2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/filter.h>
39 #include <linux/interrupt.h>
40 #include <linux/ptrace.h>
41 #include <linux/ioport.h>
45 #include <linux/icmp.h>
46 #include <linux/icmpv6.h>
47 #include <linux/tcp.h>
48 #include <linux/udp.h>
49 #include <linux/slab.h>
50 #include <linux/string.h>
51 #include <linux/init.h>
52 #include <linux/timer.h>
53 #include <linux/socket.h>
54 #include <linux/ctype.h>
55 #include <linux/inet.h>
56 #include <linux/bitops.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/phy.h>
76 #include <linux/jiffies.h>
77 #include <linux/preempt.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include <net/netns/generic.h>
81 #include <net/pkt_sched.h>
82 #include <linux/rculist.h>
83 #include <net/flow_dissector.h>
85 #include <net/bonding.h>
86 #include <net/bond_3ad.h>
87 #include <net/bond_alb.h>
88 #if IS_ENABLED(CONFIG_TLS_DEVICE)
91 #include <net/ip6_route.h>
93 #include "bonding_priv.h"
95 /*---------------------------- Module parameters ----------------------------*/
97 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
99 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
100 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
101 static int num_peer_notif = 1;
104 static int downdelay;
105 static int use_carrier = 1;
107 static char *primary;
108 static char *primary_reselect;
109 static char *lacp_rate;
110 static int min_links;
111 static char *ad_select;
112 static char *xmit_hash_policy;
113 static int arp_interval;
114 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
115 static char *arp_validate;
116 static char *arp_all_targets;
117 static char *fail_over_mac;
118 static int all_slaves_active;
119 static struct bond_params bonding_defaults;
120 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
121 static int packets_per_slave = 1;
122 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
124 module_param(max_bonds, int, 0);
125 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
126 module_param(tx_queues, int, 0);
127 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
128 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
129 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
130 "failover event (alias of num_unsol_na)");
131 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
132 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
133 "failover event (alias of num_grat_arp)");
134 module_param(miimon, int, 0);
135 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
136 module_param(updelay, int, 0);
137 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
138 module_param(downdelay, int, 0);
139 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
141 module_param(use_carrier, int, 0);
142 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
143 "0 for off, 1 for on (default)");
144 module_param(mode, charp, 0);
145 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
146 "1 for active-backup, 2 for balance-xor, "
147 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
148 "6 for balance-alb");
149 module_param(primary, charp, 0);
150 MODULE_PARM_DESC(primary, "Primary network device to use");
151 module_param(primary_reselect, charp, 0);
152 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
154 "0 for always (default), "
155 "1 for only if speed of primary is "
157 "2 for only on active slave "
159 module_param(lacp_rate, charp, 0);
160 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
161 "0 for slow, 1 for fast");
162 module_param(ad_select, charp, 0);
163 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
164 "0 for stable (default), 1 for bandwidth, "
166 module_param(min_links, int, 0);
167 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
169 module_param(xmit_hash_policy, charp, 0);
170 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
171 "0 for layer 2 (default), 1 for layer 3+4, "
172 "2 for layer 2+3, 3 for encap layer 2+3, "
173 "4 for encap layer 3+4, 5 for vlan+srcmac");
174 module_param(arp_interval, int, 0);
175 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
176 module_param_array(arp_ip_target, charp, NULL, 0);
177 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
178 module_param(arp_validate, charp, 0);
179 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
180 "0 for none (default), 1 for active, "
181 "2 for backup, 3 for all");
182 module_param(arp_all_targets, charp, 0);
183 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
184 module_param(fail_over_mac, charp, 0);
185 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
186 "the same MAC; 0 for none (default), "
187 "1 for active, 2 for follow");
188 module_param(all_slaves_active, int, 0);
189 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
190 "by setting active flag for all slaves; "
191 "0 for never (default), 1 for always.");
192 module_param(resend_igmp, int, 0);
193 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
195 module_param(packets_per_slave, int, 0);
196 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
197 "mode; 0 for a random slave, 1 packet per "
198 "slave (default), >1 packets per slave.");
199 module_param(lp_interval, uint, 0);
200 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
201 "the bonding driver sends learning packets to "
202 "each slaves peer switch. The default is 1.");
204 /*----------------------------- Global variables ----------------------------*/
206 #ifdef CONFIG_NET_POLL_CONTROLLER
207 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
210 unsigned int bond_net_id __read_mostly;
212 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
214 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
215 .offset = offsetof(struct flow_keys, control),
218 .key_id = FLOW_DISSECTOR_KEY_BASIC,
219 .offset = offsetof(struct flow_keys, basic),
222 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
223 .offset = offsetof(struct flow_keys, addrs.v4addrs),
226 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
227 .offset = offsetof(struct flow_keys, addrs.v6addrs),
230 .key_id = FLOW_DISSECTOR_KEY_TIPC,
231 .offset = offsetof(struct flow_keys, addrs.tipckey),
234 .key_id = FLOW_DISSECTOR_KEY_PORTS,
235 .offset = offsetof(struct flow_keys, ports),
238 .key_id = FLOW_DISSECTOR_KEY_ICMP,
239 .offset = offsetof(struct flow_keys, icmp),
242 .key_id = FLOW_DISSECTOR_KEY_VLAN,
243 .offset = offsetof(struct flow_keys, vlan),
246 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
247 .offset = offsetof(struct flow_keys, tags),
250 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
251 .offset = offsetof(struct flow_keys, keyid),
255 static struct flow_dissector flow_keys_bonding __read_mostly;
257 /*-------------------------- Forward declarations ---------------------------*/
259 static int bond_init(struct net_device *bond_dev);
260 static void bond_uninit(struct net_device *bond_dev);
261 static void bond_get_stats(struct net_device *bond_dev,
262 struct rtnl_link_stats64 *stats);
263 static void bond_slave_arr_handler(struct work_struct *work);
264 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
266 static void bond_netdev_notify_work(struct work_struct *work);
268 /*---------------------------- General routines -----------------------------*/
270 const char *bond_mode_name(int mode)
272 static const char *names[] = {
273 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
274 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
275 [BOND_MODE_XOR] = "load balancing (xor)",
276 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
277 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
278 [BOND_MODE_TLB] = "transmit load balancing",
279 [BOND_MODE_ALB] = "adaptive load balancing",
282 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
289 * bond_dev_queue_xmit - Prepare skb for xmit.
291 * @bond: bond device that got this skb for tx.
292 * @skb: hw accel VLAN tagged skb to transmit
293 * @slave_dev: slave that is supposed to xmit this skbuff
295 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
296 struct net_device *slave_dev)
298 skb->dev = slave_dev;
300 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
301 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
302 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
304 if (unlikely(netpoll_tx_running(bond->dev)))
305 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
307 return dev_queue_xmit(skb);
310 bool bond_sk_check(struct bonding *bond)
312 switch (BOND_MODE(bond)) {
313 case BOND_MODE_8023AD:
315 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
323 static bool bond_xdp_check(struct bonding *bond)
325 switch (BOND_MODE(bond)) {
326 case BOND_MODE_ROUNDROBIN:
327 case BOND_MODE_ACTIVEBACKUP:
329 case BOND_MODE_8023AD:
331 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
332 * payload is not in the packet due to hardware offload.
334 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
342 /*---------------------------------- VLAN -----------------------------------*/
344 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
345 * We don't protect the slave list iteration with a lock because:
346 * a. This operation is performed in IOCTL context,
347 * b. The operation is protected by the RTNL semaphore in the 8021q code,
348 * c. Holding a lock with BH disabled while directly calling a base driver
349 * entry point is generally a BAD idea.
351 * The design of synchronization/protection for this operation in the 8021q
352 * module is good for one or more VLAN devices over a single physical device
353 * and cannot be extended for a teaming solution like bonding, so there is a
354 * potential race condition here where a net device from the vlan group might
355 * be referenced (either by a base driver or the 8021q code) while it is being
356 * removed from the system. However, it turns out we're not making matters
357 * worse, and if it works for regular VLAN usage it will work here too.
361 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
362 * @bond_dev: bonding net device that got called
363 * @proto: network protocol ID
364 * @vid: vlan id being added
366 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
367 __be16 proto, u16 vid)
369 struct bonding *bond = netdev_priv(bond_dev);
370 struct slave *slave, *rollback_slave;
371 struct list_head *iter;
374 bond_for_each_slave(bond, slave, iter) {
375 res = vlan_vid_add(slave->dev, proto, vid);
383 /* unwind to the slave that failed */
384 bond_for_each_slave(bond, rollback_slave, iter) {
385 if (rollback_slave == slave)
388 vlan_vid_del(rollback_slave->dev, proto, vid);
395 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
396 * @bond_dev: bonding net device that got called
397 * @proto: network protocol ID
398 * @vid: vlan id being removed
400 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
401 __be16 proto, u16 vid)
403 struct bonding *bond = netdev_priv(bond_dev);
404 struct list_head *iter;
407 bond_for_each_slave(bond, slave, iter)
408 vlan_vid_del(slave->dev, proto, vid);
410 if (bond_is_lb(bond))
411 bond_alb_clear_vlan(bond, vid);
416 /*---------------------------------- XFRM -----------------------------------*/
418 #ifdef CONFIG_XFRM_OFFLOAD
420 * bond_ipsec_add_sa - program device with a security association
421 * @xs: pointer to transformer state struct
423 static int bond_ipsec_add_sa(struct xfrm_state *xs)
425 struct net_device *bond_dev = xs->xso.dev;
426 struct bond_ipsec *ipsec;
427 struct bonding *bond;
435 bond = netdev_priv(bond_dev);
436 slave = rcu_dereference(bond->curr_active_slave);
442 if (!slave->dev->xfrmdev_ops ||
443 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
444 netif_is_bond_master(slave->dev)) {
445 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
450 ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
455 xs->xso.real_dev = slave->dev;
457 err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
460 INIT_LIST_HEAD(&ipsec->list);
461 spin_lock_bh(&bond->ipsec_lock);
462 list_add(&ipsec->list, &bond->ipsec_list);
463 spin_unlock_bh(&bond->ipsec_lock);
471 static void bond_ipsec_add_sa_all(struct bonding *bond)
473 struct net_device *bond_dev = bond->dev;
474 struct bond_ipsec *ipsec;
478 slave = rcu_dereference(bond->curr_active_slave);
482 if (!slave->dev->xfrmdev_ops ||
483 !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
484 netif_is_bond_master(slave->dev)) {
485 spin_lock_bh(&bond->ipsec_lock);
486 if (!list_empty(&bond->ipsec_list))
487 slave_warn(bond_dev, slave->dev,
488 "%s: no slave xdo_dev_state_add\n",
490 spin_unlock_bh(&bond->ipsec_lock);
494 spin_lock_bh(&bond->ipsec_lock);
495 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
496 ipsec->xs->xso.real_dev = slave->dev;
497 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
498 slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
499 ipsec->xs->xso.real_dev = NULL;
502 spin_unlock_bh(&bond->ipsec_lock);
508 * bond_ipsec_del_sa - clear out this specific SA
509 * @xs: pointer to transformer state struct
511 static void bond_ipsec_del_sa(struct xfrm_state *xs)
513 struct net_device *bond_dev = xs->xso.dev;
514 struct bond_ipsec *ipsec;
515 struct bonding *bond;
522 bond = netdev_priv(bond_dev);
523 slave = rcu_dereference(bond->curr_active_slave);
528 if (!xs->xso.real_dev)
531 WARN_ON(xs->xso.real_dev != slave->dev);
533 if (!slave->dev->xfrmdev_ops ||
534 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
535 netif_is_bond_master(slave->dev)) {
536 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
540 slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
542 spin_lock_bh(&bond->ipsec_lock);
543 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
544 if (ipsec->xs == xs) {
545 list_del(&ipsec->list);
550 spin_unlock_bh(&bond->ipsec_lock);
554 static void bond_ipsec_del_sa_all(struct bonding *bond)
556 struct net_device *bond_dev = bond->dev;
557 struct bond_ipsec *ipsec;
561 slave = rcu_dereference(bond->curr_active_slave);
567 spin_lock_bh(&bond->ipsec_lock);
568 list_for_each_entry(ipsec, &bond->ipsec_list, list) {
569 if (!ipsec->xs->xso.real_dev)
572 if (!slave->dev->xfrmdev_ops ||
573 !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
574 netif_is_bond_master(slave->dev)) {
575 slave_warn(bond_dev, slave->dev,
576 "%s: no slave xdo_dev_state_delete\n",
579 slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
581 ipsec->xs->xso.real_dev = NULL;
583 spin_unlock_bh(&bond->ipsec_lock);
588 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
589 * @skb: current data packet
590 * @xs: pointer to transformer state struct
592 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
594 struct net_device *bond_dev = xs->xso.dev;
595 struct net_device *real_dev;
596 struct slave *curr_active;
597 struct bonding *bond;
600 bond = netdev_priv(bond_dev);
602 curr_active = rcu_dereference(bond->curr_active_slave);
603 real_dev = curr_active->dev;
605 if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
610 if (!xs->xso.real_dev) {
615 if (!real_dev->xfrmdev_ops ||
616 !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
617 netif_is_bond_master(real_dev)) {
622 err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
628 static const struct xfrmdev_ops bond_xfrmdev_ops = {
629 .xdo_dev_state_add = bond_ipsec_add_sa,
630 .xdo_dev_state_delete = bond_ipsec_del_sa,
631 .xdo_dev_offload_ok = bond_ipsec_offload_ok,
633 #endif /* CONFIG_XFRM_OFFLOAD */
635 /*------------------------------- Link status -------------------------------*/
637 /* Set the carrier state for the master according to the state of its
638 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
639 * do special 802.3ad magic.
641 * Returns zero if carrier state does not change, nonzero if it does.
643 int bond_set_carrier(struct bonding *bond)
645 struct list_head *iter;
648 if (!bond_has_slaves(bond))
651 if (BOND_MODE(bond) == BOND_MODE_8023AD)
652 return bond_3ad_set_carrier(bond);
654 bond_for_each_slave(bond, slave, iter) {
655 if (slave->link == BOND_LINK_UP) {
656 if (!netif_carrier_ok(bond->dev)) {
657 netif_carrier_on(bond->dev);
665 if (netif_carrier_ok(bond->dev)) {
666 netif_carrier_off(bond->dev);
672 /* Get link speed and duplex from the slave's base driver
673 * using ethtool. If for some reason the call fails or the
674 * values are invalid, set speed and duplex to -1,
675 * and return. Return 1 if speed or duplex settings are
676 * UNKNOWN; 0 otherwise.
678 static int bond_update_speed_duplex(struct slave *slave)
680 struct net_device *slave_dev = slave->dev;
681 struct ethtool_link_ksettings ecmd;
684 slave->speed = SPEED_UNKNOWN;
685 slave->duplex = DUPLEX_UNKNOWN;
687 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
690 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
692 switch (ecmd.base.duplex) {
700 slave->speed = ecmd.base.speed;
701 slave->duplex = ecmd.base.duplex;
706 const char *bond_slave_link_status(s8 link)
722 /* if <dev> supports MII link status reporting, check its link status.
724 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
725 * depending upon the setting of the use_carrier parameter.
727 * Return either BMSR_LSTATUS, meaning that the link is up (or we
728 * can't tell and just pretend it is), or 0, meaning that the link is
731 * If reporting is non-zero, instead of faking link up, return -1 if
732 * both ETHTOOL and MII ioctls fail (meaning the device does not
733 * support them). If use_carrier is set, return whatever it says.
734 * It'd be nice if there was a good way to tell if a driver supports
735 * netif_carrier, but there really isn't.
737 static int bond_check_dev_link(struct bonding *bond,
738 struct net_device *slave_dev, int reporting)
740 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
741 int (*ioctl)(struct net_device *, struct ifreq *, int);
743 struct mii_ioctl_data *mii;
745 if (!reporting && !netif_running(slave_dev))
748 if (bond->params.use_carrier)
749 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
751 /* Try to get link status using Ethtool first. */
752 if (slave_dev->ethtool_ops->get_link)
753 return slave_dev->ethtool_ops->get_link(slave_dev) ?
756 /* Ethtool can't be used, fallback to MII ioctls. */
757 ioctl = slave_ops->ndo_eth_ioctl;
759 /* TODO: set pointer to correct ioctl on a per team member
760 * bases to make this more efficient. that is, once
761 * we determine the correct ioctl, we will always
762 * call it and not the others for that team
766 /* We cannot assume that SIOCGMIIPHY will also read a
767 * register; not all network drivers (e.g., e100)
771 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
772 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
774 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
775 mii->reg_num = MII_BMSR;
776 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
777 return mii->val_out & BMSR_LSTATUS;
781 /* If reporting, report that either there's no ndo_eth_ioctl,
782 * or both SIOCGMIIREG and get_link failed (meaning that we
783 * cannot report link status). If not reporting, pretend
786 return reporting ? -1 : BMSR_LSTATUS;
789 /*----------------------------- Multicast list ------------------------------*/
791 /* Push the promiscuity flag down to appropriate slaves */
792 static int bond_set_promiscuity(struct bonding *bond, int inc)
794 struct list_head *iter;
797 if (bond_uses_primary(bond)) {
798 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
801 err = dev_set_promiscuity(curr_active->dev, inc);
805 bond_for_each_slave(bond, slave, iter) {
806 err = dev_set_promiscuity(slave->dev, inc);
814 /* Push the allmulti flag down to all slaves */
815 static int bond_set_allmulti(struct bonding *bond, int inc)
817 struct list_head *iter;
820 if (bond_uses_primary(bond)) {
821 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
824 err = dev_set_allmulti(curr_active->dev, inc);
828 bond_for_each_slave(bond, slave, iter) {
829 err = dev_set_allmulti(slave->dev, inc);
837 /* Retrieve the list of registered multicast addresses for the bonding
838 * device and retransmit an IGMP JOIN request to the current active
841 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
843 struct bonding *bond = container_of(work, struct bonding,
846 if (!rtnl_trylock()) {
847 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
850 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
852 if (bond->igmp_retrans > 1) {
853 bond->igmp_retrans--;
854 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
859 /* Flush bond's hardware addresses from slave */
860 static void bond_hw_addr_flush(struct net_device *bond_dev,
861 struct net_device *slave_dev)
863 struct bonding *bond = netdev_priv(bond_dev);
865 dev_uc_unsync(slave_dev, bond_dev);
866 dev_mc_unsync(slave_dev, bond_dev);
868 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
869 /* del lacpdu mc addr from mc list */
870 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
872 dev_mc_del(slave_dev, lacpdu_multicast);
876 /*--------------------------- Active slave change ---------------------------*/
878 /* Update the hardware address list and promisc/allmulti for the new and
879 * old active slaves (if any). Modes that are not using primary keep all
880 * slaves up date at all times; only the modes that use primary need to call
881 * this function to swap these settings during a failover.
883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
884 struct slave *old_active)
887 if (bond->dev->flags & IFF_PROMISC)
888 dev_set_promiscuity(old_active->dev, -1);
890 if (bond->dev->flags & IFF_ALLMULTI)
891 dev_set_allmulti(old_active->dev, -1);
893 bond_hw_addr_flush(bond->dev, old_active->dev);
897 /* FIXME: Signal errors upstream. */
898 if (bond->dev->flags & IFF_PROMISC)
899 dev_set_promiscuity(new_active->dev, 1);
901 if (bond->dev->flags & IFF_ALLMULTI)
902 dev_set_allmulti(new_active->dev, 1);
904 netif_addr_lock_bh(bond->dev);
905 dev_uc_sync(new_active->dev, bond->dev);
906 dev_mc_sync(new_active->dev, bond->dev);
907 netif_addr_unlock_bh(bond->dev);
912 * bond_set_dev_addr - clone slave's address to bond
913 * @bond_dev: bond net device
914 * @slave_dev: slave net device
916 * Should be called with RTNL held.
918 static int bond_set_dev_addr(struct net_device *bond_dev,
919 struct net_device *slave_dev)
923 slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
924 bond_dev, slave_dev, slave_dev->addr_len);
925 err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
929 __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
930 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
931 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
935 static struct slave *bond_get_old_active(struct bonding *bond,
936 struct slave *new_active)
939 struct list_head *iter;
941 bond_for_each_slave(bond, slave, iter) {
942 if (slave == new_active)
945 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
952 /* bond_do_fail_over_mac
954 * Perform special MAC address swapping for fail_over_mac settings
958 static void bond_do_fail_over_mac(struct bonding *bond,
959 struct slave *new_active,
960 struct slave *old_active)
962 u8 tmp_mac[MAX_ADDR_LEN];
963 struct sockaddr_storage ss;
966 switch (bond->params.fail_over_mac) {
967 case BOND_FOM_ACTIVE:
969 rv = bond_set_dev_addr(bond->dev, new_active->dev);
971 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
975 case BOND_FOM_FOLLOW:
976 /* if new_active && old_active, swap them
977 * if just old_active, do nothing (going to no active slave)
978 * if just new_active, set new_active to bond's MAC
984 old_active = bond_get_old_active(bond, new_active);
987 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
988 new_active->dev->addr_len);
989 bond_hw_addr_copy(ss.__data,
990 old_active->dev->dev_addr,
991 old_active->dev->addr_len);
992 ss.ss_family = new_active->dev->type;
994 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
995 bond->dev->addr_len);
996 ss.ss_family = bond->dev->type;
999 rv = dev_set_mac_address(new_active->dev,
1000 (struct sockaddr *)&ss, NULL);
1002 slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1010 bond_hw_addr_copy(ss.__data, tmp_mac,
1011 new_active->dev->addr_len);
1012 ss.ss_family = old_active->dev->type;
1014 rv = dev_set_mac_address(old_active->dev,
1015 (struct sockaddr *)&ss, NULL);
1017 slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1022 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1023 bond->params.fail_over_mac);
1029 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1031 struct slave *prim = rtnl_dereference(bond->primary_slave);
1032 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1034 if (!prim || prim->link != BOND_LINK_UP) {
1035 if (!curr || curr->link != BOND_LINK_UP)
1040 if (bond->force_primary) {
1041 bond->force_primary = false;
1045 if (!curr || curr->link != BOND_LINK_UP)
1048 /* At this point, prim and curr are both up */
1049 switch (bond->params.primary_reselect) {
1050 case BOND_PRI_RESELECT_ALWAYS:
1052 case BOND_PRI_RESELECT_BETTER:
1053 if (prim->speed < curr->speed)
1055 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1058 case BOND_PRI_RESELECT_FAILURE:
1061 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1062 bond->params.primary_reselect);
1068 * bond_find_best_slave - select the best available slave to be the active one
1069 * @bond: our bonding struct
1071 static struct slave *bond_find_best_slave(struct bonding *bond)
1073 struct slave *slave, *bestslave = NULL;
1074 struct list_head *iter;
1075 int mintime = bond->params.updelay;
1077 slave = bond_choose_primary_or_current(bond);
1081 bond_for_each_slave(bond, slave, iter) {
1082 if (slave->link == BOND_LINK_UP)
1084 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1085 slave->delay < mintime) {
1086 mintime = slave->delay;
1094 static bool bond_should_notify_peers(struct bonding *bond)
1096 struct slave *slave;
1099 slave = rcu_dereference(bond->curr_active_slave);
1102 if (!slave || !bond->send_peer_notif ||
1103 bond->send_peer_notif %
1104 max(1, bond->params.peer_notif_delay) != 0 ||
1105 !netif_carrier_ok(bond->dev) ||
1106 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1109 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1110 slave ? slave->dev->name : "NULL");
1116 * bond_change_active_slave - change the active slave into the specified one
1117 * @bond: our bonding struct
1118 * @new_active: the new slave to make the active one
1120 * Set the new slave to the bond's settings and unset them on the old
1121 * curr_active_slave.
1122 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1124 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1125 * because it is apparently the best available slave we have, even though its
1126 * updelay hasn't timed out yet.
1128 * Caller must hold RTNL.
1130 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1132 struct slave *old_active;
1136 old_active = rtnl_dereference(bond->curr_active_slave);
1138 if (old_active == new_active)
1141 #ifdef CONFIG_XFRM_OFFLOAD
1142 bond_ipsec_del_sa_all(bond);
1143 #endif /* CONFIG_XFRM_OFFLOAD */
1146 new_active->last_link_up = jiffies;
1148 if (new_active->link == BOND_LINK_BACK) {
1149 if (bond_uses_primary(bond)) {
1150 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1151 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1154 new_active->delay = 0;
1155 bond_set_slave_link_state(new_active, BOND_LINK_UP,
1156 BOND_SLAVE_NOTIFY_NOW);
1158 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1159 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1161 if (bond_is_lb(bond))
1162 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1164 if (bond_uses_primary(bond))
1165 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1169 if (bond_uses_primary(bond))
1170 bond_hw_addr_swap(bond, new_active, old_active);
1172 if (bond_is_lb(bond)) {
1173 bond_alb_handle_active_change(bond, new_active);
1175 bond_set_slave_inactive_flags(old_active,
1176 BOND_SLAVE_NOTIFY_NOW);
1178 bond_set_slave_active_flags(new_active,
1179 BOND_SLAVE_NOTIFY_NOW);
1181 rcu_assign_pointer(bond->curr_active_slave, new_active);
1184 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1186 bond_set_slave_inactive_flags(old_active,
1187 BOND_SLAVE_NOTIFY_NOW);
1190 bool should_notify_peers = false;
1192 bond_set_slave_active_flags(new_active,
1193 BOND_SLAVE_NOTIFY_NOW);
1195 if (bond->params.fail_over_mac)
1196 bond_do_fail_over_mac(bond, new_active,
1199 if (netif_running(bond->dev)) {
1200 bond->send_peer_notif =
1201 bond->params.num_peer_notif *
1202 max(1, bond->params.peer_notif_delay);
1203 should_notify_peers =
1204 bond_should_notify_peers(bond);
1207 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1208 if (should_notify_peers) {
1209 bond->send_peer_notif--;
1210 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1216 #ifdef CONFIG_XFRM_OFFLOAD
1217 bond_ipsec_add_sa_all(bond);
1218 #endif /* CONFIG_XFRM_OFFLOAD */
1220 /* resend IGMP joins since active slave has changed or
1221 * all were sent on curr_active_slave.
1222 * resend only if bond is brought up with the affected
1223 * bonding modes and the retransmission is enabled
1225 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1226 ((bond_uses_primary(bond) && new_active) ||
1227 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1228 bond->igmp_retrans = bond->params.resend_igmp;
1229 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1234 * bond_select_active_slave - select a new active slave, if needed
1235 * @bond: our bonding struct
1237 * This functions should be called when one of the following occurs:
1238 * - The old curr_active_slave has been released or lost its link.
1239 * - The primary_slave has got its link back.
1240 * - A slave has got its link back and there's no old curr_active_slave.
1242 * Caller must hold RTNL.
1244 void bond_select_active_slave(struct bonding *bond)
1246 struct slave *best_slave;
1251 best_slave = bond_find_best_slave(bond);
1252 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1253 bond_change_active_slave(bond, best_slave);
1254 rv = bond_set_carrier(bond);
1258 if (netif_carrier_ok(bond->dev))
1259 netdev_info(bond->dev, "active interface up!\n");
1261 netdev_info(bond->dev, "now running without any active interface!\n");
1265 #ifdef CONFIG_NET_POLL_CONTROLLER
1266 static inline int slave_enable_netpoll(struct slave *slave)
1271 np = kzalloc(sizeof(*np), GFP_KERNEL);
1276 err = __netpoll_setup(np, slave->dev);
1285 static inline void slave_disable_netpoll(struct slave *slave)
1287 struct netpoll *np = slave->np;
1297 static void bond_poll_controller(struct net_device *bond_dev)
1299 struct bonding *bond = netdev_priv(bond_dev);
1300 struct slave *slave = NULL;
1301 struct list_head *iter;
1302 struct ad_info ad_info;
1304 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1305 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1308 bond_for_each_slave_rcu(bond, slave, iter) {
1309 if (!bond_slave_is_up(slave))
1312 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1313 struct aggregator *agg =
1314 SLAVE_AD_INFO(slave)->port.aggregator;
1317 agg->aggregator_identifier != ad_info.aggregator_id)
1321 netpoll_poll_dev(slave->dev);
1325 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1327 struct bonding *bond = netdev_priv(bond_dev);
1328 struct list_head *iter;
1329 struct slave *slave;
1331 bond_for_each_slave(bond, slave, iter)
1332 if (bond_slave_is_up(slave))
1333 slave_disable_netpoll(slave);
1336 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1338 struct bonding *bond = netdev_priv(dev);
1339 struct list_head *iter;
1340 struct slave *slave;
1343 bond_for_each_slave(bond, slave, iter) {
1344 err = slave_enable_netpoll(slave);
1346 bond_netpoll_cleanup(dev);
1353 static inline int slave_enable_netpoll(struct slave *slave)
1357 static inline void slave_disable_netpoll(struct slave *slave)
1360 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1365 /*---------------------------------- IOCTL ----------------------------------*/
1367 static netdev_features_t bond_fix_features(struct net_device *dev,
1368 netdev_features_t features)
1370 struct bonding *bond = netdev_priv(dev);
1371 struct list_head *iter;
1372 netdev_features_t mask;
1373 struct slave *slave;
1375 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1376 if (bond_sk_check(bond))
1377 features |= BOND_TLS_FEATURES;
1379 features &= ~BOND_TLS_FEATURES;
1384 features &= ~NETIF_F_ONE_FOR_ALL;
1385 features |= NETIF_F_ALL_FOR_ALL;
1387 bond_for_each_slave(bond, slave, iter) {
1388 features = netdev_increment_features(features,
1389 slave->dev->features,
1392 features = netdev_add_tso_features(features, mask);
1397 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1398 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1399 NETIF_F_HIGHDMA | NETIF_F_LRO)
1401 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1402 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1404 #define BOND_MPLS_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1405 NETIF_F_GSO_SOFTWARE)
1408 static void bond_compute_features(struct bonding *bond)
1410 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1411 IFF_XMIT_DST_RELEASE_PERM;
1412 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1413 netdev_features_t enc_features = BOND_ENC_FEATURES;
1414 #ifdef CONFIG_XFRM_OFFLOAD
1415 netdev_features_t xfrm_features = BOND_XFRM_FEATURES;
1416 #endif /* CONFIG_XFRM_OFFLOAD */
1417 netdev_features_t mpls_features = BOND_MPLS_FEATURES;
1418 struct net_device *bond_dev = bond->dev;
1419 struct list_head *iter;
1420 struct slave *slave;
1421 unsigned short max_hard_header_len = ETH_HLEN;
1422 unsigned int tso_max_size = TSO_MAX_SIZE;
1423 u16 tso_max_segs = TSO_MAX_SEGS;
1425 if (!bond_has_slaves(bond))
1427 vlan_features &= NETIF_F_ALL_FOR_ALL;
1428 mpls_features &= NETIF_F_ALL_FOR_ALL;
1430 bond_for_each_slave(bond, slave, iter) {
1431 vlan_features = netdev_increment_features(vlan_features,
1432 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1434 enc_features = netdev_increment_features(enc_features,
1435 slave->dev->hw_enc_features,
1438 #ifdef CONFIG_XFRM_OFFLOAD
1439 xfrm_features = netdev_increment_features(xfrm_features,
1440 slave->dev->hw_enc_features,
1441 BOND_XFRM_FEATURES);
1442 #endif /* CONFIG_XFRM_OFFLOAD */
1444 mpls_features = netdev_increment_features(mpls_features,
1445 slave->dev->mpls_features,
1446 BOND_MPLS_FEATURES);
1448 dst_release_flag &= slave->dev->priv_flags;
1449 if (slave->dev->hard_header_len > max_hard_header_len)
1450 max_hard_header_len = slave->dev->hard_header_len;
1452 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1453 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1455 bond_dev->hard_header_len = max_hard_header_len;
1458 bond_dev->vlan_features = vlan_features;
1459 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1460 NETIF_F_HW_VLAN_CTAG_TX |
1461 NETIF_F_HW_VLAN_STAG_TX;
1462 #ifdef CONFIG_XFRM_OFFLOAD
1463 bond_dev->hw_enc_features |= xfrm_features;
1464 #endif /* CONFIG_XFRM_OFFLOAD */
1465 bond_dev->mpls_features = mpls_features;
1466 netif_set_tso_max_segs(bond_dev, tso_max_segs);
1467 netif_set_tso_max_size(bond_dev, tso_max_size);
1469 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1470 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1471 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1472 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1474 netdev_change_features(bond_dev);
1477 static void bond_setup_by_slave(struct net_device *bond_dev,
1478 struct net_device *slave_dev)
1480 bond_dev->header_ops = slave_dev->header_ops;
1482 bond_dev->type = slave_dev->type;
1483 bond_dev->hard_header_len = slave_dev->hard_header_len;
1484 bond_dev->needed_headroom = slave_dev->needed_headroom;
1485 bond_dev->addr_len = slave_dev->addr_len;
1487 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1488 slave_dev->addr_len);
1491 /* On bonding slaves other than the currently active slave, suppress
1492 * duplicates except for alb non-mcast/bcast.
1494 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1495 struct slave *slave,
1496 struct bonding *bond)
1498 if (bond_is_slave_inactive(slave)) {
1499 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1500 skb->pkt_type != PACKET_BROADCAST &&
1501 skb->pkt_type != PACKET_MULTICAST)
1508 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1510 struct sk_buff *skb = *pskb;
1511 struct slave *slave;
1512 struct bonding *bond;
1513 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1515 int ret = RX_HANDLER_ANOTHER;
1517 skb = skb_share_check(skb, GFP_ATOMIC);
1519 return RX_HANDLER_CONSUMED;
1523 slave = bond_slave_get_rcu(skb->dev);
1526 recv_probe = READ_ONCE(bond->recv_probe);
1528 ret = recv_probe(skb, bond, slave);
1529 if (ret == RX_HANDLER_CONSUMED) {
1536 * For packets determined by bond_should_deliver_exact_match() call to
1537 * be suppressed we want to make an exception for link-local packets.
1538 * This is necessary for e.g. LLDP daemons to be able to monitor
1539 * inactive slave links without being forced to bind to them
1542 * At the same time, packets that are passed to the bonding master
1543 * (including link-local ones) can have their originating interface
1544 * determined via PACKET_ORIGDEV socket option.
1546 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1547 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1548 return RX_HANDLER_PASS;
1549 return RX_HANDLER_EXACT;
1552 skb->dev = bond->dev;
1554 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1555 netif_is_bridge_port(bond->dev) &&
1556 skb->pkt_type == PACKET_HOST) {
1558 if (unlikely(skb_cow_head(skb,
1559 skb->data - skb_mac_header(skb)))) {
1561 return RX_HANDLER_CONSUMED;
1563 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1564 bond->dev->addr_len);
1570 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1572 switch (BOND_MODE(bond)) {
1573 case BOND_MODE_ROUNDROBIN:
1574 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1575 case BOND_MODE_ACTIVEBACKUP:
1576 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1577 case BOND_MODE_BROADCAST:
1578 return NETDEV_LAG_TX_TYPE_BROADCAST;
1580 case BOND_MODE_8023AD:
1581 return NETDEV_LAG_TX_TYPE_HASH;
1583 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1587 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1588 enum netdev_lag_tx_type type)
1590 if (type != NETDEV_LAG_TX_TYPE_HASH)
1591 return NETDEV_LAG_HASH_NONE;
1593 switch (bond->params.xmit_policy) {
1594 case BOND_XMIT_POLICY_LAYER2:
1595 return NETDEV_LAG_HASH_L2;
1596 case BOND_XMIT_POLICY_LAYER34:
1597 return NETDEV_LAG_HASH_L34;
1598 case BOND_XMIT_POLICY_LAYER23:
1599 return NETDEV_LAG_HASH_L23;
1600 case BOND_XMIT_POLICY_ENCAP23:
1601 return NETDEV_LAG_HASH_E23;
1602 case BOND_XMIT_POLICY_ENCAP34:
1603 return NETDEV_LAG_HASH_E34;
1604 case BOND_XMIT_POLICY_VLAN_SRCMAC:
1605 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1607 return NETDEV_LAG_HASH_UNKNOWN;
1611 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1612 struct netlink_ext_ack *extack)
1614 struct netdev_lag_upper_info lag_upper_info;
1615 enum netdev_lag_tx_type type;
1617 type = bond_lag_tx_type(bond);
1618 lag_upper_info.tx_type = type;
1619 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1621 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1622 &lag_upper_info, extack);
1625 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1627 netdev_upper_dev_unlink(slave->dev, bond->dev);
1628 slave->dev->flags &= ~IFF_SLAVE;
1631 static void slave_kobj_release(struct kobject *kobj)
1633 struct slave *slave = to_slave(kobj);
1634 struct bonding *bond = bond_get_bond_by_slave(slave);
1636 cancel_delayed_work_sync(&slave->notify_work);
1637 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1638 kfree(SLAVE_AD_INFO(slave));
1643 static struct kobj_type slave_ktype = {
1644 .release = slave_kobj_release,
1646 .sysfs_ops = &slave_sysfs_ops,
1650 static int bond_kobj_init(struct slave *slave)
1654 err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1655 &(slave->dev->dev.kobj), "bonding_slave");
1657 kobject_put(&slave->kobj);
1662 static struct slave *bond_alloc_slave(struct bonding *bond,
1663 struct net_device *slave_dev)
1665 struct slave *slave = NULL;
1667 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1672 slave->dev = slave_dev;
1673 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1675 if (bond_kobj_init(slave))
1678 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1679 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1681 if (!SLAVE_AD_INFO(slave)) {
1682 kobject_put(&slave->kobj);
1690 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1692 info->bond_mode = BOND_MODE(bond);
1693 info->miimon = bond->params.miimon;
1694 info->num_slaves = bond->slave_cnt;
1697 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1699 strcpy(info->slave_name, slave->dev->name);
1700 info->link = slave->link;
1701 info->state = bond_slave_state(slave);
1702 info->link_failure_count = slave->link_failure_count;
1705 static void bond_netdev_notify_work(struct work_struct *_work)
1707 struct slave *slave = container_of(_work, struct slave,
1710 if (rtnl_trylock()) {
1711 struct netdev_bonding_info binfo;
1713 bond_fill_ifslave(slave, &binfo.slave);
1714 bond_fill_ifbond(slave->bond, &binfo.master);
1715 netdev_bonding_info_change(slave->dev, &binfo);
1718 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1722 void bond_queue_slave_event(struct slave *slave)
1724 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1727 void bond_lower_state_changed(struct slave *slave)
1729 struct netdev_lag_lower_state_info info;
1731 info.link_up = slave->link == BOND_LINK_UP ||
1732 slave->link == BOND_LINK_FAIL;
1733 info.tx_enabled = bond_is_active_slave(slave);
1734 netdev_lower_state_changed(slave->dev, &info);
1737 #define BOND_NL_ERR(bond_dev, extack, errmsg) do { \
1739 NL_SET_ERR_MSG(extack, errmsg); \
1741 netdev_err(bond_dev, "Error: %s\n", errmsg); \
1744 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do { \
1746 NL_SET_ERR_MSG(extack, errmsg); \
1748 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg); \
1751 /* enslave device <slave> to bond device <master> */
1752 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1753 struct netlink_ext_ack *extack)
1755 struct bonding *bond = netdev_priv(bond_dev);
1756 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1757 struct slave *new_slave = NULL, *prev_slave;
1758 struct sockaddr_storage ss;
1762 if (slave_dev->flags & IFF_MASTER &&
1763 !netif_is_bond_master(slave_dev)) {
1764 BOND_NL_ERR(bond_dev, extack,
1765 "Device type (master device) cannot be enslaved");
1769 if (!bond->params.use_carrier &&
1770 slave_dev->ethtool_ops->get_link == NULL &&
1771 slave_ops->ndo_eth_ioctl == NULL) {
1772 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1775 /* already in-use? */
1776 if (netdev_is_rx_handler_busy(slave_dev)) {
1777 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1778 "Device is in use and cannot be enslaved");
1782 if (bond_dev == slave_dev) {
1783 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1787 /* vlan challenged mutual exclusion */
1788 /* no need to lock since we're protected by rtnl_lock */
1789 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1790 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1791 if (vlan_uses_dev(bond_dev)) {
1792 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1793 "Can not enslave VLAN challenged device to VLAN enabled bond");
1796 slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1799 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1802 if (slave_dev->features & NETIF_F_HW_ESP)
1803 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1805 /* Old ifenslave binaries are no longer supported. These can
1806 * be identified with moderate accuracy by the state of the slave:
1807 * the current ifenslave will set the interface down prior to
1808 * enslaving it; the old ifenslave will not.
1810 if (slave_dev->flags & IFF_UP) {
1811 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1812 "Device can not be enslaved while up");
1816 /* set bonding device ether type by slave - bonding netdevices are
1817 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1818 * there is a need to override some of the type dependent attribs/funcs.
1820 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1821 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1823 if (!bond_has_slaves(bond)) {
1824 if (bond_dev->type != slave_dev->type) {
1825 slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1826 bond_dev->type, slave_dev->type);
1828 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1830 res = notifier_to_errno(res);
1832 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1836 /* Flush unicast and multicast addresses */
1837 dev_uc_flush(bond_dev);
1838 dev_mc_flush(bond_dev);
1840 if (slave_dev->type != ARPHRD_ETHER)
1841 bond_setup_by_slave(bond_dev, slave_dev);
1843 ether_setup(bond_dev);
1844 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1847 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1850 } else if (bond_dev->type != slave_dev->type) {
1851 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1852 "Device type is different from other slaves");
1856 if (slave_dev->type == ARPHRD_INFINIBAND &&
1857 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1858 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1859 "Only active-backup mode is supported for infiniband slaves");
1861 goto err_undo_flags;
1864 if (!slave_ops->ndo_set_mac_address ||
1865 slave_dev->type == ARPHRD_INFINIBAND) {
1866 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1867 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1868 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1869 if (!bond_has_slaves(bond)) {
1870 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1871 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1873 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1874 "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1876 goto err_undo_flags;
1881 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1883 /* If this is the first slave, then we need to set the master's hardware
1884 * address to be the same as the slave's.
1886 if (!bond_has_slaves(bond) &&
1887 bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1888 res = bond_set_dev_addr(bond->dev, slave_dev);
1890 goto err_undo_flags;
1893 new_slave = bond_alloc_slave(bond, slave_dev);
1896 goto err_undo_flags;
1899 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1900 * is set via sysfs or module option if desired.
1902 new_slave->queue_id = 0;
1904 /* Save slave's original mtu and then set it to match the bond */
1905 new_slave->original_mtu = slave_dev->mtu;
1906 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1908 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1912 /* Save slave's original ("permanent") mac address for modes
1913 * that need it, and for restoring it upon release, and then
1914 * set it to the master's address
1916 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1917 slave_dev->addr_len);
1919 if (!bond->params.fail_over_mac ||
1920 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1921 /* Set slave to master's mac address. The application already
1922 * set the master's mac address to that of the first slave
1924 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1925 ss.ss_family = slave_dev->type;
1926 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1929 slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1930 goto err_restore_mtu;
1934 /* set slave flag before open to prevent IPv6 addrconf */
1935 slave_dev->flags |= IFF_SLAVE;
1937 /* open the slave since the application closed it */
1938 res = dev_open(slave_dev, extack);
1940 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1941 goto err_restore_mac;
1944 slave_dev->priv_flags |= IFF_BONDING;
1945 /* initialize slave stats */
1946 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1948 if (bond_is_lb(bond)) {
1949 /* bond_alb_init_slave() must be called before all other stages since
1950 * it might fail and we do not want to have to undo everything
1952 res = bond_alb_init_slave(bond, new_slave);
1957 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1959 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1963 prev_slave = bond_last_slave(bond);
1965 new_slave->delay = 0;
1966 new_slave->link_failure_count = 0;
1968 if (bond_update_speed_duplex(new_slave) &&
1969 bond_needs_speed_duplex(bond))
1970 new_slave->link = BOND_LINK_DOWN;
1972 new_slave->last_rx = jiffies -
1973 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1974 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1975 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1977 if (bond->params.miimon && !bond->params.use_carrier) {
1978 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1980 if ((link_reporting == -1) && !bond->params.arp_interval) {
1981 /* miimon is set but a bonded network driver
1982 * does not support ETHTOOL/MII and
1983 * arp_interval is not set. Note: if
1984 * use_carrier is enabled, we will never go
1985 * here (because netif_carrier is always
1986 * supported); thus, we don't need to change
1987 * the messages for netif_carrier.
1989 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");
1990 } else if (link_reporting == -1) {
1991 /* unable get link status using mii/ethtool */
1992 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");
1996 /* check for initial state */
1997 new_slave->link = BOND_LINK_NOCHANGE;
1998 if (bond->params.miimon) {
1999 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2000 if (bond->params.updelay) {
2001 bond_set_slave_link_state(new_slave,
2003 BOND_SLAVE_NOTIFY_NOW);
2004 new_slave->delay = bond->params.updelay;
2006 bond_set_slave_link_state(new_slave,
2008 BOND_SLAVE_NOTIFY_NOW);
2011 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2012 BOND_SLAVE_NOTIFY_NOW);
2014 } else if (bond->params.arp_interval) {
2015 bond_set_slave_link_state(new_slave,
2016 (netif_carrier_ok(slave_dev) ?
2017 BOND_LINK_UP : BOND_LINK_DOWN),
2018 BOND_SLAVE_NOTIFY_NOW);
2020 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2021 BOND_SLAVE_NOTIFY_NOW);
2024 if (new_slave->link != BOND_LINK_DOWN)
2025 new_slave->last_link_up = jiffies;
2026 slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2027 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2028 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2030 if (bond_uses_primary(bond) && bond->params.primary[0]) {
2031 /* if there is a primary slave, remember it */
2032 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2033 rcu_assign_pointer(bond->primary_slave, new_slave);
2034 bond->force_primary = true;
2038 switch (BOND_MODE(bond)) {
2039 case BOND_MODE_ACTIVEBACKUP:
2040 bond_set_slave_inactive_flags(new_slave,
2041 BOND_SLAVE_NOTIFY_NOW);
2043 case BOND_MODE_8023AD:
2044 /* in 802.3ad mode, the internal mechanism
2045 * will activate the slaves in the selected
2048 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2049 /* if this is the first slave */
2051 SLAVE_AD_INFO(new_slave)->id = 1;
2052 /* Initialize AD with the number of times that the AD timer is called in 1 second
2053 * can be called only after the mac address of the bond is set
2055 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2057 SLAVE_AD_INFO(new_slave)->id =
2058 SLAVE_AD_INFO(prev_slave)->id + 1;
2061 bond_3ad_bind_slave(new_slave);
2065 bond_set_active_slave(new_slave);
2066 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2069 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2071 /* always active in trunk mode */
2072 bond_set_active_slave(new_slave);
2074 /* In trunking mode there is little meaning to curr_active_slave
2075 * anyway (it holds no special properties of the bond device),
2076 * so we can change it without calling change_active_interface()
2078 if (!rcu_access_pointer(bond->curr_active_slave) &&
2079 new_slave->link == BOND_LINK_UP)
2080 rcu_assign_pointer(bond->curr_active_slave, new_slave);
2083 } /* switch(bond_mode) */
2085 #ifdef CONFIG_NET_POLL_CONTROLLER
2086 if (bond->dev->npinfo) {
2087 if (slave_enable_netpoll(new_slave)) {
2088 slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2095 if (!(bond_dev->features & NETIF_F_LRO))
2096 dev_disable_lro(slave_dev);
2098 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2101 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2105 res = bond_master_upper_dev_link(bond, new_slave, extack);
2107 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2108 goto err_unregister;
2111 bond_lower_state_changed(new_slave);
2113 res = bond_sysfs_slave_add(new_slave);
2115 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2116 goto err_upper_unlink;
2119 /* If the mode uses primary, then the following is handled by
2120 * bond_change_active_slave().
2122 if (!bond_uses_primary(bond)) {
2123 /* set promiscuity level to new slave */
2124 if (bond_dev->flags & IFF_PROMISC) {
2125 res = dev_set_promiscuity(slave_dev, 1);
2130 /* set allmulti level to new slave */
2131 if (bond_dev->flags & IFF_ALLMULTI) {
2132 res = dev_set_allmulti(slave_dev, 1);
2134 if (bond_dev->flags & IFF_PROMISC)
2135 dev_set_promiscuity(slave_dev, -1);
2140 netif_addr_lock_bh(bond_dev);
2141 dev_mc_sync_multiple(slave_dev, bond_dev);
2142 dev_uc_sync_multiple(slave_dev, bond_dev);
2143 netif_addr_unlock_bh(bond_dev);
2145 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2146 /* add lacpdu mc addr to mc list */
2147 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2149 dev_mc_add(slave_dev, lacpdu_multicast);
2154 bond_compute_features(bond);
2155 bond_set_carrier(bond);
2157 if (bond_uses_primary(bond)) {
2159 bond_select_active_slave(bond);
2160 unblock_netpoll_tx();
2163 if (bond_mode_can_use_xmit_hash(bond))
2164 bond_update_slave_arr(bond, NULL);
2167 if (!slave_dev->netdev_ops->ndo_bpf ||
2168 !slave_dev->netdev_ops->ndo_xdp_xmit) {
2169 if (bond->xdp_prog) {
2170 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2171 "Slave does not support XDP");
2175 } else if (bond->xdp_prog) {
2176 struct netdev_bpf xdp = {
2177 .command = XDP_SETUP_PROG,
2179 .prog = bond->xdp_prog,
2183 if (dev_xdp_prog_count(slave_dev) > 0) {
2184 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2185 "Slave has XDP program loaded, please unload before enslaving");
2190 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2192 /* ndo_bpf() sets extack error message */
2193 slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2197 bpf_prog_inc(bond->xdp_prog);
2200 slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2201 bond_is_active_slave(new_slave) ? "an active" : "a backup",
2202 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2204 /* enslave is successful */
2205 bond_queue_slave_event(new_slave);
2208 /* Undo stages on error */
2210 bond_sysfs_slave_del(new_slave);
2213 bond_upper_dev_unlink(bond, new_slave);
2216 netdev_rx_handler_unregister(slave_dev);
2219 vlan_vids_del_by_dev(slave_dev, bond_dev);
2220 if (rcu_access_pointer(bond->primary_slave) == new_slave)
2221 RCU_INIT_POINTER(bond->primary_slave, NULL);
2222 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2224 bond_change_active_slave(bond, NULL);
2225 bond_select_active_slave(bond);
2226 unblock_netpoll_tx();
2228 /* either primary_slave or curr_active_slave might've changed */
2230 slave_disable_netpoll(new_slave);
2233 if (!netif_is_bond_master(slave_dev))
2234 slave_dev->priv_flags &= ~IFF_BONDING;
2235 dev_close(slave_dev);
2238 slave_dev->flags &= ~IFF_SLAVE;
2239 if (!bond->params.fail_over_mac ||
2240 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2241 /* XXX TODO - fom follow mode needs to change master's
2242 * MAC if this slave's MAC is in use by the bond, or at
2243 * least print a warning.
2245 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2246 new_slave->dev->addr_len);
2247 ss.ss_family = slave_dev->type;
2248 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2252 dev_set_mtu(slave_dev, new_slave->original_mtu);
2255 kobject_put(&new_slave->kobj);
2258 /* Enslave of first slave has failed and we need to fix master's mac */
2259 if (!bond_has_slaves(bond)) {
2260 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2261 slave_dev->dev_addr))
2262 eth_hw_addr_random(bond_dev);
2263 if (bond_dev->type != ARPHRD_ETHER) {
2264 dev_close(bond_dev);
2265 ether_setup(bond_dev);
2266 bond_dev->flags |= IFF_MASTER;
2267 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2274 /* Try to release the slave device <slave> from the bond device <master>
2275 * It is legal to access curr_active_slave without a lock because all the function
2276 * is RTNL-locked. If "all" is true it means that the function is being called
2277 * while destroying a bond interface and all slaves are being released.
2279 * The rules for slave state should be:
2280 * for Active/Backup:
2281 * Active stays on all backups go down
2282 * for Bonded connections:
2283 * The first up interface should be left on and all others downed.
2285 static int __bond_release_one(struct net_device *bond_dev,
2286 struct net_device *slave_dev,
2287 bool all, bool unregister)
2289 struct bonding *bond = netdev_priv(bond_dev);
2290 struct slave *slave, *oldcurrent;
2291 struct sockaddr_storage ss;
2292 int old_flags = bond_dev->flags;
2293 netdev_features_t old_features = bond_dev->features;
2295 /* slave is not a slave or master is not master of this slave */
2296 if (!(slave_dev->flags & IFF_SLAVE) ||
2297 !netdev_has_upper_dev(slave_dev, bond_dev)) {
2298 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2304 slave = bond_get_slave_by_dev(bond, slave_dev);
2306 /* not a slave of this bond */
2307 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2308 unblock_netpoll_tx();
2312 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2314 bond_sysfs_slave_del(slave);
2316 /* recompute stats just before removing the slave */
2317 bond_get_stats(bond->dev, &bond->bond_stats);
2319 if (bond->xdp_prog) {
2320 struct netdev_bpf xdp = {
2321 .command = XDP_SETUP_PROG,
2326 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2327 slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2330 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2331 * for this slave anymore.
2333 netdev_rx_handler_unregister(slave_dev);
2335 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2336 bond_3ad_unbind_slave(slave);
2338 bond_upper_dev_unlink(bond, slave);
2340 if (bond_mode_can_use_xmit_hash(bond))
2341 bond_update_slave_arr(bond, slave);
2343 slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2344 bond_is_active_slave(slave) ? "active" : "backup");
2346 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2348 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2350 if (!all && (!bond->params.fail_over_mac ||
2351 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2352 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2353 bond_has_slaves(bond))
2354 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",
2355 slave->perm_hwaddr);
2358 if (rtnl_dereference(bond->primary_slave) == slave)
2359 RCU_INIT_POINTER(bond->primary_slave, NULL);
2361 if (oldcurrent == slave)
2362 bond_change_active_slave(bond, NULL);
2364 if (bond_is_lb(bond)) {
2365 /* Must be called only after the slave has been
2366 * detached from the list and the curr_active_slave
2367 * has been cleared (if our_slave == old_current),
2368 * but before a new active slave is selected.
2370 bond_alb_deinit_slave(bond, slave);
2374 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2375 } else if (oldcurrent == slave) {
2376 /* Note that we hold RTNL over this sequence, so there
2377 * is no concern that another slave add/remove event
2380 bond_select_active_slave(bond);
2383 bond_set_carrier(bond);
2384 if (!bond_has_slaves(bond))
2385 eth_hw_addr_random(bond_dev);
2387 unblock_netpoll_tx();
2391 if (!bond_has_slaves(bond)) {
2392 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2393 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2396 bond_compute_features(bond);
2397 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2398 (old_features & NETIF_F_VLAN_CHALLENGED))
2399 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2401 vlan_vids_del_by_dev(slave_dev, bond_dev);
2403 /* If the mode uses primary, then this case was handled above by
2404 * bond_change_active_slave(..., NULL)
2406 if (!bond_uses_primary(bond)) {
2407 /* unset promiscuity level from slave
2408 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2409 * of the IFF_PROMISC flag in the bond_dev, but we need the
2410 * value of that flag before that change, as that was the value
2411 * when this slave was attached, so we cache at the start of the
2412 * function and use it here. Same goes for ALLMULTI below
2414 if (old_flags & IFF_PROMISC)
2415 dev_set_promiscuity(slave_dev, -1);
2417 /* unset allmulti level from slave */
2418 if (old_flags & IFF_ALLMULTI)
2419 dev_set_allmulti(slave_dev, -1);
2421 bond_hw_addr_flush(bond_dev, slave_dev);
2424 slave_disable_netpoll(slave);
2426 /* close slave before restoring its mac address */
2427 dev_close(slave_dev);
2429 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2430 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2431 /* restore original ("permanent") mac address */
2432 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2433 slave->dev->addr_len);
2434 ss.ss_family = slave_dev->type;
2435 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2439 __dev_set_mtu(slave_dev, slave->original_mtu);
2441 dev_set_mtu(slave_dev, slave->original_mtu);
2443 if (!netif_is_bond_master(slave_dev))
2444 slave_dev->priv_flags &= ~IFF_BONDING;
2446 kobject_put(&slave->kobj);
2451 /* A wrapper used because of ndo_del_link */
2452 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2454 return __bond_release_one(bond_dev, slave_dev, false, false);
2457 /* First release a slave and then destroy the bond if no more slaves are left.
2458 * Must be under rtnl_lock when this function is called.
2460 static int bond_release_and_destroy(struct net_device *bond_dev,
2461 struct net_device *slave_dev)
2463 struct bonding *bond = netdev_priv(bond_dev);
2466 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2467 if (ret == 0 && !bond_has_slaves(bond) &&
2468 bond_dev->reg_state != NETREG_UNREGISTERING) {
2469 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2470 netdev_info(bond_dev, "Destroying bond\n");
2471 bond_remove_proc_entry(bond);
2472 unregister_netdevice(bond_dev);
2477 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2479 struct bonding *bond = netdev_priv(bond_dev);
2481 bond_fill_ifbond(bond, info);
2484 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2486 struct bonding *bond = netdev_priv(bond_dev);
2487 struct list_head *iter;
2488 int i = 0, res = -ENODEV;
2489 struct slave *slave;
2491 bond_for_each_slave(bond, slave, iter) {
2492 if (i++ == (int)info->slave_id) {
2494 bond_fill_ifslave(slave, info);
2502 /*-------------------------------- Monitoring -------------------------------*/
2504 /* called with rcu_read_lock() */
2505 static int bond_miimon_inspect(struct bonding *bond)
2507 int link_state, commit = 0;
2508 struct list_head *iter;
2509 struct slave *slave;
2510 bool ignore_updelay;
2512 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2514 bond_for_each_slave_rcu(bond, slave, iter) {
2515 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2517 link_state = bond_check_dev_link(bond, slave->dev, 0);
2519 switch (slave->link) {
2524 bond_propose_link_state(slave, BOND_LINK_FAIL);
2526 slave->delay = bond->params.downdelay;
2528 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2530 BOND_MODE_ACTIVEBACKUP) ?
2531 (bond_is_active_slave(slave) ?
2532 "active " : "backup ") : "",
2533 bond->params.downdelay * bond->params.miimon);
2536 case BOND_LINK_FAIL:
2538 /* recovered before downdelay expired */
2539 bond_propose_link_state(slave, BOND_LINK_UP);
2540 slave->last_link_up = jiffies;
2541 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2542 (bond->params.downdelay - slave->delay) *
2543 bond->params.miimon);
2548 if (slave->delay <= 0) {
2549 bond_propose_link_state(slave, BOND_LINK_DOWN);
2557 case BOND_LINK_DOWN:
2561 bond_propose_link_state(slave, BOND_LINK_BACK);
2563 slave->delay = bond->params.updelay;
2566 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2567 ignore_updelay ? 0 :
2568 bond->params.updelay *
2569 bond->params.miimon);
2572 case BOND_LINK_BACK:
2574 bond_propose_link_state(slave, BOND_LINK_DOWN);
2575 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2576 (bond->params.updelay - slave->delay) *
2577 bond->params.miimon);
2585 if (slave->delay <= 0) {
2586 bond_propose_link_state(slave, BOND_LINK_UP);
2588 ignore_updelay = false;
2600 static void bond_miimon_link_change(struct bonding *bond,
2601 struct slave *slave,
2604 switch (BOND_MODE(bond)) {
2605 case BOND_MODE_8023AD:
2606 bond_3ad_handle_link_change(slave, link);
2610 bond_alb_handle_link_change(bond, slave, link);
2613 bond_update_slave_arr(bond, NULL);
2618 static void bond_miimon_commit(struct bonding *bond)
2620 struct list_head *iter;
2621 struct slave *slave, *primary;
2623 bond_for_each_slave(bond, slave, iter) {
2624 switch (slave->link_new_state) {
2625 case BOND_LINK_NOCHANGE:
2626 /* For 802.3ad mode, check current slave speed and
2627 * duplex again in case its port was disabled after
2628 * invalid speed/duplex reporting but recovered before
2629 * link monitoring could make a decision on the actual
2632 if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2633 slave->link == BOND_LINK_UP)
2634 bond_3ad_adapter_speed_duplex_changed(slave);
2638 if (bond_update_speed_duplex(slave) &&
2639 bond_needs_speed_duplex(bond)) {
2640 slave->link = BOND_LINK_DOWN;
2641 if (net_ratelimit())
2642 slave_warn(bond->dev, slave->dev,
2643 "failed to get link speed/duplex\n");
2646 bond_set_slave_link_state(slave, BOND_LINK_UP,
2647 BOND_SLAVE_NOTIFY_NOW);
2648 slave->last_link_up = jiffies;
2650 primary = rtnl_dereference(bond->primary_slave);
2651 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2652 /* prevent it from being the active one */
2653 bond_set_backup_slave(slave);
2654 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2655 /* make it immediately active */
2656 bond_set_active_slave(slave);
2659 slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2660 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2661 slave->duplex ? "full" : "half");
2663 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2665 if (!bond->curr_active_slave || slave == primary)
2670 case BOND_LINK_DOWN:
2671 if (slave->link_failure_count < UINT_MAX)
2672 slave->link_failure_count++;
2674 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2675 BOND_SLAVE_NOTIFY_NOW);
2677 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2678 BOND_MODE(bond) == BOND_MODE_8023AD)
2679 bond_set_slave_inactive_flags(slave,
2680 BOND_SLAVE_NOTIFY_NOW);
2682 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2684 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2686 if (slave == rcu_access_pointer(bond->curr_active_slave))
2692 slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2693 slave->link_new_state);
2694 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2701 bond_select_active_slave(bond);
2702 unblock_netpoll_tx();
2705 bond_set_carrier(bond);
2710 * Really a wrapper that splits the mii monitor into two phases: an
2711 * inspection, then (if inspection indicates something needs to be done)
2712 * an acquisition of appropriate locks followed by a commit phase to
2713 * implement whatever link state changes are indicated.
2715 static void bond_mii_monitor(struct work_struct *work)
2717 struct bonding *bond = container_of(work, struct bonding,
2719 bool should_notify_peers = false;
2721 unsigned long delay;
2722 struct slave *slave;
2723 struct list_head *iter;
2725 delay = msecs_to_jiffies(bond->params.miimon);
2727 if (!bond_has_slaves(bond))
2731 should_notify_peers = bond_should_notify_peers(bond);
2732 commit = !!bond_miimon_inspect(bond);
2733 if (bond->send_peer_notif) {
2735 if (rtnl_trylock()) {
2736 bond->send_peer_notif--;
2744 /* Race avoidance with bond_close cancel of workqueue */
2745 if (!rtnl_trylock()) {
2747 should_notify_peers = false;
2751 bond_for_each_slave(bond, slave, iter) {
2752 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2754 bond_miimon_commit(bond);
2756 rtnl_unlock(); /* might sleep, hold no other locks */
2760 if (bond->params.miimon)
2761 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2763 if (should_notify_peers) {
2764 if (!rtnl_trylock())
2766 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2771 static int bond_upper_dev_walk(struct net_device *upper,
2772 struct netdev_nested_priv *priv)
2774 __be32 ip = *(__be32 *)priv->data;
2776 return ip == bond_confirm_addr(upper, 0, ip);
2779 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2781 struct netdev_nested_priv priv = {
2782 .data = (void *)&ip,
2786 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2790 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2797 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2798 struct sk_buff *skb)
2800 struct net_device *bond_dev = slave->bond->dev;
2801 struct net_device *slave_dev = slave->dev;
2802 struct bond_vlan_tag *outer_tag = tags;
2804 if (!tags || tags->vlan_proto == VLAN_N_VID)
2809 /* Go through all the tags backwards and add them to the packet */
2810 while (tags->vlan_proto != VLAN_N_VID) {
2811 if (!tags->vlan_id) {
2816 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2817 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2818 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2821 net_err_ratelimited("failed to insert inner VLAN tag\n");
2827 /* Set the outer tag */
2828 if (outer_tag->vlan_id) {
2829 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2830 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2831 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2832 outer_tag->vlan_id);
2838 /* We go to the (large) trouble of VLAN tagging ARP frames because
2839 * switches in VLAN mode (especially if ports are configured as
2840 * "native" to a VLAN) might not pass non-tagged frames.
2842 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2843 __be32 src_ip, struct bond_vlan_tag *tags)
2845 struct net_device *bond_dev = slave->bond->dev;
2846 struct net_device *slave_dev = slave->dev;
2847 struct sk_buff *skb;
2849 slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2850 arp_op, &dest_ip, &src_ip);
2852 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2853 NULL, slave_dev->dev_addr, NULL);
2856 net_err_ratelimited("ARP packet allocation failed\n");
2860 if (bond_handle_vlan(slave, tags, skb))
2865 /* Validate the device path between the @start_dev and the @end_dev.
2866 * The path is valid if the @end_dev is reachable through device
2868 * When the path is validated, collect any vlan information in the
2871 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2872 struct net_device *end_dev,
2875 struct bond_vlan_tag *tags;
2876 struct net_device *upper;
2877 struct list_head *iter;
2879 if (start_dev == end_dev) {
2880 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2882 return ERR_PTR(-ENOMEM);
2883 tags[level].vlan_proto = VLAN_N_VID;
2887 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2888 tags = bond_verify_device_path(upper, end_dev, level + 1);
2889 if (IS_ERR_OR_NULL(tags)) {
2894 if (is_vlan_dev(upper)) {
2895 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2896 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2905 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2908 struct bond_vlan_tag *tags;
2909 __be32 *targets = bond->params.arp_targets, addr;
2912 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2913 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2914 __func__, &targets[i]);
2917 /* Find out through which dev should the packet go */
2918 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2921 /* there's no route to target - try to send arp
2922 * probe to generate any traffic (arp_validate=0)
2924 if (bond->params.arp_validate)
2925 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2928 bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2933 /* bond device itself */
2934 if (rt->dst.dev == bond->dev)
2938 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2941 if (!IS_ERR_OR_NULL(tags))
2944 /* Not our device - skip */
2945 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2946 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2952 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2954 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2959 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2963 if (!sip || !bond_has_this_ip(bond, tip)) {
2964 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2965 __func__, &sip, &tip);
2969 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2971 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2975 slave->last_rx = jiffies;
2976 slave->target_last_arp_rx[i] = jiffies;
2979 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2980 struct slave *slave)
2982 struct arphdr *arp = (struct arphdr *)skb->data;
2983 struct slave *curr_active_slave, *curr_arp_slave;
2984 unsigned char *arp_ptr;
2988 alen = arp_hdr_len(bond->dev);
2990 if (alen > skb_headlen(skb)) {
2991 arp = kmalloc(alen, GFP_ATOMIC);
2994 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2998 if (arp->ar_hln != bond->dev->addr_len ||
2999 skb->pkt_type == PACKET_OTHERHOST ||
3000 skb->pkt_type == PACKET_LOOPBACK ||
3001 arp->ar_hrd != htons(ARPHRD_ETHER) ||
3002 arp->ar_pro != htons(ETH_P_IP) ||
3006 arp_ptr = (unsigned char *)(arp + 1);
3007 arp_ptr += bond->dev->addr_len;
3008 memcpy(&sip, arp_ptr, 4);
3009 arp_ptr += 4 + bond->dev->addr_len;
3010 memcpy(&tip, arp_ptr, 4);
3012 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3013 __func__, slave->dev->name, bond_slave_state(slave),
3014 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3017 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3018 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3020 /* We 'trust' the received ARP enough to validate it if:
3022 * (a) the slave receiving the ARP is active (which includes the
3023 * current ARP slave, if any), or
3025 * (b) the receiving slave isn't active, but there is a currently
3026 * active slave and it received valid arp reply(s) after it became
3027 * the currently active slave, or
3029 * (c) there is an ARP slave that sent an ARP during the prior ARP
3030 * interval, and we receive an ARP reply on any slave. We accept
3031 * these because switch FDB update delays may deliver the ARP
3032 * reply to a slave other than the sender of the ARP request.
3034 * Note: for (b), backup slaves are receiving the broadcast ARP
3035 * request, not a reply. This request passes from the sending
3036 * slave through the L2 switch(es) to the receiving slave. Since
3037 * this is checking the request, sip/tip are swapped for
3040 * This is done to avoid endless looping when we can't reach the
3041 * arp_ip_target and fool ourselves with our own arp requests.
3043 if (bond_is_active_slave(slave))
3044 bond_validate_arp(bond, slave, sip, tip);
3045 else if (curr_active_slave &&
3046 time_after(slave_last_rx(bond, curr_active_slave),
3047 curr_active_slave->last_link_up))
3048 bond_validate_arp(bond, slave, tip, sip);
3049 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3050 bond_time_in_interval(bond,
3051 dev_trans_start(curr_arp_slave->dev), 1))
3052 bond_validate_arp(bond, slave, sip, tip);
3055 if (arp != (struct arphdr *)skb->data)
3057 return RX_HANDLER_ANOTHER;
3060 #if IS_ENABLED(CONFIG_IPV6)
3061 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3062 const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3064 struct net_device *bond_dev = slave->bond->dev;
3065 struct net_device *slave_dev = slave->dev;
3066 struct in6_addr mcaddr;
3067 struct sk_buff *skb;
3069 slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3072 skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3074 net_err_ratelimited("NS packet allocation failed\n");
3078 addrconf_addr_solict_mult(daddr, &mcaddr);
3079 if (bond_handle_vlan(slave, tags, skb))
3080 ndisc_send_skb(skb, &mcaddr, saddr);
3083 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3085 struct in6_addr *targets = bond->params.ns_targets;
3086 struct bond_vlan_tag *tags;
3087 struct dst_entry *dst;
3088 struct in6_addr saddr;
3092 for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3093 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3094 __func__, &targets[i]);
3097 /* Find out through which dev should the packet go */
3098 memset(&fl6, 0, sizeof(struct flowi6));
3099 fl6.daddr = targets[i];
3100 fl6.flowi6_oif = bond->dev->ifindex;
3102 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3105 /* there's no route to target - try to send arp
3106 * probe to generate any traffic (arp_validate=0)
3108 if (bond->params.arp_validate)
3109 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3112 bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3116 /* bond device itself */
3117 if (dst->dev == bond->dev)
3121 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3124 if (!IS_ERR_OR_NULL(tags))
3127 /* Not our device - skip */
3128 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3129 &targets[i], dst->dev ? dst->dev->name : "NULL");
3135 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3136 bond_ns_send(slave, &targets[i], &saddr, tags);
3142 static int bond_confirm_addr6(struct net_device *dev,
3143 struct netdev_nested_priv *priv)
3145 struct in6_addr *addr = (struct in6_addr *)priv->data;
3147 return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3150 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3152 struct netdev_nested_priv priv = {
3157 if (bond_confirm_addr6(bond->dev, &priv))
3161 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3168 static void bond_validate_ns(struct bonding *bond, struct slave *slave,
3169 struct in6_addr *saddr, struct in6_addr *daddr)
3173 if (ipv6_addr_any(saddr) || !bond_has_this_ip6(bond, daddr)) {
3174 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3175 __func__, saddr, daddr);
3179 i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3181 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3185 slave->last_rx = jiffies;
3186 slave->target_last_arp_rx[i] = jiffies;
3189 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3190 struct slave *slave)
3192 struct slave *curr_active_slave, *curr_arp_slave;
3193 struct icmp6hdr *hdr = icmp6_hdr(skb);
3194 struct in6_addr *saddr, *daddr;
3196 if (skb->pkt_type == PACKET_OTHERHOST ||
3197 skb->pkt_type == PACKET_LOOPBACK ||
3198 hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
3201 saddr = &ipv6_hdr(skb)->saddr;
3202 daddr = &ipv6_hdr(skb)->daddr;
3204 slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3205 __func__, slave->dev->name, bond_slave_state(slave),
3206 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3209 curr_active_slave = rcu_dereference(bond->curr_active_slave);
3210 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3212 /* We 'trust' the received ARP enough to validate it if:
3213 * see bond_arp_rcv().
3215 if (bond_is_active_slave(slave))
3216 bond_validate_ns(bond, slave, saddr, daddr);
3217 else if (curr_active_slave &&
3218 time_after(slave_last_rx(bond, curr_active_slave),
3219 curr_active_slave->last_link_up))
3220 bond_validate_ns(bond, slave, saddr, daddr);
3221 else if (curr_arp_slave &&
3222 bond_time_in_interval(bond,
3223 dev_trans_start(curr_arp_slave->dev), 1))
3224 bond_validate_ns(bond, slave, saddr, daddr);
3227 return RX_HANDLER_ANOTHER;
3231 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3232 struct slave *slave)
3234 #if IS_ENABLED(CONFIG_IPV6)
3235 bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3237 bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3239 slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3240 __func__, skb->dev->name);
3242 /* Use arp validate logic for both ARP and NS */
3243 if (!slave_do_arp_validate(bond, slave)) {
3244 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3245 #if IS_ENABLED(CONFIG_IPV6)
3246 (slave_do_arp_validate_only(bond) && is_ipv6) ||
3248 !slave_do_arp_validate_only(bond))
3249 slave->last_rx = jiffies;
3250 return RX_HANDLER_ANOTHER;
3251 } else if (is_arp) {
3252 return bond_arp_rcv(skb, bond, slave);
3253 #if IS_ENABLED(CONFIG_IPV6)
3254 } else if (is_ipv6) {
3255 return bond_na_rcv(skb, bond, slave);
3258 return RX_HANDLER_ANOTHER;
3262 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3264 bond_arp_send_all(bond, slave);
3265 #if IS_ENABLED(CONFIG_IPV6)
3266 bond_ns_send_all(bond, slave);
3270 /* function to verify if we're in the arp_interval timeslice, returns true if
3271 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3272 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3274 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3277 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3279 return time_in_range(jiffies,
3280 last_act - delta_in_ticks,
3281 last_act + mod * delta_in_ticks + delta_in_ticks/2);
3284 /* This function is called regularly to monitor each slave's link
3285 * ensuring that traffic is being sent and received when arp monitoring
3286 * is used in load-balancing mode. if the adapter has been dormant, then an
3287 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3288 * arp monitoring in active backup mode.
3290 static void bond_loadbalance_arp_mon(struct bonding *bond)
3292 struct slave *slave, *oldcurrent;
3293 struct list_head *iter;
3294 int do_failover = 0, slave_state_changed = 0;
3296 if (!bond_has_slaves(bond))
3301 oldcurrent = rcu_dereference(bond->curr_active_slave);
3302 /* see if any of the previous devices are up now (i.e. they have
3303 * xmt and rcv traffic). the curr_active_slave does not come into
3304 * the picture unless it is null. also, slave->last_link_up is not
3305 * needed here because we send an arp on each slave and give a slave
3306 * as long as it needs to get the tx/rx within the delta.
3307 * TODO: what about up/down delay in arp mode? it wasn't here before
3310 bond_for_each_slave_rcu(bond, slave, iter) {
3311 unsigned long trans_start = dev_trans_start(slave->dev);
3313 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3315 if (slave->link != BOND_LINK_UP) {
3316 if (bond_time_in_interval(bond, trans_start, 1) &&
3317 bond_time_in_interval(bond, slave->last_rx, 1)) {
3319 bond_propose_link_state(slave, BOND_LINK_UP);
3320 slave_state_changed = 1;
3322 /* primary_slave has no meaning in round-robin
3323 * mode. the window of a slave being up and
3324 * curr_active_slave being null after enslaving
3328 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3331 slave_info(bond->dev, slave->dev, "interface is now up\n");
3335 /* slave->link == BOND_LINK_UP */
3337 /* not all switches will respond to an arp request
3338 * when the source ip is 0, so don't take the link down
3339 * if we don't know our ip yet
3341 if (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
3342 !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3344 bond_propose_link_state(slave, BOND_LINK_DOWN);
3345 slave_state_changed = 1;
3347 if (slave->link_failure_count < UINT_MAX)
3348 slave->link_failure_count++;
3350 slave_info(bond->dev, slave->dev, "interface is now down\n");
3352 if (slave == oldcurrent)
3357 /* note: if switch is in round-robin mode, all links
3358 * must tx arp to ensure all links rx an arp - otherwise
3359 * links may oscillate or not come up at all; if switch is
3360 * in something like xor mode, there is nothing we can
3361 * do - all replies will be rx'ed on same link causing slaves
3362 * to be unstable during low/no traffic periods
3364 if (bond_slave_is_up(slave))
3365 bond_send_validate(bond, slave);
3370 if (do_failover || slave_state_changed) {
3371 if (!rtnl_trylock())
3374 bond_for_each_slave(bond, slave, iter) {
3375 if (slave->link_new_state != BOND_LINK_NOCHANGE)
3376 slave->link = slave->link_new_state;
3379 if (slave_state_changed) {
3380 bond_slave_state_change(bond);
3381 if (BOND_MODE(bond) == BOND_MODE_XOR)
3382 bond_update_slave_arr(bond, NULL);
3386 bond_select_active_slave(bond);
3387 unblock_netpoll_tx();
3393 if (bond->params.arp_interval)
3394 queue_delayed_work(bond->wq, &bond->arp_work,
3395 msecs_to_jiffies(bond->params.arp_interval));
3398 /* Called to inspect slaves for active-backup mode ARP monitor link state
3399 * changes. Sets proposed link state in slaves to specify what action
3400 * should take place for the slave. Returns 0 if no changes are found, >0
3401 * if changes to link states must be committed.
3403 * Called with rcu_read_lock held.
3405 static int bond_ab_arp_inspect(struct bonding *bond)
3407 unsigned long trans_start, last_rx;
3408 struct list_head *iter;
3409 struct slave *slave;
3412 bond_for_each_slave_rcu(bond, slave, iter) {
3413 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3414 last_rx = slave_last_rx(bond, slave);
3416 if (slave->link != BOND_LINK_UP) {
3417 if (bond_time_in_interval(bond, last_rx, 1)) {
3418 bond_propose_link_state(slave, BOND_LINK_UP);
3420 } else if (slave->link == BOND_LINK_BACK) {
3421 bond_propose_link_state(slave, BOND_LINK_FAIL);
3427 /* Give slaves 2*delta after being enslaved or made
3428 * active. This avoids bouncing, as the last receive
3429 * times need a full ARP monitor cycle to be updated.
3431 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3434 /* Backup slave is down if:
3435 * - No current_arp_slave AND
3436 * - more than (missed_max+1)*delta since last receive AND
3437 * - the bond has an IP address
3439 * Note: a non-null current_arp_slave indicates
3440 * the curr_active_slave went down and we are
3441 * searching for a new one; under this condition
3442 * we only take the curr_active_slave down - this
3443 * gives each slave a chance to tx/rx traffic
3444 * before being taken out
3446 if (!bond_is_active_slave(slave) &&
3447 !rcu_access_pointer(bond->current_arp_slave) &&
3448 !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3449 bond_propose_link_state(slave, BOND_LINK_DOWN);
3453 /* Active slave is down if:
3454 * - more than missed_max*delta since transmitting OR
3455 * - (more than missed_max*delta since receive AND
3456 * the bond has an IP address)
3458 trans_start = dev_trans_start(slave->dev);
3459 if (bond_is_active_slave(slave) &&
3460 (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
3461 !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3462 bond_propose_link_state(slave, BOND_LINK_DOWN);
3470 /* Called to commit link state changes noted by inspection step of
3471 * active-backup mode ARP monitor.
3473 * Called with RTNL hold.
3475 static void bond_ab_arp_commit(struct bonding *bond)
3477 unsigned long trans_start;
3478 struct list_head *iter;
3479 struct slave *slave;
3481 bond_for_each_slave(bond, slave, iter) {
3482 switch (slave->link_new_state) {
3483 case BOND_LINK_NOCHANGE:
3487 trans_start = dev_trans_start(slave->dev);
3488 if (rtnl_dereference(bond->curr_active_slave) != slave ||
3489 (!rtnl_dereference(bond->curr_active_slave) &&
3490 bond_time_in_interval(bond, trans_start, 1))) {
3491 struct slave *current_arp_slave;
3493 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3494 bond_set_slave_link_state(slave, BOND_LINK_UP,
3495 BOND_SLAVE_NOTIFY_NOW);
3496 if (current_arp_slave) {
3497 bond_set_slave_inactive_flags(
3499 BOND_SLAVE_NOTIFY_NOW);
3500 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3503 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3505 if (!rtnl_dereference(bond->curr_active_slave) ||
3506 slave == rtnl_dereference(bond->primary_slave))
3513 case BOND_LINK_DOWN:
3514 if (slave->link_failure_count < UINT_MAX)
3515 slave->link_failure_count++;
3517 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3518 BOND_SLAVE_NOTIFY_NOW);
3519 bond_set_slave_inactive_flags(slave,
3520 BOND_SLAVE_NOTIFY_NOW);
3522 slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3524 if (slave == rtnl_dereference(bond->curr_active_slave)) {
3525 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3531 case BOND_LINK_FAIL:
3532 bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3533 BOND_SLAVE_NOTIFY_NOW);
3534 bond_set_slave_inactive_flags(slave,
3535 BOND_SLAVE_NOTIFY_NOW);
3537 /* A slave has just been enslaved and has become
3538 * the current active slave.
3540 if (rtnl_dereference(bond->curr_active_slave))
3541 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3545 slave_err(bond->dev, slave->dev,
3546 "impossible: link_new_state %d on slave\n",
3547 slave->link_new_state);
3553 bond_select_active_slave(bond);
3554 unblock_netpoll_tx();
3557 bond_set_carrier(bond);
3560 /* Send ARP probes for active-backup mode ARP monitor.
3562 * Called with rcu_read_lock held.
3564 static bool bond_ab_arp_probe(struct bonding *bond)
3566 struct slave *slave, *before = NULL, *new_slave = NULL,
3567 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3568 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3569 struct list_head *iter;
3571 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3573 if (curr_arp_slave && curr_active_slave)
3574 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3575 curr_arp_slave->dev->name,
3576 curr_active_slave->dev->name);
3578 if (curr_active_slave) {
3579 bond_send_validate(bond, curr_active_slave);
3580 return should_notify_rtnl;
3583 /* if we don't have a curr_active_slave, search for the next available
3584 * backup slave from the current_arp_slave and make it the candidate
3585 * for becoming the curr_active_slave
3588 if (!curr_arp_slave) {
3589 curr_arp_slave = bond_first_slave_rcu(bond);
3590 if (!curr_arp_slave)
3591 return should_notify_rtnl;
3594 bond_for_each_slave_rcu(bond, slave, iter) {
3595 if (!found && !before && bond_slave_is_up(slave))
3598 if (found && !new_slave && bond_slave_is_up(slave))
3600 /* if the link state is up at this point, we
3601 * mark it down - this can happen if we have
3602 * simultaneous link failures and
3603 * reselect_active_interface doesn't make this
3604 * one the current slave so it is still marked
3605 * up when it is actually down
3607 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3608 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3609 BOND_SLAVE_NOTIFY_LATER);
3610 if (slave->link_failure_count < UINT_MAX)
3611 slave->link_failure_count++;
3613 bond_set_slave_inactive_flags(slave,
3614 BOND_SLAVE_NOTIFY_LATER);
3616 slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3618 if (slave == curr_arp_slave)
3622 if (!new_slave && before)
3628 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3629 BOND_SLAVE_NOTIFY_LATER);
3630 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3631 bond_send_validate(bond, new_slave);
3632 new_slave->last_link_up = jiffies;
3633 rcu_assign_pointer(bond->current_arp_slave, new_slave);
3636 bond_for_each_slave_rcu(bond, slave, iter) {
3637 if (slave->should_notify || slave->should_notify_link) {
3638 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3642 return should_notify_rtnl;
3645 static void bond_activebackup_arp_mon(struct bonding *bond)
3647 bool should_notify_peers = false;
3648 bool should_notify_rtnl = false;
3651 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3653 if (!bond_has_slaves(bond))
3658 should_notify_peers = bond_should_notify_peers(bond);
3660 if (bond_ab_arp_inspect(bond)) {
3663 /* Race avoidance with bond_close flush of workqueue */
3664 if (!rtnl_trylock()) {
3666 should_notify_peers = false;
3670 bond_ab_arp_commit(bond);
3676 should_notify_rtnl = bond_ab_arp_probe(bond);
3680 if (bond->params.arp_interval)
3681 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3683 if (should_notify_peers || should_notify_rtnl) {
3684 if (!rtnl_trylock())
3687 if (should_notify_peers) {
3688 bond->send_peer_notif--;
3689 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3692 if (should_notify_rtnl) {
3693 bond_slave_state_notify(bond);
3694 bond_slave_link_notify(bond);
3701 static void bond_arp_monitor(struct work_struct *work)
3703 struct bonding *bond = container_of(work, struct bonding,
3706 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3707 bond_activebackup_arp_mon(bond);
3709 bond_loadbalance_arp_mon(bond);
3712 /*-------------------------- netdev event handling --------------------------*/
3714 /* Change device name */
3715 static int bond_event_changename(struct bonding *bond)
3717 bond_remove_proc_entry(bond);
3718 bond_create_proc_entry(bond);
3720 bond_debug_reregister(bond);
3725 static int bond_master_netdev_event(unsigned long event,
3726 struct net_device *bond_dev)
3728 struct bonding *event_bond = netdev_priv(bond_dev);
3730 netdev_dbg(bond_dev, "%s called\n", __func__);
3733 case NETDEV_CHANGENAME:
3734 return bond_event_changename(event_bond);
3735 case NETDEV_UNREGISTER:
3736 bond_remove_proc_entry(event_bond);
3737 #ifdef CONFIG_XFRM_OFFLOAD
3738 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3739 #endif /* CONFIG_XFRM_OFFLOAD */
3741 case NETDEV_REGISTER:
3742 bond_create_proc_entry(event_bond);
3751 static int bond_slave_netdev_event(unsigned long event,
3752 struct net_device *slave_dev)
3754 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3755 struct bonding *bond;
3756 struct net_device *bond_dev;
3758 /* A netdev event can be generated while enslaving a device
3759 * before netdev_rx_handler_register is called in which case
3760 * slave will be NULL
3763 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3767 bond_dev = slave->bond->dev;
3769 primary = rtnl_dereference(bond->primary_slave);
3771 slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3774 case NETDEV_UNREGISTER:
3775 if (bond_dev->type != ARPHRD_ETHER)
3776 bond_release_and_destroy(bond_dev, slave_dev);
3778 __bond_release_one(bond_dev, slave_dev, false, true);
3782 /* For 802.3ad mode only:
3783 * Getting invalid Speed/Duplex values here will put slave
3784 * in weird state. Mark it as link-fail if the link was
3785 * previously up or link-down if it hasn't yet come up, and
3786 * let link-monitoring (miimon) set it right when correct
3787 * speeds/duplex are available.
3789 if (bond_update_speed_duplex(slave) &&
3790 BOND_MODE(bond) == BOND_MODE_8023AD) {
3791 if (slave->last_link_up)
3792 slave->link = BOND_LINK_FAIL;
3794 slave->link = BOND_LINK_DOWN;
3797 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3798 bond_3ad_adapter_speed_duplex_changed(slave);
3801 /* Refresh slave-array if applicable!
3802 * If the setup does not use miimon or arpmon (mode-specific!),
3803 * then these events will not cause the slave-array to be
3804 * refreshed. This will cause xmit to use a slave that is not
3805 * usable. Avoid such situation by refeshing the array at these
3806 * events. If these (miimon/arpmon) parameters are configured
3807 * then array gets refreshed twice and that should be fine!
3809 if (bond_mode_can_use_xmit_hash(bond))
3810 bond_update_slave_arr(bond, NULL);
3812 case NETDEV_CHANGEMTU:
3813 /* TODO: Should slaves be allowed to
3814 * independently alter their MTU? For
3815 * an active-backup bond, slaves need
3816 * not be the same type of device, so
3817 * MTUs may vary. For other modes,
3818 * slaves arguably should have the
3819 * same MTUs. To do this, we'd need to
3820 * take over the slave's change_mtu
3821 * function for the duration of their
3825 case NETDEV_CHANGENAME:
3826 /* we don't care if we don't have primary set */
3827 if (!bond_uses_primary(bond) ||
3828 !bond->params.primary[0])
3831 if (slave == primary) {
3832 /* slave's name changed - he's no longer primary */
3833 RCU_INIT_POINTER(bond->primary_slave, NULL);
3834 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3835 /* we have a new primary slave */
3836 rcu_assign_pointer(bond->primary_slave, slave);
3837 } else { /* we didn't change primary - exit */
3841 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3842 primary ? slave_dev->name : "none");
3845 bond_select_active_slave(bond);
3846 unblock_netpoll_tx();
3848 case NETDEV_FEAT_CHANGE:
3849 bond_compute_features(bond);
3851 case NETDEV_RESEND_IGMP:
3852 /* Propagate to master device */
3853 call_netdevice_notifiers(event, slave->bond->dev);
3862 /* bond_netdev_event: handle netdev notifier chain events.
3864 * This function receives events for the netdev chain. The caller (an
3865 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3866 * locks for us to safely manipulate the slave devices (RTNL lock,
3869 static int bond_netdev_event(struct notifier_block *this,
3870 unsigned long event, void *ptr)
3872 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3874 netdev_dbg(event_dev, "%s received %s\n",
3875 __func__, netdev_cmd_to_name(event));
3877 if (!(event_dev->priv_flags & IFF_BONDING))
3880 if (event_dev->flags & IFF_MASTER) {
3883 ret = bond_master_netdev_event(event, event_dev);
3884 if (ret != NOTIFY_DONE)
3888 if (event_dev->flags & IFF_SLAVE)
3889 return bond_slave_netdev_event(event, event_dev);
3894 static struct notifier_block bond_netdev_notifier = {
3895 .notifier_call = bond_netdev_event,
3898 /*---------------------------- Hashing Policies -----------------------------*/
3900 /* Helper to access data in a packet, with or without a backing skb.
3901 * If skb is given the data is linearized if necessary via pskb_may_pull.
3903 static inline const void *bond_pull_data(struct sk_buff *skb,
3904 const void *data, int hlen, int n)
3906 if (likely(n <= hlen))
3908 else if (skb && likely(pskb_may_pull(skb, n)))
3914 /* L2 hash helper */
3915 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3919 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3923 ep = (struct ethhdr *)(data + mhoff);
3924 return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
3927 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
3928 int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
3930 const struct ipv6hdr *iph6;
3931 const struct iphdr *iph;
3933 if (l2_proto == htons(ETH_P_IP)) {
3934 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
3938 iph = (const struct iphdr *)(data + *nhoff);
3939 iph_to_flow_copy_v4addrs(fk, iph);
3940 *nhoff += iph->ihl << 2;
3941 if (!ip_is_fragment(iph))
3942 *ip_proto = iph->protocol;
3943 } else if (l2_proto == htons(ETH_P_IPV6)) {
3944 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
3948 iph6 = (const struct ipv6hdr *)(data + *nhoff);
3949 iph_to_flow_copy_v6addrs(fk, iph6);
3950 *nhoff += sizeof(*iph6);
3951 *ip_proto = iph6->nexthdr;
3956 if (l34 && *ip_proto >= 0)
3957 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
3962 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3964 u32 srcmac_vendor = 0, srcmac_dev = 0;
3965 struct ethhdr *mac_hdr;
3969 data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3972 mac_hdr = (struct ethhdr *)(data + mhoff);
3974 for (i = 0; i < 3; i++)
3975 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3977 for (i = 3; i < ETH_ALEN; i++)
3978 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3980 if (skb && skb_vlan_tag_present(skb))
3981 vlan = skb_vlan_tag_get(skb);
3983 return vlan ^ srcmac_vendor ^ srcmac_dev;
3986 /* Extract the appropriate headers based on bond's xmit policy */
3987 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
3988 __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
3990 bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3993 switch (bond->params.xmit_policy) {
3994 case BOND_XMIT_POLICY_ENCAP23:
3995 case BOND_XMIT_POLICY_ENCAP34:
3996 memset(fk, 0, sizeof(*fk));
3997 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3998 fk, data, l2_proto, nhoff, hlen, 0);
4003 fk->ports.ports = 0;
4004 memset(&fk->icmp, 0, sizeof(fk->icmp));
4005 if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4008 /* ICMP error packets contains at least 8 bytes of the header
4009 * of the packet which generated the error. Use this information
4010 * to correlate ICMP error packets within the same flow which
4011 * generated the error.
4013 if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4014 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4015 if (ip_proto == IPPROTO_ICMP) {
4016 if (!icmp_is_err(fk->icmp.type))
4019 nhoff += sizeof(struct icmphdr);
4020 } else if (ip_proto == IPPROTO_ICMPV6) {
4021 if (!icmpv6_is_err(fk->icmp.type))
4024 nhoff += sizeof(struct icmp6hdr);
4026 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4032 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4034 hash ^= (__force u32)flow_get_u32_dst(flow) ^
4035 (__force u32)flow_get_u32_src(flow);
4036 hash ^= (hash >> 16);
4037 hash ^= (hash >> 8);
4039 /* discard lowest hash bit to deal with the common even ports pattern */
4040 if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4041 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4047 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4048 * the data as required, but this function can be used without it if the data is
4049 * known to be linear (e.g. with xdp_buff).
4051 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4052 __be16 l2_proto, int mhoff, int nhoff, int hlen)
4054 struct flow_keys flow;
4057 if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4058 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4060 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4061 !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4062 return bond_eth_hash(skb, data, mhoff, hlen);
4064 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4065 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4066 hash = bond_eth_hash(skb, data, mhoff, hlen);
4069 memcpy(&hash, &flow.icmp, sizeof(hash));
4071 memcpy(&hash, &flow.ports.ports, sizeof(hash));
4074 return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4078 * bond_xmit_hash - generate a hash value based on the xmit policy
4079 * @bond: bonding device
4080 * @skb: buffer to use for headers
4082 * This function will extract the necessary headers from the skb buffer and use
4083 * them to generate a hash based on the xmit_policy set in the bonding device
4085 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4087 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4091 return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4092 skb_mac_offset(skb), skb_network_offset(skb),
4097 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4098 * @bond: bonding device
4099 * @xdp: buffer to use for headers
4101 * The XDP variant of bond_xmit_hash.
4103 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4107 if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4110 eth = (struct ethhdr *)xdp->data;
4112 return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4113 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4116 /*-------------------------- Device entry points ----------------------------*/
4118 void bond_work_init_all(struct bonding *bond)
4120 INIT_DELAYED_WORK(&bond->mcast_work,
4121 bond_resend_igmp_join_requests_delayed);
4122 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4123 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4124 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4125 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4126 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4129 static void bond_work_cancel_all(struct bonding *bond)
4131 cancel_delayed_work_sync(&bond->mii_work);
4132 cancel_delayed_work_sync(&bond->arp_work);
4133 cancel_delayed_work_sync(&bond->alb_work);
4134 cancel_delayed_work_sync(&bond->ad_work);
4135 cancel_delayed_work_sync(&bond->mcast_work);
4136 cancel_delayed_work_sync(&bond->slave_arr_work);
4139 static int bond_open(struct net_device *bond_dev)
4141 struct bonding *bond = netdev_priv(bond_dev);
4142 struct list_head *iter;
4143 struct slave *slave;
4145 /* reset slave->backup and slave->inactive */
4146 if (bond_has_slaves(bond)) {
4147 bond_for_each_slave(bond, slave, iter) {
4148 if (bond_uses_primary(bond) &&
4149 slave != rcu_access_pointer(bond->curr_active_slave)) {
4150 bond_set_slave_inactive_flags(slave,
4151 BOND_SLAVE_NOTIFY_NOW);
4152 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4153 bond_set_slave_active_flags(slave,
4154 BOND_SLAVE_NOTIFY_NOW);
4159 if (bond_is_lb(bond)) {
4160 /* bond_alb_initialize must be called before the timer
4163 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4165 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4166 queue_delayed_work(bond->wq, &bond->alb_work, 0);
4169 if (bond->params.miimon) /* link check interval, in milliseconds. */
4170 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4172 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
4173 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4174 bond->recv_probe = bond_rcv_validate;
4177 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4178 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4179 /* register to receive LACPDUs */
4180 bond->recv_probe = bond_3ad_lacpdu_recv;
4181 bond_3ad_initiate_agg_selection(bond, 1);
4184 if (bond_mode_can_use_xmit_hash(bond))
4185 bond_update_slave_arr(bond, NULL);
4190 static int bond_close(struct net_device *bond_dev)
4192 struct bonding *bond = netdev_priv(bond_dev);
4194 bond_work_cancel_all(bond);
4195 bond->send_peer_notif = 0;
4196 if (bond_is_lb(bond))
4197 bond_alb_deinitialize(bond);
4198 bond->recv_probe = NULL;
4203 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4204 * that some drivers can provide 32bit values only.
4206 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4207 const struct rtnl_link_stats64 *_new,
4208 const struct rtnl_link_stats64 *_old)
4210 const u64 *new = (const u64 *)_new;
4211 const u64 *old = (const u64 *)_old;
4212 u64 *res = (u64 *)_res;
4215 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4218 s64 delta = nv - ov;
4220 /* detects if this particular field is 32bit only */
4221 if (((nv | ov) >> 32) == 0)
4222 delta = (s64)(s32)((u32)nv - (u32)ov);
4224 /* filter anomalies, some drivers reset their stats
4225 * at down/up events.
4232 #ifdef CONFIG_LOCKDEP
4233 static int bond_get_lowest_level_rcu(struct net_device *dev)
4235 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4236 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4237 int cur = 0, max = 0;
4240 iter = &dev->adj_list.lower;
4245 ldev = netdev_next_lower_dev_rcu(now, &iter);
4250 niter = &ldev->adj_list.lower;
4251 dev_stack[cur] = now;
4252 iter_stack[cur++] = iter;
4261 next = dev_stack[--cur];
4262 niter = iter_stack[cur];
4273 static void bond_get_stats(struct net_device *bond_dev,
4274 struct rtnl_link_stats64 *stats)
4276 struct bonding *bond = netdev_priv(bond_dev);
4277 struct rtnl_link_stats64 temp;
4278 struct list_head *iter;
4279 struct slave *slave;
4284 #ifdef CONFIG_LOCKDEP
4285 nest_level = bond_get_lowest_level_rcu(bond_dev);
4288 spin_lock_nested(&bond->stats_lock, nest_level);
4289 memcpy(stats, &bond->bond_stats, sizeof(*stats));
4291 bond_for_each_slave_rcu(bond, slave, iter) {
4292 const struct rtnl_link_stats64 *new =
4293 dev_get_stats(slave->dev, &temp);
4295 bond_fold_stats(stats, new, &slave->slave_stats);
4297 /* save off the slave stats for the next run */
4298 memcpy(&slave->slave_stats, new, sizeof(*new));
4301 memcpy(&bond->bond_stats, stats, sizeof(*stats));
4302 spin_unlock(&bond->stats_lock);
4306 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4308 struct bonding *bond = netdev_priv(bond_dev);
4309 struct mii_ioctl_data *mii = NULL;
4310 const struct net_device_ops *ops;
4311 struct net_device *real_dev;
4312 struct hwtstamp_config cfg;
4316 netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4327 /* We do this again just in case we were called by SIOCGMIIREG
4328 * instead of SIOCGMIIPHY.
4334 if (mii->reg_num == 1) {
4336 if (netif_carrier_ok(bond->dev))
4337 mii->val_out = BMSR_LSTATUS;
4342 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4345 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4350 real_dev = bond_option_active_slave_get_rcu(bond);
4354 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4355 ifrr.ifr_ifru = ifr->ifr_ifru;
4357 ops = real_dev->netdev_ops;
4358 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4359 res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4363 ifr->ifr_ifru = ifrr.ifr_ifru;
4364 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4367 /* Set the BOND_PHC_INDEX flag to notify user space */
4368 cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4370 return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4381 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4383 struct bonding *bond = netdev_priv(bond_dev);
4384 struct net_device *slave_dev = NULL;
4385 struct ifbond k_binfo;
4386 struct ifbond __user *u_binfo = NULL;
4387 struct ifslave k_sinfo;
4388 struct ifslave __user *u_sinfo = NULL;
4389 struct bond_opt_value newval;
4393 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4396 case SIOCBONDINFOQUERY:
4397 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4399 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4402 bond_info_query(bond_dev, &k_binfo);
4403 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4407 case SIOCBONDSLAVEINFOQUERY:
4408 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4410 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4413 res = bond_slave_info_query(bond_dev, &k_sinfo);
4415 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4423 net = dev_net(bond_dev);
4425 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4428 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4430 slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4436 case SIOCBONDENSLAVE:
4437 res = bond_enslave(bond_dev, slave_dev, NULL);
4439 case SIOCBONDRELEASE:
4440 res = bond_release(bond_dev, slave_dev);
4442 case SIOCBONDSETHWADDR:
4443 res = bond_set_dev_addr(bond_dev, slave_dev);
4445 case SIOCBONDCHANGEACTIVE:
4446 bond_opt_initstr(&newval, slave_dev->name);
4447 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4457 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4458 void __user *data, int cmd)
4460 struct ifreq ifrdata = { .ifr_data = data };
4463 case BOND_INFO_QUERY_OLD:
4464 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4465 case BOND_SLAVE_INFO_QUERY_OLD:
4466 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4467 case BOND_ENSLAVE_OLD:
4468 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4469 case BOND_RELEASE_OLD:
4470 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4471 case BOND_SETHWADDR_OLD:
4472 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4473 case BOND_CHANGE_ACTIVE_OLD:
4474 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4480 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4482 struct bonding *bond = netdev_priv(bond_dev);
4484 if (change & IFF_PROMISC)
4485 bond_set_promiscuity(bond,
4486 bond_dev->flags & IFF_PROMISC ? 1 : -1);
4488 if (change & IFF_ALLMULTI)
4489 bond_set_allmulti(bond,
4490 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4493 static void bond_set_rx_mode(struct net_device *bond_dev)
4495 struct bonding *bond = netdev_priv(bond_dev);
4496 struct list_head *iter;
4497 struct slave *slave;
4500 if (bond_uses_primary(bond)) {
4501 slave = rcu_dereference(bond->curr_active_slave);
4503 dev_uc_sync(slave->dev, bond_dev);
4504 dev_mc_sync(slave->dev, bond_dev);
4507 bond_for_each_slave_rcu(bond, slave, iter) {
4508 dev_uc_sync_multiple(slave->dev, bond_dev);
4509 dev_mc_sync_multiple(slave->dev, bond_dev);
4515 static int bond_neigh_init(struct neighbour *n)
4517 struct bonding *bond = netdev_priv(n->dev);
4518 const struct net_device_ops *slave_ops;
4519 struct neigh_parms parms;
4520 struct slave *slave;
4524 slave = bond_first_slave_rcu(bond);
4527 slave_ops = slave->dev->netdev_ops;
4528 if (!slave_ops->ndo_neigh_setup)
4531 /* TODO: find another way [1] to implement this.
4532 * Passing a zeroed structure is fragile,
4533 * but at least we do not pass garbage.
4535 * [1] One way would be that ndo_neigh_setup() never touch
4536 * struct neigh_parms, but propagate the new neigh_setup()
4537 * back to ___neigh_create() / neigh_parms_alloc()
4539 memset(&parms, 0, sizeof(parms));
4540 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4545 if (parms.neigh_setup)
4546 ret = parms.neigh_setup(n);
4552 /* The bonding ndo_neigh_setup is called at init time beofre any
4553 * slave exists. So we must declare proxy setup function which will
4554 * be used at run time to resolve the actual slave neigh param setup.
4556 * It's also called by master devices (such as vlans) to setup their
4557 * underlying devices. In that case - do nothing, we're already set up from
4560 static int bond_neigh_setup(struct net_device *dev,
4561 struct neigh_parms *parms)
4563 /* modify only our neigh_parms */
4564 if (parms->dev == dev)
4565 parms->neigh_setup = bond_neigh_init;
4570 /* Change the MTU of all of a master's slaves to match the master */
4571 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4573 struct bonding *bond = netdev_priv(bond_dev);
4574 struct slave *slave, *rollback_slave;
4575 struct list_head *iter;
4578 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4580 bond_for_each_slave(bond, slave, iter) {
4581 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4582 slave, slave->dev->netdev_ops->ndo_change_mtu);
4584 res = dev_set_mtu(slave->dev, new_mtu);
4587 /* If we failed to set the slave's mtu to the new value
4588 * we must abort the operation even in ACTIVE_BACKUP
4589 * mode, because if we allow the backup slaves to have
4590 * different mtu values than the active slave we'll
4591 * need to change their mtu when doing a failover. That
4592 * means changing their mtu from timer context, which
4593 * is probably not a good idea.
4595 slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4601 bond_dev->mtu = new_mtu;
4606 /* unwind from head to the slave that failed */
4607 bond_for_each_slave(bond, rollback_slave, iter) {
4610 if (rollback_slave == slave)
4613 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4615 slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4622 /* Change HW address
4624 * Note that many devices must be down to change the HW address, and
4625 * downing the master releases all slaves. We can make bonds full of
4626 * bonding devices to test this, however.
4628 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4630 struct bonding *bond = netdev_priv(bond_dev);
4631 struct slave *slave, *rollback_slave;
4632 struct sockaddr_storage *ss = addr, tmp_ss;
4633 struct list_head *iter;
4636 if (BOND_MODE(bond) == BOND_MODE_ALB)
4637 return bond_alb_set_mac_address(bond_dev, addr);
4640 netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4642 /* If fail_over_mac is enabled, do nothing and return success.
4643 * Returning an error causes ifenslave to fail.
4645 if (bond->params.fail_over_mac &&
4646 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4649 if (!is_valid_ether_addr(ss->__data))
4650 return -EADDRNOTAVAIL;
4652 bond_for_each_slave(bond, slave, iter) {
4653 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4655 res = dev_set_mac_address(slave->dev, addr, NULL);
4657 /* TODO: consider downing the slave
4659 * User should expect communications
4660 * breakage anyway until ARP finish
4663 slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4670 dev_addr_set(bond_dev, ss->__data);
4674 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4675 tmp_ss.ss_family = bond_dev->type;
4677 /* unwind from head to the slave that failed */
4678 bond_for_each_slave(bond, rollback_slave, iter) {
4681 if (rollback_slave == slave)
4684 tmp_res = dev_set_mac_address(rollback_slave->dev,
4685 (struct sockaddr *)&tmp_ss, NULL);
4687 slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4696 * bond_get_slave_by_id - get xmit slave with slave_id
4697 * @bond: bonding device that is transmitting
4698 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4700 * This function tries to get slave with slave_id but in case
4701 * it fails, it tries to find the first available slave for transmission.
4703 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4706 struct list_head *iter;
4707 struct slave *slave;
4710 /* Here we start from the slave with slave_id */
4711 bond_for_each_slave_rcu(bond, slave, iter) {
4713 if (bond_slave_can_tx(slave))
4718 /* Here we start from the first slave up to slave_id */
4720 bond_for_each_slave_rcu(bond, slave, iter) {
4723 if (bond_slave_can_tx(slave))
4726 /* no slave that can tx has been found */
4731 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4732 * @bond: bonding device to use
4734 * Based on the value of the bonding device's packets_per_slave parameter
4735 * this function generates a slave id, which is usually used as the next
4736 * slave to transmit through.
4738 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4741 struct reciprocal_value reciprocal_packets_per_slave;
4742 int packets_per_slave = bond->params.packets_per_slave;
4744 switch (packets_per_slave) {
4746 slave_id = prandom_u32();
4749 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4752 reciprocal_packets_per_slave =
4753 bond->params.reciprocal_packets_per_slave;
4754 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4755 slave_id = reciprocal_divide(slave_id,
4756 reciprocal_packets_per_slave);
4763 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4764 struct sk_buff *skb)
4766 struct slave *slave;
4770 /* Start with the curr_active_slave that joined the bond as the
4771 * default for sending IGMP traffic. For failover purposes one
4772 * needs to maintain some consistency for the interface that will
4773 * send the join/membership reports. The curr_active_slave found
4774 * will send all of this type of traffic.
4776 if (skb->protocol == htons(ETH_P_IP)) {
4777 int noff = skb_network_offset(skb);
4780 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4784 if (iph->protocol == IPPROTO_IGMP) {
4785 slave = rcu_dereference(bond->curr_active_slave);
4788 return bond_get_slave_by_id(bond, 0);
4793 slave_cnt = READ_ONCE(bond->slave_cnt);
4794 if (likely(slave_cnt)) {
4795 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4796 return bond_get_slave_by_id(bond, slave_id);
4801 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4802 struct xdp_buff *xdp)
4804 struct slave *slave;
4807 const struct ethhdr *eth;
4808 void *data = xdp->data;
4810 if (data + sizeof(struct ethhdr) > xdp->data_end)
4813 eth = (struct ethhdr *)data;
4814 data += sizeof(struct ethhdr);
4816 /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4817 if (eth->h_proto == htons(ETH_P_IP)) {
4818 const struct iphdr *iph;
4820 if (data + sizeof(struct iphdr) > xdp->data_end)
4823 iph = (struct iphdr *)data;
4825 if (iph->protocol == IPPROTO_IGMP) {
4826 slave = rcu_dereference(bond->curr_active_slave);
4829 return bond_get_slave_by_id(bond, 0);
4834 slave_cnt = READ_ONCE(bond->slave_cnt);
4835 if (likely(slave_cnt)) {
4836 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4837 return bond_get_slave_by_id(bond, slave_id);
4842 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4843 struct net_device *bond_dev)
4845 struct bonding *bond = netdev_priv(bond_dev);
4846 struct slave *slave;
4848 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4850 return bond_dev_queue_xmit(bond, skb, slave->dev);
4852 return bond_tx_drop(bond_dev, skb);
4855 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4857 return rcu_dereference(bond->curr_active_slave);
4860 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4861 * the bond has a usable interface.
4863 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4864 struct net_device *bond_dev)
4866 struct bonding *bond = netdev_priv(bond_dev);
4867 struct slave *slave;
4869 slave = bond_xmit_activebackup_slave_get(bond);
4871 return bond_dev_queue_xmit(bond, skb, slave->dev);
4873 return bond_tx_drop(bond_dev, skb);
4876 /* Use this to update slave_array when (a) it's not appropriate to update
4877 * slave_array right away (note that update_slave_array() may sleep)
4878 * and / or (b) RTNL is not held.
4880 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4882 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4885 /* Slave array work handler. Holds only RTNL */
4886 static void bond_slave_arr_handler(struct work_struct *work)
4888 struct bonding *bond = container_of(work, struct bonding,
4889 slave_arr_work.work);
4892 if (!rtnl_trylock())
4895 ret = bond_update_slave_arr(bond, NULL);
4898 pr_warn_ratelimited("Failed to update slave array from WT\n");
4904 bond_slave_arr_work_rearm(bond, 1);
4907 static void bond_skip_slave(struct bond_up_slave *slaves,
4908 struct slave *skipslave)
4912 /* Rare situation where caller has asked to skip a specific
4913 * slave but allocation failed (most likely!). BTW this is
4914 * only possible when the call is initiated from
4915 * __bond_release_one(). In this situation; overwrite the
4916 * skipslave entry in the array with the last entry from the
4917 * array to avoid a situation where the xmit path may choose
4918 * this to-be-skipped slave to send a packet out.
4920 for (idx = 0; slaves && idx < slaves->count; idx++) {
4921 if (skipslave == slaves->arr[idx]) {
4923 slaves->arr[slaves->count - 1];
4930 static void bond_set_slave_arr(struct bonding *bond,
4931 struct bond_up_slave *usable_slaves,
4932 struct bond_up_slave *all_slaves)
4934 struct bond_up_slave *usable, *all;
4936 usable = rtnl_dereference(bond->usable_slaves);
4937 rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4938 kfree_rcu(usable, rcu);
4940 all = rtnl_dereference(bond->all_slaves);
4941 rcu_assign_pointer(bond->all_slaves, all_slaves);
4942 kfree_rcu(all, rcu);
4945 static void bond_reset_slave_arr(struct bonding *bond)
4947 struct bond_up_slave *usable, *all;
4949 usable = rtnl_dereference(bond->usable_slaves);
4951 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4952 kfree_rcu(usable, rcu);
4955 all = rtnl_dereference(bond->all_slaves);
4957 RCU_INIT_POINTER(bond->all_slaves, NULL);
4958 kfree_rcu(all, rcu);
4962 /* Build the usable slaves array in control path for modes that use xmit-hash
4963 * to determine the slave interface -
4964 * (a) BOND_MODE_8023AD
4966 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4968 * The caller is expected to hold RTNL only and NO other lock!
4970 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4972 struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4973 struct slave *slave;
4974 struct list_head *iter;
4980 usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4981 bond->slave_cnt), GFP_KERNEL);
4982 all_slaves = kzalloc(struct_size(all_slaves, arr,
4983 bond->slave_cnt), GFP_KERNEL);
4984 if (!usable_slaves || !all_slaves) {
4988 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4989 struct ad_info ad_info;
4991 spin_lock_bh(&bond->mode_lock);
4992 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4993 spin_unlock_bh(&bond->mode_lock);
4994 pr_debug("bond_3ad_get_active_agg_info failed\n");
4995 /* No active aggragator means it's not safe to use
4996 * the previous array.
4998 bond_reset_slave_arr(bond);
5001 spin_unlock_bh(&bond->mode_lock);
5002 agg_id = ad_info.aggregator_id;
5004 bond_for_each_slave(bond, slave, iter) {
5005 if (skipslave == slave)
5008 all_slaves->arr[all_slaves->count++] = slave;
5009 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5010 struct aggregator *agg;
5012 agg = SLAVE_AD_INFO(slave)->port.aggregator;
5013 if (!agg || agg->aggregator_identifier != agg_id)
5016 if (!bond_slave_can_tx(slave))
5019 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5020 usable_slaves->count);
5022 usable_slaves->arr[usable_slaves->count++] = slave;
5025 bond_set_slave_arr(bond, usable_slaves, all_slaves);
5028 if (ret != 0 && skipslave) {
5029 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5031 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5034 kfree_rcu(all_slaves, rcu);
5035 kfree_rcu(usable_slaves, rcu);
5040 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5041 struct sk_buff *skb,
5042 struct bond_up_slave *slaves)
5044 struct slave *slave;
5048 hash = bond_xmit_hash(bond, skb);
5049 count = slaves ? READ_ONCE(slaves->count) : 0;
5050 if (unlikely(!count))
5053 slave = slaves->arr[hash % count];
5057 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5058 struct xdp_buff *xdp)
5060 struct bond_up_slave *slaves;
5064 hash = bond_xmit_hash_xdp(bond, xdp);
5065 slaves = rcu_dereference(bond->usable_slaves);
5066 count = slaves ? READ_ONCE(slaves->count) : 0;
5067 if (unlikely(!count))
5070 return slaves->arr[hash % count];
5073 /* Use this Xmit function for 3AD as well as XOR modes. The current
5074 * usable slave array is formed in the control path. The xmit function
5075 * just calculates hash and sends the packet out.
5077 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5078 struct net_device *dev)
5080 struct bonding *bond = netdev_priv(dev);
5081 struct bond_up_slave *slaves;
5082 struct slave *slave;
5084 slaves = rcu_dereference(bond->usable_slaves);
5085 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5087 return bond_dev_queue_xmit(bond, skb, slave->dev);
5089 return bond_tx_drop(dev, skb);
5092 /* in broadcast mode, we send everything to all usable interfaces. */
5093 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5094 struct net_device *bond_dev)
5096 struct bonding *bond = netdev_priv(bond_dev);
5097 struct slave *slave = NULL;
5098 struct list_head *iter;
5099 bool xmit_suc = false;
5100 bool skb_used = false;
5102 bond_for_each_slave_rcu(bond, slave, iter) {
5103 struct sk_buff *skb2;
5105 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5108 if (bond_is_last_slave(bond, slave)) {
5112 skb2 = skb_clone(skb, GFP_ATOMIC);
5114 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5115 bond_dev->name, __func__);
5120 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5125 dev_kfree_skb_any(skb);
5128 return NETDEV_TX_OK;
5130 dev_core_stats_tx_dropped_inc(bond_dev);
5131 return NET_XMIT_DROP;
5134 /*------------------------- Device initialization ---------------------------*/
5136 /* Lookup the slave that corresponds to a qid */
5137 static inline int bond_slave_override(struct bonding *bond,
5138 struct sk_buff *skb)
5140 struct slave *slave = NULL;
5141 struct list_head *iter;
5143 if (!skb_rx_queue_recorded(skb))
5146 /* Find out if any slaves have the same mapping as this skb. */
5147 bond_for_each_slave_rcu(bond, slave, iter) {
5148 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5149 if (bond_slave_is_up(slave) &&
5150 slave->link == BOND_LINK_UP) {
5151 bond_dev_queue_xmit(bond, skb, slave->dev);
5154 /* If the slave isn't UP, use default transmit policy. */
5163 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5164 struct net_device *sb_dev)
5166 /* This helper function exists to help dev_pick_tx get the correct
5167 * destination queue. Using a helper function skips a call to
5168 * skb_tx_hash and will put the skbs in the queue we expect on their
5169 * way down to the bonding driver.
5171 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5173 /* Save the original txq to restore before passing to the driver */
5174 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5176 if (unlikely(txq >= dev->real_num_tx_queues)) {
5178 txq -= dev->real_num_tx_queues;
5179 } while (txq >= dev->real_num_tx_queues);
5184 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5185 struct sk_buff *skb,
5188 struct bonding *bond = netdev_priv(master_dev);
5189 struct bond_up_slave *slaves;
5190 struct slave *slave = NULL;
5192 switch (BOND_MODE(bond)) {
5193 case BOND_MODE_ROUNDROBIN:
5194 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5196 case BOND_MODE_ACTIVEBACKUP:
5197 slave = bond_xmit_activebackup_slave_get(bond);
5199 case BOND_MODE_8023AD:
5202 slaves = rcu_dereference(bond->all_slaves);
5204 slaves = rcu_dereference(bond->usable_slaves);
5205 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5207 case BOND_MODE_BROADCAST:
5210 slave = bond_xmit_alb_slave_get(bond, skb);
5213 slave = bond_xmit_tlb_slave_get(bond, skb);
5216 /* Should never happen, mode already checked */
5217 WARN_ONCE(true, "Unknown bonding mode");
5226 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5228 switch (sk->sk_family) {
5229 #if IS_ENABLED(CONFIG_IPV6)
5231 if (ipv6_only_sock(sk) ||
5232 ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5233 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5234 flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5235 flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5240 default: /* AF_INET */
5241 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5242 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5243 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5247 flow->ports.src = inet_sk(sk)->inet_sport;
5248 flow->ports.dst = inet_sk(sk)->inet_dport;
5252 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5253 * @sk: socket to use for headers
5255 * This function will extract the necessary field from the socket and use
5256 * them to generate a hash based on the LAYER34 xmit_policy.
5257 * Assumes that sk is a TCP or UDP socket.
5259 static u32 bond_sk_hash_l34(struct sock *sk)
5261 struct flow_keys flow;
5264 bond_sk_to_flow(sk, &flow);
5267 memcpy(&hash, &flow.ports.ports, sizeof(hash));
5269 return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5272 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5275 struct bond_up_slave *slaves;
5276 struct slave *slave;
5280 slaves = rcu_dereference(bond->usable_slaves);
5281 count = slaves ? READ_ONCE(slaves->count) : 0;
5282 if (unlikely(!count))
5285 hash = bond_sk_hash_l34(sk);
5286 slave = slaves->arr[hash % count];
5291 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5294 struct bonding *bond = netdev_priv(dev);
5295 struct net_device *lower = NULL;
5298 if (bond_sk_check(bond))
5299 lower = __bond_sk_get_lower_dev(bond, sk);
5305 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5306 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5307 struct net_device *dev)
5309 if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
5310 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
5311 return bond_tx_drop(dev, skb);
5315 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5317 struct bonding *bond = netdev_priv(dev);
5319 if (bond_should_override_tx_queue(bond) &&
5320 !bond_slave_override(bond, skb))
5321 return NETDEV_TX_OK;
5323 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5324 if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5325 return bond_tls_device_xmit(bond, skb, dev);
5328 switch (BOND_MODE(bond)) {
5329 case BOND_MODE_ROUNDROBIN:
5330 return bond_xmit_roundrobin(skb, dev);
5331 case BOND_MODE_ACTIVEBACKUP:
5332 return bond_xmit_activebackup(skb, dev);
5333 case BOND_MODE_8023AD:
5335 return bond_3ad_xor_xmit(skb, dev);
5336 case BOND_MODE_BROADCAST:
5337 return bond_xmit_broadcast(skb, dev);
5339 return bond_alb_xmit(skb, dev);
5341 return bond_tlb_xmit(skb, dev);
5343 /* Should never happen, mode already checked */
5344 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5346 return bond_tx_drop(dev, skb);
5350 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5352 struct bonding *bond = netdev_priv(dev);
5353 netdev_tx_t ret = NETDEV_TX_OK;
5355 /* If we risk deadlock from transmitting this in the
5356 * netpoll path, tell netpoll to queue the frame for later tx
5358 if (unlikely(is_netpoll_tx_blocked(dev)))
5359 return NETDEV_TX_BUSY;
5362 if (bond_has_slaves(bond))
5363 ret = __bond_start_xmit(skb, dev);
5365 ret = bond_tx_drop(dev, skb);
5371 static struct net_device *
5372 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5374 struct bonding *bond = netdev_priv(bond_dev);
5375 struct slave *slave;
5377 /* Caller needs to hold rcu_read_lock() */
5379 switch (BOND_MODE(bond)) {
5380 case BOND_MODE_ROUNDROBIN:
5381 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5384 case BOND_MODE_ACTIVEBACKUP:
5385 slave = bond_xmit_activebackup_slave_get(bond);
5388 case BOND_MODE_8023AD:
5390 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5394 /* Should never happen. Mode guarded by bond_xdp_check() */
5395 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5406 static int bond_xdp_xmit(struct net_device *bond_dev,
5407 int n, struct xdp_frame **frames, u32 flags)
5409 int nxmit, err = -ENXIO;
5413 for (nxmit = 0; nxmit < n; nxmit++) {
5414 struct xdp_frame *frame = frames[nxmit];
5415 struct xdp_frame *frames1[] = {frame};
5416 struct net_device *slave_dev;
5417 struct xdp_buff xdp;
5419 xdp_convert_frame_to_buff(frame, &xdp);
5421 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5427 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5434 /* If error happened on the first frame then we can pass the error up, otherwise
5435 * report the number of frames that were xmitted.
5438 return (nxmit == 0 ? err : nxmit);
5443 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5444 struct netlink_ext_ack *extack)
5446 struct bonding *bond = netdev_priv(dev);
5447 struct list_head *iter;
5448 struct slave *slave, *rollback_slave;
5449 struct bpf_prog *old_prog;
5450 struct netdev_bpf xdp = {
5451 .command = XDP_SETUP_PROG,
5460 if (!bond_xdp_check(bond))
5463 old_prog = bond->xdp_prog;
5464 bond->xdp_prog = prog;
5466 bond_for_each_slave(bond, slave, iter) {
5467 struct net_device *slave_dev = slave->dev;
5469 if (!slave_dev->netdev_ops->ndo_bpf ||
5470 !slave_dev->netdev_ops->ndo_xdp_xmit) {
5471 SLAVE_NL_ERR(dev, slave_dev, extack,
5472 "Slave device does not support XDP");
5477 if (dev_xdp_prog_count(slave_dev) > 0) {
5478 SLAVE_NL_ERR(dev, slave_dev, extack,
5479 "Slave has XDP program loaded, please unload before enslaving");
5484 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5486 /* ndo_bpf() sets extack error message */
5487 slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5495 static_branch_inc(&bpf_master_redirect_enabled_key);
5496 } else if (old_prog) {
5497 bpf_prog_put(old_prog);
5498 static_branch_dec(&bpf_master_redirect_enabled_key);
5504 /* unwind the program changes */
5505 bond->xdp_prog = old_prog;
5506 xdp.prog = old_prog;
5507 xdp.extack = NULL; /* do not overwrite original error */
5509 bond_for_each_slave(bond, rollback_slave, iter) {
5510 struct net_device *slave_dev = rollback_slave->dev;
5513 if (slave == rollback_slave)
5516 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5518 slave_err(dev, slave_dev,
5519 "Error %d when unwinding XDP program change\n", err_unwind);
5521 bpf_prog_inc(xdp.prog);
5526 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5528 switch (xdp->command) {
5529 case XDP_SETUP_PROG:
5530 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5536 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5538 if (speed == 0 || speed == SPEED_UNKNOWN)
5539 speed = slave->speed;
5541 speed = min(speed, slave->speed);
5546 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5547 struct ethtool_link_ksettings *cmd)
5549 struct bonding *bond = netdev_priv(bond_dev);
5550 struct list_head *iter;
5551 struct slave *slave;
5554 cmd->base.duplex = DUPLEX_UNKNOWN;
5555 cmd->base.port = PORT_OTHER;
5557 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5558 * do not need to check mode. Though link speed might not represent
5559 * the true receive or transmit bandwidth (not all modes are symmetric)
5560 * this is an accurate maximum.
5562 bond_for_each_slave(bond, slave, iter) {
5563 if (bond_slave_can_tx(slave)) {
5564 if (slave->speed != SPEED_UNKNOWN) {
5565 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5566 speed = bond_mode_bcast_speed(slave,
5569 speed += slave->speed;
5571 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5572 slave->duplex != DUPLEX_UNKNOWN)
5573 cmd->base.duplex = slave->duplex;
5576 cmd->base.speed = speed ? : SPEED_UNKNOWN;
5581 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5582 struct ethtool_drvinfo *drvinfo)
5584 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5585 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5589 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5590 struct ethtool_ts_info *info)
5592 struct bonding *bond = netdev_priv(bond_dev);
5593 const struct ethtool_ops *ops;
5594 struct net_device *real_dev;
5595 struct phy_device *phydev;
5599 real_dev = bond_option_active_slave_get_rcu(bond);
5604 ops = real_dev->ethtool_ops;
5605 phydev = real_dev->phydev;
5607 if (phy_has_tsinfo(phydev)) {
5608 ret = phy_ts_info(phydev, info);
5610 } else if (ops->get_ts_info) {
5611 ret = ops->get_ts_info(real_dev, info);
5616 info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5617 SOF_TIMESTAMPING_SOFTWARE;
5618 info->phc_index = -1;
5625 static const struct ethtool_ops bond_ethtool_ops = {
5626 .get_drvinfo = bond_ethtool_get_drvinfo,
5627 .get_link = ethtool_op_get_link,
5628 .get_link_ksettings = bond_ethtool_get_link_ksettings,
5629 .get_ts_info = bond_ethtool_get_ts_info,
5632 static const struct net_device_ops bond_netdev_ops = {
5633 .ndo_init = bond_init,
5634 .ndo_uninit = bond_uninit,
5635 .ndo_open = bond_open,
5636 .ndo_stop = bond_close,
5637 .ndo_start_xmit = bond_start_xmit,
5638 .ndo_select_queue = bond_select_queue,
5639 .ndo_get_stats64 = bond_get_stats,
5640 .ndo_eth_ioctl = bond_eth_ioctl,
5641 .ndo_siocbond = bond_do_ioctl,
5642 .ndo_siocdevprivate = bond_siocdevprivate,
5643 .ndo_change_rx_flags = bond_change_rx_flags,
5644 .ndo_set_rx_mode = bond_set_rx_mode,
5645 .ndo_change_mtu = bond_change_mtu,
5646 .ndo_set_mac_address = bond_set_mac_address,
5647 .ndo_neigh_setup = bond_neigh_setup,
5648 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
5649 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
5650 #ifdef CONFIG_NET_POLL_CONTROLLER
5651 .ndo_netpoll_setup = bond_netpoll_setup,
5652 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
5653 .ndo_poll_controller = bond_poll_controller,
5655 .ndo_add_slave = bond_enslave,
5656 .ndo_del_slave = bond_release,
5657 .ndo_fix_features = bond_fix_features,
5658 .ndo_features_check = passthru_features_check,
5659 .ndo_get_xmit_slave = bond_xmit_get_slave,
5660 .ndo_sk_get_lower_dev = bond_sk_get_lower_dev,
5661 .ndo_bpf = bond_xdp,
5662 .ndo_xdp_xmit = bond_xdp_xmit,
5663 .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5666 static const struct device_type bond_type = {
5670 static void bond_destructor(struct net_device *bond_dev)
5672 struct bonding *bond = netdev_priv(bond_dev);
5675 destroy_workqueue(bond->wq);
5677 if (bond->rr_tx_counter)
5678 free_percpu(bond->rr_tx_counter);
5681 void bond_setup(struct net_device *bond_dev)
5683 struct bonding *bond = netdev_priv(bond_dev);
5685 spin_lock_init(&bond->mode_lock);
5686 bond->params = bonding_defaults;
5688 /* Initialize pointers */
5689 bond->dev = bond_dev;
5691 /* Initialize the device entry points */
5692 ether_setup(bond_dev);
5693 bond_dev->max_mtu = ETH_MAX_MTU;
5694 bond_dev->netdev_ops = &bond_netdev_ops;
5695 bond_dev->ethtool_ops = &bond_ethtool_ops;
5697 bond_dev->needs_free_netdev = true;
5698 bond_dev->priv_destructor = bond_destructor;
5700 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5702 /* Initialize the device options */
5703 bond_dev->flags |= IFF_MASTER;
5704 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5705 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5707 #ifdef CONFIG_XFRM_OFFLOAD
5708 /* set up xfrm device ops (only supported in active-backup right now) */
5709 bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5710 INIT_LIST_HEAD(&bond->ipsec_list);
5711 spin_lock_init(&bond->ipsec_lock);
5712 #endif /* CONFIG_XFRM_OFFLOAD */
5714 /* don't acquire bond device's netif_tx_lock when transmitting */
5715 bond_dev->features |= NETIF_F_LLTX;
5717 /* By default, we declare the bond to be fully
5718 * VLAN hardware accelerated capable. Special
5719 * care is taken in the various xmit functions
5720 * when there are slaves that are not hw accel
5724 /* Don't allow bond devices to change network namespaces. */
5725 bond_dev->features |= NETIF_F_NETNS_LOCAL;
5727 bond_dev->hw_features = BOND_VLAN_FEATURES |
5728 NETIF_F_HW_VLAN_CTAG_RX |
5729 NETIF_F_HW_VLAN_CTAG_FILTER;
5731 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5732 bond_dev->features |= bond_dev->hw_features;
5733 bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5734 #ifdef CONFIG_XFRM_OFFLOAD
5735 bond_dev->hw_features |= BOND_XFRM_FEATURES;
5736 /* Only enable XFRM features if this is an active-backup config */
5737 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5738 bond_dev->features |= BOND_XFRM_FEATURES;
5739 #endif /* CONFIG_XFRM_OFFLOAD */
5740 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5741 if (bond_sk_check(bond))
5742 bond_dev->features |= BOND_TLS_FEATURES;
5746 /* Destroy a bonding device.
5747 * Must be under rtnl_lock when this function is called.
5749 static void bond_uninit(struct net_device *bond_dev)
5751 struct bonding *bond = netdev_priv(bond_dev);
5752 struct bond_up_slave *usable, *all;
5753 struct list_head *iter;
5754 struct slave *slave;
5756 bond_netpoll_cleanup(bond_dev);
5758 /* Release the bonded slaves */
5759 bond_for_each_slave(bond, slave, iter)
5760 __bond_release_one(bond_dev, slave->dev, true, true);
5761 netdev_info(bond_dev, "Released all slaves\n");
5763 usable = rtnl_dereference(bond->usable_slaves);
5765 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5766 kfree_rcu(usable, rcu);
5769 all = rtnl_dereference(bond->all_slaves);
5771 RCU_INIT_POINTER(bond->all_slaves, NULL);
5772 kfree_rcu(all, rcu);
5775 list_del(&bond->bond_list);
5777 bond_debug_unregister(bond);
5780 /*------------------------- Module initialization ---------------------------*/
5782 static int bond_check_params(struct bond_params *params)
5784 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5785 struct bond_opt_value newval;
5786 const struct bond_opt_value *valptr;
5787 int arp_all_targets_value = 0;
5788 u16 ad_actor_sys_prio = 0;
5789 u16 ad_user_port_key = 0;
5790 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5792 int bond_mode = BOND_MODE_ROUNDROBIN;
5793 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5797 /* Convert string parameters. */
5799 bond_opt_initstr(&newval, mode);
5800 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5802 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5805 bond_mode = valptr->value;
5808 if (xmit_hash_policy) {
5809 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5810 bond_mode == BOND_MODE_ACTIVEBACKUP ||
5811 bond_mode == BOND_MODE_BROADCAST) {
5812 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5813 bond_mode_name(bond_mode));
5815 bond_opt_initstr(&newval, xmit_hash_policy);
5816 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5819 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5823 xmit_hashtype = valptr->value;
5828 if (bond_mode != BOND_MODE_8023AD) {
5829 pr_info("lacp_rate param is irrelevant in mode %s\n",
5830 bond_mode_name(bond_mode));
5832 bond_opt_initstr(&newval, lacp_rate);
5833 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5836 pr_err("Error: Invalid lacp rate \"%s\"\n",
5840 lacp_fast = valptr->value;
5845 bond_opt_initstr(&newval, ad_select);
5846 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5849 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5852 params->ad_select = valptr->value;
5853 if (bond_mode != BOND_MODE_8023AD)
5854 pr_warn("ad_select param only affects 802.3ad mode\n");
5856 params->ad_select = BOND_AD_STABLE;
5859 if (max_bonds < 0) {
5860 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5861 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5862 max_bonds = BOND_DEFAULT_MAX_BONDS;
5866 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5872 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5877 if (downdelay < 0) {
5878 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5879 downdelay, INT_MAX);
5883 if ((use_carrier != 0) && (use_carrier != 1)) {
5884 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5889 if (num_peer_notif < 0 || num_peer_notif > 255) {
5890 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5895 /* reset values for 802.3ad/TLB/ALB */
5896 if (!bond_mode_uses_arp(bond_mode)) {
5898 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");
5899 pr_warn("Forcing miimon to 100msec\n");
5900 miimon = BOND_DEFAULT_MIIMON;
5904 if (tx_queues < 1 || tx_queues > 255) {
5905 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5906 tx_queues, BOND_DEFAULT_TX_QUEUES);
5907 tx_queues = BOND_DEFAULT_TX_QUEUES;
5910 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5911 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5913 all_slaves_active = 0;
5916 if (resend_igmp < 0 || resend_igmp > 255) {
5917 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5918 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5919 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5922 bond_opt_initval(&newval, packets_per_slave);
5923 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5924 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5925 packets_per_slave, USHRT_MAX);
5926 packets_per_slave = 1;
5929 if (bond_mode == BOND_MODE_ALB) {
5930 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",
5935 if (updelay || downdelay) {
5936 /* just warn the user the up/down delay will have
5937 * no effect since miimon is zero...
5939 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",
5940 updelay, downdelay);
5943 /* don't allow arp monitoring */
5945 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5946 miimon, arp_interval);
5950 if ((updelay % miimon) != 0) {
5951 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5952 updelay, miimon, (updelay / miimon) * miimon);
5957 if ((downdelay % miimon) != 0) {
5958 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5960 (downdelay / miimon) * miimon);
5963 downdelay /= miimon;
5966 if (arp_interval < 0) {
5967 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5968 arp_interval, INT_MAX);
5972 for (arp_ip_count = 0, i = 0;
5973 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5976 /* not a complete check, but good enough to catch mistakes */
5977 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5978 !bond_is_ip_target_ok(ip)) {
5979 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5983 if (bond_get_targets_ip(arp_target, ip) == -1)
5984 arp_target[arp_ip_count++] = ip;
5986 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5991 if (arp_interval && !arp_ip_count) {
5992 /* don't allow arping if no arp_ip_target given... */
5993 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5999 if (!arp_interval) {
6000 pr_err("arp_validate requires arp_interval\n");
6004 bond_opt_initstr(&newval, arp_validate);
6005 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6008 pr_err("Error: invalid arp_validate \"%s\"\n",
6012 arp_validate_value = valptr->value;
6014 arp_validate_value = 0;
6017 if (arp_all_targets) {
6018 bond_opt_initstr(&newval, arp_all_targets);
6019 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6022 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6024 arp_all_targets_value = 0;
6026 arp_all_targets_value = valptr->value;
6031 pr_info("MII link monitoring set to %d ms\n", miimon);
6032 } else if (arp_interval) {
6033 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6034 arp_validate_value);
6035 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6036 arp_interval, valptr->string, arp_ip_count);
6038 for (i = 0; i < arp_ip_count; i++)
6039 pr_cont(" %s", arp_ip_target[i]);
6043 } else if (max_bonds) {
6044 /* miimon and arp_interval not set, we need one so things
6045 * work as expected, see bonding.txt for details
6047 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");
6050 if (primary && !bond_mode_uses_primary(bond_mode)) {
6051 /* currently, using a primary only makes sense
6052 * in active backup, TLB or ALB modes
6054 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6055 primary, bond_mode_name(bond_mode));
6059 if (primary && primary_reselect) {
6060 bond_opt_initstr(&newval, primary_reselect);
6061 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6064 pr_err("Error: Invalid primary_reselect \"%s\"\n",
6068 primary_reselect_value = valptr->value;
6070 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6073 if (fail_over_mac) {
6074 bond_opt_initstr(&newval, fail_over_mac);
6075 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6078 pr_err("Error: invalid fail_over_mac \"%s\"\n",
6082 fail_over_mac_value = valptr->value;
6083 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6084 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6086 fail_over_mac_value = BOND_FOM_NONE;
6089 bond_opt_initstr(&newval, "default");
6090 valptr = bond_opt_parse(
6091 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6094 pr_err("Error: No ad_actor_sys_prio default value");
6097 ad_actor_sys_prio = valptr->value;
6099 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6102 pr_err("Error: No ad_user_port_key default value");
6105 ad_user_port_key = valptr->value;
6107 bond_opt_initstr(&newval, "default");
6108 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6110 pr_err("Error: No tlb_dynamic_lb default value");
6113 tlb_dynamic_lb = valptr->value;
6115 if (lp_interval == 0) {
6116 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6117 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6118 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6121 /* fill params struct with the proper values */
6122 params->mode = bond_mode;
6123 params->xmit_policy = xmit_hashtype;
6124 params->miimon = miimon;
6125 params->num_peer_notif = num_peer_notif;
6126 params->arp_interval = arp_interval;
6127 params->arp_validate = arp_validate_value;
6128 params->arp_all_targets = arp_all_targets_value;
6129 params->missed_max = 2;
6130 params->updelay = updelay;
6131 params->downdelay = downdelay;
6132 params->peer_notif_delay = 0;
6133 params->use_carrier = use_carrier;
6134 params->lacp_active = 1;
6135 params->lacp_fast = lacp_fast;
6136 params->primary[0] = 0;
6137 params->primary_reselect = primary_reselect_value;
6138 params->fail_over_mac = fail_over_mac_value;
6139 params->tx_queues = tx_queues;
6140 params->all_slaves_active = all_slaves_active;
6141 params->resend_igmp = resend_igmp;
6142 params->min_links = min_links;
6143 params->lp_interval = lp_interval;
6144 params->packets_per_slave = packets_per_slave;
6145 params->tlb_dynamic_lb = tlb_dynamic_lb;
6146 params->ad_actor_sys_prio = ad_actor_sys_prio;
6147 eth_zero_addr(params->ad_actor_system);
6148 params->ad_user_port_key = ad_user_port_key;
6149 if (packets_per_slave > 0) {
6150 params->reciprocal_packets_per_slave =
6151 reciprocal_value(packets_per_slave);
6153 /* reciprocal_packets_per_slave is unused if
6154 * packets_per_slave is 0 or 1, just initialize it
6156 params->reciprocal_packets_per_slave =
6157 (struct reciprocal_value) { 0 };
6161 strscpy_pad(params->primary, primary, sizeof(params->primary));
6163 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6164 #if IS_ENABLED(CONFIG_IPV6)
6165 memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6171 /* Called from registration process */
6172 static int bond_init(struct net_device *bond_dev)
6174 struct bonding *bond = netdev_priv(bond_dev);
6175 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6177 netdev_dbg(bond_dev, "Begin bond_init\n");
6179 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6183 if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
6184 bond->rr_tx_counter = alloc_percpu(u32);
6185 if (!bond->rr_tx_counter) {
6186 destroy_workqueue(bond->wq);
6192 spin_lock_init(&bond->stats_lock);
6193 netdev_lockdep_set_classes(bond_dev);
6195 list_add_tail(&bond->bond_list, &bn->dev_list);
6197 bond_prepare_sysfs_group(bond);
6199 bond_debug_register(bond);
6201 /* Ensure valid dev_addr */
6202 if (is_zero_ether_addr(bond_dev->dev_addr) &&
6203 bond_dev->addr_assign_type == NET_ADDR_PERM)
6204 eth_hw_addr_random(bond_dev);
6209 unsigned int bond_get_num_tx_queues(void)
6214 /* Create a new bond based on the specified name and bonding parameters.
6215 * If name is NULL, obtain a suitable "bond%d" name for us.
6216 * Caller must NOT hold rtnl_lock; we need to release it here before we
6217 * set up our sysfs entries.
6219 int bond_create(struct net *net, const char *name)
6221 struct net_device *bond_dev;
6222 struct bonding *bond;
6223 struct alb_bond_info *bond_info;
6228 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6229 name ? name : "bond%d", NET_NAME_UNKNOWN,
6230 bond_setup, tx_queues);
6232 pr_err("%s: eek! can't alloc netdev!\n", name);
6238 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
6239 * It is set to 0 by default which is wrong.
6241 bond = netdev_priv(bond_dev);
6242 bond_info = &(BOND_ALB_INFO(bond));
6243 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
6245 dev_net_set(bond_dev, net);
6246 bond_dev->rtnl_link_ops = &bond_link_ops;
6248 res = register_netdevice(bond_dev);
6250 free_netdev(bond_dev);
6256 netif_carrier_off(bond_dev);
6258 bond_work_init_all(bond);
6264 static int __net_init bond_net_init(struct net *net)
6266 struct bond_net *bn = net_generic(net, bond_net_id);
6269 INIT_LIST_HEAD(&bn->dev_list);
6271 bond_create_proc_dir(bn);
6272 bond_create_sysfs(bn);
6277 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6279 struct bond_net *bn;
6283 list_for_each_entry(net, net_list, exit_list) {
6284 bn = net_generic(net, bond_net_id);
6285 bond_destroy_sysfs(bn);
6288 /* Kill off any bonds created after unregistering bond rtnl ops */
6290 list_for_each_entry(net, net_list, exit_list) {
6291 struct bonding *bond, *tmp_bond;
6293 bn = net_generic(net, bond_net_id);
6294 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6295 unregister_netdevice_queue(bond->dev, &list);
6297 unregister_netdevice_many(&list);
6300 list_for_each_entry(net, net_list, exit_list) {
6301 bn = net_generic(net, bond_net_id);
6302 bond_destroy_proc_dir(bn);
6306 static struct pernet_operations bond_net_ops = {
6307 .init = bond_net_init,
6308 .exit_batch = bond_net_exit_batch,
6310 .size = sizeof(struct bond_net),
6313 static int __init bonding_init(void)
6318 res = bond_check_params(&bonding_defaults);
6322 res = register_pernet_subsys(&bond_net_ops);
6326 res = bond_netlink_init();
6330 bond_create_debugfs();
6332 for (i = 0; i < max_bonds; i++) {
6333 res = bond_create(&init_net, NULL);
6338 skb_flow_dissector_init(&flow_keys_bonding,
6339 flow_keys_bonding_keys,
6340 ARRAY_SIZE(flow_keys_bonding_keys));
6342 register_netdevice_notifier(&bond_netdev_notifier);
6346 bond_destroy_debugfs();
6347 bond_netlink_fini();
6349 unregister_pernet_subsys(&bond_net_ops);
6354 static void __exit bonding_exit(void)
6356 unregister_netdevice_notifier(&bond_netdev_notifier);
6358 bond_destroy_debugfs();
6360 bond_netlink_fini();
6361 unregister_pernet_subsys(&bond_net_ops);
6363 #ifdef CONFIG_NET_POLL_CONTROLLER
6364 /* Make sure we don't have an imbalance on our netpoll blocking */
6365 WARN_ON(atomic_read(&netpoll_block_tx));
6369 module_init(bonding_init);
6370 module_exit(bonding_exit);
6371 MODULE_LICENSE("GPL");
6372 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6373 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");