdelay * bond->params.miimon);
2563 case BOND_LINK_FAIL:
2565 /* recovered before downdelay expired */
2566 bond_propose_link_state(slave, BOND_LINK_UP);
2567 slave->last_link_up = jiffies;
2568 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2569 (bond->params.downdelay - slave->delay) *
2570 bond->params.miimon);
2575 if (slave->delay <= 0) {
2576 bond_propose_link_state(slave, BOND_LINK_DOWN);
2584 case BOND_LINK_DOWN:
2588 bond_propose_link_state(slave, BOND_LINK_BACK);
2590 slave->delay = bond->params.updelay;
2593 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2594 ignore_updelay ? 0 :
2595 bond->params.updelay *
2596 bond->params.miimon);
2599 case BOND_LINK_BACK:
2601 bond_propose_link_state(slave, BOND_LINK_DOWN);
2602 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2603 (bond->params.updelay - slave->delay) *
2604 bond->params.miimon);
2612 if (slave->delay <= 0) {
2613 bond_propose_link_state(slave, BOND_LINK_UP);
2615 ignore_updelay = false;
2627 static void bond_miimon_link_change(struct bonding *bond,
2628 struct slave *slave,
2631 switch (BOND_MODE(bond)) {
2632 case BOND_MODE_8023AD:
2633 bond_3ad_handle_link_change(slave, link);
2637 bond_alb_handle_link_change(bond, slave, link);
2640 bond_update_slave_arr(bond, NULL);
2645 static void bond_miimon_commit(struct bonding *bond)
2647 struct list_head *iter;
2648 struct slave *slave, *primary;
2650 bond_for_each_slave(bond, slave, iter) {
2651 switch (slave->link_new_state) {
2652 case BOND_LINK_NOCHANGE:
2653 /* For 802.3ad mode, check current slave speed and
2654 * duplex again in case its port was disabled after
2655 * invalid speed/duplex reporting but recovered before
2656 * link monitoring could make a decision on the actual
2659 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2660 slave->link == BOND_LINK_UP)
2661 bond_3ad_adapter_speed_duplex_changed(slave);
2665 if (bond_update_speed_duplex(slave) &&
2666 bond_needs_speed_duplex(bond)) {
2667 slave->link = BOND_LINK_DOWN;
2668 if (net_ratelimit())
2669 slave_warn(bond->dev, slave->dev,
2670 "failed to get link speed/duplex\n");
2673 bond_set_slave_link_state(slave, BOND_LINK_UP,
2674 BOND_SLAVE_NOTIFY_NOW);
2675 slave->last_link_up = jiffies;
2677 primary = rtnl_dereference(bond->primary_slave);
2678 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2679 /* prevent it from being the active one */
2680 bond_set_backup_slave(slave);
2681 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2682 /* make it immediately active */
2683 bond_set_active_slave(slave);
2686 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2687 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2688 slave->duplex ? "full" : "half");
2690 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2692 if (!bond->curr_active_slave || slave == primary)
2697 case BOND_LINK_DOWN:
2698 if (slave->link_failure_count < UINT_MAX)
2699 slave->link_failure_count++;
2701 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2702 BOND_SLAVE_NOTIFY_NOW);
2704 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2705 BOND_MODE(bond) == BOND_MODE_8023AD)
2706 bond_set_slave_inactive_flags(slave,
2707 BOND_SLAVE_NOTIFY_NOW);
2709 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2711 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2713 if (slave == rcu_access_pointer(bond->curr_active_slave))
2719 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2720 slave->link_new_state);
2721 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2728 bond_select_active_slave(bond);
2729 unblock_netpoll_tx();
2732 bond_set_carrier(bond);
2737 * Really a wrapper that splits the mii monitor into two phases: an
2738 * inspection, then (if inspection indicates something needs to be done)
2739 * an acquisition of appropriate locks followed by a commit phase to
2740 * implement whatever link state changes are indicated.
2742 static void bond_mii_monitor(struct work_struct *work)
2744 struct bonding *bond = container_of(work, struct bonding,
2746 bool should_notify_peers = false;
2748 unsigned long delay;
2749 struct slave *slave;
2750 struct list_head *iter;
2752 delay = msecs_to_jiffies(bond->params.miimon);
2754 if (!bond_has_slaves(bond))
2758 should_notify_peers = bond_should_notify_peers(bond);
2759 commit = !!bond_miimon_inspect(bond);
2760 if (bond->send_peer_notif) {
2762 if (rtnl_trylock()) {
2763 bond->send_peer_notif--;
2771 /* Race avoidance with bond_close cancel of workqueue */
2772 if (!rtnl_trylock()) {
2774 should_notify_peers = false;
2778 bond_for_each_slave(bond, slave, iter) {
2779 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2781 bond_miimon_commit(bond);
2783 rtnl_unlock(); /* might sleep, hold no other locks */
2787 if (bond->params.miimon)
2788 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2790 if (should_notify_peers) {
2791 if (!rtnl_trylock())
2793 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2798 static int bond_upper_dev_walk(struct net_device *upper,
2799 struct netdev_nested_priv *priv)
2801 __be32 ip = *(__be32 *)priv->data;
2803 return ip == bond_confirm_addr(upper, 0, ip);
2806 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2808 struct netdev_nested_priv priv = {
2809 .data = (void *)&ip,
2813 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2817 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2824 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2825 struct sk_buff *skb)
2827 struct net_device *bond_dev = slave->bond->dev;
2828 struct net_device *slave_dev = slave->dev;
2829 struct bond_vlan_tag *outer_tag = tags;
2831 if (!tags || tags->vlan_proto == VLAN_N_VID)
2836 /* Go through all the tags backwards and add them to the packet */
2837 while (tags->vlan_proto != VLAN_N_VID) {
2838 if (!tags->vlan_id) {
2843 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2844 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2845 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2848 net_err_ratelimited("failed to insert inner VLAN tag\n");
2854 /* Set the outer tag */
2855 if (outer_tag->vlan_id) {
2856 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2857 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2858 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2859 outer_tag->vlan_id);
2865 /* We go to the (large) trouble of VLAN tagging ARP frames because
2866 * switches in VLAN mode (especially if ports are configured as
2867 * "native" to a VLAN) might not pass non-tagged frames.
2869 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2870 __be32 src_ip, struct bond_vlan_tag *tags)
2872 struct net_device *bond_dev = slave->bond->dev;
2873 struct net_device *slave_dev = slave->dev;
2874 struct sk_buff *skb;
2876 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2877 arp_op, &dest_ip, &src_ip);
2879 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2880 NULL, slave_dev->dev_addr, NULL);
2883 net_err_ratelimited("ARP packet allocation failed\n");
2887 if (bond_handle_vlan(slave, tags, skb)) {
2888 slave_update_last_tx(slave);
2895 /* Validate the device path between the @start_dev and the @end_dev.
2896 * The path is valid if the @end_dev is reachable through device
2898 * When the path is validated, collect any vlan information in the
2901 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2902 struct net_device *end_dev,
2905 struct bond_vlan_tag *tags;
2906 struct net_device *upper;
2907 struct list_head *iter;
2909 if (start_dev == end_dev) {
2910 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2912 return ERR_PTR(-ENOMEM);
2913 tags[level].vlan_proto = VLAN_N_VID;
2917 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2918 tags = bond_verify_device_path(upper, end_dev, level + 1);
2919 if (IS_ERR_OR_NULL(tags)) {
2924 if (is_vlan_dev(upper)) {
2925 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2926 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2935 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2938 struct bond_vlan_tag *tags;
2939 __be32 *targets = bond->params.arp_targets, addr;
2942 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2943 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2944 __func__, &targets[i]);
2947 /* Find out through which dev should the packet go */
2948 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2951 /* there's no route to target - try to send arp
2952 * probe to generate any traffic (arp_validate=0)
2954 if (bond->params.arp_validate)
2955 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2958 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2963 /* bond device itself */
2964 if (rt->dst.dev == bond->dev)
2968 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2971 if (!IS_ERR_OR_NULL(tags))
2974 /* Not our device - skip */
2975 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2976 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2982 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2984 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2989 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2993 if (!sip || !bond_has_this_ip(bond, tip)) {
2994 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2995 __func__, &sip, &tip);
2999 i = bond_get_targets_ip(bond->params.arp_targets, sip);
3001 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3005 slave->last_rx = jiffies;
3006 slave->target_last_arp_rx[i] = jiffies;
3009 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3010 struct slave *slave)
3012 struct arphdr *arp = (struct arphdr *)skb->data;
3013 struct slave *curr_active_slave, *curr_arp_slave;
3014 unsigned char *arp_ptr;
3018 alen = arp_hdr_len(bond->dev);
3020 if (alen > skb_headlen(skb)) {
3021 arp = kmalloc(alen, GFP_ATOMIC);
3024 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3028 if (arp->ar_hln != bond->dev->addr_len ||
3029 skb->pkt_type == PACKET_OTHERHOST ||
3030 skb->pkt_type == PACKET_LOOPBACK ||
3031 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3032 arp->ar_pro != htons(ETH_P_IP) ||
3036 arp_ptr = (unsigned char *)(arp + 1);
3037 arp_ptr += bond->dev->addr_len;
3038 memcpy(&sip, arp_ptr, 4);
3039 arp_ptr += 4 + bond->dev->addr_len;
3040 memcpy(&tip, arp_ptr, 4);
3042 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3043 __func__, slave->dev->name, bond_slave_state(slave),
3044 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3047 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3048 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3050 /* We 'trust' the received ARP enough to validate it if:
3052 * (a) the slave receiving the ARP is active (which includes the
3053 * current ARP slave, if any), or
3055 * (b) the receiving slave isn't active, but there is a currently
3056 * active slave and it received valid arp reply(s) after it became
3057 * the currently active slave, or
3059 * (c) there is an ARP slave that sent an ARP during the prior ARP
3060 * interval, and we receive an ARP reply on any slave. We accept
3061 * these because switch FDB update delays may deliver the ARP
3062 * reply to a slave other than the sender of the ARP request.
3064 * Note: for (b), backup slaves are receiving the broadcast ARP
3065 * request, not a reply. This request passes from the sending
3066 * slave through the L2 switch(es) to the receiving slave. Since
3067 * this is checking the request, sip/tip are swapped for
3070 * This is done to avoid endless looping when we can't reach the
3071 * arp_ip_target and fool ourselves with our own arp requests.
3073 if (bond_is_active_slave(slave))
3074 bond_validate_arp(bond, slave, sip, tip);
3075 else if (curr_active_slave &&
3076 time_after(slave_last_rx(bond, curr_active_slave),
3077 curr_active_slave->last_link_up))
3078 bond_validate_arp(bond, slave, tip, sip);
3079 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3080 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3081 bond_validate_arp(bond, slave, sip, tip);
3084 if (arp != (struct arphdr *)skb->data)
3086 return RX_HANDLER_ANOTHER;
3089 #if IS_ENABLED(CONFIG_IPV6)
3090 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3091 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3093 struct net_device *bond_dev = slave->bond->dev;
3094 struct net_device *slave_dev = slave->dev;
3095 struct in6_addr mcaddr;
3096 struct sk_buff *skb;
3098 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3101 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3103 net_err_ratelimited("NS packet allocation failed\n");
3107 addrconf_addr_solict_mult(daddr, &mcaddr);
3108 if (bond_handle_vlan(slave, tags, skb)) {
3109 slave_update_last_tx(slave);
3110 ndisc_send_skb(skb, &mcaddr, saddr);
3114 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3116 struct in6_addr *targets = bond->params.ns_targets;
3117 struct bond_vlan_tag *tags;
3118 struct dst_entry *dst;
3119 struct in6_addr saddr;
3123 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3124 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3125 __func__, &targets[i]);
3128 /* Find out through which dev should the packet go */
3129 memset(&fl6, 0, sizeof(struct flowi6));
3130 fl6.daddr = targets[i];
3131 fl6.flowi6_oif = bond->dev->ifindex;
3133 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3136 /* there's no route to target - try to send arp
3137 * probe to generate any traffic (arp_validate=0)
3139 if (bond->params.arp_validate)
3140 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3143 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3147 /* bond device itself */
3148 if (dst->dev == bond->dev)
3152 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3155 if (!IS_ERR_OR_NULL(tags))
3158 /* Not our device - skip */
3159 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3160 &targets[i], dst->dev ? dst->dev->name : "NULL");
3166 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3167 bond_ns_send(slave, &targets[i], &saddr, tags);
3169 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3176 static int bond_confirm_addr6(struct net_device *dev,
3177 struct netdev_nested_priv *priv)
3179 struct in6_addr *addr = (struct in6_addr *)priv->data;
3181 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3184 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3186 struct netdev_nested_priv priv = {
3191 if (bond_confirm_addr6(bond->dev, &priv))
3195 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3202 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3203 struct in6_addr *saddr, struct in6_addr *daddr)
3208 * 1. Source address is unspecified address.
3209 * 2. Dest address is neither all-nodes multicast address nor
3210 * exist on bond interface.
3212 if (ipv6_addr_any(saddr) ||
3213 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3214 !bond_has_this_ip6(bond, daddr))) {
3215 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3216 __func__, saddr, daddr);
3220 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3222 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3226 slave->last_rx = jiffies;
3227 slave->target_last_arp_rx[i] = jiffies;
3230 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3231 struct slave *slave)
3233 struct slave *curr_active_slave, *curr_arp_slave;
3234 struct icmp6hdr *hdr = icmp6_hdr(skb);
3235 struct in6_addr *saddr, *daddr;
3237 if (skb->pkt_type == PACKET_OTHERHOST ||
3238 skb->pkt_type == PACKET_LOOPBACK ||
3239 hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
3242 saddr = &ipv6_hdr(skb)->saddr;
3243 daddr = &ipv6_hdr(skb)->daddr;
3245 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3246 __func__, slave->dev->name, bond_slave_state(slave),
3247 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3250 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3251 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3253 /* We 'trust' the received ARP enough to validate it if:
3254 * see bond_arp_rcv().
3256 if (bond_is_active_slave(slave))
3257 bond_validate_na(bond, slave, saddr, daddr);
3258 else if (curr_active_slave &&
3259 time_after(slave_last_rx(bond, curr_active_slave),
3260 curr_active_slave->last_link_up))
3261 bond_validate_na(bond, slave, saddr, daddr);
3262 else if (curr_arp_slave &&
3263 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3264 bond_validate_na(bond, slave, saddr, daddr);
3267 return RX_HANDLER_ANOTHER;
3271 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3272 struct slave *slave)
3274 #if IS_ENABLED(CONFIG_IPV6)
3275 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3277 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3279 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3280 __func__, skb->dev->name);
3282 /* Use arp validate logic for both ARP and NS */
3283 if (!slave_do_arp_validate(bond, slave)) {
3284 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3285 #if IS_ENABLED(CONFIG_IPV6)
3286 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3288 !slave_do_arp_validate_only(bond))
3289 slave->last_rx = jiffies;
3290 return RX_HANDLER_ANOTHER;
3291 } else if (is_arp) {
3292 return bond_arp_rcv(skb, bond, slave);
3293 #if IS_ENABLED(CONFIG_IPV6)
3294 } else if (is_ipv6) {
3295 return bond_na_rcv(skb, bond, slave);
3298 return RX_HANDLER_ANOTHER;
3302 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3304 bond_arp_send_all(bond, slave);
3305 #if IS_ENABLED(CONFIG_IPV6)
3306 bond_ns_send_all(bond, slave);
3310 /* function to verify if we're in the arp_interval timeslice, returns true if
3311 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3312 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3314 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3317 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3319 return time_in_range(jiffies,
3320 last_act - delta_in_ticks,
3321 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3324 /* This function is called regularly to monitor each slave's link
3325 * ensuring that traffic is being sent and received when arp monitoring
3326 * is used in load-balancing mode. if the adapter has been dormant, then an
3327 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3328 * arp monitoring in active backup mode.
3330 static void bond_loadbalance_arp_mon(struct bonding *bond)
3332 struct slave *slave, *oldcurrent;
3333 struct list_head *iter;
3334 int do_failover = 0, slave_state_changed = 0;
3336 if (!bond_has_slaves(bond))
3341 oldcurrent = rcu_dereference(bond->curr_active_slave);
3342 /* see if any of the previous devices are up now (i.e. they have
3343 * xmt and rcv traffic). the curr_active_slave does not come into
3344 * the picture unless it is null. also, slave->last_link_up is not
3345 * needed here because we send an arp on each slave and give a slave
3346 * as long as it needs to get the tx/rx within the delta.
3347 * TODO: what about up/down delay in arp mode? it wasn't here before
3350 bond_for_each_slave_rcu(bond, slave, iter) {
3351 unsigned long last_tx = slave_last_tx(slave);
3353 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3355 if (slave->link != BOND_LINK_UP) {
3356 if (bond_time_in_interval(bond, last_tx, 1) &&
3357 bond_time_in_interval(bond, slave->last_rx, 1)) {
3359 bond_propose_link_state(slave, BOND_LINK_UP);
3360 slave_state_changed = 1;
3362 /* primary_slave has no meaning in round-robin
3363 * mode. the window of a slave being up and
3364 * curr_active_slave being null after enslaving
3368 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3371 slave_info(bond->dev, slave->dev, "interface is now up\n");
3375 /* slave->link == BOND_LINK_UP */
3377 /* not all switches will respond to an arp request
3378 * when the source ip is 0, so don't take the link down
3379 * if we don't know our ip yet
3381 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3382 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3384 bond_propose_link_state(slave, BOND_LINK_DOWN);
3385 slave_state_changed = 1;
3387 if (slave->link_failure_count < UINT_MAX)
3388 slave->link_failure_count++;
3390 slave_info(bond->dev, slave->dev, "interface is now down\n");
3392 if (slave == oldcurrent)
3397 /* note: if switch is in round-robin mode, all links
3398 * must tx arp to ensure all links rx an arp - otherwise
3399 * links may oscillate or not come up at all; if switch is
3400 * in something like xor mode, there is nothing we can
3401 * do - all replies will be rx'ed on same link causing slaves
3402 * to be unstable during low/no traffic periods
3404 if (bond_slave_is_up(slave))
3405 bond_send_validate(bond, slave);
3410 if (do_failover || slave_state_changed) {
3411 if (!rtnl_trylock())
3414 bond_for_each_slave(bond, slave, iter) {
3415 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3416 slave->link = slave->link_new_state;
3419 if (slave_state_changed) {
3420 bond_slave_state_change(bond);
3421 if (BOND_MODE(bond) == BOND_MODE_XOR)
3422 bond_update_slave_arr(bond, NULL);
3426 bond_select_active_slave(bond);
3427 unblock_netpoll_tx();
3433 if (bond->params.arp_interval)
3434 queue_delayed_work(bond->wq, &bond->arp_work,
3435 msecs_to_jiffies(bond->params.arp_interval));
3438 /* Called to inspect slaves for active-backup mode ARP monitor link state
3439 * changes. Sets proposed link state in slaves to specify what action
3440 * should take place for the slave. Returns 0 if no changes are found, >0
3441 * if changes to link states must be committed.
3443 * Called with rcu_read_lock held.
3445 static int bond_ab_arp_inspect(struct bonding *bond)
3447 unsigned long last_tx, last_rx;
3448 struct list_head *iter;
3449 struct slave *slave;
3452 bond_for_each_slave_rcu(bond, slave, iter) {
3453 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3454 last_rx = slave_last_rx(bond, slave);
3456 if (slave->link != BOND_LINK_UP) {
3457 if (bond_time_in_interval(bond, last_rx, 1)) {
3458 bond_propose_link_state(slave, BOND_LINK_UP);
3460 } else if (slave->link == BOND_LINK_BACK) {
3461 bond_propose_link_state(slave, BOND_LINK_FAIL);
3467 /* Give slaves 2*delta after being enslaved or made
3468 * active. This avoids bouncing, as the last receive
3469 * times need a full ARP monitor cycle to be updated.
3471 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3474 /* Backup slave is down if:
3475 * - No current_arp_slave AND
3476 * - more than (missed_max+1)*delta since last receive AND
3477 * - the bond has an IP address
3479 * Note: a non-null current_arp_slave indicates
3480 * the curr_active_slave went down and we are
3481 * searching for a new one; under this condition
3482 * we only take the curr_active_slave down - this
3483 * gives each slave a chance to tx/rx traffic
3484 * before being taken out
3486 if (!bond_is_active_slave(slave) &&
3487 !rcu_access_pointer(bond->current_arp_slave) &&
3488 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3489 bond_propose_link_state(slave, BOND_LINK_DOWN);
3493 /* Active slave is down if:
3494 * - more than missed_max*delta since transmitting OR
3495 * - (more than missed_max*delta since receive AND
3496 * the bond has an IP address)
3498 last_tx = slave_last_tx(slave);
3499 if (bond_is_active_slave(slave) &&
3500 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3501 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3502 bond_propose_link_state(slave, BOND_LINK_DOWN);
3510 /* Called to commit link state changes noted by inspection step of
3511 * active-backup mode ARP monitor.
3513 * Called with RTNL hold.
3515 static void bond_ab_arp_commit(struct bonding *bond)
3517 struct list_head *iter;
3518 unsigned long last_tx;
3519 struct slave *slave;
3521 bond_for_each_slave(bond, slave, iter) {
3522 switch (slave->link_new_state) {
3523 case BOND_LINK_NOCHANGE:
3527 last_tx = slave_last_tx(slave);
3528 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3529 (!rtnl_dereference(bond->curr_active_slave) &&
3530 bond_time_in_interval(bond, last_tx, 1))) {
3531 struct slave *current_arp_slave;
3533 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3534 bond_set_slave_link_state(slave, BOND_LINK_UP,
3535 BOND_SLAVE_NOTIFY_NOW);
3536 if (current_arp_slave) {
3537 bond_set_slave_inactive_flags(
3539 BOND_SLAVE_NOTIFY_NOW);
3540 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3543 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3545 if (!rtnl_dereference(bond->curr_active_slave) ||
3546 slave == rtnl_dereference(bond->primary_slave))
3553 case BOND_LINK_DOWN:
3554 if (slave->link_failure_count < UINT_MAX)
3555 slave->link_failure_count++;
3557 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3558 BOND_SLAVE_NOTIFY_NOW);
3559 bond_set_slave_inactive_flags(slave,
3560 BOND_SLAVE_NOTIFY_NOW);
3562 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3564 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3565 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3571 case BOND_LINK_FAIL:
3572 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3573 BOND_SLAVE_NOTIFY_NOW);
3574 bond_set_slave_inactive_flags(slave,
3575 BOND_SLAVE_NOTIFY_NOW);
3577 /* A slave has just been enslaved and has become
3578 * the current active slave.
3580 if (rtnl_dereference(bond->curr_active_slave))
3581 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3585 slave_err(bond->dev, slave->dev,
3586 "impossible: link_new_state %d on slave\n",
3587 slave->link_new_state);
3593 bond_select_active_slave(bond);
3594 unblock_netpoll_tx();
3597 bond_set_carrier(bond);
3600 /* Send ARP probes for active-backup mode ARP monitor.
3602 * Called with rcu_read_lock held.
3604 static bool bond_ab_arp_probe(struct bonding *bond)
3606 struct slave *slave, *before = NULL, *new_slave = NULL,
3607 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3608 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3609 struct list_head *iter;
3611 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3613 if (curr_arp_slave && curr_active_slave)
3614 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3615 curr_arp_slave->dev->name,
3616 curr_active_slave->dev->name);
3618 if (curr_active_slave) {
3619 bond_send_validate(bond, curr_active_slave);
3620 return should_notify_rtnl;
3623 /* if we don't have a curr_active_slave, search for the next available
3624 * backup slave from the current_arp_slave and make it the candidate
3625 * for becoming the curr_active_slave
3628 if (!curr_arp_slave) {
3629 curr_arp_slave = bond_first_slave_rcu(bond);
3630 if (!curr_arp_slave)
3631 return should_notify_rtnl;
3634 bond_for_each_slave_rcu(bond, slave, iter) {
3635 if (!found && !before && bond_slave_is_up(slave))
3638 if (found && !new_slave && bond_slave_is_up(slave))
3640 /* if the link state is up at this point, we
3641 * mark it down - this can happen if we have
3642 * simultaneous link failures and
3643 * reselect_active_interface doesn't make this
3644 * one the current slave so it is still marked
3645 * up when it is actually down
3647 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3648 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3649 BOND_SLAVE_NOTIFY_LATER);
3650 if (slave->link_failure_count < UINT_MAX)
3651 slave->link_failure_count++;
3653 bond_set_slave_inactive_flags(slave,
3654 BOND_SLAVE_NOTIFY_LATER);
3656 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3658 if (slave == curr_arp_slave)
3662 if (!new_slave && before)
3668 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3669 BOND_SLAVE_NOTIFY_LATER);
3670 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3671 bond_send_validate(bond, new_slave);
3672 new_slave->last_link_up = jiffies;
3673 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3676 bond_for_each_slave_rcu(bond, slave, iter) {
3677 if (slave->should_notify || slave->should_notify_link) {
3678 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3682 return should_notify_rtnl;
3685 static void bond_activebackup_arp_mon(struct bonding *bond)
3687 bool should_notify_peers = false;
3688 bool should_notify_rtnl = false;
3691 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3693 if (!bond_has_slaves(bond))
3698 should_notify_peers = bond_should_notify_peers(bond);
3700 if (bond_ab_arp_inspect(bond)) {
3703 /* Race avoidance with bond_close flush of workqueue */
3704 if (!rtnl_trylock()) {
3706 should_notify_peers = false;
3710 bond_ab_arp_commit(bond);
3716 should_notify_rtnl = bond_ab_arp_probe(bond);
3720 if (bond->params.arp_interval)
3721 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3723 if (should_notify_peers || should_notify_rtnl) {
3724 if (!rtnl_trylock())
3727 if (should_notify_peers) {
3728 bond->send_peer_notif--;
3729 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3732 if (should_notify_rtnl) {
3733 bond_slave_state_notify(bond);
3734 bond_slave_link_notify(bond);
3741 static void bond_arp_monitor(struct work_struct *work)
3743 struct bonding *bond = container_of(work, struct bonding,
3746 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3747 bond_activebackup_arp_mon(bond);
3749 bond_loadbalance_arp_mon(bond);
3752 /*-------------------------- netdev event handling --------------------------*/
3754 /* Change device name */
3755 static int bond_event_changename(struct bonding *bond)
3757 bond_remove_proc_entry(bond);
3758 bond_create_proc_entry(bond);
3760 bond_debug_reregister(bond);
3765 static int bond_master_netdev_event(unsigned long event,
3766 struct net_device *bond_dev)
3768 struct bonding *event_bond = netdev_priv(bond_dev);
3770 netdev_dbg(bond_dev, "%s called\n", __func__);
3773 case NETDEV_CHANGENAME:
3774 return bond_event_changename(event_bond);
3775 case NETDEV_UNREGISTER:
3776 bond_remove_proc_entry(event_bond);
3777 #ifdef CONFIG_XFRM_OFFLOAD
3778 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3779 #endif /* CONFIG_XFRM_OFFLOAD */
3781 case NETDEV_REGISTER:
3782 bond_create_proc_entry(event_bond);
3791 static int bond_slave_netdev_event(unsigned long event,
3792 struct net_device *slave_dev)
3794 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3795 struct bonding *bond;
3796 struct net_device *bond_dev;
3798 /* A netdev event can be generated while enslaving a device
3799 * before netdev_rx_handler_register is called in which case
3800 * slave will be NULL
3803 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3807 bond_dev = slave->bond->dev;
3809 primary = rtnl_dereference(bond->primary_slave);
3811 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3814 case NETDEV_UNREGISTER:
3815 if (bond_dev->type != ARPHRD_ETHER)
3816 bond_release_and_destroy(bond_dev, slave_dev);
3818 __bond_release_one(bond_dev, slave_dev, false, true);
3822 /* For 802.3ad mode only:
3823 * Getting invalid Speed/Duplex values here will put slave
3824 * in weird state. Mark it as link-fail if the link was
3825 * previously up or link-down if it hasn't yet come up, and
3826 * let link-monitoring (miimon) set it right when correct
3827 * speeds/duplex are available.
3829 if (bond_update_speed_duplex(slave) &&
3830 BOND_MODE(bond) == BOND_MODE_8023AD) {
3831 if (slave->last_link_up)
3832 slave->link = BOND_LINK_FAIL;
3834 slave->link = BOND_LINK_DOWN;
3837 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3838 bond_3ad_adapter_speed_duplex_changed(slave);
3841 /* Refresh slave-array if applicable!
3842 * If the setup does not use miimon or arpmon (mode-specific!),
3843 * then these events will not cause the slave-array to be
3844 * refreshed. This will cause xmit to use a slave that is not
3845 * usable. Avoid such situation by refeshing the array at these
3846 * events. If these (miimon/arpmon) parameters are configured
3847 * then array gets refreshed twice and that should be fine!
3849 if (bond_mode_can_use_xmit_hash(bond))
3850 bond_update_slave_arr(bond, NULL);
3852 case NETDEV_CHANGEMTU:
3853 /* TODO: Should slaves be allowed to
3854 * independently alter their MTU? For
3855 * an active-backup bond, slaves need
3856 * not be the same type of device, so
3857 * MTUs may vary. For other modes,
3858 * slaves arguably should have the
3859 * same MTUs. To do this, we'd need to
3860 * take over the slave's change_mtu
3861 * function for the duration of their
3865 case NETDEV_CHANGENAME:
3866 /* we don't care if we don't have primary set */
3867 if (!bond_uses_primary(bond) ||
3868 !bond->params.primary[0])
3871 if (slave == primary) {
3872 /* slave's name changed - he's no longer primary */
3873 RCU_INIT_POINTER(bond->primary_slave, NULL);
3874 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3875 /* we have a new primary slave */
3876 rcu_assign_pointer(bond->primary_slave, slave);
3877 } else { /* we didn't change primary - exit */
3881 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3882 primary ? slave_dev->name : "none");
3885 bond_select_active_slave(bond);
3886 unblock_netpoll_tx();
3888 case NETDEV_FEAT_CHANGE:
3889 bond_compute_features(bond);
3891 case NETDEV_RESEND_IGMP:
3892 /* Propagate to master device */
3893 call_netdevice_notifiers(event, slave->bond->dev);
3902 /* bond_netdev_event: handle netdev notifier chain events.
3904 * This function receives events for the netdev chain. The caller (an
3905 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3906 * locks for us to safely manipulate the slave devices (RTNL lock,
3909 static int bond_netdev_event(struct notifier_block *this,
3910 unsigned long event, void *ptr)
3912 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3914 netdev_dbg(event_dev, "%s received %s\n",
3915 __func__, netdev_cmd_to_name(event));
3917 if (!(event_dev->priv_flags & IFF_BONDING))
3920 if (event_dev->flags & IFF_MASTER) {
3923 ret = bond_master_netdev_event(event, event_dev);
3924 if (ret != NOTIFY_DONE)
3928 if (event_dev->flags & IFF_SLAVE)
3929 return bond_slave_netdev_event(event, event_dev);
3934 static struct notifier_block bond_netdev_notifier = {
3935 .notifier_call = bond_netdev_event,
3938 /*---------------------------- Hashing Policies -----------------------------*/
3940 /* Helper to access data in a packet, with or without a backing skb.
3941 * If skb is given the data is linearized if necessary via pskb_may_pull.
3943 static inline const void *bond_pull_data(struct sk_buff *skb,
3944 const void *data, int hlen, int n)
3946 if (likely(n <= hlen))
3948 else if (skb && likely(pskb_may_pull(skb, n)))
3954 /* L2 hash helper */
3955 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3959 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3963 ep = (struct ethhdr *)(data + mhoff);
3964 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
3967 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
3968 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
3970 const struct ipv6hdr *iph6;
3971 const struct iphdr *iph;
3973 if (l2_proto == htons(ETH_P_IP)) {
3974 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
3978 iph = (const struct iphdr *)(data + *nhoff);
3979 iph_to_flow_copy_v4addrs(fk, iph);
3980 *nhoff += iph->ihl << 2;
3981 if (!ip_is_fragment(iph))
3982 *ip_proto = iph->protocol;
3983 } else if (l2_proto == htons(ETH_P_IPV6)) {
3984 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
3988 iph6 = (const struct ipv6hdr *)(data + *nhoff);
3989 iph_to_flow_copy_v6addrs(fk, iph6);
3990 *nhoff += sizeof(*iph6);
3991 *ip_proto = iph6->nexthdr;
3996 if (l34 && *ip_proto >= 0)
3997 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4002 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4004 u32 srcmac_vendor = 0, srcmac_dev = 0;
4005 struct ethhdr *mac_hdr;
4009 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4012 mac_hdr = (struct ethhdr *)(data + mhoff);
4014 for (i = 0; i < 3; i++)
4015 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4017 for (i = 3; i < ETH_ALEN; i++)
4018 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4020 if (skb && skb_vlan_tag_present(skb))
4021 vlan = skb_vlan_tag_get(skb);
4023 return vlan ^ srcmac_vendor ^ srcmac_dev;
4026 /* Extract the appropriate headers based on bond's xmit policy */
4027 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4028 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4030 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4033 switch (bond->params.xmit_policy) {
4034 case BOND_XMIT_POLICY_ENCAP23:
4035 case BOND_XMIT_POLICY_ENCAP34:
4036 memset(fk, 0, sizeof(*fk));
4037 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4038 fk, data, l2_proto, nhoff, hlen, 0);
4043 fk->ports.ports = 0;
4044 memset(&fk->icmp, 0, sizeof(fk->icmp));
4045 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4048 /* ICMP error packets contains at least 8 bytes of the header
4049 * of the packet which generated the error. Use this information
4050 * to correlate ICMP error packets within the same flow which
4051 * generated the error.
4053 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4054 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4055 if (ip_proto == IPPROTO_ICMP) {
4056 if (!icmp_is_err(fk->icmp.type))
4059 nhoff += sizeof(struct icmphdr);
4060 } else if (ip_proto == IPPROTO_ICMPV6) {
4061 if (!icmpv6_is_err(fk->icmp.type))
4064 nhoff += sizeof(struct icmp6hdr);
4066 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4072 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4074 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4075 (__force u32)flow_get_u32_src(flow);
4076 hash ^= (hash >> 16);
4077 hash ^= (hash >> 8);
4079 /* discard lowest hash bit to deal with the common even ports pattern */
4080 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4081 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4087 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4088 * the data as required, but this function can be used without it if the data is
4089 * known to be linear (e.g. with xdp_buff).
4091 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4092 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4094 struct flow_keys flow;
4097 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4098 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4100 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4101 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4102 return bond_eth_hash(skb, data, mhoff, hlen);
4104 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4105 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4106 hash = bond_eth_hash(skb, data, mhoff, hlen);
4109 memcpy(&hash, &flow.icmp, sizeof(hash));
4111 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4114 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4118 * bond_xmit_hash - generate a hash value based on the xmit policy
4119 * @bond: bonding device
4120 * @skb: buffer to use for headers
4122 * This function will extract the necessary headers from the skb buffer and use
4123 * them to generate a hash based on the xmit_policy set in the bonding device
4125 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4127 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4131 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4132 skb_mac_offset(skb), skb_network_offset(skb),
4137 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4138 * @bond: bonding device
4139 * @xdp: buffer to use for headers
4141 * The XDP variant of bond_xmit_hash.
4143 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4147 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4150 eth = (struct ethhdr *)xdp->data;
4152 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4153 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4156 /*-------------------------- Device entry points ----------------------------*/
4158 void bond_work_init_all(struct bonding *bond)
4160 INIT_DELAYED_WORK(&bond->mcast_work,
4161 bond_resend_igmp_join_requests_delayed);
4162 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4163 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4164 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4165 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4166 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4169 static void bond_work_cancel_all(struct bonding *bond)
4171 cancel_delayed_work_sync(&bond->mii_work);
4172 cancel_delayed_work_sync(&bond->arp_work);
4173 cancel_delayed_work_sync(&bond->alb_work);
4174 cancel_delayed_work_sync(&bond->ad_work);
4175 cancel_delayed_work_sync(&bond->mcast_work);
4176 cancel_delayed_work_sync(&bond->slave_arr_work);
4179 static int bond_open(struct net_device *bond_dev)
4181 struct bonding *bond = netdev_priv(bond_dev);
4182 struct list_head *iter;
4183 struct slave *slave;
4185 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4186 bond->rr_tx_counter = alloc_percpu(u32);
4187 if (!bond->rr_tx_counter)
4191 /* reset slave->backup and slave->inactive */
4192 if (bond_has_slaves(bond)) {
4193 bond_for_each_slave(bond, slave, iter) {
4194 if (bond_uses_primary(bond) &&
4195 slave != rcu_access_pointer(bond->curr_active_slave)) {
4196 bond_set_slave_inactive_flags(slave,
4197 BOND_SLAVE_NOTIFY_NOW);
4198 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4199 bond_set_slave_active_flags(slave,
4200 BOND_SLAVE_NOTIFY_NOW);
4205 if (bond_is_lb(bond)) {
4206 /* bond_alb_initialize must be called before the timer
4209 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4211 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4212 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4215 if (bond->params.miimon) /* link check interval, in milliseconds. */
4216 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4218 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4219 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4220 bond->recv_probe = bond_rcv_validate;
4223 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4224 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4225 /* register to receive LACPDUs */
4226 bond->recv_probe = bond_3ad_lacpdu_recv;
4227 bond_3ad_initiate_agg_selection(bond, 1);
4229 bond_for_each_slave(bond, slave, iter)
4230 dev_mc_add(slave->dev, lacpdu_mcast_addr);
4233 if (bond_mode_can_use_xmit_hash(bond))
4234 bond_update_slave_arr(bond, NULL);
4239 static int bond_close(struct net_device *bond_dev)
4241 struct bonding *bond = netdev_priv(bond_dev);
4242 struct slave *slave;
4244 bond_work_cancel_all(bond);
4245 bond->send_peer_notif = 0;
4246 if (bond_is_lb(bond))
4247 bond_alb_deinitialize(bond);
4248 bond->recv_probe = NULL;
4250 if (bond_uses_primary(bond)) {
4252 slave = rcu_dereference(bond->curr_active_slave);
4254 bond_hw_addr_flush(bond_dev, slave->dev);
4257 struct list_head *iter;
4259 bond_for_each_slave(bond, slave, iter)
4260 bond_hw_addr_flush(bond_dev, slave->dev);
4266 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4267 * that some drivers can provide 32bit values only.
4269 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4270 const struct rtnl_link_stats64 *_new,
4271 const struct rtnl_link_stats64 *_old)
4273 const u64 *new = (const u64 *)_new;
4274 const u64 *old = (const u64 *)_old;
4275 u64 *res = (u64 *)_res;
4278 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4281 s64 delta = nv - ov;
4283 /* detects if this particular field is 32bit only */
4284 if (((nv | ov) >> 32) == 0)
4285 delta = (s64)(s32)((u32)nv - (u32)ov);
4287 /* filter anomalies, some drivers reset their stats
4288 * at down/up events.
4295 #ifdef CONFIG_LOCKDEP
4296 static int bond_get_lowest_level_rcu(struct net_device *dev)
4298 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4299 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4300 int cur = 0, max = 0;
4303 iter = &dev->adj_list.lower;
4308 ldev = netdev_next_lower_dev_rcu(now, &iter);
4313 niter = &ldev->adj_list.lower;
4314 dev_stack[cur] = now;
4315 iter_stack[cur++] = iter;
4324 next = dev_stack[--cur];
4325 niter = iter_stack[cur];
4336 static void bond_get_stats(struct net_device *bond_dev,
4337 struct rtnl_link_stats64 *stats)
4339 struct bonding *bond = netdev_priv(bond_dev);
4340 struct rtnl_link_stats64 temp;
4341 struct list_head *iter;
4342 struct slave *slave;
4347 #ifdef CONFIG_LOCKDEP
4348 nest_level = bond_get_lowest_level_rcu(bond_dev);
4351 spin_lock_nested(&bond->stats_lock, nest_level);
4352 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4354 bond_for_each_slave_rcu(bond, slave, iter) {
4355 const struct rtnl_link_stats64 *new =
4356 dev_get_stats(slave->dev, &temp);
4358 bond_fold_stats(stats, new, &slave->slave_stats);
4360 /* save off the slave stats for the next run */
4361 memcpy(&slave->slave_stats, new, sizeof(*new));
4364 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4365 spin_unlock(&bond->stats_lock);
4369 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4371 struct bonding *bond = netdev_priv(bond_dev);
4372 struct mii_ioctl_data *mii = NULL;
4373 const struct net_device_ops *ops;
4374 struct net_device *real_dev;
4375 struct hwtstamp_config cfg;
4379 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4390 /* We do this again just in case we were called by SIOCGMIIREG
4391 * instead of SIOCGMIIPHY.
4397 if (mii->reg_num == 1) {
4399 if (netif_carrier_ok(bond->dev))
4400 mii->val_out = BMSR_LSTATUS;
4405 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4408 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4413 real_dev = bond_option_active_slave_get_rcu(bond);
4417 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4418 ifrr.ifr_ifru = ifr->ifr_ifru;
4420 ops = real_dev->netdev_ops;
4421 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4422 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4426 ifr->ifr_ifru = ifrr.ifr_ifru;
4427 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4430 /* Set the BOND_PHC_INDEX flag to notify user space */
4431 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4433 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4444 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4446 struct bonding *bond = netdev_priv(bond_dev);
4447 struct net_device *slave_dev = NULL;
4448 struct ifbond k_binfo;
4449 struct ifbond __user *u_binfo = NULL;
4450 struct ifslave k_sinfo;
4451 struct ifslave __user *u_sinfo = NULL;
4452 struct bond_opt_value newval;
4456 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4459 case SIOCBONDINFOQUERY:
4460 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4462 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4465 bond_info_query(bond_dev, &k_binfo);
4466 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4470 case SIOCBONDSLAVEINFOQUERY:
4471 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4473 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4476 res = bond_slave_info_query(bond_dev, &k_sinfo);
4478 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4486 net = dev_net(bond_dev);
4488 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4491 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4493 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4499 case SIOCBONDENSLAVE:
4500 res = bond_enslave(bond_dev, slave_dev, NULL);
4502 case SIOCBONDRELEASE:
4503 res = bond_release(bond_dev, slave_dev);
4505 case SIOCBONDSETHWADDR:
4506 res = bond_set_dev_addr(bond_dev, slave_dev);
4508 case SIOCBONDCHANGEACTIVE:
4509 bond_opt_initstr(&newval, slave_dev->name);
4510 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4520 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4521 void __user *data, int cmd)
4523 struct ifreq ifrdata = { .ifr_data = data };
4526 case BOND_INFO_QUERY_OLD:
4527 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4528 case BOND_SLAVE_INFO_QUERY_OLD:
4529 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4530 case BOND_ENSLAVE_OLD:
4531 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4532 case BOND_RELEASE_OLD:
4533 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4534 case BOND_SETHWADDR_OLD:
4535 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4536 case BOND_CHANGE_ACTIVE_OLD:
4537 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4543 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4545 struct bonding *bond = netdev_priv(bond_dev);
4547 if (change & IFF_PROMISC)
4548 bond_set_promiscuity(bond,
4549 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4551 if (change & IFF_ALLMULTI)
4552 bond_set_allmulti(bond,
4553 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4556 static void bond_set_rx_mode(struct net_device *bond_dev)
4558 struct bonding *bond = netdev_priv(bond_dev);
4559 struct list_head *iter;
4560 struct slave *slave;
4563 if (bond_uses_primary(bond)) {
4564 slave = rcu_dereference(bond->curr_active_slave);
4566 dev_uc_sync(slave->dev, bond_dev);
4567 dev_mc_sync(slave->dev, bond_dev);
4570 bond_for_each_slave_rcu(bond, slave, iter) {
4571 dev_uc_sync_multiple(slave->dev, bond_dev);
4572 dev_mc_sync_multiple(slave->dev, bond_dev);
4578 static int bond_neigh_init(struct neighbour *n)
4580 struct bonding *bond = netdev_priv(n->dev);
4581 const struct net_device_ops *slave_ops;
4582 struct neigh_parms parms;
4583 struct slave *slave;
4587 slave = bond_first_slave_rcu(bond);
4590 slave_ops = slave->dev->netdev_ops;
4591 if (!slave_ops->ndo_neigh_setup)
4594 /* TODO: find another way [1] to implement this.
4595 * Passing a zeroed structure is fragile,
4596 * but at least we do not pass garbage.
4598 * [1] One way would be that ndo_neigh_setup() never touch
4599 * struct neigh_parms, but propagate the new neigh_setup()
4600 * back to ___neigh_create() / neigh_parms_alloc()
4602 memset(&parms, 0, sizeof(parms));
4603 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4608 if (parms.neigh_setup)
4609 ret = parms.neigh_setup(n);
4615 /* The bonding ndo_neigh_setup is called at init time beofre any
4616 * slave exists. So we must declare proxy setup function which will
4617 * be used at run time to resolve the actual slave neigh param setup.
4619 * It's also called by master devices (such as vlans) to setup their
4620 * underlying devices. In that case - do nothing, we're already set up from
4623 static int bond_neigh_setup(struct net_device *dev,
4624 struct neigh_parms *parms)
4626 /* modify only our neigh_parms */
4627 if (parms->dev == dev)
4628 parms->neigh_setup = bond_neigh_init;
4633 /* Change the MTU of all of a master's slaves to match the master */
4634 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4636 struct bonding *bond = netdev_priv(bond_dev);
4637 struct slave *slave, *rollback_slave;
4638 struct list_head *iter;
4641 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4643 bond_for_each_slave(bond, slave, iter) {
4644 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4645 slave, slave->dev->netdev_ops->ndo_change_mtu);
4647 res = dev_set_mtu(slave->dev, new_mtu);
4650 /* If we failed to set the slave's mtu to the new value
4651 * we must abort the operation even in ACTIVE_BACKUP
4652 * mode, because if we allow the backup slaves to have
4653 * different mtu values than the active slave we'll
4654 * need to change their mtu when doing a failover. That
4655 * means changing their mtu from timer context, which
4656 * is probably not a good idea.
4658 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4664 bond_dev->mtu = new_mtu;
4669 /* unwind from head to the slave that failed */
4670 bond_for_each_slave(bond, rollback_slave, iter) {
4673 if (rollback_slave == slave)
4676 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4678 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4685 /* Change HW address
4687 * Note that many devices must be down to change the HW address, and
4688 * downing the master releases all slaves. We can make bonds full of
4689 * bonding devices to test this, however.
4691 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4693 struct bonding *bond = netdev_priv(bond_dev);
4694 struct slave *slave, *rollback_slave;
4695 struct sockaddr_storage *ss = addr, tmp_ss;
4696 struct list_head *iter;
4699 if (BOND_MODE(bond) == BOND_MODE_ALB)
4700 return bond_alb_set_mac_address(bond_dev, addr);
4703 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4705 /* If fail_over_mac is enabled, do nothing and return success.
4706 * Returning an error causes ifenslave to fail.
4708 if (bond->params.fail_over_mac &&
4709 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4712 if (!is_valid_ether_addr(ss->__data))
4713 return -EADDRNOTAVAIL;
4715 bond_for_each_slave(bond, slave, iter) {
4716 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4718 res = dev_set_mac_address(slave->dev, addr, NULL);
4720 /* TODO: consider downing the slave
4722 * User should expect communications
4723 * breakage anyway until ARP finish
4726 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4733 dev_addr_set(bond_dev, ss->__data);
4737 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4738 tmp_ss.ss_family = bond_dev->type;
4740 /* unwind from head to the slave that failed */
4741 bond_for_each_slave(bond, rollback_slave, iter) {
4744 if (rollback_slave == slave)
4747 tmp_res = dev_set_mac_address(rollback_slave->dev,
4748 (struct sockaddr *)&tmp_ss, NULL);
4750 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4759 * bond_get_slave_by_id - get xmit slave with slave_id
4760 * @bond: bonding device that is transmitting
4761 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4763 * This function tries to get slave with slave_id but in case
4764 * it fails, it tries to find the first available slave for transmission.
4766 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4769 struct list_head *iter;
4770 struct slave *slave;
4773 /* Here we start from the slave with slave_id */
4774 bond_for_each_slave_rcu(bond, slave, iter) {
4776 if (bond_slave_can_tx(slave))
4781 /* Here we start from the first slave up to slave_id */
4783 bond_for_each_slave_rcu(bond, slave, iter) {
4786 if (bond_slave_can_tx(slave))
4789 /* no slave that can tx has been found */
4794 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4795 * @bond: bonding device to use
4797 * Based on the value of the bonding device's packets_per_slave parameter
4798 * this function generates a slave id, which is usually used as the next
4799 * slave to transmit through.
4801 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4804 struct reciprocal_value reciprocal_packets_per_slave;
4805 int packets_per_slave = bond->params.packets_per_slave;
4807 switch (packets_per_slave) {
4809 slave_id = get_random_u32();
4812 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4815 reciprocal_packets_per_slave =
4816 bond->params.reciprocal_packets_per_slave;
4817 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4818 slave_id = reciprocal_divide(slave_id,
4819 reciprocal_packets_per_slave);
4826 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4827 struct sk_buff *skb)
4829 struct slave *slave;
4833 /* Start with the curr_active_slave that joined the bond as the
4834 * default for sending IGMP traffic. For failover purposes one
4835 * needs to maintain some consistency for the interface that will
4836 * send the join/membership reports. The curr_active_slave found
4837 * will send all of this type of traffic.
4839 if (skb->protocol == htons(ETH_P_IP)) {
4840 int noff = skb_network_offset(skb);
4843 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4847 if (iph->protocol == IPPROTO_IGMP) {
4848 slave = rcu_dereference(bond->curr_active_slave);
4851 return bond_get_slave_by_id(bond, 0);
4856 slave_cnt = READ_ONCE(bond->slave_cnt);
4857 if (likely(slave_cnt)) {
4858 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4859 return bond_get_slave_by_id(bond, slave_id);
4864 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4865 struct xdp_buff *xdp)
4867 struct slave *slave;
4870 const struct ethhdr *eth;
4871 void *data = xdp->data;
4873 if (data + sizeof(struct ethhdr) > xdp->data_end)
4876 eth = (struct ethhdr *)data;
4877 data += sizeof(struct ethhdr);
4879 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4880 if (eth->h_proto == htons(ETH_P_IP)) {
4881 const struct iphdr *iph;
4883 if (data + sizeof(struct iphdr) > xdp->data_end)
4886 iph = (struct iphdr *)data;
4888 if (iph->protocol == IPPROTO_IGMP) {
4889 slave = rcu_dereference(bond->curr_active_slave);
4892 return bond_get_slave_by_id(bond, 0);
4897 slave_cnt = READ_ONCE(bond->slave_cnt);
4898 if (likely(slave_cnt)) {
4899 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4900 return bond_get_slave_by_id(bond, slave_id);
4905 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4906 struct net_device *bond_dev)
4908 struct bonding *bond = netdev_priv(bond_dev);
4909 struct slave *slave;
4911 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4913 return bond_dev_queue_xmit(bond, skb, slave->dev);
4915 return bond_tx_drop(bond_dev, skb);
4918 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4920 return rcu_dereference(bond->curr_active_slave);
4923 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4924 * the bond has a usable interface.
4926 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4927 struct net_device *bond_dev)
4929 struct bonding *bond = netdev_priv(bond_dev);
4930 struct slave *slave;
4932 slave = bond_xmit_activebackup_slave_get(bond);
4934 return bond_dev_queue_xmit(bond, skb, slave->dev);
4936 return bond_tx_drop(bond_dev, skb);
4939 /* Use this to update slave_array when (a) it's not appropriate to update
4940 * slave_array right away (note that update_slave_array() may sleep)
4941 * and / or (b) RTNL is not held.
4943 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4945 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4948 /* Slave array work handler. Holds only RTNL */
4949 static void bond_slave_arr_handler(struct work_struct *work)
4951 struct bonding *bond = container_of(work, struct bonding,
4952 slave_arr_work.work);
4955 if (!rtnl_trylock())
4958 ret = bond_update_slave_arr(bond, NULL);
4961 pr_warn_ratelimited("Failed to update slave array from WT\n");
4967 bond_slave_arr_work_rearm(bond, 1);
4970 static void bond_skip_slave(struct bond_up_slave *slaves,
4971 struct slave *skipslave)
4975 /* Rare situation where caller has asked to skip a specific
4976 * slave but allocation failed (most likely!). BTW this is
4977 * only possible when the call is initiated from
4978 * __bond_release_one(). In this situation; overwrite the
4979 * skipslave entry in the array with the last entry from the
4980 * array to avoid a situation where the xmit path may choose
4981 * this to-be-skipped slave to send a packet out.
4983 for (idx = 0; slaves && idx < slaves->count; idx++) {
4984 if (skipslave == slaves->arr[idx]) {
4986 slaves->arr[slaves->count - 1];
4993 static void bond_set_slave_arr(struct bonding *bond,
4994 struct bond_up_slave *usable_slaves,
4995 struct bond_up_slave *all_slaves)
4997 struct bond_up_slave *usable, *all;
4999 usable = rtnl_dereference(bond->usable_slaves);
5000 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5001 kfree_rcu(usable, rcu);
5003 all = rtnl_dereference(bond->all_slaves);
5004 rcu_assign_pointer(bond->all_slaves, all_slaves);
5005 kfree_rcu(all, rcu);
5008 static void bond_reset_slave_arr(struct bonding *bond)
5010 struct bond_up_slave *usable, *all;
5012 usable = rtnl_dereference(bond->usable_slaves);
5014 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5015 kfree_rcu(usable, rcu);
5018 all = rtnl_dereference(bond->all_slaves);
5020 RCU_INIT_POINTER(bond->all_slaves, NULL);
5021 kfree_rcu(all, rcu);
5025 /* Build the usable slaves array in control path for modes that use xmit-hash
5026 * to determine the slave interface -
5027 * (a) BOND_MODE_8023AD
5029 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5031 * The caller is expected to hold RTNL only and NO other lock!
5033 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5035 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5036 struct slave *slave;
5037 struct list_head *iter;
5043 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5044 bond->slave_cnt), GFP_KERNEL);
5045 all_slaves = kzalloc(struct_size(all_slaves, arr,
5046 bond->slave_cnt), GFP_KERNEL);
5047 if (!usable_slaves || !all_slaves) {
5051 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5052 struct ad_info ad_info;
5054 spin_lock_bh(&bond->mode_lock);
5055 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5056 spin_unlock_bh(&bond->mode_lock);
5057 pr_debug("bond_3ad_get_active_agg_info failed\n");
5058 /* No active aggragator means it's not safe to use
5059 * the previous array.
5061 bond_reset_slave_arr(bond);
5064 spin_unlock_bh(&bond->mode_lock);
5065 agg_id = ad_info.aggregator_id;
5067 bond_for_each_slave(bond, slave, iter) {
5068 if (skipslave == slave)
5071 all_slaves->arr[all_slaves->count++] = slave;
5072 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5073 struct aggregator *agg;
5075 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5076 if (!agg || agg->aggregator_identifier != agg_id)
5079 if (!bond_slave_can_tx(slave))
5082 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5083 usable_slaves->count);
5085 usable_slaves->arr[usable_slaves->count++] = slave;
5088 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5091 if (ret != 0 && skipslave) {
5092 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5094 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5097 kfree_rcu(all_slaves, rcu);
5098 kfree_rcu(usable_slaves, rcu);
5103 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5104 struct sk_buff *skb,
5105 struct bond_up_slave *slaves)
5107 struct slave *slave;
5111 hash = bond_xmit_hash(bond, skb);
5112 count = slaves ? READ_ONCE(slaves->count) : 0;
5113 if (unlikely(!count))
5116 slave = slaves->arr[hash % count];
5120 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5121 struct xdp_buff *xdp)
5123 struct bond_up_slave *slaves;
5127 hash = bond_xmit_hash_xdp(bond, xdp);
5128 slaves = rcu_dereference(bond->usable_slaves);
5129 count = slaves ? READ_ONCE(slaves->count) : 0;
5130 if (unlikely(!count))
5133 return slaves->arr[hash % count];
5136 /* Use this Xmit function for 3AD as well as XOR modes. The current
5137 * usable slave array is formed in the control path. The xmit function
5138 * just calculates hash and sends the packet out.
5140 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5141 struct net_device *dev)
5143 struct bonding *bond = netdev_priv(dev);
5144 struct bond_up_slave *slaves;
5145 struct slave *slave;
5147 slaves = rcu_dereference(bond->usable_slaves);
5148 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5150 return bond_dev_queue_xmit(bond, skb, slave->dev);
5152 return bond_tx_drop(dev, skb);
5155 /* in broadcast mode, we send everything to all usable interfaces. */
5156 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5157 struct net_device *bond_dev)
5159 struct bonding *bond = netdev_priv(bond_dev);
5160 struct slave *slave = NULL;
5161 struct list_head *iter;
5162 bool xmit_suc = false;
5163 bool skb_used = false;
5165 bond_for_each_slave_rcu(bond, slave, iter) {
5166 struct sk_buff *skb2;
5168 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5171 if (bond_is_last_slave(bond, slave)) {
5175 skb2 = skb_clone(skb, GFP_ATOMIC);
5177 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5178 bond_dev->name, __func__);
5183 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5188 dev_kfree_skb_any(skb);
5191 return NETDEV_TX_OK;
5193 dev_core_stats_tx_dropped_inc(bond_dev);
5194 return NET_XMIT_DROP;
5197 /*------------------------- Device initialization ---------------------------*/
5199 /* Lookup the slave that corresponds to a qid */
5200 static inline int bond_slave_override(struct bonding *bond,
5201 struct sk_buff *skb)
5203 struct slave *slave = NULL;
5204 struct list_head *iter;
5206 if (!skb_rx_queue_recorded(skb))
5209 /* Find out if any slaves have the same mapping as this skb. */
5210 bond_for_each_slave_rcu(bond, slave, iter) {
5211 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5212 if (bond_slave_is_up(slave) &&
5213 slave->link == BOND_LINK_UP) {
5214 bond_dev_queue_xmit(bond, skb, slave->dev);
5217 /* If the slave isn't UP, use default transmit policy. */
5226 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5227 struct net_device *sb_dev)
5229 /* This helper function exists to help dev_pick_tx get the correct
5230 * destination queue. Using a helper function skips a call to
5231 * skb_tx_hash and will put the skbs in the queue we expect on their
5232 * way down to the bonding driver.
5234 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5236 /* Save the original txq to restore before passing to the driver */
5237 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5239 if (unlikely(txq >= dev->real_num_tx_queues)) {
5241 txq -= dev->real_num_tx_queues;
5242 } while (txq >= dev->real_num_tx_queues);
5247 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5248 struct sk_buff *skb,
5251 struct bonding *bond = netdev_priv(master_dev);
5252 struct bond_up_slave *slaves;
5253 struct slave *slave = NULL;
5255 switch (BOND_MODE(bond)) {
5256 case BOND_MODE_ROUNDROBIN:
5257 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5259 case BOND_MODE_ACTIVEBACKUP:
5260 slave = bond_xmit_activebackup_slave_get(bond);
5262 case BOND_MODE_8023AD:
5265 slaves = rcu_dereference(bond->all_slaves);
5267 slaves = rcu_dereference(bond->usable_slaves);
5268 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5270 case BOND_MODE_BROADCAST:
5273 slave = bond_xmit_alb_slave_get(bond, skb);
5276 slave = bond_xmit_tlb_slave_get(bond, skb);
5279 /* Should never happen, mode already checked */
5280 WARN_ONCE(true, "Unknown bonding mode");
5289 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5291 switch (sk->sk_family) {
5292 #if IS_ENABLED(CONFIG_IPV6)
5294 if (ipv6_only_sock(sk) ||
5295 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5296 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5297 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5298 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5303 default: /* AF_INET */
5304 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5305 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5306 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5310 flow->ports.src = inet_sk(sk)->inet_sport;
5311 flow->ports.dst = inet_sk(sk)->inet_dport;
5315 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5316 * @sk: socket to use for headers
5318 * This function will extract the necessary field from the socket and use
5319 * them to generate a hash based on the LAYER34 xmit_policy.
5320 * Assumes that sk is a TCP or UDP socket.
5322 static u32 bond_sk_hash_l34(struct sock *sk)
5324 struct flow_keys flow;
5327 bond_sk_to_flow(sk, &flow);
5330 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5332 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5335 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5338 struct bond_up_slave *slaves;
5339 struct slave *slave;
5343 slaves = rcu_dereference(bond->usable_slaves);
5344 count = slaves ? READ_ONCE(slaves->count) : 0;
5345 if (unlikely(!count))
5348 hash = bond_sk_hash_l34(sk);
5349 slave = slaves->arr[hash % count];
5354 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5357 struct bonding *bond = netdev_priv(dev);
5358 struct net_device *lower = NULL;
5361 if (bond_sk_check(bond))
5362 lower = __bond_sk_get_lower_dev(bond, sk);
5368 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5369 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5370 struct net_device *dev)
5372 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5374 /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded
5375 * was true, if tls_device_down is running in parallel, but it's OK,
5376 * because bond_get_slave_by_dev has a NULL check.
5378 if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5379 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5380 return bond_tx_drop(dev, skb);
5384 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5386 struct bonding *bond = netdev_priv(dev);
5388 if (bond_should_override_tx_queue(bond) &&
5389 !bond_slave_override(bond, skb))
5390 return NETDEV_TX_OK;
5392 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5393 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5394 return bond_tls_device_xmit(bond, skb, dev);
5397 switch (BOND_MODE(bond)) {
5398 case BOND_MODE_ROUNDROBIN:
5399 return bond_xmit_roundrobin(skb, dev);
5400 case BOND_MODE_ACTIVEBACKUP:
5401 return bond_xmit_activebackup(skb, dev);
5402 case BOND_MODE_8023AD:
5404 return bond_3ad_xor_xmit(skb, dev);
5405 case BOND_MODE_BROADCAST:
5406 return bond_xmit_broadcast(skb, dev);
5408 return bond_alb_xmit(skb, dev);
5410 return bond_tlb_xmit(skb, dev);
5412 /* Should never happen, mode already checked */
5413 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5415 return bond_tx_drop(dev, skb);
5419 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5421 struct bonding *bond = netdev_priv(dev);
5422 netdev_tx_t ret = NETDEV_TX_OK;
5424 /* If we risk deadlock from transmitting this in the
5425 * netpoll path, tell netpoll to queue the frame for later tx
5427 if (unlikely(is_netpoll_tx_blocked(dev)))
5428 return NETDEV_TX_BUSY;
5431 if (bond_has_slaves(bond))
5432 ret = __bond_start_xmit(skb, dev);
5434 ret = bond_tx_drop(dev, skb);
5440 static struct net_device *
5441 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5443 struct bonding *bond = netdev_priv(bond_dev);
5444 struct slave *slave;
5446 /* Caller needs to hold rcu_read_lock() */
5448 switch (BOND_MODE(bond)) {
5449 case BOND_MODE_ROUNDROBIN:
5450 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5453 case BOND_MODE_ACTIVEBACKUP:
5454 slave = bond_xmit_activebackup_slave_get(bond);
5457 case BOND_MODE_8023AD:
5459 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5463 /* Should never happen. Mode guarded by bond_xdp_check() */
5464 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5475 static int bond_xdp_xmit(struct net_device *bond_dev,
5476 int n, struct xdp_frame **frames, u32 flags)
5478 int nxmit, err = -ENXIO;
5482 for (nxmit = 0; nxmit < n; nxmit++) {
5483 struct xdp_frame *frame = frames[nxmit];
5484 struct xdp_frame *frames1[] = {frame};
5485 struct net_device *slave_dev;
5486 struct xdp_buff xdp;
5488 xdp_convert_frame_to_buff(frame, &xdp);
5490 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5496 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5503 /* If error happened on the first frame then we can pass the error up, otherwise
5504 * report the number of frames that were xmitted.
5507 return (nxmit == 0 ? err : nxmit);
5512 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5513 struct netlink_ext_ack *extack)
5515 struct bonding *bond = netdev_priv(dev);
5516 struct list_head *iter;
5517 struct slave *slave, *rollback_slave;
5518 struct bpf_prog *old_prog;
5519 struct netdev_bpf xdp = {
5520 .command = XDP_SETUP_PROG,
5529 if (!bond_xdp_check(bond))
5532 old_prog = bond->xdp_prog;
5533 bond->xdp_prog = prog;
5535 bond_for_each_slave(bond, slave, iter) {
5536 struct net_device *slave_dev = slave->dev;
5538 if (!slave_dev->netdev_ops->ndo_bpf ||
5539 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5540 SLAVE_NL_ERR(dev, slave_dev, extack,
5541 "Slave device does not support XDP");
5546 if (dev_xdp_prog_count(slave_dev) > 0) {
5547 SLAVE_NL_ERR(dev, slave_dev, extack,
5548 "Slave has XDP program loaded, please unload before enslaving");
5553 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5555 /* ndo_bpf() sets extack error message */
5556 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5564 static_branch_inc(&bpf_master_redirect_enabled_key);
5565 } else if (old_prog) {
5566 bpf_prog_put(old_prog);
5567 static_branch_dec(&bpf_master_redirect_enabled_key);
5573 /* unwind the program changes */
5574 bond->xdp_prog = old_prog;
5575 xdp.prog = old_prog;
5576 xdp.extack = NULL; /* do not overwrite original error */
5578 bond_for_each_slave(bond, rollback_slave, iter) {
5579 struct net_device *slave_dev = rollback_slave->dev;
5582 if (slave == rollback_slave)
5585 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5587 slave_err(dev, slave_dev,
5588 "Error %d when unwinding XDP program change\n", err_unwind);
5590 bpf_prog_inc(xdp.prog);
5595 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5597 switch (xdp->command) {
5598 case XDP_SETUP_PROG:
5599 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5605 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5607 if (speed == 0 || speed == SPEED_UNKNOWN)
5608 speed = slave->speed;
5610 speed = min(speed, slave->speed);
5615 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5616 struct ethtool_link_ksettings *cmd)
5618 struct bonding *bond = netdev_priv(bond_dev);
5619 struct list_head *iter;
5620 struct slave *slave;
5623 cmd->base.duplex = DUPLEX_UNKNOWN;
5624 cmd->base.port = PORT_OTHER;
5626 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5627 * do not need to check mode. Though link speed might not represent
5628 * the true receive or transmit bandwidth (not all modes are symmetric)
5629 * this is an accurate maximum.
5631 bond_for_each_slave(bond, slave, iter) {
5632 if (bond_slave_can_tx(slave)) {
5633 if (slave->speed != SPEED_UNKNOWN) {
5634 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5635 speed = bond_mode_bcast_speed(slave,
5638 speed += slave->speed;
5640 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5641 slave->duplex != DUPLEX_UNKNOWN)
5642 cmd->base.duplex = slave->duplex;
5645 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5650 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5651 struct ethtool_drvinfo *drvinfo)
5653 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5654 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5658 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5659 struct ethtool_ts_info *info)
5661 struct bonding *bond = netdev_priv(bond_dev);
5662 const struct ethtool_ops *ops;
5663 struct net_device *real_dev;
5664 struct phy_device *phydev;
5668 real_dev = bond_option_active_slave_get_rcu(bond);
5673 ops = real_dev->ethtool_ops;
5674 phydev = real_dev->phydev;
5676 if (phy_has_tsinfo(phydev)) {
5677 ret = phy_ts_info(phydev, info);
5679 } else if (ops->get_ts_info) {
5680 ret = ops->get_ts_info(real_dev, info);
5685 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5686 SOF_TIMESTAMPING_SOFTWARE;
5687 info->phc_index = -1;
5694 static const struct ethtool_ops bond_ethtool_ops = {
5695 .get_drvinfo = bond_ethtool_get_drvinfo,
5696 .get_link = ethtool_op_get_link,
5697 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5698 .get_ts_info = bond_ethtool_get_ts_info,
5701 static const struct net_device_ops bond_netdev_ops = {
5702 .ndo_init = bond_init,
5703 .ndo_uninit = bond_uninit,
5704 .ndo_open = bond_open,
5705 .ndo_stop = bond_close,
5706 .ndo_start_xmit = bond_start_xmit,
5707 .ndo_select_queue = bond_select_queue,
5708 .ndo_get_stats64 = bond_get_stats,
5709 .ndo_eth_ioctl = bond_eth_ioctl,
5710 .ndo_siocbond = bond_do_ioctl,
5711 .ndo_siocdevprivate = bond_siocdevprivate,
5712 .ndo_change_rx_flags = bond_change_rx_flags,
5713 .ndo_set_rx_mode = bond_set_rx_mode,
5714 .ndo_change_mtu = bond_change_mtu,
5715 .ndo_set_mac_address = bond_set_mac_address,
5716 .ndo_neigh_setup = bond_neigh_setup,
5717 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5718 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5719 #ifdef CONFIG_NET_POLL_CONTROLLER
5720 .ndo_netpoll_setup = bond_netpoll_setup,
5721 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5722 .ndo_poll_controller = bond_poll_controller,
5724 .ndo_add_slave = bond_enslave,
5725 .ndo_del_slave = bond_release,
5726 .ndo_fix_features = bond_fix_features,
5727 .ndo_features_check = passthru_features_check,
5728 .ndo_get_xmit_slave = bond_xmit_get_slave,
5729 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5730 .ndo_bpf = bond_xdp,
5731 .ndo_xdp_xmit = bond_xdp_xmit,
5732 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5735 static const struct device_type bond_type = {
5739 static void bond_destructor(struct net_device *bond_dev)
5741 struct bonding *bond = netdev_priv(bond_dev);
5744 destroy_workqueue(bond->wq);
5746 if (bond->rr_tx_counter)
5747 free_percpu(bond->rr_tx_counter);
5750 void bond_setup(struct net_device *bond_dev)
5752 struct bonding *bond = netdev_priv(bond_dev);
5754 spin_lock_init(&bond->mode_lock);
5755 bond->params = bonding_defaults;
5757 /* Initialize pointers */
5758 bond->dev = bond_dev;
5760 /* Initialize the device entry points */
5761 ether_setup(bond_dev);
5762 bond_dev->max_mtu = ETH_MAX_MTU;
5763 bond_dev->netdev_ops = &bond_netdev_ops;
5764 bond_dev->ethtool_ops = &bond_ethtool_ops;
5766 bond_dev->needs_free_netdev = true;
5767 bond_dev->priv_destructor = bond_destructor;
5769 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5771 /* Initialize the device options */
5772 bond_dev->flags |= IFF_MASTER;
5773 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5774 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5776 #ifdef CONFIG_XFRM_OFFLOAD
5777 /* set up xfrm device ops (only supported in active-backup right now) */
5778 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5779 INIT_LIST_HEAD(&bond->ipsec_list);
5780 spin_lock_init(&bond->ipsec_lock);
5781 #endif /* CONFIG_XFRM_OFFLOAD */
5783 /* don't acquire bond device's netif_tx_lock when transmitting */
5784 bond_dev->features |= NETIF_F_LLTX;
5786 /* By default, we declare the bond to be fully
5787 * VLAN hardware accelerated capable. Special
5788 * care is taken in the various xmit functions
5789 * when there are slaves that are not hw accel
5793 /* Don't allow bond devices to change network namespaces. */
5794 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5796 bond_dev->hw_features = BOND_VLAN_FEATURES |
5797 NETIF_F_HW_VLAN_CTAG_RX |
5798 NETIF_F_HW_VLAN_CTAG_FILTER;
5800 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5801 bond_dev->features |= bond_dev->hw_features;
5802 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5803 #ifdef CONFIG_XFRM_OFFLOAD
5804 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5805 /* Only enable XFRM features if this is an active-backup config */
5806 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5807 bond_dev->features |= BOND_XFRM_FEATURES;
5808 #endif /* CONFIG_XFRM_OFFLOAD */
5809 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5810 if (bond_sk_check(bond))
5811 bond_dev->features |= BOND_TLS_FEATURES;
5815 /* Destroy a bonding device.
5816 * Must be under rtnl_lock when this function is called.
5818 static void bond_uninit(struct net_device *bond_dev)
5820 struct bonding *bond = netdev_priv(bond_dev);
5821 struct bond_up_slave *usable, *all;
5822 struct list_head *iter;
5823 struct slave *slave;
5825 bond_netpoll_cleanup(bond_dev);
5827 /* Release the bonded slaves */
5828 bond_for_each_slave(bond, slave, iter)
5829 __bond_release_one(bond_dev, slave->dev, true, true);
5830 netdev_info(bond_dev, "Released all slaves\n");
5832 usable = rtnl_dereference(bond->usable_slaves);
5834 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5835 kfree_rcu(usable, rcu);
5838 all = rtnl_dereference(bond->all_slaves);
5840 RCU_INIT_POINTER(bond->all_slaves, NULL);
5841 kfree_rcu(all, rcu);
5844 list_del(&bond->bond_list);
5846 bond_debug_unregister(bond);
5849 /*------------------------- Module initialization ---------------------------*/
5851 static int bond_check_params(struct bond_params *params)
5853 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5854 struct bond_opt_value newval;
5855 const struct bond_opt_value *valptr;
5856 int arp_all_targets_value = 0;
5857 u16 ad_actor_sys_prio = 0;
5858 u16 ad_user_port_key = 0;
5859 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5861 int bond_mode = BOND_MODE_ROUNDROBIN;
5862 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5866 /* Convert string parameters. */
5868 bond_opt_initstr(&newval, mode);
5869 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5871 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5874 bond_mode = valptr->value;
5877 if (xmit_hash_policy) {
5878 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5879 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5880 bond_mode == BOND_MODE_BROADCAST) {
5881 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5882 bond_mode_name(bond_mode));
5884 bond_opt_initstr(&newval, xmit_hash_policy);
5885 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5888 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5892 xmit_hashtype = valptr->value;
5897 if (bond_mode != BOND_MODE_8023AD) {
5898 pr_info("lacp_rate param is irrelevant in mode %s\n",
5899 bond_mode_name(bond_mode));
5901 bond_opt_initstr(&newval, lacp_rate);
5902 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5905 pr_err("Error: Invalid lacp rate \"%s\"\n",
5909 lacp_fast = valptr->value;
5914 bond_opt_initstr(&newval, ad_select);
5915 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5918 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5921 params->ad_select = valptr->value;
5922 if (bond_mode != BOND_MODE_8023AD)
5923 pr_warn("ad_select param only affects 802.3ad mode\n");
5925 params->ad_select = BOND_AD_STABLE;
5928 if (max_bonds < 0) {
5929 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5930 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5931 max_bonds = BOND_DEFAULT_MAX_BONDS;
5935 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5941 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5946 if (downdelay < 0) {
5947 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5948 downdelay, INT_MAX);
5952 if ((use_carrier != 0) && (use_carrier != 1)) {
5953 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5958 if (num_peer_notif < 0 || num_peer_notif > 255) {
5959 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5964 /* reset values for 802.3ad/TLB/ALB */
5965 if (!bond_mode_uses_arp(bond_mode)) {
5967 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");
5968 pr_warn("Forcing miimon to 100msec\n");
5969 miimon = BOND_DEFAULT_MIIMON;
5973 if (tx_queues < 1 || tx_queues > 255) {
5974 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5975 tx_queues, BOND_DEFAULT_TX_QUEUES);
5976 tx_queues = BOND_DEFAULT_TX_QUEUES;
5979 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5980 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5982 all_slaves_active = 0;
5985 if (resend_igmp < 0 || resend_igmp > 255) {
5986 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5987 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5988 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5991 bond_opt_initval(&newval, packets_per_slave);
5992 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5993 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5994 packets_per_slave, USHRT_MAX);
5995 packets_per_slave = 1;
5998 if (bond_mode == BOND_MODE_ALB) {
5999 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",
6004 if (updelay || downdelay) {
6005 /* just warn the user the up/down delay will have
6006 * no effect since miimon is zero...
6008 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",
6009 updelay, downdelay);
6012 /* don't allow arp monitoring */
6014 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6015 miimon, arp_interval);
6019 if ((updelay % miimon) != 0) {
6020 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6021 updelay, miimon, (updelay / miimon) * miimon);
6026 if ((downdelay % miimon) != 0) {
6027 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6029 (downdelay / miimon) * miimon);
6032 downdelay /= miimon;
6035 if (arp_interval < 0) {
6036 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6037 arp_interval, INT_MAX);
6041 for (arp_ip_count = 0, i = 0;
6042 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6045 /* not a complete check, but good enough to catch mistakes */
6046 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6047 !bond_is_ip_target_ok(ip)) {
6048 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6052 if (bond_get_targets_ip(arp_target, ip) == -1)
6053 arp_target[arp_ip_count++] = ip;
6055 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6060 if (arp_interval && !arp_ip_count) {
6061 /* don't allow arping if no arp_ip_target given... */
6062 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6068 if (!arp_interval) {
6069 pr_err("arp_validate requires arp_interval\n");
6073 bond_opt_initstr(&newval, arp_validate);
6074 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6077 pr_err("Error: invalid arp_validate \"%s\"\n",
6081 arp_validate_value = valptr->value;
6083 arp_validate_value = 0;
6086 if (arp_all_targets) {
6087 bond_opt_initstr(&newval, arp_all_targets);
6088 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6091 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6093 arp_all_targets_value = 0;
6095 arp_all_targets_value = valptr->value;
6100 pr_info("MII link monitoring set to %d ms\n", miimon);
6101 } else if (arp_interval) {
6102 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6103 arp_validate_value);
6104 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6105 arp_interval, valptr->string, arp_ip_count);
6107 for (i = 0; i < arp_ip_count; i++)
6108 pr_cont(" %s", arp_ip_target[i]);
6112 } else if (max_bonds) {
6113 /* miimon and arp_interval not set, we need one so things
6114 * work as expected, see bonding.txt for details
6116 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");
6119 if (primary && !bond_mode_uses_primary(bond_mode)) {
6120 /* currently, using a primary only makes sense
6121 * in active backup, TLB or ALB modes
6123 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6124 primary, bond_mode_name(bond_mode));
6128 if (primary && primary_reselect) {
6129 bond_opt_initstr(&newval, primary_reselect);
6130 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6133 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6137 primary_reselect_value = valptr->value;
6139 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6142 if (fail_over_mac) {
6143 bond_opt_initstr(&newval, fail_over_mac);
6144 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6147 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6151 fail_over_mac_value = valptr->value;
6152 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6153 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6155 fail_over_mac_value = BOND_FOM_NONE;
6158 bond_opt_initstr(&newval, "default");
6159 valptr = bond_opt_parse(
6160 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6163 pr_err("Error: No ad_actor_sys_prio default value");
6166 ad_actor_sys_prio = valptr->value;
6168 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6171 pr_err("Error: No ad_user_port_key default value");
6174 ad_user_port_key = valptr->value;
6176 bond_opt_initstr(&newval, "default");
6177 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6179 pr_err("Error: No tlb_dynamic_lb default value");
6182 tlb_dynamic_lb = valptr->value;
6184 if (lp_interval == 0) {
6185 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6186 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6187 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6190 /* fill params struct with the proper values */
6191 params->mode = bond_mode;
6192 params->xmit_policy = xmit_hashtype;
6193 params->miimon = miimon;
6194 params->num_peer_notif = num_peer_notif;
6195 params->arp_interval = arp_interval;
6196 params->arp_validate = arp_validate_value;
6197 params->arp_all_targets = arp_all_targets_value;
6198 params->missed_max = 2;
6199 params->updelay = updelay;
6200 params->downdelay = downdelay;
6201 params->peer_notif_delay = 0;
6202 params->use_carrier = use_carrier;
6203 params->lacp_active = 1;
6204 params->lacp_fast = lacp_fast;
6205 params->primary[0] = 0;
6206 params->primary_reselect = primary_reselect_value;
6207 params->fail_over_mac = fail_over_mac_value;
6208 params->tx_queues = tx_queues;
6209 params->all_slaves_active = all_slaves_active;
6210 params->resend_igmp = resend_igmp;
6211 params->min_links = min_links;
6212 params->lp_interval = lp_interval;
6213 params->packets_per_slave = packets_per_slave;
6214 params->tlb_dynamic_lb = tlb_dynamic_lb;
6215 params->ad_actor_sys_prio = ad_actor_sys_prio;
6216 eth_zero_addr(params->ad_actor_system);
6217 params->ad_user_port_key = ad_user_port_key;
6218 if (packets_per_slave > 0) {
6219 params->reciprocal_packets_per_slave =
6220 reciprocal_value(packets_per_slave);
6222 /* reciprocal_packets_per_slave is unused if
6223 * packets_per_slave is 0 or 1, just initialize it
6225 params->reciprocal_packets_per_slave =
6226 (struct reciprocal_value) { 0 };
6230 strscpy_pad(params->primary, primary, sizeof(params->primary));
6232 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6233 #if IS_ENABLED(CONFIG_IPV6)
6234 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6240 /* Called from registration process */
6241 static int bond_init(struct net_device *bond_dev)
6243 struct bonding *bond = netdev_priv(bond_dev);
6244 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6246 netdev_dbg(bond_dev, "Begin bond_init\n");
6248 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6252 spin_lock_init(&bond->stats_lock);
6253 netdev_lockdep_set_classes(bond_dev);
6255 list_add_tail(&bond->bond_list, &bn->dev_list);
6257 bond_prepare_sysfs_group(bond);
6259 bond_debug_register(bond);
6261 /* Ensure valid dev_addr */
6262 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6263 bond_dev->addr_assign_type == NET_ADDR_PERM)
6264 eth_hw_addr_random(bond_dev);
6269 unsigned int bond_get_num_tx_queues(void)
6274 /* Create a new bond based on the specified name and bonding parameters.
6275 * If name is NULL, obtain a suitable "bond%d" name for us.
6276 * Caller must NOT hold rtnl_lock; we need to release it here before we
6277 * set up our sysfs entries.
6279 int bond_create(struct net *net, const char *name)
6281 struct net_device *bond_dev;
6282 struct bonding *bond;
6287 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6288 name ? name : "bond%d", NET_NAME_UNKNOWN,
6289 bond_setup, tx_queues);
6293 bond = netdev_priv(bond_dev);
6294 dev_net_set(bond_dev, net);
6295 bond_dev->rtnl_link_ops = &bond_link_ops;
6297 res = register_netdevice(bond_dev);
6299 free_netdev(bond_dev);
6303 netif_carrier_off(bond_dev);
6305 bond_work_init_all(bond);
6312 static int __net_init bond_net_init(struct net *net)
6314 struct bond_net *bn = net_generic(net, bond_net_id);
6317 INIT_LIST_HEAD(&bn->dev_list);
6319 bond_create_proc_dir(bn);
6320 bond_create_sysfs(bn);
6325 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6327 struct bond_net *bn;
6331 list_for_each_entry(net, net_list, exit_list) {
6332 bn = net_generic(net, bond_net_id);
6333 bond_destroy_sysfs(bn);
6336 /* Kill off any bonds created after unregistering bond rtnl ops */
6338 list_for_each_entry(net, net_list, exit_list) {
6339 struct bonding *bond, *tmp_bond;
6341 bn = net_generic(net, bond_net_id);
6342 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6343 unregister_netdevice_queue(bond->dev, &list);
6345 unregister_netdevice_many(&list);
6348 list_for_each_entry(net, net_list, exit_list) {
6349 bn = net_generic(net, bond_net_id);
6350 bond_destroy_proc_dir(bn);
6354 static struct pernet_operations bond_net_ops = {
6355 .init = bond_net_init,
6356 .exit_batch = bond_net_exit_batch,
6358 .size = sizeof(struct bond_net),
6361 static int __init bonding_init(void)
6366 res = bond_check_params(&bonding_defaults);
6370 res = register_pernet_subsys(&bond_net_ops);
6374 res = bond_netlink_init();
6378 bond_create_debugfs();
6380 for (i = 0; i < max_bonds; i++) {
6381 res = bond_create(&init_net, NULL);
6386 skb_flow_dissector_init(&flow_keys_bonding,
6387 flow_keys_bonding_keys,
6388 ARRAY_SIZE(flow_keys_bonding_keys));
6390 register_netdevice_notifier(&bond_netdev_notifier);
6394 bond_destroy_debugfs();
6395 bond_netlink_fini();
6397 unregister_pernet_subsys(&bond_net_ops);
6402 static void __exit bonding_exit(void)
6404 unregister_netdevice_notifier(&bond_netdev_notifier);
6406 bond_destroy_debugfs();
6408 bond_netlink_fini();
6409 unregister_pernet_subsys(&bond_net_ops);
6411 #ifdef CONFIG_NET_POLL_CONTROLLER
6412 /* Make sure we don't have an imbalance on our netpoll blocking */
6413 WARN_ON(atomic_read(&netpoll_block_tx));
6417 module_init(bonding_init);
6418 module_exit(bonding_exit);
6419 MODULE_LICENSE("GPL");
6420 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6421 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");