1 // SPDX-License-Identifier: GPL-1.0+
3 * originally based on the dummy device.
5 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
6 * Based on dummy.c, and eql.c devices.
8 * bonding.c: an Ethernet Bonding driver
10 * This is useful to talk to a Cisco EtherChannel compatible equipment:
12 * Sun Trunking (Solaris)
13 * Alteon AceDirector Trunks
15 * and probably many L2 switches ...
18 * ifconfig bond0 ipaddress netmask up
19 * will setup a network device, with an ip address. No mac address
20 * will be assigned at this time. The hw mac address will come from
21 * the first slave bonded to the channel. All slaves will then use
22 * this hw mac address.
25 * will release all slaves, marking them as down.
27 * ifenslave bond0 eth0
28 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
29 * a: be used as initial mac address
30 * b: if a hw mac address already is there, eth0's hw mac address
31 * will then be set from bond0.
35 #include <linux/kernel.h>
36 #include <linux/module.h>
37 #include <linux/types.h>
38 #include <linux/fcntl.h>
39 #include <linux/filter.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/icmp.h>
47 #include <linux/icmpv6.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
60 #include <linux/uaccess.h>
61 #include <linux/errno.h>
62 #include <linux/netdevice.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/smp.h>
70 #include <linux/if_ether.h>
72 #include <linux/mii.h>
73 #include <linux/ethtool.h>
74 #include <linux/if_vlan.h>
75 #include <linux/if_bonding.h>
76 #include <linux/phy.h>
77 #include <linux/jiffies.h>
78 #include <linux/preempt.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include <net/pkt_sched.h>
83 #include <linux/rculist.h>
84 #include <net/flow_dissector.h>
86 #include <net/bonding.h>
87 #include <net/bond_3ad.h>
88 #include <net/bond_alb.h>
89 #if IS_ENABLED(CONFIG_TLS_DEVICE)
92 #include <net/ip6_route.h>
95 #include "bonding_priv.h"
97 /*---------------------------- Module parameters ----------------------------*/
99 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
101 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
102 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
103 static int num_peer_notif = 1;
106 static int downdelay;
107 static int use_carrier = 1;
109 static char *primary;
110 static char *primary_reselect;
111 static char *lacp_rate;
112 static int min_links;
113 static char *ad_select;
114 static char *xmit_hash_policy;
115 static int arp_interval;
116 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
117 static char *arp_validate;
118 static char *arp_all_targets;
119 static char *fail_over_mac;
120 static int all_slaves_active;
121 static struct bond_params bonding_defaults;
122 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
123 static int packets_per_slave = 1;
124 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
126 module_param(max_bonds, int, 0);
127 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
128 module_param(tx_queues, int, 0);
129 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
130 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
131 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
132 "failover event (alias of num_unsol_na)");
133 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
134 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
135 "failover event (alias of num_grat_arp)");
136 module_param(miimon, int, 0);
137 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
138 module_param(updelay, int, 0);
139 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
140 module_param(downdelay, int, 0);
141 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
143 module_param(use_carrier, int, 0);
144 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
145 "0 for off, 1 for on (default)");
146 module_param(mode, charp, 0);
147 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
148 "1 for active-backup, 2 for balance-xor, "
149 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
150 "6 for balance-alb");
151 module_param(primary, charp, 0);
152 MODULE_PARM_DESC(primary, "Primary network device to use");
153 module_param(primary_reselect, charp, 0);
154 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
156 "0 for always (default), "
157 "1 for only if speed of primary is "
159 "2 for only on active slave "
161 module_param(lacp_rate, charp, 0);
162 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
163 "0 for slow, 1 for fast");
164 module_param(ad_select, charp, 0);
165 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
166 "0 for stable (default), 1 for bandwidth, "
168 module_param(min_links, int, 0);
169 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
171 module_param(xmit_hash_policy, charp, 0);
172 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
173 "0 for layer 2 (default), 1 for layer 3+4, "
174 "2 for layer 2+3, 3 for encap layer 2+3, "
175 "4 for encap layer 3+4, 5 for vlan+srcmac");
176 module_param(arp_interval, int, 0);
177 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
178 module_param_array(arp_ip_target, charp, NULL, 0);
179 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
180 module_param(arp_validate, charp, 0);
181 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
182 "0 for none (default), 1 for active, "
183 "2 for backup, 3 for all");
184 module_param(arp_all_targets, charp, 0);
185 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
186 module_param(fail_over_mac, charp, 0);
187 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
188 "the same MAC; 0 for none (default), "
189 "1 for active, 2 for follow");
190 module_param(all_slaves_active, int, 0);
191 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
192 "by setting active flag for all slaves; "
193 "0 for never (default), 1 for always.");
194 module_param(resend_igmp, int, 0);
195 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
197 module_param(packets_per_slave, int, 0);
198 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
199 "mode; 0 for a random slave, 1 packet per "
200 "slave (default), >1 packets per slave.");
201 module_param(lp_interval, uint, 0);
202 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
203 "the bonding driver sends learning packets to "
204 "each slaves peer switch. The default is 1.");
206 /*----------------------------- Global variables ----------------------------*/
208 #ifdef CONFIG_NET_POLL_CONTROLLER
209 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
212 unsigned int bond_net_id __read_mostly;
214 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
216 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
217 .offset = offsetof(struct flow_keys, control),
220 .key_id = FLOW_DISSECTOR_KEY_BASIC,
221 .offset = offsetof(struct flow_keys, basic),
224 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
225 .offset = offsetof(struct flow_keys, addrs.v4addrs),
228 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
229 .offset = offsetof(struct flow_keys, addrs.v6addrs),
232 .key_id = FLOW_DISSECTOR_KEY_TIPC,
233 .offset = offsetof(struct flow_keys, addrs.tipckey),
236 .key_id = FLOW_DISSECTOR_KEY_PORTS,
237 .offset = offsetof(struct flow_keys, ports),
240 .key_id = FLOW_DISSECTOR_KEY_ICMP,
241 .offset = offsetof(struct flow_keys, icmp),
244 .key_id = FLOW_DISSECTOR_KEY_VLAN,
245 .offset = offsetof(struct flow_keys, vlan),
248 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
249 .offset = offsetof(struct flow_keys, tags),
252 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
253 .offset = offsetof(struct flow_keys, keyid),
257 static struct flow_dissector flow_keys_bonding __read_mostly;
259 /*-------------------------- Forward declarations ---------------------------*/
261 static int bond_init(struct net_device *bond_dev);
262 static void bond_uninit(struct net_device *bond_dev);
263 static void bond_get_stats(struct net_device *bond_dev,
264 struct rtnl_link_stats64 *stats);
265 static void bond_slave_arr_handler(struct work_struct *work);
266 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
268 static void bond_netdev_notify_work(struct work_struct *work);
270 /*---------------------------- General routines -----------------------------*/
272 const char *bond_mode_name(int mode)
274 static const char *names[] = {
275 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
276 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
277 [BOND_MODE_XOR] = "load balancing (xor)",
278 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
279 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
280 [BOND_MODE_TLB] = "transmit load balancing",
281 [BOND_MODE_ALB] = "adaptive load balancing",
284 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
291 * bond_dev_queue_xmit - Prepare skb for xmit.
293 * @bond: bond device that got this skb for tx.
294 * @skb: hw accel VLAN tagged skb to transmit
295 * @slave_dev: slave that is supposed to xmit this skbuff
297 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
298 struct net_device *slave_dev)
300 skb->dev = slave_dev;
302 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
303 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
304 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
306 if (unlikely(netpoll_tx_running(bond->dev)))
307 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
309 return dev_queue_xmit(skb);
312 static bool bond_sk_check(struct bonding *bond)
314 switch (BOND_MODE(bond)) {
315 case BOND_MODE_8023AD:
317 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
325 static bool bond_xdp_check(struct bonding *bond)
327 switch (BOND_MODE(bond)) {
328 case BOND_MODE_ROUNDROBIN:
329 case BOND_MODE_ACTIVEBACKUP:
331 case BOND_MODE_8023AD:
333 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
334 * payload is not in the packet due to hardware offload.
336 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
344 /*---------------------------------- VLAN -----------------------------------*/
346 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
347 * We don't protect the slave list iteration with a lock because:
348 * a. This operation is performed in IOCTL context,
349 * b. The operation is protected by the RTNL semaphore in the 8021q code,
350 * c. Holding a lock with BH disabled while directly calling a base driver
351 * entry point is generally a BAD idea.
353 * The design of synchronization/protection for this operation in the 8021q
354 * module is good for one or more VLAN devices over a single physical device
355 * and cannot be extended for a teaming solution like bonding, so there is a
356 * potential race condition here where a net device from the vlan group might
357 * be referenced (either by a base driver or the 8021q code) while it is being
358 * removed from the system. However, it turns out we're not making matters
359 * worse, and if it works for regular VLAN usage it will work here too.
363 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
364 * @bond_dev: bonding net device that got called
365 * @proto: network protocol ID
366 * @vid: vlan id being added
368 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
369 __be16 proto, u16 vid)
371 struct bonding *bond = netdev_priv(bond_dev);
372 struct slave *slave, *rollback_slave;
373 struct list_head *iter;
376 bond_for_each_slave(bond, slave, iter) {
377 res = vlan_vid_add(slave->dev, proto, vid);
385 /* unwind to the slave that failed */
386 bond_for_each_slave(bond, rollback_slave, iter) {
387 if (rollback_slave == slave)
390 vlan_vid_del(rollback_slave->dev, proto, vid);
397 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
398 * @bond_dev: bonding net device that got called
399 * @proto: network protocol ID
400 * @vid: vlan id being removed
402 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
403 __be16 proto, u16 vid)
405 struct bonding *bond = netdev_priv(bond_dev);
406 struct list_head *iter;
409 bond_for_each_slave(bond, slave, iter)
410 vlan_vid_del(slave->dev, proto, vid);
412 if (bond_is_lb(bond))
413 bond_alb_clear_vlan(bond, vid);
418 /*---------------------------------- XFRM -----------------------------------*/
420 #ifdef CONFIG_XFRM_OFFLOAD
422 * bond_ipsec_add_sa - program device with a security association
423 * @xs: pointer to transformer state struct
424 * @extack: extack point to fill failure reason
426 static int bond_ipsec_add_sa(struct xfrm_state *xs,
427 struct netlink_ext_ack *extack)
429 struct net_device *bond_dev = xs->xso.dev;
430 struct bond_ipsec *ipsec;
431 struct bonding *bond;
439 bond = netdev_priv(bond_dev);
440 slave = rcu_dereference(bond->curr_active_slave);
446 if (!slave->dev->xfrmdev_ops ||
447 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
448 netif_is_bond_master(slave->dev)) {
449 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
454 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
459 xs->xso.real_dev = slave->dev;
461 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
464 INIT_LIST_HEAD(&ipsec->list);
465 spin_lock_bh(&bond->ipsec_lock);
466 list_add(&ipsec->list, &bond->ipsec_list);
467 spin_unlock_bh(&bond->ipsec_lock);
475 static void bond_ipsec_add_sa_all(struct bonding *bond)
477 struct net_device *bond_dev = bond->dev;
478 struct bond_ipsec *ipsec;
482 slave = rcu_dereference(bond->curr_active_slave);
486 if (!slave->dev->xfrmdev_ops ||
487 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
488 netif_is_bond_master(slave->dev)) {
489 spin_lock_bh(&bond->ipsec_lock);
490 if (!list_empty(&bond->ipsec_list))
491 slave_warn(bond_dev, slave->dev,
492 "%s: no slave xdo_dev_state_add\n",
494 spin_unlock_bh(&bond->ipsec_lock);
498 spin_lock_bh(&bond->ipsec_lock);
499 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
500 ipsec->xs->xso.real_dev = slave->dev;
501 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
502 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
503 ipsec->xs->xso.real_dev = NULL;
506 spin_unlock_bh(&bond->ipsec_lock);
512 * bond_ipsec_del_sa - clear out this specific SA
513 * @xs: pointer to transformer state struct
515 static void bond_ipsec_del_sa(struct xfrm_state *xs)
517 struct net_device *bond_dev = xs->xso.dev;
518 struct bond_ipsec *ipsec;
519 struct bonding *bond;
526 bond = netdev_priv(bond_dev);
527 slave = rcu_dereference(bond->curr_active_slave);
532 if (!xs->xso.real_dev)
535 WARN_ON(xs->xso.real_dev != slave->dev);
537 if (!slave->dev->xfrmdev_ops ||
538 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
539 netif_is_bond_master(slave->dev)) {
540 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
544 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
546 spin_lock_bh(&bond->ipsec_lock);
547 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
548 if (ipsec->xs == xs) {
549 list_del(&ipsec->list);
554 spin_unlock_bh(&bond->ipsec_lock);
558 static void bond_ipsec_del_sa_all(struct bonding *bond)
560 struct net_device *bond_dev = bond->dev;
561 struct bond_ipsec *ipsec;
565 slave = rcu_dereference(bond->curr_active_slave);
571 spin_lock_bh(&bond->ipsec_lock);
572 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
573 if (!ipsec->xs->xso.real_dev)
576 if (!slave->dev->xfrmdev_ops ||
577 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
578 netif_is_bond_master(slave->dev)) {
579 slave_warn(bond_dev, slave->dev,
580 "%s: no slave xdo_dev_state_delete\n",
583 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
585 ipsec->xs->xso.real_dev = NULL;
587 spin_unlock_bh(&bond->ipsec_lock);
592 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
593 * @skb: current data packet
594 * @xs: pointer to transformer state struct
596 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
598 struct net_device *bond_dev = xs->xso.dev;
599 struct net_device *real_dev;
600 struct slave *curr_active;
601 struct bonding *bond;
604 bond = netdev_priv(bond_dev);
606 curr_active = rcu_dereference(bond->curr_active_slave);
607 real_dev = curr_active->dev;
609 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
614 if (!xs->xso.real_dev) {
619 if (!real_dev->xfrmdev_ops ||
620 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
621 netif_is_bond_master(real_dev)) {
626 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
632 static const struct xfrmdev_ops bond_xfrmdev_ops = {
633 .xdo_dev_state_add = bond_ipsec_add_sa,
634 .xdo_dev_state_delete = bond_ipsec_del_sa,
635 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
637 #endif /* CONFIG_XFRM_OFFLOAD */
639 /*------------------------------- Link status -------------------------------*/
641 /* Set the carrier state for the master according to the state of its
642 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
643 * do special 802.3ad magic.
645 * Returns zero if carrier state does not change, nonzero if it does.
647 int bond_set_carrier(struct bonding *bond)
649 struct list_head *iter;
652 if (!bond_has_slaves(bond))
655 if (BOND_MODE(bond) == BOND_MODE_8023AD)
656 return bond_3ad_set_carrier(bond);
658 bond_for_each_slave(bond, slave, iter) {
659 if (slave->link == BOND_LINK_UP) {
660 if (!netif_carrier_ok(bond->dev)) {
661 netif_carrier_on(bond->dev);
669 if (netif_carrier_ok(bond->dev)) {
670 netif_carrier_off(bond->dev);
676 /* Get link speed and duplex from the slave's base driver
677 * using ethtool. If for some reason the call fails or the
678 * values are invalid, set speed and duplex to -1,
679 * and return. Return 1 if speed or duplex settings are
680 * UNKNOWN; 0 otherwise.
682 static int bond_update_speed_duplex(struct slave *slave)
684 struct net_device *slave_dev = slave->dev;
685 struct ethtool_link_ksettings ecmd;
688 slave->speed = SPEED_UNKNOWN;
689 slave->duplex = DUPLEX_UNKNOWN;
691 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
694 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
696 switch (ecmd.base.duplex) {
704 slave->speed = ecmd.base.speed;
705 slave->duplex = ecmd.base.duplex;
710 const char *bond_slave_link_status(s8 link)
726 /* if <dev> supports MII link status reporting, check its link status.
728 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
729 * depending upon the setting of the use_carrier parameter.
731 * Return either BMSR_LSTATUS, meaning that the link is up (or we
732 * can't tell and just pretend it is), or 0, meaning that the link is
735 * If reporting is non-zero, instead of faking link up, return -1 if
736 * both ETHTOOL and MII ioctls fail (meaning the device does not
737 * support them). If use_carrier is set, return whatever it says.
738 * It'd be nice if there was a good way to tell if a driver supports
739 * netif_carrier, but there really isn't.
741 static int bond_check_dev_link(struct bonding *bond,
742 struct net_device *slave_dev, int reporting)
744 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
745 int (*ioctl)(struct net_device *, struct ifreq *, int);
747 struct mii_ioctl_data *mii;
749 if (!reporting && !netif_running(slave_dev))
752 if (bond->params.use_carrier)
753 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
755 /* Try to get link status using Ethtool first. */
756 if (slave_dev->ethtool_ops->get_link)
757 return slave_dev->ethtool_ops->get_link(slave_dev) ?
760 /* Ethtool can't be used, fallback to MII ioctls. */
761 ioctl = slave_ops->ndo_eth_ioctl;
763 /* TODO: set pointer to correct ioctl on a per team member
764 * bases to make this more efficient. that is, once
765 * we determine the correct ioctl, we will always
766 * call it and not the others for that team
770 /* We cannot assume that SIOCGMIIPHY will also read a
771 * register; not all network drivers (e.g., e100)
775 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
776 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
778 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
779 mii->reg_num = MII_BMSR;
780 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
781 return mii->val_out & BMSR_LSTATUS;
785 /* If reporting, report that either there's no ndo_eth_ioctl,
786 * or both SIOCGMIIREG and get_link failed (meaning that we
787 * cannot report link status). If not reporting, pretend
790 return reporting ? -1 : BMSR_LSTATUS;
793 /*----------------------------- Multicast list ------------------------------*/
795 /* Push the promiscuity flag down to appropriate slaves */
796 static int bond_set_promiscuity(struct bonding *bond, int inc)
798 struct list_head *iter;
801 if (bond_uses_primary(bond)) {
802 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
805 err = dev_set_promiscuity(curr_active->dev, inc);
809 bond_for_each_slave(bond, slave, iter) {
810 err = dev_set_promiscuity(slave->dev, inc);
818 /* Push the allmulti flag down to all slaves */
819 static int bond_set_allmulti(struct bonding *bond, int inc)
821 struct list_head *iter;
824 if (bond_uses_primary(bond)) {
825 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
828 err = dev_set_allmulti(curr_active->dev, inc);
832 bond_for_each_slave(bond, slave, iter) {
833 err = dev_set_allmulti(slave->dev, inc);
841 /* Retrieve the list of registered multicast addresses for the bonding
842 * device and retransmit an IGMP JOIN request to the current active
845 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
847 struct bonding *bond = container_of(work, struct bonding,
850 if (!rtnl_trylock()) {
851 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
854 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
856 if (bond->igmp_retrans > 1) {
857 bond->igmp_retrans--;
858 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
863 /* Flush bond's hardware addresses from slave */
864 static void bond_hw_addr_flush(struct net_device *bond_dev,
865 struct net_device *slave_dev)
867 struct bonding *bond = netdev_priv(bond_dev);
869 dev_uc_unsync(slave_dev, bond_dev);
870 dev_mc_unsync(slave_dev, bond_dev);
872 if (BOND_MODE(bond) == BOND_MODE_8023AD)
873 dev_mc_del(slave_dev, lacpdu_mcast_addr);
876 /*--------------------------- Active slave change ---------------------------*/
878 /* Update the hardware address list and promisc/allmulti for the new and
879 * old active slaves (if any). Modes that are not using primary keep all
880 * slaves up date at all times; only the modes that use primary need to call
881 * this function to swap these settings during a failover.
883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
884 struct slave *old_active)
887 if (bond->dev->flags & IFF_PROMISC)
888 dev_set_promiscuity(old_active->dev, -1);
890 if (bond->dev->flags & IFF_ALLMULTI)
891 dev_set_allmulti(old_active->dev, -1);
893 if (bond->dev->flags & IFF_UP)
894 bond_hw_addr_flush(bond->dev, old_active->dev);
898 /* FIXME: Signal errors upstream. */
899 if (bond->dev->flags & IFF_PROMISC)
900 dev_set_promiscuity(new_active->dev, 1);
902 if (bond->dev->flags & IFF_ALLMULTI)
903 dev_set_allmulti(new_active->dev, 1);
905 if (bond->dev->flags & IFF_UP) {
906 netif_addr_lock_bh(bond->dev);
907 dev_uc_sync(new_active->dev, bond->dev);
908 dev_mc_sync(new_active->dev, bond->dev);
909 netif_addr_unlock_bh(bond->dev);
915 * bond_set_dev_addr - clone slave's address to bond
916 * @bond_dev: bond net device
917 * @slave_dev: slave net device
919 * Should be called with RTNL held.
921 static int bond_set_dev_addr(struct net_device *bond_dev,
922 struct net_device *slave_dev)
926 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
927 bond_dev, slave_dev, slave_dev->addr_len);
928 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
932 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
933 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
934 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
938 static struct slave *bond_get_old_active(struct bonding *bond,
939 struct slave *new_active)
942 struct list_head *iter;
944 bond_for_each_slave(bond, slave, iter) {
945 if (slave == new_active)
948 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
955 /* bond_do_fail_over_mac
957 * Perform special MAC address swapping for fail_over_mac settings
961 static void bond_do_fail_over_mac(struct bonding *bond,
962 struct slave *new_active,
963 struct slave *old_active)
965 u8 tmp_mac[MAX_ADDR_LEN];
966 struct sockaddr_storage ss;
969 switch (bond->params.fail_over_mac) {
970 case BOND_FOM_ACTIVE:
972 rv = bond_set_dev_addr(bond->dev, new_active->dev);
974 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
978 case BOND_FOM_FOLLOW:
979 /* if new_active && old_active, swap them
980 * if just old_active, do nothing (going to no active slave)
981 * if just new_active, set new_active to bond's MAC
987 old_active = bond_get_old_active(bond, new_active);
990 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
991 new_active->dev->addr_len);
992 bond_hw_addr_copy(ss.__data,
993 old_active->dev->dev_addr,
994 old_active->dev->addr_len);
995 ss.ss_family = new_active->dev->type;
997 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
998 bond->dev->addr_len);
999 ss.ss_family = bond->dev->type;
1002 rv = dev_set_mac_address(new_active->dev,
1003 (struct sockaddr *)&ss, NULL);
1005 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1013 bond_hw_addr_copy(ss.__data, tmp_mac,
1014 new_active->dev->addr_len);
1015 ss.ss_family = old_active->dev->type;
1017 rv = dev_set_mac_address(old_active->dev,
1018 (struct sockaddr *)&ss, NULL);
1020 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1025 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1026 bond->params.fail_over_mac);
1033 * bond_choose_primary_or_current - select the primary or high priority slave
1034 * @bond: our bonding struct
1036 * - Check if there is a primary link. If the primary link was set and is up,
1037 * go on and do link reselection.
1039 * - If primary link is not set or down, find the highest priority link.
1040 * If the highest priority link is not current slave, set it as primary
1041 * link and do link reselection.
1043 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1045 struct slave *prim = rtnl_dereference(bond->primary_slave);
1046 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1047 struct slave *slave, *hprio = NULL;
1048 struct list_head *iter;
1050 if (!prim || prim->link != BOND_LINK_UP) {
1051 bond_for_each_slave(bond, slave, iter) {
1052 if (slave->link == BOND_LINK_UP) {
1053 hprio = hprio ?: slave;
1054 if (slave->prio > hprio->prio)
1059 if (hprio && hprio != curr) {
1064 if (!curr || curr->link != BOND_LINK_UP)
1069 if (bond->force_primary) {
1070 bond->force_primary = false;
1075 if (!curr || curr->link != BOND_LINK_UP)
1078 /* At this point, prim and curr are both up */
1079 switch (bond->params.primary_reselect) {
1080 case BOND_PRI_RESELECT_ALWAYS:
1082 case BOND_PRI_RESELECT_BETTER:
1083 if (prim->speed < curr->speed)
1085 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1088 case BOND_PRI_RESELECT_FAILURE:
1091 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1092 bond->params.primary_reselect);
1098 * bond_find_best_slave - select the best available slave to be the active one
1099 * @bond: our bonding struct
1101 static struct slave *bond_find_best_slave(struct bonding *bond)
1103 struct slave *slave, *bestslave = NULL;
1104 struct list_head *iter;
1105 int mintime = bond->params.updelay;
1107 slave = bond_choose_primary_or_current(bond);
1111 bond_for_each_slave(bond, slave, iter) {
1112 if (slave->link == BOND_LINK_UP)
1114 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1115 slave->delay < mintime) {
1116 mintime = slave->delay;
1124 static bool bond_should_notify_peers(struct bonding *bond)
1126 struct slave *slave;
1129 slave = rcu_dereference(bond->curr_active_slave);
1132 if (!slave || !bond->send_peer_notif ||
1133 bond->send_peer_notif %
1134 max(1, bond->params.peer_notif_delay) != 0 ||
1135 !netif_carrier_ok(bond->dev) ||
1136 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1139 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1140 slave ? slave->dev->name : "NULL");
1146 * bond_change_active_slave - change the active slave into the specified one
1147 * @bond: our bonding struct
1148 * @new_active: the new slave to make the active one
1150 * Set the new slave to the bond's settings and unset them on the old
1151 * curr_active_slave.
1152 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1154 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1155 * because it is apparently the best available slave we have, even though its
1156 * updelay hasn't timed out yet.
1158 * Caller must hold RTNL.
1160 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1162 struct slave *old_active;
1166 old_active = rtnl_dereference(bond->curr_active_slave);
1168 if (old_active == new_active)
1171 #ifdef CONFIG_XFRM_OFFLOAD
1172 bond_ipsec_del_sa_all(bond);
1173 #endif /* CONFIG_XFRM_OFFLOAD */
1176 new_active->last_link_up = jiffies;
1178 if (new_active->link == BOND_LINK_BACK) {
1179 if (bond_uses_primary(bond)) {
1180 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1181 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1184 new_active->delay = 0;
1185 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1186 BOND_SLAVE_NOTIFY_NOW);
1188 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1189 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1191 if (bond_is_lb(bond))
1192 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1194 if (bond_uses_primary(bond))
1195 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1199 if (bond_uses_primary(bond))
1200 bond_hw_addr_swap(bond, new_active, old_active);
1202 if (bond_is_lb(bond)) {
1203 bond_alb_handle_active_change(bond, new_active);
1205 bond_set_slave_inactive_flags(old_active,
1206 BOND_SLAVE_NOTIFY_NOW);
1208 bond_set_slave_active_flags(new_active,
1209 BOND_SLAVE_NOTIFY_NOW);
1211 rcu_assign_pointer(bond->curr_active_slave, new_active);
1214 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1216 bond_set_slave_inactive_flags(old_active,
1217 BOND_SLAVE_NOTIFY_NOW);
1220 bool should_notify_peers = false;
1222 bond_set_slave_active_flags(new_active,
1223 BOND_SLAVE_NOTIFY_NOW);
1225 if (bond->params.fail_over_mac)
1226 bond_do_fail_over_mac(bond, new_active,
1229 if (netif_running(bond->dev)) {
1230 bond->send_peer_notif =
1231 bond->params.num_peer_notif *
1232 max(1, bond->params.peer_notif_delay);
1233 should_notify_peers =
1234 bond_should_notify_peers(bond);
1237 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1238 if (should_notify_peers) {
1239 bond->send_peer_notif--;
1240 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1246 #ifdef CONFIG_XFRM_OFFLOAD
1247 bond_ipsec_add_sa_all(bond);
1248 #endif /* CONFIG_XFRM_OFFLOAD */
1250 /* resend IGMP joins since active slave has changed or
1251 * all were sent on curr_active_slave.
1252 * resend only if bond is brought up with the affected
1253 * bonding modes and the retransmission is enabled
1255 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1256 ((bond_uses_primary(bond) && new_active) ||
1257 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1258 bond->igmp_retrans = bond->params.resend_igmp;
1259 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1264 * bond_select_active_slave - select a new active slave, if needed
1265 * @bond: our bonding struct
1267 * This functions should be called when one of the following occurs:
1268 * - The old curr_active_slave has been released or lost its link.
1269 * - The primary_slave has got its link back.
1270 * - A slave has got its link back and there's no old curr_active_slave.
1272 * Caller must hold RTNL.
1274 void bond_select_active_slave(struct bonding *bond)
1276 struct slave *best_slave;
1281 best_slave = bond_find_best_slave(bond);
1282 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1283 bond_change_active_slave(bond, best_slave);
1284 rv = bond_set_carrier(bond);
1288 if (netif_carrier_ok(bond->dev))
1289 netdev_info(bond->dev, "active interface up!\n");
1291 netdev_info(bond->dev, "now running without any active interface!\n");
1295 #ifdef CONFIG_NET_POLL_CONTROLLER
1296 static inline int slave_enable_netpoll(struct slave *slave)
1301 np = kzalloc(sizeof(*np), GFP_KERNEL);
1306 err = __netpoll_setup(np, slave->dev);
1315 static inline void slave_disable_netpoll(struct slave *slave)
1317 struct netpoll *np = slave->np;
1327 static void bond_poll_controller(struct net_device *bond_dev)
1329 struct bonding *bond = netdev_priv(bond_dev);
1330 struct slave *slave = NULL;
1331 struct list_head *iter;
1332 struct ad_info ad_info;
1334 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1335 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1338 bond_for_each_slave_rcu(bond, slave, iter) {
1339 if (!bond_slave_is_up(slave))
1342 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1343 struct aggregator *agg =
1344 SLAVE_AD_INFO(slave)->port.aggregator;
1347 agg->aggregator_identifier != ad_info.aggregator_id)
1351 netpoll_poll_dev(slave->dev);
1355 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1357 struct bonding *bond = netdev_priv(bond_dev);
1358 struct list_head *iter;
1359 struct slave *slave;
1361 bond_for_each_slave(bond, slave, iter)
1362 if (bond_slave_is_up(slave))
1363 slave_disable_netpoll(slave);
1366 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1368 struct bonding *bond = netdev_priv(dev);
1369 struct list_head *iter;
1370 struct slave *slave;
1373 bond_for_each_slave(bond, slave, iter) {
1374 err = slave_enable_netpoll(slave);
1376 bond_netpoll_cleanup(dev);
1383 static inline int slave_enable_netpoll(struct slave *slave)
1387 static inline void slave_disable_netpoll(struct slave *slave)
1390 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1395 /*---------------------------------- IOCTL ----------------------------------*/
1397 static netdev_features_t bond_fix_features(struct net_device *dev,
1398 netdev_features_t features)
1400 struct bonding *bond = netdev_priv(dev);
1401 struct list_head *iter;
1402 netdev_features_t mask;
1403 struct slave *slave;
1407 features &= ~NETIF_F_ONE_FOR_ALL;
1408 features |= NETIF_F_ALL_FOR_ALL;
1410 bond_for_each_slave(bond, slave, iter) {
1411 features = netdev_increment_features(features,
1412 slave->dev->features,
1415 features = netdev_add_tso_features(features, mask);
1420 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1421 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1422 NETIF_F_HIGHDMA | NETIF_F_LRO)
1424 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1425 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1427 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1428 NETIF_F_GSO_SOFTWARE)
1431 static void bond_compute_features(struct bonding *bond)
1433 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1434 IFF_XMIT_DST_RELEASE_PERM;
1435 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1436 netdev_features_t enc_features = BOND_ENC_FEATURES;
1437 #ifdef CONFIG_XFRM_OFFLOAD
1438 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1439 #endif /* CONFIG_XFRM_OFFLOAD */
1440 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1441 struct net_device *bond_dev = bond->dev;
1442 struct list_head *iter;
1443 struct slave *slave;
1444 unsigned short max_hard_header_len = ETH_HLEN;
1445 unsigned int tso_max_size = TSO_MAX_SIZE;
1446 u16 tso_max_segs = TSO_MAX_SEGS;
1448 if (!bond_has_slaves(bond))
1450 vlan_features &= NETIF_F_ALL_FOR_ALL;
1451 mpls_features &= NETIF_F_ALL_FOR_ALL;
1453 bond_for_each_slave(bond, slave, iter) {
1454 vlan_features = netdev_increment_features(vlan_features,
1455 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1457 enc_features = netdev_increment_features(enc_features,
1458 slave->dev->hw_enc_features,
1461 #ifdef CONFIG_XFRM_OFFLOAD
1462 xfrm_features = netdev_increment_features(xfrm_features,
1463 slave->dev->hw_enc_features,
1464 BOND_XFRM_FEATURES);
1465 #endif /* CONFIG_XFRM_OFFLOAD */
1467 mpls_features = netdev_increment_features(mpls_features,
1468 slave->dev->mpls_features,
1469 BOND_MPLS_FEATURES);
1471 dst_release_flag &= slave->dev->priv_flags;
1472 if (slave->dev->hard_header_len > max_hard_header_len)
1473 max_hard_header_len = slave->dev->hard_header_len;
1475 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1476 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1478 bond_dev->hard_header_len = max_hard_header_len;
1481 bond_dev->vlan_features = vlan_features;
1482 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1483 NETIF_F_HW_VLAN_CTAG_TX |
1484 NETIF_F_HW_VLAN_STAG_TX;
1485 #ifdef CONFIG_XFRM_OFFLOAD
1486 bond_dev->hw_enc_features |= xfrm_features;
1487 #endif /* CONFIG_XFRM_OFFLOAD */
1488 bond_dev->mpls_features = mpls_features;
1489 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1490 netif_set_tso_max_size(bond_dev, tso_max_size);
1492 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1493 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1494 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1495 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1497 netdev_change_features(bond_dev);
1500 static void bond_setup_by_slave(struct net_device *bond_dev,
1501 struct net_device *slave_dev)
1503 bool was_up = !!(bond_dev->flags & IFF_UP);
1505 dev_close(bond_dev);
1507 bond_dev->header_ops = slave_dev->header_ops;
1509 bond_dev->type = slave_dev->type;
1510 bond_dev->hard_header_len = slave_dev->hard_header_len;
1511 bond_dev->needed_headroom = slave_dev->needed_headroom;
1512 bond_dev->addr_len = slave_dev->addr_len;
1514 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1515 slave_dev->addr_len);
1517 if (slave_dev->flags & IFF_POINTOPOINT) {
1518 bond_dev->flags &= ~(IFF_BROADCAST | IFF_MULTICAST);
1519 bond_dev->flags |= (IFF_POINTOPOINT | IFF_NOARP);
1522 dev_open(bond_dev, NULL);
1525 /* On bonding slaves other than the currently active slave, suppress
1526 * duplicates except for alb non-mcast/bcast.
1528 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1529 struct slave *slave,
1530 struct bonding *bond)
1532 if (bond_is_slave_inactive(slave)) {
1533 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1534 skb->pkt_type != PACKET_BROADCAST &&
1535 skb->pkt_type != PACKET_MULTICAST)
1542 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1544 struct sk_buff *skb = *pskb;
1545 struct slave *slave;
1546 struct bonding *bond;
1547 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1549 int ret = RX_HANDLER_ANOTHER;
1551 skb = skb_share_check(skb, GFP_ATOMIC);
1553 return RX_HANDLER_CONSUMED;
1557 slave = bond_slave_get_rcu(skb->dev);
1560 recv_probe = READ_ONCE(bond->recv_probe);
1562 ret = recv_probe(skb, bond, slave);
1563 if (ret == RX_HANDLER_CONSUMED) {
1570 * For packets determined by bond_should_deliver_exact_match() call to
1571 * be suppressed we want to make an exception for link-local packets.
1572 * This is necessary for e.g. LLDP daemons to be able to monitor
1573 * inactive slave links without being forced to bind to them
1576 * At the same time, packets that are passed to the bonding master
1577 * (including link-local ones) can have their originating interface
1578 * determined via PACKET_ORIGDEV socket option.
1580 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1581 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1582 return RX_HANDLER_PASS;
1583 return RX_HANDLER_EXACT;
1586 skb->dev = bond->dev;
1588 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1589 netif_is_bridge_port(bond->dev) &&
1590 skb->pkt_type == PACKET_HOST) {
1592 if (unlikely(skb_cow_head(skb,
1593 skb->data - skb_mac_header(skb)))) {
1595 return RX_HANDLER_CONSUMED;
1597 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1598 bond->dev->addr_len);
1604 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1606 switch (BOND_MODE(bond)) {
1607 case BOND_MODE_ROUNDROBIN:
1608 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1609 case BOND_MODE_ACTIVEBACKUP:
1610 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1611 case BOND_MODE_BROADCAST:
1612 return NETDEV_LAG_TX_TYPE_BROADCAST;
1614 case BOND_MODE_8023AD:
1615 return NETDEV_LAG_TX_TYPE_HASH;
1617 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1621 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1622 enum netdev_lag_tx_type type)
1624 if (type != NETDEV_LAG_TX_TYPE_HASH)
1625 return NETDEV_LAG_HASH_NONE;
1627 switch (bond->params.xmit_policy) {
1628 case BOND_XMIT_POLICY_LAYER2:
1629 return NETDEV_LAG_HASH_L2;
1630 case BOND_XMIT_POLICY_LAYER34:
1631 return NETDEV_LAG_HASH_L34;
1632 case BOND_XMIT_POLICY_LAYER23:
1633 return NETDEV_LAG_HASH_L23;
1634 case BOND_XMIT_POLICY_ENCAP23:
1635 return NETDEV_LAG_HASH_E23;
1636 case BOND_XMIT_POLICY_ENCAP34:
1637 return NETDEV_LAG_HASH_E34;
1638 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1639 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1641 return NETDEV_LAG_HASH_UNKNOWN;
1645 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1646 struct netlink_ext_ack *extack)
1648 struct netdev_lag_upper_info lag_upper_info;
1649 enum netdev_lag_tx_type type;
1652 type = bond_lag_tx_type(bond);
1653 lag_upper_info.tx_type = type;
1654 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1656 err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1657 &lag_upper_info, extack);
1661 slave->dev->flags |= IFF_SLAVE;
1665 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1667 netdev_upper_dev_unlink(slave->dev, bond->dev);
1668 slave->dev->flags &= ~IFF_SLAVE;
1671 static void slave_kobj_release(struct kobject *kobj)
1673 struct slave *slave = to_slave(kobj);
1674 struct bonding *bond = bond_get_bond_by_slave(slave);
1676 cancel_delayed_work_sync(&slave->notify_work);
1677 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1678 kfree(SLAVE_AD_INFO(slave));
1683 static struct kobj_type slave_ktype = {
1684 .release = slave_kobj_release,
1686 .sysfs_ops = &slave_sysfs_ops,
1690 static int bond_kobj_init(struct slave *slave)
1694 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1695 &(slave->dev->dev.kobj), "bonding_slave");
1697 kobject_put(&slave->kobj);
1702 static struct slave *bond_alloc_slave(struct bonding *bond,
1703 struct net_device *slave_dev)
1705 struct slave *slave = NULL;
1707 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1712 slave->dev = slave_dev;
1713 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1715 if (bond_kobj_init(slave))
1718 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1719 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1721 if (!SLAVE_AD_INFO(slave)) {
1722 kobject_put(&slave->kobj);
1730 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1732 info->bond_mode = BOND_MODE(bond);
1733 info->miimon = bond->params.miimon;
1734 info->num_slaves = bond->slave_cnt;
1737 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1739 strcpy(info->slave_name, slave->dev->name);
1740 info->link = slave->link;
1741 info->state = bond_slave_state(slave);
1742 info->link_failure_count = slave->link_failure_count;
1745 static void bond_netdev_notify_work(struct work_struct *_work)
1747 struct slave *slave = container_of(_work, struct slave,
1750 if (rtnl_trylock()) {
1751 struct netdev_bonding_info binfo;
1753 bond_fill_ifslave(slave, &binfo.slave);
1754 bond_fill_ifbond(slave->bond, &binfo.master);
1755 netdev_bonding_info_change(slave->dev, &binfo);
1758 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1762 void bond_queue_slave_event(struct slave *slave)
1764 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1767 void bond_lower_state_changed(struct slave *slave)
1769 struct netdev_lag_lower_state_info info;
1771 info.link_up = slave->link == BOND_LINK_UP ||
1772 slave->link == BOND_LINK_FAIL;
1773 info.tx_enabled = bond_is_active_slave(slave);
1774 netdev_lower_state_changed(slave->dev, &info);
1777 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1779 NL_SET_ERR_MSG(extack, errmsg); \
1781 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1784 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1786 NL_SET_ERR_MSG(extack, errmsg); \
1788 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1791 /* The bonding driver uses ether_setup() to convert a master bond device
1792 * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1793 * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE and IFF_UP
1796 static void bond_ether_setup(struct net_device *bond_dev)
1798 unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1800 ether_setup(bond_dev);
1801 bond_dev->flags |= IFF_MASTER | flags;
1802 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1805 void bond_xdp_set_features(struct net_device *bond_dev)
1807 struct bonding *bond = netdev_priv(bond_dev);
1808 xdp_features_t val = NETDEV_XDP_ACT_MASK;
1809 struct list_head *iter;
1810 struct slave *slave;
1814 if (!bond_xdp_check(bond) || !bond_has_slaves(bond)) {
1815 xdp_clear_features_flag(bond_dev);
1819 bond_for_each_slave(bond, slave, iter)
1820 val &= slave->dev->xdp_features;
1822 val &= ~NETDEV_XDP_ACT_XSK_ZEROCOPY;
1824 xdp_set_features_flag(bond_dev, val);
1827 /* enslave device <slave> to bond device <master> */
1828 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1829 struct netlink_ext_ack *extack)
1831 struct bonding *bond = netdev_priv(bond_dev);
1832 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1833 struct slave *new_slave = NULL, *prev_slave;
1834 struct sockaddr_storage ss;
1838 if (slave_dev->flags & IFF_MASTER &&
1839 !netif_is_bond_master(slave_dev)) {
1840 BOND_NL_ERR(bond_dev, extack,
1841 "Device type (master device) cannot be enslaved");
1845 if (!bond->params.use_carrier &&
1846 slave_dev->ethtool_ops->get_link == NULL &&
1847 slave_ops->ndo_eth_ioctl == NULL) {
1848 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1851 /* already in-use? */
1852 if (netdev_is_rx_handler_busy(slave_dev)) {
1853 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1854 "Device is in use and cannot be enslaved");
1858 if (bond_dev == slave_dev) {
1859 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1863 /* vlan challenged mutual exclusion */
1864 /* no need to lock since we're protected by rtnl_lock */
1865 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1866 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1867 if (vlan_uses_dev(bond_dev)) {
1868 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1869 "Can not enslave VLAN challenged device to VLAN enabled bond");
1872 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1875 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1878 if (slave_dev->features & NETIF_F_HW_ESP)
1879 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1881 /* Old ifenslave binaries are no longer supported. These can
1882 * be identified with moderate accuracy by the state of the slave:
1883 * the current ifenslave will set the interface down prior to
1884 * enslaving it; the old ifenslave will not.
1886 if (slave_dev->flags & IFF_UP) {
1887 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1888 "Device can not be enslaved while up");
1892 /* set bonding device ether type by slave - bonding netdevices are
1893 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1894 * there is a need to override some of the type dependent attribs/funcs.
1896 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1897 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1899 if (!bond_has_slaves(bond)) {
1900 if (bond_dev->type != slave_dev->type) {
1901 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1902 bond_dev->type, slave_dev->type);
1904 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1906 res = notifier_to_errno(res);
1908 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1912 /* Flush unicast and multicast addresses */
1913 dev_uc_flush(bond_dev);
1914 dev_mc_flush(bond_dev);
1916 if (slave_dev->type != ARPHRD_ETHER)
1917 bond_setup_by_slave(bond_dev, slave_dev);
1919 bond_ether_setup(bond_dev);
1921 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1924 } else if (bond_dev->type != slave_dev->type) {
1925 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1926 "Device type is different from other slaves");
1930 if (slave_dev->type == ARPHRD_INFINIBAND &&
1931 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1932 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1933 "Only active-backup mode is supported for infiniband slaves");
1935 goto err_undo_flags;
1938 if (!slave_ops->ndo_set_mac_address ||
1939 slave_dev->type == ARPHRD_INFINIBAND) {
1940 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1941 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1942 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1943 if (!bond_has_slaves(bond)) {
1944 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1945 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1947 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1948 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1950 goto err_undo_flags;
1955 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1957 /* If this is the first slave, then we need to set the master's hardware
1958 * address to be the same as the slave's.
1960 if (!bond_has_slaves(bond) &&
1961 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1962 res = bond_set_dev_addr(bond->dev, slave_dev);
1964 goto err_undo_flags;
1967 new_slave = bond_alloc_slave(bond, slave_dev);
1970 goto err_undo_flags;
1973 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1974 * is set via sysfs or module option if desired.
1976 new_slave->queue_id = 0;
1978 /* Save slave's original mtu and then set it to match the bond */
1979 new_slave->original_mtu = slave_dev->mtu;
1980 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1982 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1986 /* Save slave's original ("permanent") mac address for modes
1987 * that need it, and for restoring it upon release, and then
1988 * set it to the master's address
1990 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1991 slave_dev->addr_len);
1993 if (!bond->params.fail_over_mac ||
1994 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1995 /* Set slave to master's mac address. The application already
1996 * set the master's mac address to that of the first slave
1998 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1999 ss.ss_family = slave_dev->type;
2000 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
2003 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
2004 goto err_restore_mtu;
2008 /* set no_addrconf flag before open to prevent IPv6 addrconf */
2009 slave_dev->priv_flags |= IFF_NO_ADDRCONF;
2011 /* open the slave since the application closed it */
2012 res = dev_open(slave_dev, extack);
2014 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
2015 goto err_restore_mac;
2018 slave_dev->priv_flags |= IFF_BONDING;
2019 /* initialize slave stats */
2020 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
2022 if (bond_is_lb(bond)) {
2023 /* bond_alb_init_slave() must be called before all other stages since
2024 * it might fail and we do not want to have to undo everything
2026 res = bond_alb_init_slave(bond, new_slave);
2031 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
2033 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
2037 prev_slave = bond_last_slave(bond);
2039 new_slave->delay = 0;
2040 new_slave->link_failure_count = 0;
2042 if (bond_update_speed_duplex(new_slave) &&
2043 bond_needs_speed_duplex(bond))
2044 new_slave->link = BOND_LINK_DOWN;
2046 new_slave->last_rx = jiffies -
2047 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2048 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2049 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2051 new_slave->last_tx = new_slave->last_rx;
2053 if (bond->params.miimon && !bond->params.use_carrier) {
2054 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2056 if ((link_reporting == -1) && !bond->params.arp_interval) {
2057 /* miimon is set but a bonded network driver
2058 * does not support ETHTOOL/MII and
2059 * arp_interval is not set. Note: if
2060 * use_carrier is enabled, we will never go
2061 * here (because netif_carrier is always
2062 * supported); thus, we don't need to change
2063 * the messages for netif_carrier.
2065 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");
2066 } else if (link_reporting == -1) {
2067 /* unable get link status using mii/ethtool */
2068 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");
2072 /* check for initial state */
2073 new_slave->link = BOND_LINK_NOCHANGE;
2074 if (bond->params.miimon) {
2075 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2076 if (bond->params.updelay) {
2077 bond_set_slave_link_state(new_slave,
2079 BOND_SLAVE_NOTIFY_NOW);
2080 new_slave->delay = bond->params.updelay;
2082 bond_set_slave_link_state(new_slave,
2084 BOND_SLAVE_NOTIFY_NOW);
2087 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2088 BOND_SLAVE_NOTIFY_NOW);
2090 } else if (bond->params.arp_interval) {
2091 bond_set_slave_link_state(new_slave,
2092 (netif_carrier_ok(slave_dev) ?
2093 BOND_LINK_UP : BOND_LINK_DOWN),
2094 BOND_SLAVE_NOTIFY_NOW);
2096 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2097 BOND_SLAVE_NOTIFY_NOW);
2100 if (new_slave->link != BOND_LINK_DOWN)
2101 new_slave->last_link_up = jiffies;
2102 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2103 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2104 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2106 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2107 /* if there is a primary slave, remember it */
2108 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2109 rcu_assign_pointer(bond->primary_slave, new_slave);
2110 bond->force_primary = true;
2114 switch (BOND_MODE(bond)) {
2115 case BOND_MODE_ACTIVEBACKUP:
2116 bond_set_slave_inactive_flags(new_slave,
2117 BOND_SLAVE_NOTIFY_NOW);
2119 case BOND_MODE_8023AD:
2120 /* in 802.3ad mode, the internal mechanism
2121 * will activate the slaves in the selected
2124 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2125 /* if this is the first slave */
2127 SLAVE_AD_INFO(new_slave)->id = 1;
2128 /* Initialize AD with the number of times that the AD timer is called in 1 second
2129 * can be called only after the mac address of the bond is set
2131 bond_3ad_initialize(bond);
2133 SLAVE_AD_INFO(new_slave)->id =
2134 SLAVE_AD_INFO(prev_slave)->id + 1;
2137 bond_3ad_bind_slave(new_slave);
2141 bond_set_active_slave(new_slave);
2142 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2145 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2147 /* always active in trunk mode */
2148 bond_set_active_slave(new_slave);
2150 /* In trunking mode there is little meaning to curr_active_slave
2151 * anyway (it holds no special properties of the bond device),
2152 * so we can change it without calling change_active_interface()
2154 if (!rcu_access_pointer(bond->curr_active_slave) &&
2155 new_slave->link == BOND_LINK_UP)
2156 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2159 } /* switch(bond_mode) */
2161 #ifdef CONFIG_NET_POLL_CONTROLLER
2162 if (bond->dev->npinfo) {
2163 if (slave_enable_netpoll(new_slave)) {
2164 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2171 if (!(bond_dev->features & NETIF_F_LRO))
2172 dev_disable_lro(slave_dev);
2174 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2177 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2181 res = bond_master_upper_dev_link(bond, new_slave, extack);
2183 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2184 goto err_unregister;
2187 bond_lower_state_changed(new_slave);
2189 res = bond_sysfs_slave_add(new_slave);
2191 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2192 goto err_upper_unlink;
2195 /* If the mode uses primary, then the following is handled by
2196 * bond_change_active_slave().
2198 if (!bond_uses_primary(bond)) {
2199 /* set promiscuity level to new slave */
2200 if (bond_dev->flags & IFF_PROMISC) {
2201 res = dev_set_promiscuity(slave_dev, 1);
2206 /* set allmulti level to new slave */
2207 if (bond_dev->flags & IFF_ALLMULTI) {
2208 res = dev_set_allmulti(slave_dev, 1);
2210 if (bond_dev->flags & IFF_PROMISC)
2211 dev_set_promiscuity(slave_dev, -1);
2216 if (bond_dev->flags & IFF_UP) {
2217 netif_addr_lock_bh(bond_dev);
2218 dev_mc_sync_multiple(slave_dev, bond_dev);
2219 dev_uc_sync_multiple(slave_dev, bond_dev);
2220 netif_addr_unlock_bh(bond_dev);
2222 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2223 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2228 bond_compute_features(bond);
2229 bond_set_carrier(bond);
2231 if (bond_uses_primary(bond)) {
2233 bond_select_active_slave(bond);
2234 unblock_netpoll_tx();
2237 if (bond_mode_can_use_xmit_hash(bond))
2238 bond_update_slave_arr(bond, NULL);
2241 if (!slave_dev->netdev_ops->ndo_bpf ||
2242 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2243 if (bond->xdp_prog) {
2244 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2245 "Slave does not support XDP");
2249 } else if (bond->xdp_prog) {
2250 struct netdev_bpf xdp = {
2251 .command = XDP_SETUP_PROG,
2253 .prog = bond->xdp_prog,
2257 if (dev_xdp_prog_count(slave_dev) > 0) {
2258 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2259 "Slave has XDP program loaded, please unload before enslaving");
2264 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2266 /* ndo_bpf() sets extack error message */
2267 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2271 bpf_prog_inc(bond->xdp_prog);
2274 bond_xdp_set_features(bond_dev);
2276 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2277 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2278 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2280 /* enslave is successful */
2281 bond_queue_slave_event(new_slave);
2284 /* Undo stages on error */
2286 bond_sysfs_slave_del(new_slave);
2289 bond_upper_dev_unlink(bond, new_slave);
2292 netdev_rx_handler_unregister(slave_dev);
2295 vlan_vids_del_by_dev(slave_dev, bond_dev);
2296 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2297 RCU_INIT_POINTER(bond->primary_slave, NULL);
2298 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2300 bond_change_active_slave(bond, NULL);
2301 bond_select_active_slave(bond);
2302 unblock_netpoll_tx();
2304 /* either primary_slave or curr_active_slave might've changed */
2306 slave_disable_netpoll(new_slave);
2309 if (!netif_is_bond_master(slave_dev))
2310 slave_dev->priv_flags &= ~IFF_BONDING;
2311 dev_close(slave_dev);
2314 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2315 if (!bond->params.fail_over_mac ||
2316 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2317 /* XXX TODO - fom follow mode needs to change master's
2318 * MAC if this slave's MAC is in use by the bond, or at
2319 * least print a warning.
2321 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2322 new_slave->dev->addr_len);
2323 ss.ss_family = slave_dev->type;
2324 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2328 dev_set_mtu(slave_dev, new_slave->original_mtu);
2331 kobject_put(&new_slave->kobj);
2334 /* Enslave of first slave has failed and we need to fix master's mac */
2335 if (!bond_has_slaves(bond)) {
2336 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2337 slave_dev->dev_addr))
2338 eth_hw_addr_random(bond_dev);
2339 if (bond_dev->type != ARPHRD_ETHER) {
2340 dev_close(bond_dev);
2341 bond_ether_setup(bond_dev);
2348 /* Try to release the slave device <slave> from the bond device <master>
2349 * It is legal to access curr_active_slave without a lock because all the function
2350 * is RTNL-locked. If "all" is true it means that the function is being called
2351 * while destroying a bond interface and all slaves are being released.
2353 * The rules for slave state should be:
2354 * for Active/Backup:
2355 * Active stays on all backups go down
2356 * for Bonded connections:
2357 * The first up interface should be left on and all others downed.
2359 static int __bond_release_one(struct net_device *bond_dev,
2360 struct net_device *slave_dev,
2361 bool all, bool unregister)
2363 struct bonding *bond = netdev_priv(bond_dev);
2364 struct slave *slave, *oldcurrent;
2365 struct sockaddr_storage ss;
2366 int old_flags = bond_dev->flags;
2367 netdev_features_t old_features = bond_dev->features;
2369 /* slave is not a slave or master is not master of this slave */
2370 if (!(slave_dev->flags & IFF_SLAVE) ||
2371 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2372 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2378 slave = bond_get_slave_by_dev(bond, slave_dev);
2380 /* not a slave of this bond */
2381 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2382 unblock_netpoll_tx();
2386 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2388 bond_sysfs_slave_del(slave);
2390 /* recompute stats just before removing the slave */
2391 bond_get_stats(bond->dev, &bond->bond_stats);
2393 if (bond->xdp_prog) {
2394 struct netdev_bpf xdp = {
2395 .command = XDP_SETUP_PROG,
2400 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2401 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2404 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2405 * for this slave anymore.
2407 netdev_rx_handler_unregister(slave_dev);
2409 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2410 bond_3ad_unbind_slave(slave);
2412 bond_upper_dev_unlink(bond, slave);
2414 if (bond_mode_can_use_xmit_hash(bond))
2415 bond_update_slave_arr(bond, slave);
2417 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2418 bond_is_active_slave(slave) ? "active" : "backup");
2420 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2422 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2424 if (!all && (!bond->params.fail_over_mac ||
2425 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2426 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2427 bond_has_slaves(bond))
2428 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",
2429 slave->perm_hwaddr);
2432 if (rtnl_dereference(bond->primary_slave) == slave)
2433 RCU_INIT_POINTER(bond->primary_slave, NULL);
2435 if (oldcurrent == slave)
2436 bond_change_active_slave(bond, NULL);
2438 if (bond_is_lb(bond)) {
2439 /* Must be called only after the slave has been
2440 * detached from the list and the curr_active_slave
2441 * has been cleared (if our_slave == old_current),
2442 * but before a new active slave is selected.
2444 bond_alb_deinit_slave(bond, slave);
2448 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2449 } else if (oldcurrent == slave) {
2450 /* Note that we hold RTNL over this sequence, so there
2451 * is no concern that another slave add/remove event
2454 bond_select_active_slave(bond);
2457 bond_set_carrier(bond);
2458 if (!bond_has_slaves(bond))
2459 eth_hw_addr_random(bond_dev);
2461 unblock_netpoll_tx();
2465 if (!bond_has_slaves(bond)) {
2466 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2467 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2470 bond_compute_features(bond);
2471 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2472 (old_features & NETIF_F_VLAN_CHALLENGED))
2473 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2475 vlan_vids_del_by_dev(slave_dev, bond_dev);
2477 /* If the mode uses primary, then this case was handled above by
2478 * bond_change_active_slave(..., NULL)
2480 if (!bond_uses_primary(bond)) {
2481 /* unset promiscuity level from slave
2482 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2483 * of the IFF_PROMISC flag in the bond_dev, but we need the
2484 * value of that flag before that change, as that was the value
2485 * when this slave was attached, so we cache at the start of the
2486 * function and use it here. Same goes for ALLMULTI below
2488 if (old_flags & IFF_PROMISC)
2489 dev_set_promiscuity(slave_dev, -1);
2491 /* unset allmulti level from slave */
2492 if (old_flags & IFF_ALLMULTI)
2493 dev_set_allmulti(slave_dev, -1);
2495 if (old_flags & IFF_UP)
2496 bond_hw_addr_flush(bond_dev, slave_dev);
2499 slave_disable_netpoll(slave);
2501 /* close slave before restoring its mac address */
2502 dev_close(slave_dev);
2504 slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2506 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2507 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2508 /* restore original ("permanent") mac address */
2509 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2510 slave->dev->addr_len);
2511 ss.ss_family = slave_dev->type;
2512 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2516 __dev_set_mtu(slave_dev, slave->original_mtu);
2518 dev_set_mtu(slave_dev, slave->original_mtu);
2520 if (!netif_is_bond_master(slave_dev))
2521 slave_dev->priv_flags &= ~IFF_BONDING;
2523 bond_xdp_set_features(bond_dev);
2524 kobject_put(&slave->kobj);
2529 /* A wrapper used because of ndo_del_link */
2530 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2532 return __bond_release_one(bond_dev, slave_dev, false, false);
2535 /* First release a slave and then destroy the bond if no more slaves are left.
2536 * Must be under rtnl_lock when this function is called.
2538 static int bond_release_and_destroy(struct net_device *bond_dev,
2539 struct net_device *slave_dev)
2541 struct bonding *bond = netdev_priv(bond_dev);
2544 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2545 if (ret == 0 && !bond_has_slaves(bond) &&
2546 bond_dev->reg_state != NETREG_UNREGISTERING) {
2547 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2548 netdev_info(bond_dev, "Destroying bond\n");
2549 bond_remove_proc_entry(bond);
2550 unregister_netdevice(bond_dev);
2555 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2557 struct bonding *bond = netdev_priv(bond_dev);
2559 bond_fill_ifbond(bond, info);
2562 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2564 struct bonding *bond = netdev_priv(bond_dev);
2565 struct list_head *iter;
2566 int i = 0, res = -ENODEV;
2567 struct slave *slave;
2569 bond_for_each_slave(bond, slave, iter) {
2570 if (i++ == (int)info->slave_id) {
2572 bond_fill_ifslave(slave, info);
2580 /*-------------------------------- Monitoring -------------------------------*/
2582 /* called with rcu_read_lock() */
2583 static int bond_miimon_inspect(struct bonding *bond)
2585 bool ignore_updelay = false;
2586 int link_state, commit = 0;
2587 struct list_head *iter;
2588 struct slave *slave;
2590 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2591 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2593 struct bond_up_slave *usable_slaves;
2595 usable_slaves = rcu_dereference(bond->usable_slaves);
2597 if (usable_slaves && usable_slaves->count == 0)
2598 ignore_updelay = true;
2601 bond_for_each_slave_rcu(bond, slave, iter) {
2602 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2604 link_state = bond_check_dev_link(bond, slave->dev, 0);
2606 switch (slave->link) {
2611 bond_propose_link_state(slave, BOND_LINK_FAIL);
2613 slave->delay = bond->params.downdelay;
2614 if (slave->delay && net_ratelimit()) {
2615 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2617 BOND_MODE_ACTIVEBACKUP) ?
2618 (bond_is_active_slave(slave) ?
2619 "active " : "backup ") : "",
2620 bond->params.downdelay * bond->params.miimon);
2623 case BOND_LINK_FAIL:
2625 /* recovered before downdelay expired */
2626 bond_propose_link_state(slave, BOND_LINK_UP);
2627 slave->last_link_up = jiffies;
2628 if (net_ratelimit())
2629 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2630 (bond->params.downdelay - slave->delay) *
2631 bond->params.miimon);
2636 if (slave->delay <= 0) {
2637 bond_propose_link_state(slave, BOND_LINK_DOWN);
2645 case BOND_LINK_DOWN:
2649 bond_propose_link_state(slave, BOND_LINK_BACK);
2651 slave->delay = bond->params.updelay;
2653 if (slave->delay && net_ratelimit()) {
2654 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2655 ignore_updelay ? 0 :
2656 bond->params.updelay *
2657 bond->params.miimon);
2660 case BOND_LINK_BACK:
2662 bond_propose_link_state(slave, BOND_LINK_DOWN);
2663 if (net_ratelimit())
2664 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2665 (bond->params.updelay - slave->delay) *
2666 bond->params.miimon);
2674 if (slave->delay <= 0) {
2675 bond_propose_link_state(slave, BOND_LINK_UP);
2677 ignore_updelay = false;
2689 static void bond_miimon_link_change(struct bonding *bond,
2690 struct slave *slave,
2693 switch (BOND_MODE(bond)) {
2694 case BOND_MODE_8023AD:
2695 bond_3ad_handle_link_change(slave, link);
2699 bond_alb_handle_link_change(bond, slave, link);
2702 bond_update_slave_arr(bond, NULL);
2707 static void bond_miimon_commit(struct bonding *bond)
2709 struct slave *slave, *primary, *active;
2710 bool do_failover = false;
2711 struct list_head *iter;
2715 bond_for_each_slave(bond, slave, iter) {
2716 switch (slave->link_new_state) {
2717 case BOND_LINK_NOCHANGE:
2718 /* For 802.3ad mode, check current slave speed and
2719 * duplex again in case its port was disabled after
2720 * invalid speed/duplex reporting but recovered before
2721 * link monitoring could make a decision on the actual
2724 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2725 slave->link == BOND_LINK_UP)
2726 bond_3ad_adapter_speed_duplex_changed(slave);
2730 if (bond_update_speed_duplex(slave) &&
2731 bond_needs_speed_duplex(bond)) {
2732 slave->link = BOND_LINK_DOWN;
2733 if (net_ratelimit())
2734 slave_warn(bond->dev, slave->dev,
2735 "failed to get link speed/duplex\n");
2738 bond_set_slave_link_state(slave, BOND_LINK_UP,
2739 BOND_SLAVE_NOTIFY_NOW);
2740 slave->last_link_up = jiffies;
2742 primary = rtnl_dereference(bond->primary_slave);
2743 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2744 /* prevent it from being the active one */
2745 bond_set_backup_slave(slave);
2746 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2747 /* make it immediately active */
2748 bond_set_active_slave(slave);
2751 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2752 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2753 slave->duplex ? "full" : "half");
2755 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2757 active = rtnl_dereference(bond->curr_active_slave);
2758 if (!active || slave == primary || slave->prio > active->prio)
2763 case BOND_LINK_DOWN:
2764 if (slave->link_failure_count < UINT_MAX)
2765 slave->link_failure_count++;
2767 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2768 BOND_SLAVE_NOTIFY_NOW);
2770 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2771 BOND_MODE(bond) == BOND_MODE_8023AD)
2772 bond_set_slave_inactive_flags(slave,
2773 BOND_SLAVE_NOTIFY_NOW);
2775 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2777 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2779 if (slave == rcu_access_pointer(bond->curr_active_slave))
2785 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2786 slave->link_new_state);
2787 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2795 bond_select_active_slave(bond);
2796 unblock_netpoll_tx();
2799 bond_set_carrier(bond);
2804 * Really a wrapper that splits the mii monitor into two phases: an
2805 * inspection, then (if inspection indicates something needs to be done)
2806 * an acquisition of appropriate locks followed by a commit phase to
2807 * implement whatever link state changes are indicated.
2809 static void bond_mii_monitor(struct work_struct *work)
2811 struct bonding *bond = container_of(work, struct bonding,
2813 bool should_notify_peers = false;
2815 unsigned long delay;
2816 struct slave *slave;
2817 struct list_head *iter;
2819 delay = msecs_to_jiffies(bond->params.miimon);
2821 if (!bond_has_slaves(bond))
2825 should_notify_peers = bond_should_notify_peers(bond);
2826 commit = !!bond_miimon_inspect(bond);
2827 if (bond->send_peer_notif) {
2829 if (rtnl_trylock()) {
2830 bond->send_peer_notif--;
2838 /* Race avoidance with bond_close cancel of workqueue */
2839 if (!rtnl_trylock()) {
2841 should_notify_peers = false;
2845 bond_for_each_slave(bond, slave, iter) {
2846 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2848 bond_miimon_commit(bond);
2850 rtnl_unlock(); /* might sleep, hold no other locks */
2854 if (bond->params.miimon)
2855 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2857 if (should_notify_peers) {
2858 if (!rtnl_trylock())
2860 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2865 static int bond_upper_dev_walk(struct net_device *upper,
2866 struct netdev_nested_priv *priv)
2868 __be32 ip = *(__be32 *)priv->data;
2870 return ip == bond_confirm_addr(upper, 0, ip);
2873 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2875 struct netdev_nested_priv priv = {
2876 .data = (void *)&ip,
2880 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2884 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2891 #define BOND_VLAN_PROTO_NONE cpu_to_be16(0xffff)
2893 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2894 struct sk_buff *skb)
2896 struct net_device *bond_dev = slave->bond->dev;
2897 struct net_device *slave_dev = slave->dev;
2898 struct bond_vlan_tag *outer_tag = tags;
2900 if (!tags || tags->vlan_proto == BOND_VLAN_PROTO_NONE)
2905 /* Go through all the tags backwards and add them to the packet */
2906 while (tags->vlan_proto != BOND_VLAN_PROTO_NONE) {
2907 if (!tags->vlan_id) {
2912 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2913 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2914 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2917 net_err_ratelimited("failed to insert inner VLAN tag\n");
2923 /* Set the outer tag */
2924 if (outer_tag->vlan_id) {
2925 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2926 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2927 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2928 outer_tag->vlan_id);
2934 /* We go to the (large) trouble of VLAN tagging ARP frames because
2935 * switches in VLAN mode (especially if ports are configured as
2936 * "native" to a VLAN) might not pass non-tagged frames.
2938 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2939 __be32 src_ip, struct bond_vlan_tag *tags)
2941 struct net_device *bond_dev = slave->bond->dev;
2942 struct net_device *slave_dev = slave->dev;
2943 struct sk_buff *skb;
2945 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2946 arp_op, &dest_ip, &src_ip);
2948 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2949 NULL, slave_dev->dev_addr, NULL);
2952 net_err_ratelimited("ARP packet allocation failed\n");
2956 if (bond_handle_vlan(slave, tags, skb)) {
2957 slave_update_last_tx(slave);
2964 /* Validate the device path between the @start_dev and the @end_dev.
2965 * The path is valid if the @end_dev is reachable through device
2967 * When the path is validated, collect any vlan information in the
2970 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2971 struct net_device *end_dev,
2974 struct bond_vlan_tag *tags;
2975 struct net_device *upper;
2976 struct list_head *iter;
2978 if (start_dev == end_dev) {
2979 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2981 return ERR_PTR(-ENOMEM);
2982 tags[level].vlan_proto = BOND_VLAN_PROTO_NONE;
2986 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2987 tags = bond_verify_device_path(upper, end_dev, level + 1);
2988 if (IS_ERR_OR_NULL(tags)) {
2993 if (is_vlan_dev(upper)) {
2994 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2995 tags[level].vlan_id = vlan_dev_vlan_id(upper);
3004 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
3007 struct bond_vlan_tag *tags;
3008 __be32 *targets = bond->params.arp_targets, addr;
3011 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
3012 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
3013 __func__, &targets[i]);
3016 /* Find out through which dev should the packet go */
3017 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
3020 /* there's no route to target - try to send arp
3021 * probe to generate any traffic (arp_validate=0)
3023 if (bond->params.arp_validate)
3024 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
3027 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
3032 /* bond device itself */
3033 if (rt->dst.dev == bond->dev)
3037 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
3040 if (!IS_ERR_OR_NULL(tags))
3043 /* Not our device - skip */
3044 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3045 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3051 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3053 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3058 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3062 if (!sip || !bond_has_this_ip(bond, tip)) {
3063 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3064 __func__, &sip, &tip);
3068 i = bond_get_targets_ip(bond->params.arp_targets, sip);
3070 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3074 slave->last_rx = jiffies;
3075 slave->target_last_arp_rx[i] = jiffies;
3078 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3079 struct slave *slave)
3081 struct arphdr *arp = (struct arphdr *)skb->data;
3082 struct slave *curr_active_slave, *curr_arp_slave;
3083 unsigned char *arp_ptr;
3087 alen = arp_hdr_len(bond->dev);
3089 if (alen > skb_headlen(skb)) {
3090 arp = kmalloc(alen, GFP_ATOMIC);
3093 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3097 if (arp->ar_hln != bond->dev->addr_len ||
3098 skb->pkt_type == PACKET_OTHERHOST ||
3099 skb->pkt_type == PACKET_LOOPBACK ||
3100 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3101 arp->ar_pro != htons(ETH_P_IP) ||
3105 arp_ptr = (unsigned char *)(arp + 1);
3106 arp_ptr += bond->dev->addr_len;
3107 memcpy(&sip, arp_ptr, 4);
3108 arp_ptr += 4 + bond->dev->addr_len;
3109 memcpy(&tip, arp_ptr, 4);
3111 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3112 __func__, slave->dev->name, bond_slave_state(slave),
3113 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3116 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3117 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3119 /* We 'trust' the received ARP enough to validate it if:
3121 * (a) the slave receiving the ARP is active (which includes the
3122 * current ARP slave, if any), or
3124 * (b) the receiving slave isn't active, but there is a currently
3125 * active slave and it received valid arp reply(s) after it became
3126 * the currently active slave, or
3128 * (c) there is an ARP slave that sent an ARP during the prior ARP
3129 * interval, and we receive an ARP reply on any slave. We accept
3130 * these because switch FDB update delays may deliver the ARP
3131 * reply to a slave other than the sender of the ARP request.
3133 * Note: for (b), backup slaves are receiving the broadcast ARP
3134 * request, not a reply. This request passes from the sending
3135 * slave through the L2 switch(es) to the receiving slave. Since
3136 * this is checking the request, sip/tip are swapped for
3139 * This is done to avoid endless looping when we can't reach the
3140 * arp_ip_target and fool ourselves with our own arp requests.
3142 if (bond_is_active_slave(slave))
3143 bond_validate_arp(bond, slave, sip, tip);
3144 else if (curr_active_slave &&
3145 time_after(slave_last_rx(bond, curr_active_slave),
3146 curr_active_slave->last_link_up))
3147 bond_validate_arp(bond, slave, tip, sip);
3148 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3149 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3150 bond_validate_arp(bond, slave, sip, tip);
3153 if (arp != (struct arphdr *)skb->data)
3155 return RX_HANDLER_ANOTHER;
3158 #if IS_ENABLED(CONFIG_IPV6)
3159 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3160 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3162 struct net_device *bond_dev = slave->bond->dev;
3163 struct net_device *slave_dev = slave->dev;
3164 struct in6_addr mcaddr;
3165 struct sk_buff *skb;
3167 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3170 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3172 net_err_ratelimited("NS packet allocation failed\n");
3176 addrconf_addr_solict_mult(daddr, &mcaddr);
3177 if (bond_handle_vlan(slave, tags, skb)) {
3178 slave_update_last_tx(slave);
3179 ndisc_send_skb(skb, &mcaddr, saddr);
3183 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3185 struct in6_addr *targets = bond->params.ns_targets;
3186 struct bond_vlan_tag *tags;
3187 struct dst_entry *dst;
3188 struct in6_addr saddr;
3192 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3193 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3194 __func__, &targets[i]);
3197 /* Find out through which dev should the packet go */
3198 memset(&fl6, 0, sizeof(struct flowi6));
3199 fl6.daddr = targets[i];
3200 fl6.flowi6_oif = bond->dev->ifindex;
3202 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3205 /* there's no route to target - try to send arp
3206 * probe to generate any traffic (arp_validate=0)
3208 if (bond->params.arp_validate)
3209 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3212 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3216 /* bond device itself */
3217 if (dst->dev == bond->dev)
3221 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3224 if (!IS_ERR_OR_NULL(tags))
3227 /* Not our device - skip */
3228 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3229 &targets[i], dst->dev ? dst->dev->name : "NULL");
3235 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3236 bond_ns_send(slave, &targets[i], &saddr, tags);
3238 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3245 static int bond_confirm_addr6(struct net_device *dev,
3246 struct netdev_nested_priv *priv)
3248 struct in6_addr *addr = (struct in6_addr *)priv->data;
3250 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3253 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3255 struct netdev_nested_priv priv = {
3260 if (bond_confirm_addr6(bond->dev, &priv))
3264 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3271 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3272 struct in6_addr *saddr, struct in6_addr *daddr)
3277 * 1. Source address is unspecified address.
3278 * 2. Dest address is neither all-nodes multicast address nor
3279 * exist on bond interface.
3281 if (ipv6_addr_any(saddr) ||
3282 (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3283 !bond_has_this_ip6(bond, daddr))) {
3284 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3285 __func__, saddr, daddr);
3289 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3291 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3295 slave->last_rx = jiffies;
3296 slave->target_last_arp_rx[i] = jiffies;
3299 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3300 struct slave *slave)
3302 struct slave *curr_active_slave, *curr_arp_slave;
3303 struct in6_addr *saddr, *daddr;
3306 struct icmp6hdr icmp6;
3307 } *combined, _combined;
3309 if (skb->pkt_type == PACKET_OTHERHOST ||
3310 skb->pkt_type == PACKET_LOOPBACK)
3313 combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3314 if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3315 (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3316 combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3319 saddr = &combined->ip6.saddr;
3320 daddr = &combined->ip6.daddr;
3322 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3323 __func__, slave->dev->name, bond_slave_state(slave),
3324 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3327 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3328 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3330 /* We 'trust' the received ARP enough to validate it if:
3331 * see bond_arp_rcv().
3333 if (bond_is_active_slave(slave))
3334 bond_validate_na(bond, slave, saddr, daddr);
3335 else if (curr_active_slave &&
3336 time_after(slave_last_rx(bond, curr_active_slave),
3337 curr_active_slave->last_link_up))
3338 bond_validate_na(bond, slave, daddr, saddr);
3339 else if (curr_arp_slave &&
3340 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3341 bond_validate_na(bond, slave, saddr, daddr);
3344 return RX_HANDLER_ANOTHER;
3348 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3349 struct slave *slave)
3351 #if IS_ENABLED(CONFIG_IPV6)
3352 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3354 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3356 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3357 __func__, skb->dev->name);
3359 /* Use arp validate logic for both ARP and NS */
3360 if (!slave_do_arp_validate(bond, slave)) {
3361 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3362 #if IS_ENABLED(CONFIG_IPV6)
3363 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3365 !slave_do_arp_validate_only(bond))
3366 slave->last_rx = jiffies;
3367 return RX_HANDLER_ANOTHER;
3368 } else if (is_arp) {
3369 return bond_arp_rcv(skb, bond, slave);
3370 #if IS_ENABLED(CONFIG_IPV6)
3371 } else if (is_ipv6) {
3372 return bond_na_rcv(skb, bond, slave);
3375 return RX_HANDLER_ANOTHER;
3379 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3381 bond_arp_send_all(bond, slave);
3382 #if IS_ENABLED(CONFIG_IPV6)
3383 bond_ns_send_all(bond, slave);
3387 /* function to verify if we're in the arp_interval timeslice, returns true if
3388 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3389 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3391 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3394 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3396 return time_in_range(jiffies,
3397 last_act - delta_in_ticks,
3398 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3401 /* This function is called regularly to monitor each slave's link
3402 * ensuring that traffic is being sent and received when arp monitoring
3403 * is used in load-balancing mode. if the adapter has been dormant, then an
3404 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3405 * arp monitoring in active backup mode.
3407 static void bond_loadbalance_arp_mon(struct bonding *bond)
3409 struct slave *slave, *oldcurrent;
3410 struct list_head *iter;
3411 int do_failover = 0, slave_state_changed = 0;
3413 if (!bond_has_slaves(bond))
3418 oldcurrent = rcu_dereference(bond->curr_active_slave);
3419 /* see if any of the previous devices are up now (i.e. they have
3420 * xmt and rcv traffic). the curr_active_slave does not come into
3421 * the picture unless it is null. also, slave->last_link_up is not
3422 * needed here because we send an arp on each slave and give a slave
3423 * as long as it needs to get the tx/rx within the delta.
3424 * TODO: what about up/down delay in arp mode? it wasn't here before
3427 bond_for_each_slave_rcu(bond, slave, iter) {
3428 unsigned long last_tx = slave_last_tx(slave);
3430 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3432 if (slave->link != BOND_LINK_UP) {
3433 if (bond_time_in_interval(bond, last_tx, 1) &&
3434 bond_time_in_interval(bond, slave->last_rx, 1)) {
3436 bond_propose_link_state(slave, BOND_LINK_UP);
3437 slave_state_changed = 1;
3439 /* primary_slave has no meaning in round-robin
3440 * mode. the window of a slave being up and
3441 * curr_active_slave being null after enslaving
3445 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3448 slave_info(bond->dev, slave->dev, "interface is now up\n");
3452 /* slave->link == BOND_LINK_UP */
3454 /* not all switches will respond to an arp request
3455 * when the source ip is 0, so don't take the link down
3456 * if we don't know our ip yet
3458 if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3459 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3461 bond_propose_link_state(slave, BOND_LINK_DOWN);
3462 slave_state_changed = 1;
3464 if (slave->link_failure_count < UINT_MAX)
3465 slave->link_failure_count++;
3467 slave_info(bond->dev, slave->dev, "interface is now down\n");
3469 if (slave == oldcurrent)
3474 /* note: if switch is in round-robin mode, all links
3475 * must tx arp to ensure all links rx an arp - otherwise
3476 * links may oscillate or not come up at all; if switch is
3477 * in something like xor mode, there is nothing we can
3478 * do - all replies will be rx'ed on same link causing slaves
3479 * to be unstable during low/no traffic periods
3481 if (bond_slave_is_up(slave))
3482 bond_send_validate(bond, slave);
3487 if (do_failover || slave_state_changed) {
3488 if (!rtnl_trylock())
3491 bond_for_each_slave(bond, slave, iter) {
3492 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3493 slave->link = slave->link_new_state;
3496 if (slave_state_changed) {
3497 bond_slave_state_change(bond);
3498 if (BOND_MODE(bond) == BOND_MODE_XOR)
3499 bond_update_slave_arr(bond, NULL);
3503 bond_select_active_slave(bond);
3504 unblock_netpoll_tx();
3510 if (bond->params.arp_interval)
3511 queue_delayed_work(bond->wq, &bond->arp_work,
3512 msecs_to_jiffies(bond->params.arp_interval));
3515 /* Called to inspect slaves for active-backup mode ARP monitor link state
3516 * changes. Sets proposed link state in slaves to specify what action
3517 * should take place for the slave. Returns 0 if no changes are found, >0
3518 * if changes to link states must be committed.
3520 * Called with rcu_read_lock held.
3522 static int bond_ab_arp_inspect(struct bonding *bond)
3524 unsigned long last_tx, last_rx;
3525 struct list_head *iter;
3526 struct slave *slave;
3529 bond_for_each_slave_rcu(bond, slave, iter) {
3530 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3531 last_rx = slave_last_rx(bond, slave);
3533 if (slave->link != BOND_LINK_UP) {
3534 if (bond_time_in_interval(bond, last_rx, 1)) {
3535 bond_propose_link_state(slave, BOND_LINK_UP);
3537 } else if (slave->link == BOND_LINK_BACK) {
3538 bond_propose_link_state(slave, BOND_LINK_FAIL);
3544 /* Give slaves 2*delta after being enslaved or made
3545 * active. This avoids bouncing, as the last receive
3546 * times need a full ARP monitor cycle to be updated.
3548 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3551 /* Backup slave is down if:
3552 * - No current_arp_slave AND
3553 * - more than (missed_max+1)*delta since last receive AND
3554 * - the bond has an IP address
3556 * Note: a non-null current_arp_slave indicates
3557 * the curr_active_slave went down and we are
3558 * searching for a new one; under this condition
3559 * we only take the curr_active_slave down - this
3560 * gives each slave a chance to tx/rx traffic
3561 * before being taken out
3563 if (!bond_is_active_slave(slave) &&
3564 !rcu_access_pointer(bond->current_arp_slave) &&
3565 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3566 bond_propose_link_state(slave, BOND_LINK_DOWN);
3570 /* Active slave is down if:
3571 * - more than missed_max*delta since transmitting OR
3572 * - (more than missed_max*delta since receive AND
3573 * the bond has an IP address)
3575 last_tx = slave_last_tx(slave);
3576 if (bond_is_active_slave(slave) &&
3577 (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3578 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3579 bond_propose_link_state(slave, BOND_LINK_DOWN);
3587 /* Called to commit link state changes noted by inspection step of
3588 * active-backup mode ARP monitor.
3590 * Called with RTNL hold.
3592 static void bond_ab_arp_commit(struct bonding *bond)
3594 bool do_failover = false;
3595 struct list_head *iter;
3596 unsigned long last_tx;
3597 struct slave *slave;
3599 bond_for_each_slave(bond, slave, iter) {
3600 switch (slave->link_new_state) {
3601 case BOND_LINK_NOCHANGE:
3605 last_tx = slave_last_tx(slave);
3606 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3607 (!rtnl_dereference(bond->curr_active_slave) &&
3608 bond_time_in_interval(bond, last_tx, 1))) {
3609 struct slave *current_arp_slave;
3611 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3612 bond_set_slave_link_state(slave, BOND_LINK_UP,
3613 BOND_SLAVE_NOTIFY_NOW);
3614 if (current_arp_slave) {
3615 bond_set_slave_inactive_flags(
3617 BOND_SLAVE_NOTIFY_NOW);
3618 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3621 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3623 if (!rtnl_dereference(bond->curr_active_slave) ||
3624 slave == rtnl_dereference(bond->primary_slave) ||
3625 slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3632 case BOND_LINK_DOWN:
3633 if (slave->link_failure_count < UINT_MAX)
3634 slave->link_failure_count++;
3636 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3637 BOND_SLAVE_NOTIFY_NOW);
3638 bond_set_slave_inactive_flags(slave,
3639 BOND_SLAVE_NOTIFY_NOW);
3641 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3643 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3644 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3650 case BOND_LINK_FAIL:
3651 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3652 BOND_SLAVE_NOTIFY_NOW);
3653 bond_set_slave_inactive_flags(slave,
3654 BOND_SLAVE_NOTIFY_NOW);
3656 /* A slave has just been enslaved and has become
3657 * the current active slave.
3659 if (rtnl_dereference(bond->curr_active_slave))
3660 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3664 slave_err(bond->dev, slave->dev,
3665 "impossible: link_new_state %d on slave\n",
3666 slave->link_new_state);
3673 bond_select_active_slave(bond);
3674 unblock_netpoll_tx();
3677 bond_set_carrier(bond);
3680 /* Send ARP probes for active-backup mode ARP monitor.
3682 * Called with rcu_read_lock held.
3684 static bool bond_ab_arp_probe(struct bonding *bond)
3686 struct slave *slave, *before = NULL, *new_slave = NULL,
3687 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3688 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3689 struct list_head *iter;
3691 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3693 if (curr_arp_slave && curr_active_slave)
3694 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3695 curr_arp_slave->dev->name,
3696 curr_active_slave->dev->name);
3698 if (curr_active_slave) {
3699 bond_send_validate(bond, curr_active_slave);
3700 return should_notify_rtnl;
3703 /* if we don't have a curr_active_slave, search for the next available
3704 * backup slave from the current_arp_slave and make it the candidate
3705 * for becoming the curr_active_slave
3708 if (!curr_arp_slave) {
3709 curr_arp_slave = bond_first_slave_rcu(bond);
3710 if (!curr_arp_slave)
3711 return should_notify_rtnl;
3714 bond_for_each_slave_rcu(bond, slave, iter) {
3715 if (!found && !before && bond_slave_is_up(slave))
3718 if (found && !new_slave && bond_slave_is_up(slave))
3720 /* if the link state is up at this point, we
3721 * mark it down - this can happen if we have
3722 * simultaneous link failures and
3723 * reselect_active_interface doesn't make this
3724 * one the current slave so it is still marked
3725 * up when it is actually down
3727 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3728 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3729 BOND_SLAVE_NOTIFY_LATER);
3730 if (slave->link_failure_count < UINT_MAX)
3731 slave->link_failure_count++;
3733 bond_set_slave_inactive_flags(slave,
3734 BOND_SLAVE_NOTIFY_LATER);
3736 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3738 if (slave == curr_arp_slave)
3742 if (!new_slave && before)
3748 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3749 BOND_SLAVE_NOTIFY_LATER);
3750 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3751 bond_send_validate(bond, new_slave);
3752 new_slave->last_link_up = jiffies;
3753 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3756 bond_for_each_slave_rcu(bond, slave, iter) {
3757 if (slave->should_notify || slave->should_notify_link) {
3758 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3762 return should_notify_rtnl;
3765 static void bond_activebackup_arp_mon(struct bonding *bond)
3767 bool should_notify_peers = false;
3768 bool should_notify_rtnl = false;
3771 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3773 if (!bond_has_slaves(bond))
3778 should_notify_peers = bond_should_notify_peers(bond);
3780 if (bond_ab_arp_inspect(bond)) {
3783 /* Race avoidance with bond_close flush of workqueue */
3784 if (!rtnl_trylock()) {
3786 should_notify_peers = false;
3790 bond_ab_arp_commit(bond);
3796 should_notify_rtnl = bond_ab_arp_probe(bond);
3800 if (bond->params.arp_interval)
3801 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3803 if (should_notify_peers || should_notify_rtnl) {
3804 if (!rtnl_trylock())
3807 if (should_notify_peers) {
3808 bond->send_peer_notif--;
3809 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3812 if (should_notify_rtnl) {
3813 bond_slave_state_notify(bond);
3814 bond_slave_link_notify(bond);
3821 static void bond_arp_monitor(struct work_struct *work)
3823 struct bonding *bond = container_of(work, struct bonding,
3826 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3827 bond_activebackup_arp_mon(bond);
3829 bond_loadbalance_arp_mon(bond);
3832 /*-------------------------- netdev event handling --------------------------*/
3834 /* Change device name */
3835 static int bond_event_changename(struct bonding *bond)
3837 bond_remove_proc_entry(bond);
3838 bond_create_proc_entry(bond);
3840 bond_debug_reregister(bond);
3845 static int bond_master_netdev_event(unsigned long event,
3846 struct net_device *bond_dev)
3848 struct bonding *event_bond = netdev_priv(bond_dev);
3850 netdev_dbg(bond_dev, "%s called\n", __func__);
3853 case NETDEV_CHANGENAME:
3854 return bond_event_changename(event_bond);
3855 case NETDEV_UNREGISTER:
3856 bond_remove_proc_entry(event_bond);
3857 #ifdef CONFIG_XFRM_OFFLOAD
3858 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3859 #endif /* CONFIG_XFRM_OFFLOAD */
3861 case NETDEV_REGISTER:
3862 bond_create_proc_entry(event_bond);
3871 static int bond_slave_netdev_event(unsigned long event,
3872 struct net_device *slave_dev)
3874 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3875 struct bonding *bond;
3876 struct net_device *bond_dev;
3878 /* A netdev event can be generated while enslaving a device
3879 * before netdev_rx_handler_register is called in which case
3880 * slave will be NULL
3883 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3887 bond_dev = slave->bond->dev;
3889 primary = rtnl_dereference(bond->primary_slave);
3891 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3894 case NETDEV_UNREGISTER:
3895 if (bond_dev->type != ARPHRD_ETHER)
3896 bond_release_and_destroy(bond_dev, slave_dev);
3898 __bond_release_one(bond_dev, slave_dev, false, true);
3902 /* For 802.3ad mode only:
3903 * Getting invalid Speed/Duplex values here will put slave
3904 * in weird state. Mark it as link-fail if the link was
3905 * previously up or link-down if it hasn't yet come up, and
3906 * let link-monitoring (miimon) set it right when correct
3907 * speeds/duplex are available.
3909 if (bond_update_speed_duplex(slave) &&
3910 BOND_MODE(bond) == BOND_MODE_8023AD) {
3911 if (slave->last_link_up)
3912 slave->link = BOND_LINK_FAIL;
3914 slave->link = BOND_LINK_DOWN;
3917 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3918 bond_3ad_adapter_speed_duplex_changed(slave);
3921 /* Refresh slave-array if applicable!
3922 * If the setup does not use miimon or arpmon (mode-specific!),
3923 * then these events will not cause the slave-array to be
3924 * refreshed. This will cause xmit to use a slave that is not
3925 * usable. Avoid such situation by refeshing the array at these
3926 * events. If these (miimon/arpmon) parameters are configured
3927 * then array gets refreshed twice and that should be fine!
3929 if (bond_mode_can_use_xmit_hash(bond))
3930 bond_update_slave_arr(bond, NULL);
3932 case NETDEV_CHANGEMTU:
3933 /* TODO: Should slaves be allowed to
3934 * independently alter their MTU? For
3935 * an active-backup bond, slaves need
3936 * not be the same type of device, so
3937 * MTUs may vary. For other modes,
3938 * slaves arguably should have the
3939 * same MTUs. To do this, we'd need to
3940 * take over the slave's change_mtu
3941 * function for the duration of their
3945 case NETDEV_CHANGENAME:
3946 /* we don't care if we don't have primary set */
3947 if (!bond_uses_primary(bond) ||
3948 !bond->params.primary[0])
3951 if (slave == primary) {
3952 /* slave's name changed - he's no longer primary */
3953 RCU_INIT_POINTER(bond->primary_slave, NULL);
3954 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3955 /* we have a new primary slave */
3956 rcu_assign_pointer(bond->primary_slave, slave);
3957 } else { /* we didn't change primary - exit */
3961 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3962 primary ? slave_dev->name : "none");
3965 bond_select_active_slave(bond);
3966 unblock_netpoll_tx();
3968 case NETDEV_FEAT_CHANGE:
3969 if (!bond->notifier_ctx) {
3970 bond->notifier_ctx = true;
3971 bond_compute_features(bond);
3972 bond->notifier_ctx = false;
3975 case NETDEV_RESEND_IGMP:
3976 /* Propagate to master device */
3977 call_netdevice_notifiers(event, slave->bond->dev);
3979 case NETDEV_XDP_FEAT_CHANGE:
3980 bond_xdp_set_features(bond_dev);
3989 /* bond_netdev_event: handle netdev notifier chain events.
3991 * This function receives events for the netdev chain. The caller (an
3992 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3993 * locks for us to safely manipulate the slave devices (RTNL lock,
3996 static int bond_netdev_event(struct notifier_block *this,
3997 unsigned long event, void *ptr)
3999 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
4001 netdev_dbg(event_dev, "%s received %s\n",
4002 __func__, netdev_cmd_to_name(event));
4004 if (!(event_dev->priv_flags & IFF_BONDING))
4007 if (event_dev->flags & IFF_MASTER) {
4010 ret = bond_master_netdev_event(event, event_dev);
4011 if (ret != NOTIFY_DONE)
4015 if (event_dev->flags & IFF_SLAVE)
4016 return bond_slave_netdev_event(event, event_dev);
4021 static struct notifier_block bond_netdev_notifier = {
4022 .notifier_call = bond_netdev_event,
4025 /*---------------------------- Hashing Policies -----------------------------*/
4027 /* Helper to access data in a packet, with or without a backing skb.
4028 * If skb is given the data is linearized if necessary via pskb_may_pull.
4030 static inline const void *bond_pull_data(struct sk_buff *skb,
4031 const void *data, int hlen, int n)
4033 if (likely(n <= hlen))
4035 else if (skb && likely(pskb_may_pull(skb, n)))
4041 /* L2 hash helper */
4042 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4046 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4050 ep = (struct ethhdr *)(data + mhoff);
4051 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4054 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4055 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4057 const struct ipv6hdr *iph6;
4058 const struct iphdr *iph;
4060 if (l2_proto == htons(ETH_P_IP)) {
4061 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4065 iph = (const struct iphdr *)(data + *nhoff);
4066 iph_to_flow_copy_v4addrs(fk, iph);
4067 *nhoff += iph->ihl << 2;
4068 if (!ip_is_fragment(iph))
4069 *ip_proto = iph->protocol;
4070 } else if (l2_proto == htons(ETH_P_IPV6)) {
4071 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4075 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4076 iph_to_flow_copy_v6addrs(fk, iph6);
4077 *nhoff += sizeof(*iph6);
4078 *ip_proto = iph6->nexthdr;
4083 if (l34 && *ip_proto >= 0)
4084 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4089 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4091 u32 srcmac_vendor = 0, srcmac_dev = 0;
4092 struct ethhdr *mac_hdr;
4096 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4099 mac_hdr = (struct ethhdr *)(data + mhoff);
4101 for (i = 0; i < 3; i++)
4102 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4104 for (i = 3; i < ETH_ALEN; i++)
4105 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4107 if (skb && skb_vlan_tag_present(skb))
4108 vlan = skb_vlan_tag_get(skb);
4110 return vlan ^ srcmac_vendor ^ srcmac_dev;
4113 /* Extract the appropriate headers based on bond's xmit policy */
4114 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4115 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4117 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4120 switch (bond->params.xmit_policy) {
4121 case BOND_XMIT_POLICY_ENCAP23:
4122 case BOND_XMIT_POLICY_ENCAP34:
4123 memset(fk, 0, sizeof(*fk));
4124 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4125 fk, data, l2_proto, nhoff, hlen, 0);
4130 fk->ports.ports = 0;
4131 memset(&fk->icmp, 0, sizeof(fk->icmp));
4132 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4135 /* ICMP error packets contains at least 8 bytes of the header
4136 * of the packet which generated the error. Use this information
4137 * to correlate ICMP error packets within the same flow which
4138 * generated the error.
4140 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4141 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4142 if (ip_proto == IPPROTO_ICMP) {
4143 if (!icmp_is_err(fk->icmp.type))
4146 nhoff += sizeof(struct icmphdr);
4147 } else if (ip_proto == IPPROTO_ICMPV6) {
4148 if (!icmpv6_is_err(fk->icmp.type))
4151 nhoff += sizeof(struct icmp6hdr);
4153 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4159 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4161 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4162 (__force u32)flow_get_u32_src(flow);
4163 hash ^= (hash >> 16);
4164 hash ^= (hash >> 8);
4166 /* discard lowest hash bit to deal with the common even ports pattern */
4167 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4168 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4174 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4175 * the data as required, but this function can be used without it if the data is
4176 * known to be linear (e.g. with xdp_buff).
4178 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4179 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4181 struct flow_keys flow;
4184 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4185 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4187 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4188 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4189 return bond_eth_hash(skb, data, mhoff, hlen);
4191 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4192 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4193 hash = bond_eth_hash(skb, data, mhoff, hlen);
4196 memcpy(&hash, &flow.icmp, sizeof(hash));
4198 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4201 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4205 * bond_xmit_hash - generate a hash value based on the xmit policy
4206 * @bond: bonding device
4207 * @skb: buffer to use for headers
4209 * This function will extract the necessary headers from the skb buffer and use
4210 * them to generate a hash based on the xmit_policy set in the bonding device
4212 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4214 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4218 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4219 0, skb_network_offset(skb),
4224 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4225 * @bond: bonding device
4226 * @xdp: buffer to use for headers
4228 * The XDP variant of bond_xmit_hash.
4230 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4234 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4237 eth = (struct ethhdr *)xdp->data;
4239 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4240 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4243 /*-------------------------- Device entry points ----------------------------*/
4245 void bond_work_init_all(struct bonding *bond)
4247 INIT_DELAYED_WORK(&bond->mcast_work,
4248 bond_resend_igmp_join_requests_delayed);
4249 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4250 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4251 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4252 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4253 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4256 static void bond_work_cancel_all(struct bonding *bond)
4258 cancel_delayed_work_sync(&bond->mii_work);
4259 cancel_delayed_work_sync(&bond->arp_work);
4260 cancel_delayed_work_sync(&bond->alb_work);
4261 cancel_delayed_work_sync(&bond->ad_work);
4262 cancel_delayed_work_sync(&bond->mcast_work);
4263 cancel_delayed_work_sync(&bond->slave_arr_work);
4266 static int bond_open(struct net_device *bond_dev)
4268 struct bonding *bond = netdev_priv(bond_dev);
4269 struct list_head *iter;
4270 struct slave *slave;
4272 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4273 bond->rr_tx_counter = alloc_percpu(u32);
4274 if (!bond->rr_tx_counter)
4278 /* reset slave->backup and slave->inactive */
4279 if (bond_has_slaves(bond)) {
4280 bond_for_each_slave(bond, slave, iter) {
4281 if (bond_uses_primary(bond) &&
4282 slave != rcu_access_pointer(bond->curr_active_slave)) {
4283 bond_set_slave_inactive_flags(slave,
4284 BOND_SLAVE_NOTIFY_NOW);
4285 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4286 bond_set_slave_active_flags(slave,
4287 BOND_SLAVE_NOTIFY_NOW);
4292 if (bond_is_lb(bond)) {
4293 /* bond_alb_initialize must be called before the timer
4296 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4298 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4299 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4302 if (bond->params.miimon) /* link check interval, in milliseconds. */
4303 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4305 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4306 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4307 bond->recv_probe = bond_rcv_validate;
4310 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4311 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4312 /* register to receive LACPDUs */
4313 bond->recv_probe = bond_3ad_lacpdu_recv;
4314 bond_3ad_initiate_agg_selection(bond, 1);
4316 bond_for_each_slave(bond, slave, iter)
4317 dev_mc_add(slave->dev, lacpdu_mcast_addr);
4320 if (bond_mode_can_use_xmit_hash(bond))
4321 bond_update_slave_arr(bond, NULL);
4326 static int bond_close(struct net_device *bond_dev)
4328 struct bonding *bond = netdev_priv(bond_dev);
4329 struct slave *slave;
4331 bond_work_cancel_all(bond);
4332 bond->send_peer_notif = 0;
4333 if (bond_is_lb(bond))
4334 bond_alb_deinitialize(bond);
4335 bond->recv_probe = NULL;
4337 if (bond_uses_primary(bond)) {
4339 slave = rcu_dereference(bond->curr_active_slave);
4341 bond_hw_addr_flush(bond_dev, slave->dev);
4344 struct list_head *iter;
4346 bond_for_each_slave(bond, slave, iter)
4347 bond_hw_addr_flush(bond_dev, slave->dev);
4353 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4354 * that some drivers can provide 32bit values only.
4356 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4357 const struct rtnl_link_stats64 *_new,
4358 const struct rtnl_link_stats64 *_old)
4360 const u64 *new = (const u64 *)_new;
4361 const u64 *old = (const u64 *)_old;
4362 u64 *res = (u64 *)_res;
4365 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4368 s64 delta = nv - ov;
4370 /* detects if this particular field is 32bit only */
4371 if (((nv | ov) >> 32) == 0)
4372 delta = (s64)(s32)((u32)nv - (u32)ov);
4374 /* filter anomalies, some drivers reset their stats
4375 * at down/up events.
4382 #ifdef CONFIG_LOCKDEP
4383 static int bond_get_lowest_level_rcu(struct net_device *dev)
4385 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4386 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4387 int cur = 0, max = 0;
4390 iter = &dev->adj_list.lower;
4395 ldev = netdev_next_lower_dev_rcu(now, &iter);
4400 niter = &ldev->adj_list.lower;
4401 dev_stack[cur] = now;
4402 iter_stack[cur++] = iter;
4411 next = dev_stack[--cur];
4412 niter = iter_stack[cur];
4423 static void bond_get_stats(struct net_device *bond_dev,
4424 struct rtnl_link_stats64 *stats)
4426 struct bonding *bond = netdev_priv(bond_dev);
4427 struct rtnl_link_stats64 temp;
4428 struct list_head *iter;
4429 struct slave *slave;
4434 #ifdef CONFIG_LOCKDEP
4435 nest_level = bond_get_lowest_level_rcu(bond_dev);
4438 spin_lock_nested(&bond->stats_lock, nest_level);
4439 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4441 bond_for_each_slave_rcu(bond, slave, iter) {
4442 const struct rtnl_link_stats64 *new =
4443 dev_get_stats(slave->dev, &temp);
4445 bond_fold_stats(stats, new, &slave->slave_stats);
4447 /* save off the slave stats for the next run */
4448 memcpy(&slave->slave_stats, new, sizeof(*new));
4451 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4452 spin_unlock(&bond->stats_lock);
4456 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4458 struct bonding *bond = netdev_priv(bond_dev);
4459 struct mii_ioctl_data *mii = NULL;
4461 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4472 /* We do this again just in case we were called by SIOCGMIIREG
4473 * instead of SIOCGMIIPHY.
4479 if (mii->reg_num == 1) {
4481 if (netif_carrier_ok(bond->dev))
4482 mii->val_out = BMSR_LSTATUS;
4493 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4495 struct bonding *bond = netdev_priv(bond_dev);
4496 struct net_device *slave_dev = NULL;
4497 struct ifbond k_binfo;
4498 struct ifbond __user *u_binfo = NULL;
4499 struct ifslave k_sinfo;
4500 struct ifslave __user *u_sinfo = NULL;
4501 struct bond_opt_value newval;
4505 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4508 case SIOCBONDINFOQUERY:
4509 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4511 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4514 bond_info_query(bond_dev, &k_binfo);
4515 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4519 case SIOCBONDSLAVEINFOQUERY:
4520 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4522 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4525 res = bond_slave_info_query(bond_dev, &k_sinfo);
4527 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4535 net = dev_net(bond_dev);
4537 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4540 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4542 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4548 case SIOCBONDENSLAVE:
4549 res = bond_enslave(bond_dev, slave_dev, NULL);
4551 case SIOCBONDRELEASE:
4552 res = bond_release(bond_dev, slave_dev);
4554 case SIOCBONDSETHWADDR:
4555 res = bond_set_dev_addr(bond_dev, slave_dev);
4557 case SIOCBONDCHANGEACTIVE:
4558 bond_opt_initstr(&newval, slave_dev->name);
4559 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4569 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4570 void __user *data, int cmd)
4572 struct ifreq ifrdata = { .ifr_data = data };
4575 case BOND_INFO_QUERY_OLD:
4576 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4577 case BOND_SLAVE_INFO_QUERY_OLD:
4578 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4579 case BOND_ENSLAVE_OLD:
4580 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4581 case BOND_RELEASE_OLD:
4582 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4583 case BOND_SETHWADDR_OLD:
4584 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4585 case BOND_CHANGE_ACTIVE_OLD:
4586 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4592 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4594 struct bonding *bond = netdev_priv(bond_dev);
4596 if (change & IFF_PROMISC)
4597 bond_set_promiscuity(bond,
4598 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4600 if (change & IFF_ALLMULTI)
4601 bond_set_allmulti(bond,
4602 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4605 static void bond_set_rx_mode(struct net_device *bond_dev)
4607 struct bonding *bond = netdev_priv(bond_dev);
4608 struct list_head *iter;
4609 struct slave *slave;
4612 if (bond_uses_primary(bond)) {
4613 slave = rcu_dereference(bond->curr_active_slave);
4615 dev_uc_sync(slave->dev, bond_dev);
4616 dev_mc_sync(slave->dev, bond_dev);
4619 bond_for_each_slave_rcu(bond, slave, iter) {
4620 dev_uc_sync_multiple(slave->dev, bond_dev);
4621 dev_mc_sync_multiple(slave->dev, bond_dev);
4627 static int bond_neigh_init(struct neighbour *n)
4629 struct bonding *bond = netdev_priv(n->dev);
4630 const struct net_device_ops *slave_ops;
4631 struct neigh_parms parms;
4632 struct slave *slave;
4636 slave = bond_first_slave_rcu(bond);
4639 slave_ops = slave->dev->netdev_ops;
4640 if (!slave_ops->ndo_neigh_setup)
4643 /* TODO: find another way [1] to implement this.
4644 * Passing a zeroed structure is fragile,
4645 * but at least we do not pass garbage.
4647 * [1] One way would be that ndo_neigh_setup() never touch
4648 * struct neigh_parms, but propagate the new neigh_setup()
4649 * back to ___neigh_create() / neigh_parms_alloc()
4651 memset(&parms, 0, sizeof(parms));
4652 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4657 if (parms.neigh_setup)
4658 ret = parms.neigh_setup(n);
4664 /* The bonding ndo_neigh_setup is called at init time beofre any
4665 * slave exists. So we must declare proxy setup function which will
4666 * be used at run time to resolve the actual slave neigh param setup.
4668 * It's also called by master devices (such as vlans) to setup their
4669 * underlying devices. In that case - do nothing, we're already set up from
4672 static int bond_neigh_setup(struct net_device *dev,
4673 struct neigh_parms *parms)
4675 /* modify only our neigh_parms */
4676 if (parms->dev == dev)
4677 parms->neigh_setup = bond_neigh_init;
4682 /* Change the MTU of all of a master's slaves to match the master */
4683 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4685 struct bonding *bond = netdev_priv(bond_dev);
4686 struct slave *slave, *rollback_slave;
4687 struct list_head *iter;
4690 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4692 bond_for_each_slave(bond, slave, iter) {
4693 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4694 slave, slave->dev->netdev_ops->ndo_change_mtu);
4696 res = dev_set_mtu(slave->dev, new_mtu);
4699 /* If we failed to set the slave's mtu to the new value
4700 * we must abort the operation even in ACTIVE_BACKUP
4701 * mode, because if we allow the backup slaves to have
4702 * different mtu values than the active slave we'll
4703 * need to change their mtu when doing a failover. That
4704 * means changing their mtu from timer context, which
4705 * is probably not a good idea.
4707 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4713 bond_dev->mtu = new_mtu;
4718 /* unwind from head to the slave that failed */
4719 bond_for_each_slave(bond, rollback_slave, iter) {
4722 if (rollback_slave == slave)
4725 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4727 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4734 /* Change HW address
4736 * Note that many devices must be down to change the HW address, and
4737 * downing the master releases all slaves. We can make bonds full of
4738 * bonding devices to test this, however.
4740 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4742 struct bonding *bond = netdev_priv(bond_dev);
4743 struct slave *slave, *rollback_slave;
4744 struct sockaddr_storage *ss = addr, tmp_ss;
4745 struct list_head *iter;
4748 if (BOND_MODE(bond) == BOND_MODE_ALB)
4749 return bond_alb_set_mac_address(bond_dev, addr);
4752 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4754 /* If fail_over_mac is enabled, do nothing and return success.
4755 * Returning an error causes ifenslave to fail.
4757 if (bond->params.fail_over_mac &&
4758 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4761 if (!is_valid_ether_addr(ss->__data))
4762 return -EADDRNOTAVAIL;
4764 bond_for_each_slave(bond, slave, iter) {
4765 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4767 res = dev_set_mac_address(slave->dev, addr, NULL);
4769 /* TODO: consider downing the slave
4771 * User should expect communications
4772 * breakage anyway until ARP finish
4775 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4782 dev_addr_set(bond_dev, ss->__data);
4786 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4787 tmp_ss.ss_family = bond_dev->type;
4789 /* unwind from head to the slave that failed */
4790 bond_for_each_slave(bond, rollback_slave, iter) {
4793 if (rollback_slave == slave)
4796 tmp_res = dev_set_mac_address(rollback_slave->dev,
4797 (struct sockaddr *)&tmp_ss, NULL);
4799 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4808 * bond_get_slave_by_id - get xmit slave with slave_id
4809 * @bond: bonding device that is transmitting
4810 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4812 * This function tries to get slave with slave_id but in case
4813 * it fails, it tries to find the first available slave for transmission.
4815 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4818 struct list_head *iter;
4819 struct slave *slave;
4822 /* Here we start from the slave with slave_id */
4823 bond_for_each_slave_rcu(bond, slave, iter) {
4825 if (bond_slave_can_tx(slave))
4830 /* Here we start from the first slave up to slave_id */
4832 bond_for_each_slave_rcu(bond, slave, iter) {
4835 if (bond_slave_can_tx(slave))
4838 /* no slave that can tx has been found */
4843 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4844 * @bond: bonding device to use
4846 * Based on the value of the bonding device's packets_per_slave parameter
4847 * this function generates a slave id, which is usually used as the next
4848 * slave to transmit through.
4850 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4853 struct reciprocal_value reciprocal_packets_per_slave;
4854 int packets_per_slave = bond->params.packets_per_slave;
4856 switch (packets_per_slave) {
4858 slave_id = get_random_u32();
4861 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4864 reciprocal_packets_per_slave =
4865 bond->params.reciprocal_packets_per_slave;
4866 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4867 slave_id = reciprocal_divide(slave_id,
4868 reciprocal_packets_per_slave);
4875 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4876 struct sk_buff *skb)
4878 struct slave *slave;
4882 /* Start with the curr_active_slave that joined the bond as the
4883 * default for sending IGMP traffic. For failover purposes one
4884 * needs to maintain some consistency for the interface that will
4885 * send the join/membership reports. The curr_active_slave found
4886 * will send all of this type of traffic.
4888 if (skb->protocol == htons(ETH_P_IP)) {
4889 int noff = skb_network_offset(skb);
4892 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4896 if (iph->protocol == IPPROTO_IGMP) {
4897 slave = rcu_dereference(bond->curr_active_slave);
4900 return bond_get_slave_by_id(bond, 0);
4905 slave_cnt = READ_ONCE(bond->slave_cnt);
4906 if (likely(slave_cnt)) {
4907 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4908 return bond_get_slave_by_id(bond, slave_id);
4913 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4914 struct xdp_buff *xdp)
4916 struct slave *slave;
4919 const struct ethhdr *eth;
4920 void *data = xdp->data;
4922 if (data + sizeof(struct ethhdr) > xdp->data_end)
4925 eth = (struct ethhdr *)data;
4926 data += sizeof(struct ethhdr);
4928 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4929 if (eth->h_proto == htons(ETH_P_IP)) {
4930 const struct iphdr *iph;
4932 if (data + sizeof(struct iphdr) > xdp->data_end)
4935 iph = (struct iphdr *)data;
4937 if (iph->protocol == IPPROTO_IGMP) {
4938 slave = rcu_dereference(bond->curr_active_slave);
4941 return bond_get_slave_by_id(bond, 0);
4946 slave_cnt = READ_ONCE(bond->slave_cnt);
4947 if (likely(slave_cnt)) {
4948 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4949 return bond_get_slave_by_id(bond, slave_id);
4954 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4955 struct net_device *bond_dev)
4957 struct bonding *bond = netdev_priv(bond_dev);
4958 struct slave *slave;
4960 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4962 return bond_dev_queue_xmit(bond, skb, slave->dev);
4964 return bond_tx_drop(bond_dev, skb);
4967 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4969 return rcu_dereference(bond->curr_active_slave);
4972 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4973 * the bond has a usable interface.
4975 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4976 struct net_device *bond_dev)
4978 struct bonding *bond = netdev_priv(bond_dev);
4979 struct slave *slave;
4981 slave = bond_xmit_activebackup_slave_get(bond);
4983 return bond_dev_queue_xmit(bond, skb, slave->dev);
4985 return bond_tx_drop(bond_dev, skb);
4988 /* Use this to update slave_array when (a) it's not appropriate to update
4989 * slave_array right away (note that update_slave_array() may sleep)
4990 * and / or (b) RTNL is not held.
4992 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4994 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4997 /* Slave array work handler. Holds only RTNL */
4998 static void bond_slave_arr_handler(struct work_struct *work)
5000 struct bonding *bond = container_of(work, struct bonding,
5001 slave_arr_work.work);
5004 if (!rtnl_trylock())
5007 ret = bond_update_slave_arr(bond, NULL);
5010 pr_warn_ratelimited("Failed to update slave array from WT\n");
5016 bond_slave_arr_work_rearm(bond, 1);
5019 static void bond_skip_slave(struct bond_up_slave *slaves,
5020 struct slave *skipslave)
5024 /* Rare situation where caller has asked to skip a specific
5025 * slave but allocation failed (most likely!). BTW this is
5026 * only possible when the call is initiated from
5027 * __bond_release_one(). In this situation; overwrite the
5028 * skipslave entry in the array with the last entry from the
5029 * array to avoid a situation where the xmit path may choose
5030 * this to-be-skipped slave to send a packet out.
5032 for (idx = 0; slaves && idx < slaves->count; idx++) {
5033 if (skipslave == slaves->arr[idx]) {
5035 slaves->arr[slaves->count - 1];
5042 static void bond_set_slave_arr(struct bonding *bond,
5043 struct bond_up_slave *usable_slaves,
5044 struct bond_up_slave *all_slaves)
5046 struct bond_up_slave *usable, *all;
5048 usable = rtnl_dereference(bond->usable_slaves);
5049 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5050 kfree_rcu(usable, rcu);
5052 all = rtnl_dereference(bond->all_slaves);
5053 rcu_assign_pointer(bond->all_slaves, all_slaves);
5054 kfree_rcu(all, rcu);
5057 static void bond_reset_slave_arr(struct bonding *bond)
5059 bond_set_slave_arr(bond, NULL, NULL);
5062 /* Build the usable slaves array in control path for modes that use xmit-hash
5063 * to determine the slave interface -
5064 * (a) BOND_MODE_8023AD
5066 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5068 * The caller is expected to hold RTNL only and NO other lock!
5070 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5072 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5073 struct slave *slave;
5074 struct list_head *iter;
5080 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5081 bond->slave_cnt), GFP_KERNEL);
5082 all_slaves = kzalloc(struct_size(all_slaves, arr,
5083 bond->slave_cnt), GFP_KERNEL);
5084 if (!usable_slaves || !all_slaves) {
5088 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5089 struct ad_info ad_info;
5091 spin_lock_bh(&bond->mode_lock);
5092 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5093 spin_unlock_bh(&bond->mode_lock);
5094 pr_debug("bond_3ad_get_active_agg_info failed\n");
5095 /* No active aggragator means it's not safe to use
5096 * the previous array.
5098 bond_reset_slave_arr(bond);
5101 spin_unlock_bh(&bond->mode_lock);
5102 agg_id = ad_info.aggregator_id;
5104 bond_for_each_slave(bond, slave, iter) {
5105 if (skipslave == slave)
5108 all_slaves->arr[all_slaves->count++] = slave;
5109 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5110 struct aggregator *agg;
5112 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5113 if (!agg || agg->aggregator_identifier != agg_id)
5116 if (!bond_slave_can_tx(slave))
5119 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5120 usable_slaves->count);
5122 usable_slaves->arr[usable_slaves->count++] = slave;
5125 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5128 if (ret != 0 && skipslave) {
5129 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5131 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5134 kfree_rcu(all_slaves, rcu);
5135 kfree_rcu(usable_slaves, rcu);
5140 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5141 struct sk_buff *skb,
5142 struct bond_up_slave *slaves)
5144 struct slave *slave;
5148 hash = bond_xmit_hash(bond, skb);
5149 count = slaves ? READ_ONCE(slaves->count) : 0;
5150 if (unlikely(!count))
5153 slave = slaves->arr[hash % count];
5157 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5158 struct xdp_buff *xdp)
5160 struct bond_up_slave *slaves;
5164 hash = bond_xmit_hash_xdp(bond, xdp);
5165 slaves = rcu_dereference(bond->usable_slaves);
5166 count = slaves ? READ_ONCE(slaves->count) : 0;
5167 if (unlikely(!count))
5170 return slaves->arr[hash % count];
5173 /* Use this Xmit function for 3AD as well as XOR modes. The current
5174 * usable slave array is formed in the control path. The xmit function
5175 * just calculates hash and sends the packet out.
5177 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5178 struct net_device *dev)
5180 struct bonding *bond = netdev_priv(dev);
5181 struct bond_up_slave *slaves;
5182 struct slave *slave;
5184 slaves = rcu_dereference(bond->usable_slaves);
5185 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5187 return bond_dev_queue_xmit(bond, skb, slave->dev);
5189 return bond_tx_drop(dev, skb);
5192 /* in broadcast mode, we send everything to all usable interfaces. */
5193 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5194 struct net_device *bond_dev)
5196 struct bonding *bond = netdev_priv(bond_dev);
5197 struct slave *slave = NULL;
5198 struct list_head *iter;
5199 bool xmit_suc = false;
5200 bool skb_used = false;
5202 bond_for_each_slave_rcu(bond, slave, iter) {
5203 struct sk_buff *skb2;
5205 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5208 if (bond_is_last_slave(bond, slave)) {
5212 skb2 = skb_clone(skb, GFP_ATOMIC);
5214 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5215 bond_dev->name, __func__);
5220 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5225 dev_kfree_skb_any(skb);
5228 return NETDEV_TX_OK;
5230 dev_core_stats_tx_dropped_inc(bond_dev);
5231 return NET_XMIT_DROP;
5234 /*------------------------- Device initialization ---------------------------*/
5236 /* Lookup the slave that corresponds to a qid */
5237 static inline int bond_slave_override(struct bonding *bond,
5238 struct sk_buff *skb)
5240 struct slave *slave = NULL;
5241 struct list_head *iter;
5243 if (!skb_rx_queue_recorded(skb))
5246 /* Find out if any slaves have the same mapping as this skb. */
5247 bond_for_each_slave_rcu(bond, slave, iter) {
5248 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5249 if (bond_slave_is_up(slave) &&
5250 slave->link == BOND_LINK_UP) {
5251 bond_dev_queue_xmit(bond, skb, slave->dev);
5254 /* If the slave isn't UP, use default transmit policy. */
5263 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5264 struct net_device *sb_dev)
5266 /* This helper function exists to help dev_pick_tx get the correct
5267 * destination queue. Using a helper function skips a call to
5268 * skb_tx_hash and will put the skbs in the queue we expect on their
5269 * way down to the bonding driver.
5271 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5273 /* Save the original txq to restore before passing to the driver */
5274 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5276 if (unlikely(txq >= dev->real_num_tx_queues)) {
5278 txq -= dev->real_num_tx_queues;
5279 } while (txq >= dev->real_num_tx_queues);
5284 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5285 struct sk_buff *skb,
5288 struct bonding *bond = netdev_priv(master_dev);
5289 struct bond_up_slave *slaves;
5290 struct slave *slave = NULL;
5292 switch (BOND_MODE(bond)) {
5293 case BOND_MODE_ROUNDROBIN:
5294 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5296 case BOND_MODE_ACTIVEBACKUP:
5297 slave = bond_xmit_activebackup_slave_get(bond);
5299 case BOND_MODE_8023AD:
5302 slaves = rcu_dereference(bond->all_slaves);
5304 slaves = rcu_dereference(bond->usable_slaves);
5305 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5307 case BOND_MODE_BROADCAST:
5310 slave = bond_xmit_alb_slave_get(bond, skb);
5313 slave = bond_xmit_tlb_slave_get(bond, skb);
5316 /* Should never happen, mode already checked */
5317 WARN_ONCE(true, "Unknown bonding mode");
5326 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5328 switch (sk->sk_family) {
5329 #if IS_ENABLED(CONFIG_IPV6)
5331 if (ipv6_only_sock(sk) ||
5332 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5333 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5334 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5335 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5340 default: /* AF_INET */
5341 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5342 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5343 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5347 flow->ports.src = inet_sk(sk)->inet_sport;
5348 flow->ports.dst = inet_sk(sk)->inet_dport;
5352 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5353 * @sk: socket to use for headers
5355 * This function will extract the necessary field from the socket and use
5356 * them to generate a hash based on the LAYER34 xmit_policy.
5357 * Assumes that sk is a TCP or UDP socket.
5359 static u32 bond_sk_hash_l34(struct sock *sk)
5361 struct flow_keys flow;
5364 bond_sk_to_flow(sk, &flow);
5367 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5369 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5372 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5375 struct bond_up_slave *slaves;
5376 struct slave *slave;
5380 slaves = rcu_dereference(bond->usable_slaves);
5381 count = slaves ? READ_ONCE(slaves->count) : 0;
5382 if (unlikely(!count))
5385 hash = bond_sk_hash_l34(sk);
5386 slave = slaves->arr[hash % count];
5391 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5394 struct bonding *bond = netdev_priv(dev);
5395 struct net_device *lower = NULL;
5398 if (bond_sk_check(bond))
5399 lower = __bond_sk_get_lower_dev(bond, sk);
5405 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5406 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5407 struct net_device *dev)
5409 struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5411 /* tls_netdev might become NULL, even if tls_is_skb_tx_device_offloaded
5412 * was true, if tls_device_down is running in parallel, but it's OK,
5413 * because bond_get_slave_by_dev has a NULL check.
5415 if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5416 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5417 return bond_tx_drop(dev, skb);
5421 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5423 struct bonding *bond = netdev_priv(dev);
5425 if (bond_should_override_tx_queue(bond) &&
5426 !bond_slave_override(bond, skb))
5427 return NETDEV_TX_OK;
5429 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5430 if (tls_is_skb_tx_device_offloaded(skb))
5431 return bond_tls_device_xmit(bond, skb, dev);
5434 switch (BOND_MODE(bond)) {
5435 case BOND_MODE_ROUNDROBIN:
5436 return bond_xmit_roundrobin(skb, dev);
5437 case BOND_MODE_ACTIVEBACKUP:
5438 return bond_xmit_activebackup(skb, dev);
5439 case BOND_MODE_8023AD:
5441 return bond_3ad_xor_xmit(skb, dev);
5442 case BOND_MODE_BROADCAST:
5443 return bond_xmit_broadcast(skb, dev);
5445 return bond_alb_xmit(skb, dev);
5447 return bond_tlb_xmit(skb, dev);
5449 /* Should never happen, mode already checked */
5450 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5452 return bond_tx_drop(dev, skb);
5456 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5458 struct bonding *bond = netdev_priv(dev);
5459 netdev_tx_t ret = NETDEV_TX_OK;
5461 /* If we risk deadlock from transmitting this in the
5462 * netpoll path, tell netpoll to queue the frame for later tx
5464 if (unlikely(is_netpoll_tx_blocked(dev)))
5465 return NETDEV_TX_BUSY;
5468 if (bond_has_slaves(bond))
5469 ret = __bond_start_xmit(skb, dev);
5471 ret = bond_tx_drop(dev, skb);
5477 static struct net_device *
5478 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5480 struct bonding *bond = netdev_priv(bond_dev);
5481 struct slave *slave;
5483 /* Caller needs to hold rcu_read_lock() */
5485 switch (BOND_MODE(bond)) {
5486 case BOND_MODE_ROUNDROBIN:
5487 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5490 case BOND_MODE_ACTIVEBACKUP:
5491 slave = bond_xmit_activebackup_slave_get(bond);
5494 case BOND_MODE_8023AD:
5496 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5500 /* Should never happen. Mode guarded by bond_xdp_check() */
5501 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5512 static int bond_xdp_xmit(struct net_device *bond_dev,
5513 int n, struct xdp_frame **frames, u32 flags)
5515 int nxmit, err = -ENXIO;
5519 for (nxmit = 0; nxmit < n; nxmit++) {
5520 struct xdp_frame *frame = frames[nxmit];
5521 struct xdp_frame *frames1[] = {frame};
5522 struct net_device *slave_dev;
5523 struct xdp_buff xdp;
5525 xdp_convert_frame_to_buff(frame, &xdp);
5527 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5533 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5540 /* If error happened on the first frame then we can pass the error up, otherwise
5541 * report the number of frames that were xmitted.
5544 return (nxmit == 0 ? err : nxmit);
5549 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5550 struct netlink_ext_ack *extack)
5552 struct bonding *bond = netdev_priv(dev);
5553 struct list_head *iter;
5554 struct slave *slave, *rollback_slave;
5555 struct bpf_prog *old_prog;
5556 struct netdev_bpf xdp = {
5557 .command = XDP_SETUP_PROG,
5566 if (!bond_xdp_check(bond))
5569 old_prog = bond->xdp_prog;
5570 bond->xdp_prog = prog;
5572 bond_for_each_slave(bond, slave, iter) {
5573 struct net_device *slave_dev = slave->dev;
5575 if (!slave_dev->netdev_ops->ndo_bpf ||
5576 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5577 SLAVE_NL_ERR(dev, slave_dev, extack,
5578 "Slave device does not support XDP");
5583 if (dev_xdp_prog_count(slave_dev) > 0) {
5584 SLAVE_NL_ERR(dev, slave_dev, extack,
5585 "Slave has XDP program loaded, please unload before enslaving");
5590 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5592 /* ndo_bpf() sets extack error message */
5593 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5601 static_branch_inc(&bpf_master_redirect_enabled_key);
5602 } else if (old_prog) {
5603 bpf_prog_put(old_prog);
5604 static_branch_dec(&bpf_master_redirect_enabled_key);
5610 /* unwind the program changes */
5611 bond->xdp_prog = old_prog;
5612 xdp.prog = old_prog;
5613 xdp.extack = NULL; /* do not overwrite original error */
5615 bond_for_each_slave(bond, rollback_slave, iter) {
5616 struct net_device *slave_dev = rollback_slave->dev;
5619 if (slave == rollback_slave)
5622 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5624 slave_err(dev, slave_dev,
5625 "Error %d when unwinding XDP program change\n", err_unwind);
5627 bpf_prog_inc(xdp.prog);
5632 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5634 switch (xdp->command) {
5635 case XDP_SETUP_PROG:
5636 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5642 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5644 if (speed == 0 || speed == SPEED_UNKNOWN)
5645 speed = slave->speed;
5647 speed = min(speed, slave->speed);
5652 /* Set the BOND_PHC_INDEX flag to notify user space */
5653 static int bond_set_phc_index_flag(struct kernel_hwtstamp_config *kernel_cfg)
5655 struct ifreq *ifr = kernel_cfg->ifr;
5656 struct hwtstamp_config cfg;
5658 if (kernel_cfg->copied_to_user) {
5659 /* Lower device has a legacy implementation */
5660 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
5663 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5664 if (copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)))
5667 kernel_cfg->flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
5673 static int bond_hwtstamp_get(struct net_device *dev,
5674 struct kernel_hwtstamp_config *cfg)
5676 struct bonding *bond = netdev_priv(dev);
5677 struct net_device *real_dev;
5680 real_dev = bond_option_active_slave_get_rcu(bond);
5684 err = generic_hwtstamp_get_lower(real_dev, cfg);
5688 return bond_set_phc_index_flag(cfg);
5691 static int bond_hwtstamp_set(struct net_device *dev,
5692 struct kernel_hwtstamp_config *cfg,
5693 struct netlink_ext_ack *extack)
5695 struct bonding *bond = netdev_priv(dev);
5696 struct net_device *real_dev;
5699 if (!(cfg->flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
5702 real_dev = bond_option_active_slave_get_rcu(bond);
5706 err = generic_hwtstamp_set_lower(real_dev, cfg, extack);
5710 return bond_set_phc_index_flag(cfg);
5713 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5714 struct ethtool_link_ksettings *cmd)
5716 struct bonding *bond = netdev_priv(bond_dev);
5717 struct list_head *iter;
5718 struct slave *slave;
5721 cmd->base.duplex = DUPLEX_UNKNOWN;
5722 cmd->base.port = PORT_OTHER;
5724 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5725 * do not need to check mode. Though link speed might not represent
5726 * the true receive or transmit bandwidth (not all modes are symmetric)
5727 * this is an accurate maximum.
5729 bond_for_each_slave(bond, slave, iter) {
5730 if (bond_slave_can_tx(slave)) {
5731 bond_update_speed_duplex(slave);
5732 if (slave->speed != SPEED_UNKNOWN) {
5733 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5734 speed = bond_mode_bcast_speed(slave,
5737 speed += slave->speed;
5739 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5740 slave->duplex != DUPLEX_UNKNOWN)
5741 cmd->base.duplex = slave->duplex;
5744 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5749 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5750 struct ethtool_drvinfo *drvinfo)
5752 strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5753 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5757 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5758 struct ethtool_ts_info *info)
5760 struct bonding *bond = netdev_priv(bond_dev);
5761 struct ethtool_ts_info ts_info;
5762 struct net_device *real_dev;
5763 bool sw_tx_support = false;
5764 struct list_head *iter;
5765 struct slave *slave;
5769 real_dev = bond_option_active_slave_get_rcu(bond);
5774 ret = ethtool_get_ts_info_by_layer(real_dev, info);
5776 /* Check if all slaves support software tx timestamping */
5778 bond_for_each_slave_rcu(bond, slave, iter) {
5779 ret = ethtool_get_ts_info_by_layer(slave->dev, &ts_info);
5780 if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5781 sw_tx_support = true;
5785 sw_tx_support = false;
5792 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5798 static const struct ethtool_ops bond_ethtool_ops = {
5799 .get_drvinfo = bond_ethtool_get_drvinfo,
5800 .get_link = ethtool_op_get_link,
5801 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5802 .get_ts_info = bond_ethtool_get_ts_info,
5805 static const struct net_device_ops bond_netdev_ops = {
5806 .ndo_init = bond_init,
5807 .ndo_uninit = bond_uninit,
5808 .ndo_open = bond_open,
5809 .ndo_stop = bond_close,
5810 .ndo_start_xmit = bond_start_xmit,
5811 .ndo_select_queue = bond_select_queue,
5812 .ndo_get_stats64 = bond_get_stats,
5813 .ndo_eth_ioctl = bond_eth_ioctl,
5814 .ndo_siocbond = bond_do_ioctl,
5815 .ndo_siocdevprivate = bond_siocdevprivate,
5816 .ndo_change_rx_flags = bond_change_rx_flags,
5817 .ndo_set_rx_mode = bond_set_rx_mode,
5818 .ndo_change_mtu = bond_change_mtu,
5819 .ndo_set_mac_address = bond_set_mac_address,
5820 .ndo_neigh_setup = bond_neigh_setup,
5821 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5822 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5823 #ifdef CONFIG_NET_POLL_CONTROLLER
5824 .ndo_netpoll_setup = bond_netpoll_setup,
5825 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5826 .ndo_poll_controller = bond_poll_controller,
5828 .ndo_add_slave = bond_enslave,
5829 .ndo_del_slave = bond_release,
5830 .ndo_fix_features = bond_fix_features,
5831 .ndo_features_check = passthru_features_check,
5832 .ndo_get_xmit_slave = bond_xmit_get_slave,
5833 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5834 .ndo_bpf = bond_xdp,
5835 .ndo_xdp_xmit = bond_xdp_xmit,
5836 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5837 .ndo_hwtstamp_get = bond_hwtstamp_get,
5838 .ndo_hwtstamp_set = bond_hwtstamp_set,
5841 static const struct device_type bond_type = {
5845 static void bond_destructor(struct net_device *bond_dev)
5847 struct bonding *bond = netdev_priv(bond_dev);
5850 destroy_workqueue(bond->wq);
5852 free_percpu(bond->rr_tx_counter);
5855 void bond_setup(struct net_device *bond_dev)
5857 struct bonding *bond = netdev_priv(bond_dev);
5859 spin_lock_init(&bond->mode_lock);
5860 bond->params = bonding_defaults;
5862 /* Initialize pointers */
5863 bond->dev = bond_dev;
5865 /* Initialize the device entry points */
5866 ether_setup(bond_dev);
5867 bond_dev->max_mtu = ETH_MAX_MTU;
5868 bond_dev->netdev_ops = &bond_netdev_ops;
5869 bond_dev->ethtool_ops = &bond_ethtool_ops;
5871 bond_dev->needs_free_netdev = true;
5872 bond_dev->priv_destructor = bond_destructor;
5874 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5876 /* Initialize the device options */
5877 bond_dev->flags |= IFF_MASTER;
5878 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5879 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5881 #ifdef CONFIG_XFRM_OFFLOAD
5882 /* set up xfrm device ops (only supported in active-backup right now) */
5883 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5884 INIT_LIST_HEAD(&bond->ipsec_list);
5885 spin_lock_init(&bond->ipsec_lock);
5886 #endif /* CONFIG_XFRM_OFFLOAD */
5888 /* don't acquire bond device's netif_tx_lock when transmitting */
5889 bond_dev->features |= NETIF_F_LLTX;
5891 /* By default, we declare the bond to be fully
5892 * VLAN hardware accelerated capable. Special
5893 * care is taken in the various xmit functions
5894 * when there are slaves that are not hw accel
5898 /* Don't allow bond devices to change network namespaces. */
5899 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5901 bond_dev->hw_features = BOND_VLAN_FEATURES |
5902 NETIF_F_HW_VLAN_CTAG_RX |
5903 NETIF_F_HW_VLAN_CTAG_FILTER |
5904 NETIF_F_HW_VLAN_STAG_RX |
5905 NETIF_F_HW_VLAN_STAG_FILTER;
5907 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5908 bond_dev->features |= bond_dev->hw_features;
5909 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5910 #ifdef CONFIG_XFRM_OFFLOAD
5911 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5912 /* Only enable XFRM features if this is an active-backup config */
5913 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5914 bond_dev->features |= BOND_XFRM_FEATURES;
5915 #endif /* CONFIG_XFRM_OFFLOAD */
5918 /* Destroy a bonding device.
5919 * Must be under rtnl_lock when this function is called.
5921 static void bond_uninit(struct net_device *bond_dev)
5923 struct bonding *bond = netdev_priv(bond_dev);
5924 struct list_head *iter;
5925 struct slave *slave;
5927 bond_netpoll_cleanup(bond_dev);
5929 /* Release the bonded slaves */
5930 bond_for_each_slave(bond, slave, iter)
5931 __bond_release_one(bond_dev, slave->dev, true, true);
5932 netdev_info(bond_dev, "Released all slaves\n");
5934 bond_set_slave_arr(bond, NULL, NULL);
5936 list_del(&bond->bond_list);
5938 bond_debug_unregister(bond);
5941 /*------------------------- Module initialization ---------------------------*/
5943 static int __init bond_check_params(struct bond_params *params)
5945 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5946 struct bond_opt_value newval;
5947 const struct bond_opt_value *valptr;
5948 int arp_all_targets_value = 0;
5949 u16 ad_actor_sys_prio = 0;
5950 u16 ad_user_port_key = 0;
5951 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5953 int bond_mode = BOND_MODE_ROUNDROBIN;
5954 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5958 /* Convert string parameters. */
5960 bond_opt_initstr(&newval, mode);
5961 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5963 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5966 bond_mode = valptr->value;
5969 if (xmit_hash_policy) {
5970 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5971 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5972 bond_mode == BOND_MODE_BROADCAST) {
5973 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5974 bond_mode_name(bond_mode));
5976 bond_opt_initstr(&newval, xmit_hash_policy);
5977 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5980 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5984 xmit_hashtype = valptr->value;
5989 if (bond_mode != BOND_MODE_8023AD) {
5990 pr_info("lacp_rate param is irrelevant in mode %s\n",
5991 bond_mode_name(bond_mode));
5993 bond_opt_initstr(&newval, lacp_rate);
5994 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5997 pr_err("Error: Invalid lacp rate \"%s\"\n",
6001 lacp_fast = valptr->value;
6006 bond_opt_initstr(&newval, ad_select);
6007 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
6010 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
6013 params->ad_select = valptr->value;
6014 if (bond_mode != BOND_MODE_8023AD)
6015 pr_warn("ad_select param only affects 802.3ad mode\n");
6017 params->ad_select = BOND_AD_STABLE;
6020 if (max_bonds < 0) {
6021 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
6022 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
6023 max_bonds = BOND_DEFAULT_MAX_BONDS;
6027 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6033 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6038 if (downdelay < 0) {
6039 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6040 downdelay, INT_MAX);
6044 if ((use_carrier != 0) && (use_carrier != 1)) {
6045 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6050 if (num_peer_notif < 0 || num_peer_notif > 255) {
6051 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6056 /* reset values for 802.3ad/TLB/ALB */
6057 if (!bond_mode_uses_arp(bond_mode)) {
6059 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");
6060 pr_warn("Forcing miimon to 100msec\n");
6061 miimon = BOND_DEFAULT_MIIMON;
6065 if (tx_queues < 1 || tx_queues > 255) {
6066 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6067 tx_queues, BOND_DEFAULT_TX_QUEUES);
6068 tx_queues = BOND_DEFAULT_TX_QUEUES;
6071 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6072 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6074 all_slaves_active = 0;
6077 if (resend_igmp < 0 || resend_igmp > 255) {
6078 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6079 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6080 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6083 bond_opt_initval(&newval, packets_per_slave);
6084 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6085 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6086 packets_per_slave, USHRT_MAX);
6087 packets_per_slave = 1;
6090 if (bond_mode == BOND_MODE_ALB) {
6091 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",
6096 if (updelay || downdelay) {
6097 /* just warn the user the up/down delay will have
6098 * no effect since miimon is zero...
6100 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",
6101 updelay, downdelay);
6104 /* don't allow arp monitoring */
6106 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6107 miimon, arp_interval);
6111 if ((updelay % miimon) != 0) {
6112 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6113 updelay, miimon, (updelay / miimon) * miimon);
6118 if ((downdelay % miimon) != 0) {
6119 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6121 (downdelay / miimon) * miimon);
6124 downdelay /= miimon;
6127 if (arp_interval < 0) {
6128 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6129 arp_interval, INT_MAX);
6133 for (arp_ip_count = 0, i = 0;
6134 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6137 /* not a complete check, but good enough to catch mistakes */
6138 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6139 !bond_is_ip_target_ok(ip)) {
6140 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6144 if (bond_get_targets_ip(arp_target, ip) == -1)
6145 arp_target[arp_ip_count++] = ip;
6147 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6152 if (arp_interval && !arp_ip_count) {
6153 /* don't allow arping if no arp_ip_target given... */
6154 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6160 if (!arp_interval) {
6161 pr_err("arp_validate requires arp_interval\n");
6165 bond_opt_initstr(&newval, arp_validate);
6166 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6169 pr_err("Error: invalid arp_validate \"%s\"\n",
6173 arp_validate_value = valptr->value;
6175 arp_validate_value = 0;
6178 if (arp_all_targets) {
6179 bond_opt_initstr(&newval, arp_all_targets);
6180 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6183 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6185 arp_all_targets_value = 0;
6187 arp_all_targets_value = valptr->value;
6192 pr_info("MII link monitoring set to %d ms\n", miimon);
6193 } else if (arp_interval) {
6194 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6195 arp_validate_value);
6196 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6197 arp_interval, valptr->string, arp_ip_count);
6199 for (i = 0; i < arp_ip_count; i++)
6200 pr_cont(" %s", arp_ip_target[i]);
6204 } else if (max_bonds) {
6205 /* miimon and arp_interval not set, we need one so things
6206 * work as expected, see bonding.txt for details
6208 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");
6211 if (primary && !bond_mode_uses_primary(bond_mode)) {
6212 /* currently, using a primary only makes sense
6213 * in active backup, TLB or ALB modes
6215 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6216 primary, bond_mode_name(bond_mode));
6220 if (primary && primary_reselect) {
6221 bond_opt_initstr(&newval, primary_reselect);
6222 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6225 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6229 primary_reselect_value = valptr->value;
6231 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6234 if (fail_over_mac) {
6235 bond_opt_initstr(&newval, fail_over_mac);
6236 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6239 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6243 fail_over_mac_value = valptr->value;
6244 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6245 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6247 fail_over_mac_value = BOND_FOM_NONE;
6250 bond_opt_initstr(&newval, "default");
6251 valptr = bond_opt_parse(
6252 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6255 pr_err("Error: No ad_actor_sys_prio default value");
6258 ad_actor_sys_prio = valptr->value;
6260 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6263 pr_err("Error: No ad_user_port_key default value");
6266 ad_user_port_key = valptr->value;
6268 bond_opt_initstr(&newval, "default");
6269 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6271 pr_err("Error: No tlb_dynamic_lb default value");
6274 tlb_dynamic_lb = valptr->value;
6276 if (lp_interval == 0) {
6277 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6278 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6279 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6282 /* fill params struct with the proper values */
6283 params->mode = bond_mode;
6284 params->xmit_policy = xmit_hashtype;
6285 params->miimon = miimon;
6286 params->num_peer_notif = num_peer_notif;
6287 params->arp_interval = arp_interval;
6288 params->arp_validate = arp_validate_value;
6289 params->arp_all_targets = arp_all_targets_value;
6290 params->missed_max = 2;
6291 params->updelay = updelay;
6292 params->downdelay = downdelay;
6293 params->peer_notif_delay = 0;
6294 params->use_carrier = use_carrier;
6295 params->lacp_active = 1;
6296 params->lacp_fast = lacp_fast;
6297 params->primary[0] = 0;
6298 params->primary_reselect = primary_reselect_value;
6299 params->fail_over_mac = fail_over_mac_value;
6300 params->tx_queues = tx_queues;
6301 params->all_slaves_active = all_slaves_active;
6302 params->resend_igmp = resend_igmp;
6303 params->min_links = min_links;
6304 params->lp_interval = lp_interval;
6305 params->packets_per_slave = packets_per_slave;
6306 params->tlb_dynamic_lb = tlb_dynamic_lb;
6307 params->ad_actor_sys_prio = ad_actor_sys_prio;
6308 eth_zero_addr(params->ad_actor_system);
6309 params->ad_user_port_key = ad_user_port_key;
6310 params->coupled_control = 1;
6311 if (packets_per_slave > 0) {
6312 params->reciprocal_packets_per_slave =
6313 reciprocal_value(packets_per_slave);
6315 /* reciprocal_packets_per_slave is unused if
6316 * packets_per_slave is 0 or 1, just initialize it
6318 params->reciprocal_packets_per_slave =
6319 (struct reciprocal_value) { 0 };
6323 strscpy_pad(params->primary, primary, sizeof(params->primary));
6325 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6326 #if IS_ENABLED(CONFIG_IPV6)
6327 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6333 /* Called from registration process */
6334 static int bond_init(struct net_device *bond_dev)
6336 struct bonding *bond = netdev_priv(bond_dev);
6337 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6339 netdev_dbg(bond_dev, "Begin bond_init\n");
6341 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6345 bond->notifier_ctx = false;
6347 spin_lock_init(&bond->stats_lock);
6348 netdev_lockdep_set_classes(bond_dev);
6350 list_add_tail(&bond->bond_list, &bn->dev_list);
6352 bond_prepare_sysfs_group(bond);
6354 bond_debug_register(bond);
6356 /* Ensure valid dev_addr */
6357 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6358 bond_dev->addr_assign_type == NET_ADDR_PERM)
6359 eth_hw_addr_random(bond_dev);
6364 unsigned int bond_get_num_tx_queues(void)
6369 /* Create a new bond based on the specified name and bonding parameters.
6370 * If name is NULL, obtain a suitable "bond%d" name for us.
6371 * Caller must NOT hold rtnl_lock; we need to release it here before we
6372 * set up our sysfs entries.
6374 int bond_create(struct net *net, const char *name)
6376 struct net_device *bond_dev;
6377 struct bonding *bond;
6382 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6383 name ? name : "bond%d", NET_NAME_UNKNOWN,
6384 bond_setup, tx_queues);
6388 bond = netdev_priv(bond_dev);
6389 dev_net_set(bond_dev, net);
6390 bond_dev->rtnl_link_ops = &bond_link_ops;
6392 res = register_netdevice(bond_dev);
6394 free_netdev(bond_dev);
6398 netif_carrier_off(bond_dev);
6400 bond_work_init_all(bond);
6407 static int __net_init bond_net_init(struct net *net)
6409 struct bond_net *bn = net_generic(net, bond_net_id);
6412 INIT_LIST_HEAD(&bn->dev_list);
6414 bond_create_proc_dir(bn);
6415 bond_create_sysfs(bn);
6420 /* According to commit 69b0216ac255 ("bonding: fix bonding_masters
6421 * race condition in bond unloading") we need to remove sysfs files
6422 * before we remove our devices (done later in bond_net_exit_batch_rtnl())
6424 static void __net_exit bond_net_pre_exit(struct net *net)
6426 struct bond_net *bn = net_generic(net, bond_net_id);
6428 bond_destroy_sysfs(bn);
6431 static void __net_exit bond_net_exit_batch_rtnl(struct list_head *net_list,
6432 struct list_head *dev_kill_list)
6434 struct bond_net *bn;
6437 /* Kill off any bonds created after unregistering bond rtnl ops */
6438 list_for_each_entry(net, net_list, exit_list) {
6439 struct bonding *bond, *tmp_bond;
6441 bn = net_generic(net, bond_net_id);
6442 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6443 unregister_netdevice_queue(bond->dev, dev_kill_list);
6447 /* According to commit 23fa5c2caae0 ("bonding: destroy proc directory
6448 * only after all bonds are gone") bond_destroy_proc_dir() is called
6449 * after bond_net_exit_batch_rtnl() has completed.
6451 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6453 struct bond_net *bn;
6456 list_for_each_entry(net, net_list, exit_list) {
6457 bn = net_generic(net, bond_net_id);
6458 bond_destroy_proc_dir(bn);
6462 static struct pernet_operations bond_net_ops = {
6463 .init = bond_net_init,
6464 .pre_exit = bond_net_pre_exit,
6465 .exit_batch_rtnl = bond_net_exit_batch_rtnl,
6466 .exit_batch = bond_net_exit_batch,
6468 .size = sizeof(struct bond_net),
6471 static int __init bonding_init(void)
6476 res = bond_check_params(&bonding_defaults);
6480 res = register_pernet_subsys(&bond_net_ops);
6484 res = bond_netlink_init();
6488 bond_create_debugfs();
6490 for (i = 0; i < max_bonds; i++) {
6491 res = bond_create(&init_net, NULL);
6496 skb_flow_dissector_init(&flow_keys_bonding,
6497 flow_keys_bonding_keys,
6498 ARRAY_SIZE(flow_keys_bonding_keys));
6500 register_netdevice_notifier(&bond_netdev_notifier);
6504 bond_destroy_debugfs();
6505 bond_netlink_fini();
6507 unregister_pernet_subsys(&bond_net_ops);
6512 static void __exit bonding_exit(void)
6514 unregister_netdevice_notifier(&bond_netdev_notifier);
6516 bond_destroy_debugfs();
6518 bond_netlink_fini();
6519 unregister_pernet_subsys(&bond_net_ops);
6521 #ifdef CONFIG_NET_POLL_CONTROLLER
6522 /* Make sure we don't have an imbalance on our netpoll blocking */
6523 WARN_ON(atomic_read(&netpoll_block_tx));
6527 module_init(bonding_init);
6528 module_exit(bonding_exit);
6529 MODULE_LICENSE("GPL");
6530 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6531 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");