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/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
149 "2 for only on active slave "
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
202 int bond_net_id __read_mostly;
204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
205 static int arp_ip_count;
206 static int bond_mode = BOND_MODE_ROUNDROBIN;
207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
208 static int lacp_fast;
210 /*-------------------------- Forward declarations ---------------------------*/
212 static int bond_init(struct net_device *bond_dev);
213 static void bond_uninit(struct net_device *bond_dev);
214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
215 struct rtnl_link_stats64 *stats);
216 static void bond_slave_arr_handler(struct work_struct *work);
218 /*---------------------------- General routines -----------------------------*/
220 const char *bond_mode_name(int mode)
222 static const char *names[] = {
223 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
224 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
225 [BOND_MODE_XOR] = "load balancing (xor)",
226 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
227 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
228 [BOND_MODE_TLB] = "transmit load balancing",
229 [BOND_MODE_ALB] = "adaptive load balancing",
232 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
238 /*---------------------------------- VLAN -----------------------------------*/
241 * bond_dev_queue_xmit - Prepare skb for xmit.
243 * @bond: bond device that got this skb for tx.
244 * @skb: hw accel VLAN tagged skb to transmit
245 * @slave_dev: slave that is supposed to xmit this skbuff
247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
248 struct net_device *slave_dev)
250 skb->dev = slave_dev;
252 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
253 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
254 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
256 if (unlikely(netpoll_tx_running(bond->dev)))
257 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
263 * We don't protect the slave list iteration with a lock because:
264 * a. This operation is performed in IOCTL context,
265 * b. The operation is protected by the RTNL semaphore in the 8021q code,
266 * c. Holding a lock with BH disabled while directly calling a base driver
267 * entry point is generally a BAD idea.
269 * The design of synchronization/protection for this operation in the 8021q
270 * module is good for one or more VLAN devices over a single physical device
271 * and cannot be extended for a teaming solution like bonding, so there is a
272 * potential race condition here where a net device from the vlan group might
273 * be referenced (either by a base driver or the 8021q code) while it is being
274 * removed from the system. However, it turns out we're not making matters
275 * worse, and if it works for regular VLAN usage it will work here too.
279 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
280 * @bond_dev: bonding net device that got called
281 * @vid: vlan id being added
283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
284 __be16 proto, u16 vid)
286 struct bonding *bond = netdev_priv(bond_dev);
287 struct slave *slave, *rollback_slave;
288 struct list_head *iter;
291 bond_for_each_slave(bond, slave, iter) {
292 res = vlan_vid_add(slave->dev, proto, vid);
300 /* unwind to the slave that failed */
301 bond_for_each_slave(bond, rollback_slave, iter) {
302 if (rollback_slave == slave)
305 vlan_vid_del(rollback_slave->dev, proto, vid);
312 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
313 * @bond_dev: bonding net device that got called
314 * @vid: vlan id being removed
316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
317 __be16 proto, u16 vid)
319 struct bonding *bond = netdev_priv(bond_dev);
320 struct list_head *iter;
323 bond_for_each_slave(bond, slave, iter)
324 vlan_vid_del(slave->dev, proto, vid);
326 if (bond_is_lb(bond))
327 bond_alb_clear_vlan(bond, vid);
332 /*------------------------------- Link status -------------------------------*/
334 /* Set the carrier state for the master according to the state of its
335 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
336 * do special 802.3ad magic.
338 * Returns zero if carrier state does not change, nonzero if it does.
340 int bond_set_carrier(struct bonding *bond)
342 struct list_head *iter;
345 if (!bond_has_slaves(bond))
348 if (BOND_MODE(bond) == BOND_MODE_8023AD)
349 return bond_3ad_set_carrier(bond);
351 bond_for_each_slave(bond, slave, iter) {
352 if (slave->link == BOND_LINK_UP) {
353 if (!netif_carrier_ok(bond->dev)) {
354 netif_carrier_on(bond->dev);
362 if (netif_carrier_ok(bond->dev)) {
363 netif_carrier_off(bond->dev);
369 /* Get link speed and duplex from the slave's base driver
370 * using ethtool. If for some reason the call fails or the
371 * values are invalid, set speed and duplex to -1,
374 static void bond_update_speed_duplex(struct slave *slave)
376 struct net_device *slave_dev = slave->dev;
377 struct ethtool_cmd ecmd;
381 slave->speed = SPEED_UNKNOWN;
382 slave->duplex = DUPLEX_UNKNOWN;
384 res = __ethtool_get_settings(slave_dev, &ecmd);
388 slave_speed = ethtool_cmd_speed(&ecmd);
389 if (slave_speed == 0 || slave_speed == ((__u32) -1))
392 switch (ecmd.duplex) {
400 slave->speed = slave_speed;
401 slave->duplex = ecmd.duplex;
406 const char *bond_slave_link_status(s8 link)
422 /* if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
443 struct mii_ioctl_data *mii;
445 if (!reporting && !netif_running(slave_dev))
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
459 /* TODO: set pointer to correct ioctl on a per team member
460 * bases to make this more efficient. that is, once
461 * we determine the correct ioctl, we will always
462 * call it and not the others for that team
466 /* We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
481 /* If reporting, report that either there's no dev->do_ioctl,
482 * or both SIOCGMIIREG and get_link failed (meaning that we
483 * cannot report link status). If not reporting, pretend
486 return reporting ? -1 : BMSR_LSTATUS;
489 /*----------------------------- Multicast list ------------------------------*/
491 /* Push the promiscuity flag down to appropriate slaves */
492 static int bond_set_promiscuity(struct bonding *bond, int inc)
494 struct list_head *iter;
497 if (bond_uses_primary(bond)) {
498 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
501 err = dev_set_promiscuity(curr_active->dev, inc);
505 bond_for_each_slave(bond, slave, iter) {
506 err = dev_set_promiscuity(slave->dev, inc);
514 /* Push the allmulti flag down to all slaves */
515 static int bond_set_allmulti(struct bonding *bond, int inc)
517 struct list_head *iter;
520 if (bond_uses_primary(bond)) {
521 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
524 err = dev_set_allmulti(curr_active->dev, inc);
528 bond_for_each_slave(bond, slave, iter) {
529 err = dev_set_allmulti(slave->dev, inc);
537 /* Retrieve the list of registered multicast addresses for the bonding
538 * device and retransmit an IGMP JOIN request to the current active
541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
543 struct bonding *bond = container_of(work, struct bonding,
546 if (!rtnl_trylock()) {
547 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
550 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
552 if (bond->igmp_retrans > 1) {
553 bond->igmp_retrans--;
554 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
559 /* Flush bond's hardware addresses from slave */
560 static void bond_hw_addr_flush(struct net_device *bond_dev,
561 struct net_device *slave_dev)
563 struct bonding *bond = netdev_priv(bond_dev);
565 dev_uc_unsync(slave_dev, bond_dev);
566 dev_mc_unsync(slave_dev, bond_dev);
568 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
569 /* del lacpdu mc addr from mc list */
570 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
572 dev_mc_del(slave_dev, lacpdu_multicast);
576 /*--------------------------- Active slave change ---------------------------*/
578 /* Update the hardware address list and promisc/allmulti for the new and
579 * old active slaves (if any). Modes that are not using primary keep all
580 * slaves up date at all times; only the modes that use primary need to call
581 * this function to swap these settings during a failover.
583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
584 struct slave *old_active)
587 if (bond->dev->flags & IFF_PROMISC)
588 dev_set_promiscuity(old_active->dev, -1);
590 if (bond->dev->flags & IFF_ALLMULTI)
591 dev_set_allmulti(old_active->dev, -1);
593 bond_hw_addr_flush(bond->dev, old_active->dev);
597 /* FIXME: Signal errors upstream. */
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(new_active->dev, 1);
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(new_active->dev, 1);
604 netif_addr_lock_bh(bond->dev);
605 dev_uc_sync(new_active->dev, bond->dev);
606 dev_mc_sync(new_active->dev, bond->dev);
607 netif_addr_unlock_bh(bond->dev);
612 * bond_set_dev_addr - clone slave's address to bond
613 * @bond_dev: bond net device
614 * @slave_dev: slave net device
616 * Should be called with RTNL held.
618 static void bond_set_dev_addr(struct net_device *bond_dev,
619 struct net_device *slave_dev)
621 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
622 bond_dev, slave_dev, slave_dev->addr_len);
623 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
624 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
625 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
628 /* bond_do_fail_over_mac
630 * Perform special MAC address swapping for fail_over_mac settings
634 static void bond_do_fail_over_mac(struct bonding *bond,
635 struct slave *new_active,
636 struct slave *old_active)
638 u8 tmp_mac[ETH_ALEN];
639 struct sockaddr saddr;
642 switch (bond->params.fail_over_mac) {
643 case BOND_FOM_ACTIVE:
645 bond_set_dev_addr(bond->dev, new_active->dev);
647 case BOND_FOM_FOLLOW:
648 /* if new_active && old_active, swap them
649 * if just old_active, do nothing (going to no active slave)
650 * if just new_active, set new_active to bond's MAC
656 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
657 ether_addr_copy(saddr.sa_data,
658 old_active->dev->dev_addr);
659 saddr.sa_family = new_active->dev->type;
661 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
662 saddr.sa_family = bond->dev->type;
665 rv = dev_set_mac_address(new_active->dev, &saddr);
667 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
668 -rv, new_active->dev->name);
675 ether_addr_copy(saddr.sa_data, tmp_mac);
676 saddr.sa_family = old_active->dev->type;
678 rv = dev_set_mac_address(old_active->dev, &saddr);
680 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
681 -rv, new_active->dev->name);
685 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
686 bond->params.fail_over_mac);
692 static bool bond_should_change_active(struct bonding *bond)
694 struct slave *prim = rtnl_dereference(bond->primary_slave);
695 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
697 if (!prim || !curr || curr->link != BOND_LINK_UP)
699 if (bond->force_primary) {
700 bond->force_primary = false;
703 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
704 (prim->speed < curr->speed ||
705 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
707 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
713 * find_best_interface - select the best available slave to be the active one
714 * @bond: our bonding struct
716 static struct slave *bond_find_best_slave(struct bonding *bond)
718 struct slave *slave, *bestslave = NULL, *primary;
719 struct list_head *iter;
720 int mintime = bond->params.updelay;
722 primary = rtnl_dereference(bond->primary_slave);
723 if (primary && primary->link == BOND_LINK_UP &&
724 bond_should_change_active(bond))
727 bond_for_each_slave(bond, slave, iter) {
728 if (slave->link == BOND_LINK_UP)
730 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
731 slave->delay < mintime) {
732 mintime = slave->delay;
740 static bool bond_should_notify_peers(struct bonding *bond)
745 slave = rcu_dereference(bond->curr_active_slave);
748 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
749 slave ? slave->dev->name : "NULL");
751 if (!slave || !bond->send_peer_notif ||
752 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
759 * change_active_interface - change the active slave into the specified one
760 * @bond: our bonding struct
761 * @new: the new slave to make the active one
763 * Set the new slave to the bond's settings and unset them on the old
765 * Setting include flags, mc-list, promiscuity, allmulti, etc.
767 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
768 * because it is apparently the best available slave we have, even though its
769 * updelay hasn't timed out yet.
771 * Caller must hold RTNL.
773 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
775 struct slave *old_active;
779 old_active = rtnl_dereference(bond->curr_active_slave);
781 if (old_active == new_active)
785 new_active->last_link_up = jiffies;
787 if (new_active->link == BOND_LINK_BACK) {
788 if (bond_uses_primary(bond)) {
789 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
790 new_active->dev->name,
791 (bond->params.updelay - new_active->delay) * bond->params.miimon);
794 new_active->delay = 0;
795 bond_set_slave_link_state(new_active, BOND_LINK_UP);
797 if (BOND_MODE(bond) == BOND_MODE_8023AD)
798 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
800 if (bond_is_lb(bond))
801 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
803 if (bond_uses_primary(bond)) {
804 netdev_info(bond->dev, "making interface %s the new active one\n",
805 new_active->dev->name);
810 if (bond_uses_primary(bond))
811 bond_hw_addr_swap(bond, new_active, old_active);
813 if (bond_is_lb(bond)) {
814 bond_alb_handle_active_change(bond, new_active);
816 bond_set_slave_inactive_flags(old_active,
817 BOND_SLAVE_NOTIFY_NOW);
819 bond_set_slave_active_flags(new_active,
820 BOND_SLAVE_NOTIFY_NOW);
822 rcu_assign_pointer(bond->curr_active_slave, new_active);
825 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
827 bond_set_slave_inactive_flags(old_active,
828 BOND_SLAVE_NOTIFY_NOW);
831 bool should_notify_peers = false;
833 bond_set_slave_active_flags(new_active,
834 BOND_SLAVE_NOTIFY_NOW);
836 if (bond->params.fail_over_mac)
837 bond_do_fail_over_mac(bond, new_active,
840 if (netif_running(bond->dev)) {
841 bond->send_peer_notif =
842 bond->params.num_peer_notif;
843 should_notify_peers =
844 bond_should_notify_peers(bond);
847 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
848 if (should_notify_peers)
849 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
854 /* resend IGMP joins since active slave has changed or
855 * all were sent on curr_active_slave.
856 * resend only if bond is brought up with the affected
857 * bonding modes and the retransmission is enabled
859 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
860 ((bond_uses_primary(bond) && new_active) ||
861 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
862 bond->igmp_retrans = bond->params.resend_igmp;
863 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
868 * bond_select_active_slave - select a new active slave, if needed
869 * @bond: our bonding struct
871 * This functions should be called when one of the following occurs:
872 * - The old curr_active_slave has been released or lost its link.
873 * - The primary_slave has got its link back.
874 * - A slave has got its link back and there's no old curr_active_slave.
876 * Caller must hold RTNL.
878 void bond_select_active_slave(struct bonding *bond)
880 struct slave *best_slave;
885 best_slave = bond_find_best_slave(bond);
886 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
887 bond_change_active_slave(bond, best_slave);
888 rv = bond_set_carrier(bond);
892 if (netif_carrier_ok(bond->dev)) {
893 netdev_info(bond->dev, "first active interface up!\n");
895 netdev_info(bond->dev, "now running without any active interface!\n");
900 #ifdef CONFIG_NET_POLL_CONTROLLER
901 static inline int slave_enable_netpoll(struct slave *slave)
906 np = kzalloc(sizeof(*np), GFP_KERNEL);
911 err = __netpoll_setup(np, slave->dev);
920 static inline void slave_disable_netpoll(struct slave *slave)
922 struct netpoll *np = slave->np;
928 __netpoll_free_async(np);
931 static void bond_poll_controller(struct net_device *bond_dev)
933 struct bonding *bond = netdev_priv(bond_dev);
934 struct slave *slave = NULL;
935 struct list_head *iter;
936 struct ad_info ad_info;
937 struct netpoll_info *ni;
938 const struct net_device_ops *ops;
940 if (BOND_MODE(bond) == BOND_MODE_8023AD)
941 if (bond_3ad_get_active_agg_info(bond, &ad_info))
945 bond_for_each_slave_rcu(bond, slave, iter) {
946 ops = slave->dev->netdev_ops;
947 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
950 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
951 struct aggregator *agg =
952 SLAVE_AD_INFO(slave)->port.aggregator;
955 agg->aggregator_identifier != ad_info.aggregator_id)
959 ni = rcu_dereference_bh(slave->dev->npinfo);
960 if (down_trylock(&ni->dev_lock))
962 ops->ndo_poll_controller(slave->dev);
965 rcu_read_unlock_bh();
968 static void bond_netpoll_cleanup(struct net_device *bond_dev)
970 struct bonding *bond = netdev_priv(bond_dev);
971 struct list_head *iter;
974 bond_for_each_slave(bond, slave, iter)
975 if (bond_slave_is_up(slave))
976 slave_disable_netpoll(slave);
979 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
981 struct bonding *bond = netdev_priv(dev);
982 struct list_head *iter;
986 bond_for_each_slave(bond, slave, iter) {
987 err = slave_enable_netpoll(slave);
989 bond_netpoll_cleanup(dev);
996 static inline int slave_enable_netpoll(struct slave *slave)
1000 static inline void slave_disable_netpoll(struct slave *slave)
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1008 /*---------------------------------- IOCTL ----------------------------------*/
1010 static netdev_features_t bond_fix_features(struct net_device *dev,
1011 netdev_features_t features)
1013 struct bonding *bond = netdev_priv(dev);
1014 struct list_head *iter;
1015 netdev_features_t mask;
1016 struct slave *slave;
1020 features &= ~NETIF_F_ONE_FOR_ALL;
1021 features |= NETIF_F_ALL_FOR_ALL;
1023 bond_for_each_slave(bond, slave, iter) {
1024 features = netdev_increment_features(features,
1025 slave->dev->features,
1028 features = netdev_add_tso_features(features, mask);
1033 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1034 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1035 NETIF_F_HIGHDMA | NETIF_F_LRO)
1037 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1040 static void bond_compute_features(struct bonding *bond)
1042 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1043 IFF_XMIT_DST_RELEASE_PERM;
1044 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1045 netdev_features_t enc_features = BOND_ENC_FEATURES;
1046 struct net_device *bond_dev = bond->dev;
1047 struct list_head *iter;
1048 struct slave *slave;
1049 unsigned short max_hard_header_len = ETH_HLEN;
1050 unsigned int gso_max_size = GSO_MAX_SIZE;
1051 u16 gso_max_segs = GSO_MAX_SEGS;
1053 if (!bond_has_slaves(bond))
1055 vlan_features &= NETIF_F_ALL_FOR_ALL;
1057 bond_for_each_slave(bond, slave, iter) {
1058 vlan_features = netdev_increment_features(vlan_features,
1059 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1061 enc_features = netdev_increment_features(enc_features,
1062 slave->dev->hw_enc_features,
1064 dst_release_flag &= slave->dev->priv_flags;
1065 if (slave->dev->hard_header_len > max_hard_header_len)
1066 max_hard_header_len = slave->dev->hard_header_len;
1068 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1069 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1073 bond_dev->vlan_features = vlan_features;
1074 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1075 bond_dev->hard_header_len = max_hard_header_len;
1076 bond_dev->gso_max_segs = gso_max_segs;
1077 netif_set_gso_max_size(bond_dev, gso_max_size);
1079 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1080 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1081 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1082 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1084 netdev_change_features(bond_dev);
1087 static void bond_setup_by_slave(struct net_device *bond_dev,
1088 struct net_device *slave_dev)
1090 bond_dev->header_ops = slave_dev->header_ops;
1092 bond_dev->type = slave_dev->type;
1093 bond_dev->hard_header_len = slave_dev->hard_header_len;
1094 bond_dev->addr_len = slave_dev->addr_len;
1096 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1097 slave_dev->addr_len);
1100 /* On bonding slaves other than the currently active slave, suppress
1101 * duplicates except for alb non-mcast/bcast.
1103 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1104 struct slave *slave,
1105 struct bonding *bond)
1107 if (bond_is_slave_inactive(slave)) {
1108 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1109 skb->pkt_type != PACKET_BROADCAST &&
1110 skb->pkt_type != PACKET_MULTICAST)
1117 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1119 struct sk_buff *skb = *pskb;
1120 struct slave *slave;
1121 struct bonding *bond;
1122 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1124 int ret = RX_HANDLER_ANOTHER;
1126 skb = skb_share_check(skb, GFP_ATOMIC);
1128 return RX_HANDLER_CONSUMED;
1132 slave = bond_slave_get_rcu(skb->dev);
1135 recv_probe = ACCESS_ONCE(bond->recv_probe);
1137 ret = recv_probe(skb, bond, slave);
1138 if (ret == RX_HANDLER_CONSUMED) {
1144 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1145 return RX_HANDLER_EXACT;
1148 skb->dev = bond->dev;
1150 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1151 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1152 skb->pkt_type == PACKET_HOST) {
1154 if (unlikely(skb_cow_head(skb,
1155 skb->data - skb_mac_header(skb)))) {
1157 return RX_HANDLER_CONSUMED;
1159 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1165 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1166 struct net_device *slave_dev,
1167 struct slave *slave)
1171 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1174 slave_dev->flags |= IFF_SLAVE;
1175 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1179 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1180 struct net_device *slave_dev)
1182 netdev_upper_dev_unlink(slave_dev, bond_dev);
1183 slave_dev->flags &= ~IFF_SLAVE;
1184 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1187 static struct slave *bond_alloc_slave(struct bonding *bond)
1189 struct slave *slave = NULL;
1191 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1195 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1196 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1198 if (!SLAVE_AD_INFO(slave)) {
1206 static void bond_free_slave(struct slave *slave)
1208 struct bonding *bond = bond_get_bond_by_slave(slave);
1210 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1211 kfree(SLAVE_AD_INFO(slave));
1216 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1218 info->bond_mode = BOND_MODE(bond);
1219 info->miimon = bond->params.miimon;
1220 info->num_slaves = bond->slave_cnt;
1223 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1225 strcpy(info->slave_name, slave->dev->name);
1226 info->link = slave->link;
1227 info->state = bond_slave_state(slave);
1228 info->link_failure_count = slave->link_failure_count;
1231 static void bond_netdev_notify(struct net_device *dev,
1232 struct netdev_bonding_info *info)
1235 netdev_bonding_info_change(dev, info);
1239 static void bond_netdev_notify_work(struct work_struct *_work)
1241 struct netdev_notify_work *w =
1242 container_of(_work, struct netdev_notify_work, work.work);
1244 bond_netdev_notify(w->dev, &w->bonding_info);
1249 void bond_queue_slave_event(struct slave *slave)
1251 struct bonding *bond = slave->bond;
1252 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1257 dev_hold(slave->dev);
1258 nnw->dev = slave->dev;
1259 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1260 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1261 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1263 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1266 /* enslave device <slave> to bond device <master> */
1267 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1269 struct bonding *bond = netdev_priv(bond_dev);
1270 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1271 struct slave *new_slave = NULL, *prev_slave;
1272 struct sockaddr addr;
1276 if (!bond->params.use_carrier &&
1277 slave_dev->ethtool_ops->get_link == NULL &&
1278 slave_ops->ndo_do_ioctl == NULL) {
1279 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1283 /* already enslaved */
1284 if (slave_dev->flags & IFF_SLAVE) {
1285 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1289 if (bond_dev == slave_dev) {
1290 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1294 /* vlan challenged mutual exclusion */
1295 /* no need to lock since we're protected by rtnl_lock */
1296 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1297 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1299 if (vlan_uses_dev(bond_dev)) {
1300 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1301 slave_dev->name, bond_dev->name);
1304 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1305 slave_dev->name, slave_dev->name,
1309 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1313 /* Old ifenslave binaries are no longer supported. These can
1314 * be identified with moderate accuracy by the state of the slave:
1315 * the current ifenslave will set the interface down prior to
1316 * enslaving it; the old ifenslave will not.
1318 if ((slave_dev->flags & IFF_UP)) {
1319 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1322 goto err_undo_flags;
1325 /* set bonding device ether type by slave - bonding netdevices are
1326 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1327 * there is a need to override some of the type dependent attribs/funcs.
1329 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1330 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1332 if (!bond_has_slaves(bond)) {
1333 if (bond_dev->type != slave_dev->type) {
1334 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1335 bond_dev->type, slave_dev->type);
1337 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1339 res = notifier_to_errno(res);
1341 netdev_err(bond_dev, "refused to change device type\n");
1343 goto err_undo_flags;
1346 /* Flush unicast and multicast addresses */
1347 dev_uc_flush(bond_dev);
1348 dev_mc_flush(bond_dev);
1350 if (slave_dev->type != ARPHRD_ETHER)
1351 bond_setup_by_slave(bond_dev, slave_dev);
1353 ether_setup(bond_dev);
1354 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1357 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1360 } else if (bond_dev->type != slave_dev->type) {
1361 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1362 slave_dev->name, slave_dev->type, bond_dev->type);
1364 goto err_undo_flags;
1367 if (slave_ops->ndo_set_mac_address == NULL) {
1368 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1369 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1370 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1371 if (!bond_has_slaves(bond)) {
1372 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1373 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1375 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1377 goto err_undo_flags;
1382 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1384 /* If this is the first slave, then we need to set the master's hardware
1385 * address to be the same as the slave's.
1387 if (!bond_has_slaves(bond) &&
1388 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1389 bond_set_dev_addr(bond->dev, slave_dev);
1391 new_slave = bond_alloc_slave(bond);
1394 goto err_undo_flags;
1397 new_slave->bond = bond;
1398 new_slave->dev = slave_dev;
1399 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1400 * is set via sysfs or module option if desired.
1402 new_slave->queue_id = 0;
1404 /* Save slave's original mtu and then set it to match the bond */
1405 new_slave->original_mtu = slave_dev->mtu;
1406 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1408 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1412 /* Save slave's original ("permanent") mac address for modes
1413 * that need it, and for restoring it upon release, and then
1414 * set it to the master's address
1416 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1418 if (!bond->params.fail_over_mac ||
1419 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1420 /* Set slave to master's mac address. The application already
1421 * set the master's mac address to that of the first slave
1423 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1424 addr.sa_family = slave_dev->type;
1425 res = dev_set_mac_address(slave_dev, &addr);
1427 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1428 goto err_restore_mtu;
1432 /* open the slave since the application closed it */
1433 res = dev_open(slave_dev);
1435 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1436 goto err_restore_mac;
1439 slave_dev->priv_flags |= IFF_BONDING;
1440 /* initialize slave stats */
1441 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1443 if (bond_is_lb(bond)) {
1444 /* bond_alb_init_slave() must be called before all other stages since
1445 * it might fail and we do not want to have to undo everything
1447 res = bond_alb_init_slave(bond, new_slave);
1452 /* If the mode uses primary, then the following is handled by
1453 * bond_change_active_slave().
1455 if (!bond_uses_primary(bond)) {
1456 /* set promiscuity level to new slave */
1457 if (bond_dev->flags & IFF_PROMISC) {
1458 res = dev_set_promiscuity(slave_dev, 1);
1463 /* set allmulti level to new slave */
1464 if (bond_dev->flags & IFF_ALLMULTI) {
1465 res = dev_set_allmulti(slave_dev, 1);
1470 netif_addr_lock_bh(bond_dev);
1472 dev_mc_sync_multiple(slave_dev, bond_dev);
1473 dev_uc_sync_multiple(slave_dev, bond_dev);
1475 netif_addr_unlock_bh(bond_dev);
1478 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1479 /* add lacpdu mc addr to mc list */
1480 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1482 dev_mc_add(slave_dev, lacpdu_multicast);
1485 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1487 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1492 prev_slave = bond_last_slave(bond);
1494 new_slave->delay = 0;
1495 new_slave->link_failure_count = 0;
1497 bond_update_speed_duplex(new_slave);
1499 new_slave->last_rx = jiffies -
1500 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1501 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1502 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1504 if (bond->params.miimon && !bond->params.use_carrier) {
1505 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1507 if ((link_reporting == -1) && !bond->params.arp_interval) {
1508 /* miimon is set but a bonded network driver
1509 * does not support ETHTOOL/MII and
1510 * arp_interval is not set. Note: if
1511 * use_carrier is enabled, we will never go
1512 * here (because netif_carrier is always
1513 * supported); thus, we don't need to change
1514 * the messages for netif_carrier.
1516 netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
1518 } else if (link_reporting == -1) {
1519 /* unable get link status using mii/ethtool */
1520 netdev_warn(bond_dev, "can't get link status from interface %s; 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",
1525 /* check for initial state */
1526 if (bond->params.miimon) {
1527 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1528 if (bond->params.updelay) {
1529 bond_set_slave_link_state(new_slave,
1531 new_slave->delay = bond->params.updelay;
1533 bond_set_slave_link_state(new_slave,
1537 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1539 } else if (bond->params.arp_interval) {
1540 bond_set_slave_link_state(new_slave,
1541 (netif_carrier_ok(slave_dev) ?
1542 BOND_LINK_UP : BOND_LINK_DOWN));
1544 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1547 if (new_slave->link != BOND_LINK_DOWN)
1548 new_slave->last_link_up = jiffies;
1549 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1550 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1551 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1553 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1554 /* if there is a primary slave, remember it */
1555 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1556 rcu_assign_pointer(bond->primary_slave, new_slave);
1557 bond->force_primary = true;
1561 switch (BOND_MODE(bond)) {
1562 case BOND_MODE_ACTIVEBACKUP:
1563 bond_set_slave_inactive_flags(new_slave,
1564 BOND_SLAVE_NOTIFY_NOW);
1566 case BOND_MODE_8023AD:
1567 /* in 802.3ad mode, the internal mechanism
1568 * will activate the slaves in the selected
1571 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1572 /* if this is the first slave */
1574 SLAVE_AD_INFO(new_slave)->id = 1;
1575 /* Initialize AD with the number of times that the AD timer is called in 1 second
1576 * can be called only after the mac address of the bond is set
1578 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1580 SLAVE_AD_INFO(new_slave)->id =
1581 SLAVE_AD_INFO(prev_slave)->id + 1;
1584 bond_3ad_bind_slave(new_slave);
1588 bond_set_active_slave(new_slave);
1589 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1592 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1594 /* always active in trunk mode */
1595 bond_set_active_slave(new_slave);
1597 /* In trunking mode there is little meaning to curr_active_slave
1598 * anyway (it holds no special properties of the bond device),
1599 * so we can change it without calling change_active_interface()
1601 if (!rcu_access_pointer(bond->curr_active_slave) &&
1602 new_slave->link == BOND_LINK_UP)
1603 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1606 } /* switch(bond_mode) */
1608 #ifdef CONFIG_NET_POLL_CONTROLLER
1609 slave_dev->npinfo = bond->dev->npinfo;
1610 if (slave_dev->npinfo) {
1611 if (slave_enable_netpoll(new_slave)) {
1612 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1619 if (!(bond_dev->features & NETIF_F_LRO))
1620 dev_disable_lro(slave_dev);
1622 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1625 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1629 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1631 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1632 goto err_unregister;
1635 res = bond_sysfs_slave_add(new_slave);
1637 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1638 goto err_upper_unlink;
1642 bond_compute_features(bond);
1643 bond_set_carrier(bond);
1645 if (bond_uses_primary(bond)) {
1647 bond_select_active_slave(bond);
1648 unblock_netpoll_tx();
1651 if (bond_mode_uses_xmit_hash(bond))
1652 bond_update_slave_arr(bond, NULL);
1654 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1656 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1657 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1659 /* enslave is successful */
1660 bond_queue_slave_event(new_slave);
1663 /* Undo stages on error */
1665 bond_upper_dev_unlink(bond_dev, slave_dev);
1668 netdev_rx_handler_unregister(slave_dev);
1671 if (!bond_uses_primary(bond))
1672 bond_hw_addr_flush(bond_dev, slave_dev);
1674 vlan_vids_del_by_dev(slave_dev, bond_dev);
1675 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1676 RCU_INIT_POINTER(bond->primary_slave, NULL);
1677 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1679 bond_change_active_slave(bond, NULL);
1680 bond_select_active_slave(bond);
1681 unblock_netpoll_tx();
1683 /* either primary_slave or curr_active_slave might've changed */
1685 slave_disable_netpoll(new_slave);
1688 slave_dev->priv_flags &= ~IFF_BONDING;
1689 dev_close(slave_dev);
1692 if (!bond->params.fail_over_mac ||
1693 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1694 /* XXX TODO - fom follow mode needs to change master's
1695 * MAC if this slave's MAC is in use by the bond, or at
1696 * least print a warning.
1698 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1699 addr.sa_family = slave_dev->type;
1700 dev_set_mac_address(slave_dev, &addr);
1704 dev_set_mtu(slave_dev, new_slave->original_mtu);
1707 bond_free_slave(new_slave);
1710 /* Enslave of first slave has failed and we need to fix master's mac */
1711 if (!bond_has_slaves(bond) &&
1712 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1713 eth_hw_addr_random(bond_dev);
1718 /* Try to release the slave device <slave> from the bond device <master>
1719 * It is legal to access curr_active_slave without a lock because all the function
1720 * is RTNL-locked. If "all" is true it means that the function is being called
1721 * while destroying a bond interface and all slaves are being released.
1723 * The rules for slave state should be:
1724 * for Active/Backup:
1725 * Active stays on all backups go down
1726 * for Bonded connections:
1727 * The first up interface should be left on and all others downed.
1729 static int __bond_release_one(struct net_device *bond_dev,
1730 struct net_device *slave_dev,
1733 struct bonding *bond = netdev_priv(bond_dev);
1734 struct slave *slave, *oldcurrent;
1735 struct sockaddr addr;
1736 int old_flags = bond_dev->flags;
1737 netdev_features_t old_features = bond_dev->features;
1739 /* slave is not a slave or master is not master of this slave */
1740 if (!(slave_dev->flags & IFF_SLAVE) ||
1741 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1742 netdev_dbg(bond_dev, "cannot release %s\n",
1749 slave = bond_get_slave_by_dev(bond, slave_dev);
1751 /* not a slave of this bond */
1752 netdev_info(bond_dev, "%s not enslaved\n",
1754 unblock_netpoll_tx();
1758 bond_sysfs_slave_del(slave);
1760 /* recompute stats just before removing the slave */
1761 bond_get_stats(bond->dev, &bond->bond_stats);
1763 bond_upper_dev_unlink(bond_dev, slave_dev);
1764 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1765 * for this slave anymore.
1767 netdev_rx_handler_unregister(slave_dev);
1769 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1770 bond_3ad_unbind_slave(slave);
1772 if (bond_mode_uses_xmit_hash(bond))
1773 bond_update_slave_arr(bond, slave);
1775 netdev_info(bond_dev, "Releasing %s interface %s\n",
1776 bond_is_active_slave(slave) ? "active" : "backup",
1779 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1781 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1783 if (!all && (!bond->params.fail_over_mac ||
1784 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1785 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1786 bond_has_slaves(bond))
1787 netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
1788 slave_dev->name, slave->perm_hwaddr,
1789 bond_dev->name, slave_dev->name);
1792 if (rtnl_dereference(bond->primary_slave) == slave)
1793 RCU_INIT_POINTER(bond->primary_slave, NULL);
1795 if (oldcurrent == slave)
1796 bond_change_active_slave(bond, NULL);
1798 if (bond_is_lb(bond)) {
1799 /* Must be called only after the slave has been
1800 * detached from the list and the curr_active_slave
1801 * has been cleared (if our_slave == old_current),
1802 * but before a new active slave is selected.
1804 bond_alb_deinit_slave(bond, slave);
1808 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1809 } else if (oldcurrent == slave) {
1810 /* Note that we hold RTNL over this sequence, so there
1811 * is no concern that another slave add/remove event
1814 bond_select_active_slave(bond);
1817 if (!bond_has_slaves(bond)) {
1818 bond_set_carrier(bond);
1819 eth_hw_addr_random(bond_dev);
1822 unblock_netpoll_tx();
1826 if (!bond_has_slaves(bond)) {
1827 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1828 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1831 bond_compute_features(bond);
1832 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1833 (old_features & NETIF_F_VLAN_CHALLENGED))
1834 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1835 slave_dev->name, bond_dev->name);
1837 vlan_vids_del_by_dev(slave_dev, bond_dev);
1839 /* If the mode uses primary, then this case was handled above by
1840 * bond_change_active_slave(..., NULL)
1842 if (!bond_uses_primary(bond)) {
1843 /* unset promiscuity level from slave
1844 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1845 * of the IFF_PROMISC flag in the bond_dev, but we need the
1846 * value of that flag before that change, as that was the value
1847 * when this slave was attached, so we cache at the start of the
1848 * function and use it here. Same goes for ALLMULTI below
1850 if (old_flags & IFF_PROMISC)
1851 dev_set_promiscuity(slave_dev, -1);
1853 /* unset allmulti level from slave */
1854 if (old_flags & IFF_ALLMULTI)
1855 dev_set_allmulti(slave_dev, -1);
1857 bond_hw_addr_flush(bond_dev, slave_dev);
1860 slave_disable_netpoll(slave);
1862 /* close slave before restoring its mac address */
1863 dev_close(slave_dev);
1865 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1866 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1867 /* restore original ("permanent") mac address */
1868 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1869 addr.sa_family = slave_dev->type;
1870 dev_set_mac_address(slave_dev, &addr);
1873 dev_set_mtu(slave_dev, slave->original_mtu);
1875 slave_dev->priv_flags &= ~IFF_BONDING;
1877 bond_free_slave(slave);
1882 /* A wrapper used because of ndo_del_link */
1883 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1885 return __bond_release_one(bond_dev, slave_dev, false);
1888 /* First release a slave and then destroy the bond if no more slaves are left.
1889 * Must be under rtnl_lock when this function is called.
1891 static int bond_release_and_destroy(struct net_device *bond_dev,
1892 struct net_device *slave_dev)
1894 struct bonding *bond = netdev_priv(bond_dev);
1897 ret = bond_release(bond_dev, slave_dev);
1898 if (ret == 0 && !bond_has_slaves(bond)) {
1899 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1900 netdev_info(bond_dev, "Destroying bond %s\n",
1902 unregister_netdevice(bond_dev);
1907 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1909 struct bonding *bond = netdev_priv(bond_dev);
1910 bond_fill_ifbond(bond, info);
1914 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1916 struct bonding *bond = netdev_priv(bond_dev);
1917 struct list_head *iter;
1918 int i = 0, res = -ENODEV;
1919 struct slave *slave;
1921 bond_for_each_slave(bond, slave, iter) {
1922 if (i++ == (int)info->slave_id) {
1924 bond_fill_ifslave(slave, info);
1932 /*-------------------------------- Monitoring -------------------------------*/
1934 /* called with rcu_read_lock() */
1935 static int bond_miimon_inspect(struct bonding *bond)
1937 int link_state, commit = 0;
1938 struct list_head *iter;
1939 struct slave *slave;
1940 bool ignore_updelay;
1942 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1944 bond_for_each_slave_rcu(bond, slave, iter) {
1945 slave->new_link = BOND_LINK_NOCHANGE;
1947 link_state = bond_check_dev_link(bond, slave->dev, 0);
1949 switch (slave->link) {
1954 bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1955 slave->delay = bond->params.downdelay;
1957 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1959 BOND_MODE_ACTIVEBACKUP) ?
1960 (bond_is_active_slave(slave) ?
1961 "active " : "backup ") : "",
1963 bond->params.downdelay * bond->params.miimon);
1966 case BOND_LINK_FAIL:
1968 /* recovered before downdelay expired */
1969 bond_set_slave_link_state(slave, BOND_LINK_UP);
1970 slave->last_link_up = jiffies;
1971 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1972 (bond->params.downdelay - slave->delay) *
1973 bond->params.miimon,
1978 if (slave->delay <= 0) {
1979 slave->new_link = BOND_LINK_DOWN;
1987 case BOND_LINK_DOWN:
1991 bond_set_slave_link_state(slave, BOND_LINK_BACK);
1992 slave->delay = bond->params.updelay;
1995 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1997 ignore_updelay ? 0 :
1998 bond->params.updelay *
1999 bond->params.miimon);
2002 case BOND_LINK_BACK:
2004 bond_set_slave_link_state(slave,
2006 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2007 (bond->params.updelay - slave->delay) *
2008 bond->params.miimon,
2017 if (slave->delay <= 0) {
2018 slave->new_link = BOND_LINK_UP;
2020 ignore_updelay = false;
2032 static void bond_miimon_commit(struct bonding *bond)
2034 struct list_head *iter;
2035 struct slave *slave, *primary;
2037 bond_for_each_slave(bond, slave, iter) {
2038 switch (slave->new_link) {
2039 case BOND_LINK_NOCHANGE:
2043 bond_set_slave_link_state(slave, BOND_LINK_UP);
2044 slave->last_link_up = jiffies;
2046 primary = rtnl_dereference(bond->primary_slave);
2047 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2048 /* prevent it from being the active one */
2049 bond_set_backup_slave(slave);
2050 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2051 /* make it immediately active */
2052 bond_set_active_slave(slave);
2053 } else if (slave != primary) {
2054 /* prevent it from being the active one */
2055 bond_set_backup_slave(slave);
2058 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2060 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2061 slave->duplex ? "full" : "half");
2063 /* notify ad that the link status has changed */
2064 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2065 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2067 if (bond_is_lb(bond))
2068 bond_alb_handle_link_change(bond, slave,
2071 if (BOND_MODE(bond) == BOND_MODE_XOR)
2072 bond_update_slave_arr(bond, NULL);
2074 if (!bond->curr_active_slave || slave == primary)
2079 case BOND_LINK_DOWN:
2080 if (slave->link_failure_count < UINT_MAX)
2081 slave->link_failure_count++;
2083 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2085 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2086 BOND_MODE(bond) == BOND_MODE_8023AD)
2087 bond_set_slave_inactive_flags(slave,
2088 BOND_SLAVE_NOTIFY_NOW);
2090 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2093 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2094 bond_3ad_handle_link_change(slave,
2097 if (bond_is_lb(bond))
2098 bond_alb_handle_link_change(bond, slave,
2101 if (BOND_MODE(bond) == BOND_MODE_XOR)
2102 bond_update_slave_arr(bond, NULL);
2104 if (slave == rcu_access_pointer(bond->curr_active_slave))
2110 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2111 slave->new_link, slave->dev->name);
2112 slave->new_link = BOND_LINK_NOCHANGE;
2119 bond_select_active_slave(bond);
2120 unblock_netpoll_tx();
2123 bond_set_carrier(bond);
2128 * Really a wrapper that splits the mii monitor into two phases: an
2129 * inspection, then (if inspection indicates something needs to be done)
2130 * an acquisition of appropriate locks followed by a commit phase to
2131 * implement whatever link state changes are indicated.
2133 static void bond_mii_monitor(struct work_struct *work)
2135 struct bonding *bond = container_of(work, struct bonding,
2137 bool should_notify_peers = false;
2138 unsigned long delay;
2140 delay = msecs_to_jiffies(bond->params.miimon);
2142 if (!bond_has_slaves(bond))
2147 should_notify_peers = bond_should_notify_peers(bond);
2149 if (bond_miimon_inspect(bond)) {
2152 /* Race avoidance with bond_close cancel of workqueue */
2153 if (!rtnl_trylock()) {
2155 should_notify_peers = false;
2159 bond_miimon_commit(bond);
2161 rtnl_unlock(); /* might sleep, hold no other locks */
2166 if (bond->params.miimon)
2167 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2169 if (should_notify_peers) {
2170 if (!rtnl_trylock())
2172 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2177 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2179 struct net_device *upper;
2180 struct list_head *iter;
2183 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2187 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2188 if (ip == bond_confirm_addr(upper, 0, ip)) {
2198 /* We go to the (large) trouble of VLAN tagging ARP frames because
2199 * switches in VLAN mode (especially if ports are configured as
2200 * "native" to a VLAN) might not pass non-tagged frames.
2202 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2203 __be32 dest_ip, __be32 src_ip,
2204 struct bond_vlan_tag *tags)
2206 struct sk_buff *skb;
2207 struct bond_vlan_tag *outer_tag = tags;
2209 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2210 arp_op, slave_dev->name, &dest_ip, &src_ip);
2212 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2213 NULL, slave_dev->dev_addr, NULL);
2216 net_err_ratelimited("ARP packet allocation failed\n");
2220 if (!tags || tags->vlan_proto == VLAN_N_VID)
2225 /* Go through all the tags backwards and add them to the packet */
2226 while (tags->vlan_proto != VLAN_N_VID) {
2227 if (!tags->vlan_id) {
2232 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2233 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2234 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2237 net_err_ratelimited("failed to insert inner VLAN tag\n");
2243 /* Set the outer tag */
2244 if (outer_tag->vlan_id) {
2245 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2246 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2247 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2248 outer_tag->vlan_id);
2255 /* Validate the device path between the @start_dev and the @end_dev.
2256 * The path is valid if the @end_dev is reachable through device
2258 * When the path is validated, collect any vlan information in the
2261 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2262 struct net_device *end_dev,
2265 struct bond_vlan_tag *tags;
2266 struct net_device *upper;
2267 struct list_head *iter;
2269 if (start_dev == end_dev) {
2270 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2272 return ERR_PTR(-ENOMEM);
2273 tags[level].vlan_proto = VLAN_N_VID;
2277 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2278 tags = bond_verify_device_path(upper, end_dev, level + 1);
2279 if (IS_ERR_OR_NULL(tags)) {
2284 if (is_vlan_dev(upper)) {
2285 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2286 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2295 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2298 struct bond_vlan_tag *tags;
2299 __be32 *targets = bond->params.arp_targets, addr;
2302 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2303 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2306 /* Find out through which dev should the packet go */
2307 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2310 /* there's no route to target - try to send arp
2311 * probe to generate any traffic (arp_validate=0)
2313 if (bond->params.arp_validate)
2314 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2317 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2322 /* bond device itself */
2323 if (rt->dst.dev == bond->dev)
2327 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2330 if (!IS_ERR_OR_NULL(tags))
2333 /* Not our device - skip */
2334 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2335 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2341 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2343 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2349 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2353 if (!sip || !bond_has_this_ip(bond, tip)) {
2354 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2359 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2361 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2365 slave->last_rx = jiffies;
2366 slave->target_last_arp_rx[i] = jiffies;
2369 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2370 struct slave *slave)
2372 struct arphdr *arp = (struct arphdr *)skb->data;
2373 struct slave *curr_active_slave;
2374 unsigned char *arp_ptr;
2376 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2378 if (!slave_do_arp_validate(bond, slave)) {
2379 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2380 !slave_do_arp_validate_only(bond))
2381 slave->last_rx = jiffies;
2382 return RX_HANDLER_ANOTHER;
2383 } else if (!is_arp) {
2384 return RX_HANDLER_ANOTHER;
2387 alen = arp_hdr_len(bond->dev);
2389 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2392 if (alen > skb_headlen(skb)) {
2393 arp = kmalloc(alen, GFP_ATOMIC);
2396 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2400 if (arp->ar_hln != bond->dev->addr_len ||
2401 skb->pkt_type == PACKET_OTHERHOST ||
2402 skb->pkt_type == PACKET_LOOPBACK ||
2403 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2404 arp->ar_pro != htons(ETH_P_IP) ||
2408 arp_ptr = (unsigned char *)(arp + 1);
2409 arp_ptr += bond->dev->addr_len;
2410 memcpy(&sip, arp_ptr, 4);
2411 arp_ptr += 4 + bond->dev->addr_len;
2412 memcpy(&tip, arp_ptr, 4);
2414 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2415 slave->dev->name, bond_slave_state(slave),
2416 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2419 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2421 /* Backup slaves won't see the ARP reply, but do come through
2422 * here for each ARP probe (so we swap the sip/tip to validate
2423 * the probe). In a "redundant switch, common router" type of
2424 * configuration, the ARP probe will (hopefully) travel from
2425 * the active, through one switch, the router, then the other
2426 * switch before reaching the backup.
2428 * We 'trust' the arp requests if there is an active slave and
2429 * it received valid arp reply(s) after it became active. This
2430 * is done to avoid endless looping when we can't reach the
2431 * arp_ip_target and fool ourselves with our own arp requests.
2434 if (bond_is_active_slave(slave))
2435 bond_validate_arp(bond, slave, sip, tip);
2436 else if (curr_active_slave &&
2437 time_after(slave_last_rx(bond, curr_active_slave),
2438 curr_active_slave->last_link_up))
2439 bond_validate_arp(bond, slave, tip, sip);
2442 if (arp != (struct arphdr *)skb->data)
2444 return RX_HANDLER_ANOTHER;
2447 /* function to verify if we're in the arp_interval timeslice, returns true if
2448 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2449 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2451 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2454 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2456 return time_in_range(jiffies,
2457 last_act - delta_in_ticks,
2458 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2461 /* This function is called regularly to monitor each slave's link
2462 * ensuring that traffic is being sent and received when arp monitoring
2463 * is used in load-balancing mode. if the adapter has been dormant, then an
2464 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2465 * arp monitoring in active backup mode.
2467 static void bond_loadbalance_arp_mon(struct work_struct *work)
2469 struct bonding *bond = container_of(work, struct bonding,
2471 struct slave *slave, *oldcurrent;
2472 struct list_head *iter;
2473 int do_failover = 0, slave_state_changed = 0;
2475 if (!bond_has_slaves(bond))
2480 oldcurrent = rcu_dereference(bond->curr_active_slave);
2481 /* see if any of the previous devices are up now (i.e. they have
2482 * xmt and rcv traffic). the curr_active_slave does not come into
2483 * the picture unless it is null. also, slave->last_link_up is not
2484 * needed here because we send an arp on each slave and give a slave
2485 * as long as it needs to get the tx/rx within the delta.
2486 * TODO: what about up/down delay in arp mode? it wasn't here before
2489 bond_for_each_slave_rcu(bond, slave, iter) {
2490 unsigned long trans_start = dev_trans_start(slave->dev);
2492 if (slave->link != BOND_LINK_UP) {
2493 if (bond_time_in_interval(bond, trans_start, 1) &&
2494 bond_time_in_interval(bond, slave->last_rx, 1)) {
2496 slave->link = BOND_LINK_UP;
2497 slave_state_changed = 1;
2499 /* primary_slave has no meaning in round-robin
2500 * mode. the window of a slave being up and
2501 * curr_active_slave being null after enslaving
2505 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2509 netdev_info(bond->dev, "interface %s is now up\n",
2514 /* slave->link == BOND_LINK_UP */
2516 /* not all switches will respond to an arp request
2517 * when the source ip is 0, so don't take the link down
2518 * if we don't know our ip yet
2520 if (!bond_time_in_interval(bond, trans_start, 2) ||
2521 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2523 slave->link = BOND_LINK_DOWN;
2524 slave_state_changed = 1;
2526 if (slave->link_failure_count < UINT_MAX)
2527 slave->link_failure_count++;
2529 netdev_info(bond->dev, "interface %s is now down\n",
2532 if (slave == oldcurrent)
2537 /* note: if switch is in round-robin mode, all links
2538 * must tx arp to ensure all links rx an arp - otherwise
2539 * links may oscillate or not come up at all; if switch is
2540 * in something like xor mode, there is nothing we can
2541 * do - all replies will be rx'ed on same link causing slaves
2542 * to be unstable during low/no traffic periods
2544 if (bond_slave_is_up(slave))
2545 bond_arp_send_all(bond, slave);
2550 if (do_failover || slave_state_changed) {
2551 if (!rtnl_trylock())
2554 if (slave_state_changed) {
2555 bond_slave_state_change(bond);
2556 if (BOND_MODE(bond) == BOND_MODE_XOR)
2557 bond_update_slave_arr(bond, NULL);
2561 bond_select_active_slave(bond);
2562 unblock_netpoll_tx();
2568 if (bond->params.arp_interval)
2569 queue_delayed_work(bond->wq, &bond->arp_work,
2570 msecs_to_jiffies(bond->params.arp_interval));
2573 /* Called to inspect slaves for active-backup mode ARP monitor link state
2574 * changes. Sets new_link in slaves to specify what action should take
2575 * place for the slave. Returns 0 if no changes are found, >0 if changes
2576 * to link states must be committed.
2578 * Called with rcu_read_lock held.
2580 static int bond_ab_arp_inspect(struct bonding *bond)
2582 unsigned long trans_start, last_rx;
2583 struct list_head *iter;
2584 struct slave *slave;
2587 bond_for_each_slave_rcu(bond, slave, iter) {
2588 slave->new_link = BOND_LINK_NOCHANGE;
2589 last_rx = slave_last_rx(bond, slave);
2591 if (slave->link != BOND_LINK_UP) {
2592 if (bond_time_in_interval(bond, last_rx, 1)) {
2593 slave->new_link = BOND_LINK_UP;
2599 /* Give slaves 2*delta after being enslaved or made
2600 * active. This avoids bouncing, as the last receive
2601 * times need a full ARP monitor cycle to be updated.
2603 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2606 /* Backup slave is down if:
2607 * - No current_arp_slave AND
2608 * - more than 3*delta since last receive AND
2609 * - the bond has an IP address
2611 * Note: a non-null current_arp_slave indicates
2612 * the curr_active_slave went down and we are
2613 * searching for a new one; under this condition
2614 * we only take the curr_active_slave down - this
2615 * gives each slave a chance to tx/rx traffic
2616 * before being taken out
2618 if (!bond_is_active_slave(slave) &&
2619 !rcu_access_pointer(bond->current_arp_slave) &&
2620 !bond_time_in_interval(bond, last_rx, 3)) {
2621 slave->new_link = BOND_LINK_DOWN;
2625 /* Active slave is down if:
2626 * - more than 2*delta since transmitting OR
2627 * - (more than 2*delta since receive AND
2628 * the bond has an IP address)
2630 trans_start = dev_trans_start(slave->dev);
2631 if (bond_is_active_slave(slave) &&
2632 (!bond_time_in_interval(bond, trans_start, 2) ||
2633 !bond_time_in_interval(bond, last_rx, 2))) {
2634 slave->new_link = BOND_LINK_DOWN;
2642 /* Called to commit link state changes noted by inspection step of
2643 * active-backup mode ARP monitor.
2645 * Called with RTNL hold.
2647 static void bond_ab_arp_commit(struct bonding *bond)
2649 unsigned long trans_start;
2650 struct list_head *iter;
2651 struct slave *slave;
2653 bond_for_each_slave(bond, slave, iter) {
2654 switch (slave->new_link) {
2655 case BOND_LINK_NOCHANGE:
2659 trans_start = dev_trans_start(slave->dev);
2660 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2661 (!rtnl_dereference(bond->curr_active_slave) &&
2662 bond_time_in_interval(bond, trans_start, 1))) {
2663 struct slave *current_arp_slave;
2665 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2666 bond_set_slave_link_state(slave, BOND_LINK_UP);
2667 if (current_arp_slave) {
2668 bond_set_slave_inactive_flags(
2670 BOND_SLAVE_NOTIFY_NOW);
2671 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2674 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2677 if (!rtnl_dereference(bond->curr_active_slave) ||
2678 slave == rtnl_dereference(bond->primary_slave))
2685 case BOND_LINK_DOWN:
2686 if (slave->link_failure_count < UINT_MAX)
2687 slave->link_failure_count++;
2689 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2690 bond_set_slave_inactive_flags(slave,
2691 BOND_SLAVE_NOTIFY_NOW);
2693 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2696 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2697 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2704 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2705 slave->new_link, slave->dev->name);
2711 bond_select_active_slave(bond);
2712 unblock_netpoll_tx();
2715 bond_set_carrier(bond);
2718 /* Send ARP probes for active-backup mode ARP monitor.
2720 * Called with rcu_read_lock held.
2722 static bool bond_ab_arp_probe(struct bonding *bond)
2724 struct slave *slave, *before = NULL, *new_slave = NULL,
2725 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2726 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2727 struct list_head *iter;
2729 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2731 if (curr_arp_slave && curr_active_slave)
2732 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2733 curr_arp_slave->dev->name,
2734 curr_active_slave->dev->name);
2736 if (curr_active_slave) {
2737 bond_arp_send_all(bond, curr_active_slave);
2738 return should_notify_rtnl;
2741 /* if we don't have a curr_active_slave, search for the next available
2742 * backup slave from the current_arp_slave and make it the candidate
2743 * for becoming the curr_active_slave
2746 if (!curr_arp_slave) {
2747 curr_arp_slave = bond_first_slave_rcu(bond);
2748 if (!curr_arp_slave)
2749 return should_notify_rtnl;
2752 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2754 bond_for_each_slave_rcu(bond, slave, iter) {
2755 if (!found && !before && bond_slave_is_up(slave))
2758 if (found && !new_slave && bond_slave_is_up(slave))
2760 /* if the link state is up at this point, we
2761 * mark it down - this can happen if we have
2762 * simultaneous link failures and
2763 * reselect_active_interface doesn't make this
2764 * one the current slave so it is still marked
2765 * up when it is actually down
2767 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2768 bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2769 if (slave->link_failure_count < UINT_MAX)
2770 slave->link_failure_count++;
2772 bond_set_slave_inactive_flags(slave,
2773 BOND_SLAVE_NOTIFY_LATER);
2775 netdev_info(bond->dev, "backup interface %s is now down\n",
2778 if (slave == curr_arp_slave)
2782 if (!new_slave && before)
2788 bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2789 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2790 bond_arp_send_all(bond, new_slave);
2791 new_slave->last_link_up = jiffies;
2792 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2795 bond_for_each_slave_rcu(bond, slave, iter) {
2796 if (slave->should_notify) {
2797 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2801 return should_notify_rtnl;
2804 static void bond_activebackup_arp_mon(struct work_struct *work)
2806 struct bonding *bond = container_of(work, struct bonding,
2808 bool should_notify_peers = false;
2809 bool should_notify_rtnl = false;
2812 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2814 if (!bond_has_slaves(bond))
2819 should_notify_peers = bond_should_notify_peers(bond);
2821 if (bond_ab_arp_inspect(bond)) {
2824 /* Race avoidance with bond_close flush of workqueue */
2825 if (!rtnl_trylock()) {
2827 should_notify_peers = false;
2831 bond_ab_arp_commit(bond);
2837 should_notify_rtnl = bond_ab_arp_probe(bond);
2841 if (bond->params.arp_interval)
2842 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2844 if (should_notify_peers || should_notify_rtnl) {
2845 if (!rtnl_trylock())
2848 if (should_notify_peers)
2849 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2851 if (should_notify_rtnl)
2852 bond_slave_state_notify(bond);
2858 /*-------------------------- netdev event handling --------------------------*/
2860 /* Change device name */
2861 static int bond_event_changename(struct bonding *bond)
2863 bond_remove_proc_entry(bond);
2864 bond_create_proc_entry(bond);
2866 bond_debug_reregister(bond);
2871 static int bond_master_netdev_event(unsigned long event,
2872 struct net_device *bond_dev)
2874 struct bonding *event_bond = netdev_priv(bond_dev);
2877 case NETDEV_CHANGENAME:
2878 return bond_event_changename(event_bond);
2879 case NETDEV_UNREGISTER:
2880 bond_remove_proc_entry(event_bond);
2882 case NETDEV_REGISTER:
2883 bond_create_proc_entry(event_bond);
2885 case NETDEV_NOTIFY_PEERS:
2886 if (event_bond->send_peer_notif)
2887 event_bond->send_peer_notif--;
2896 static int bond_slave_netdev_event(unsigned long event,
2897 struct net_device *slave_dev)
2899 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2900 struct bonding *bond;
2901 struct net_device *bond_dev;
2905 /* A netdev event can be generated while enslaving a device
2906 * before netdev_rx_handler_register is called in which case
2907 * slave will be NULL
2911 bond_dev = slave->bond->dev;
2913 primary = rtnl_dereference(bond->primary_slave);
2916 case NETDEV_UNREGISTER:
2917 if (bond_dev->type != ARPHRD_ETHER)
2918 bond_release_and_destroy(bond_dev, slave_dev);
2920 bond_release(bond_dev, slave_dev);
2924 old_speed = slave->speed;
2925 old_duplex = slave->duplex;
2927 bond_update_speed_duplex(slave);
2929 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2930 if (old_speed != slave->speed)
2931 bond_3ad_adapter_speed_changed(slave);
2932 if (old_duplex != slave->duplex)
2933 bond_3ad_adapter_duplex_changed(slave);
2937 /* Refresh slave-array if applicable!
2938 * If the setup does not use miimon or arpmon (mode-specific!),
2939 * then these events will not cause the slave-array to be
2940 * refreshed. This will cause xmit to use a slave that is not
2941 * usable. Avoid such situation by refeshing the array at these
2942 * events. If these (miimon/arpmon) parameters are configured
2943 * then array gets refreshed twice and that should be fine!
2945 if (bond_mode_uses_xmit_hash(bond))
2946 bond_update_slave_arr(bond, NULL);
2948 case NETDEV_CHANGEMTU:
2949 /* TODO: Should slaves be allowed to
2950 * independently alter their MTU? For
2951 * an active-backup bond, slaves need
2952 * not be the same type of device, so
2953 * MTUs may vary. For other modes,
2954 * slaves arguably should have the
2955 * same MTUs. To do this, we'd need to
2956 * take over the slave's change_mtu
2957 * function for the duration of their
2961 case NETDEV_CHANGENAME:
2962 /* we don't care if we don't have primary set */
2963 if (!bond_uses_primary(bond) ||
2964 !bond->params.primary[0])
2967 if (slave == primary) {
2968 /* slave's name changed - he's no longer primary */
2969 RCU_INIT_POINTER(bond->primary_slave, NULL);
2970 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2971 /* we have a new primary slave */
2972 rcu_assign_pointer(bond->primary_slave, slave);
2973 } else { /* we didn't change primary - exit */
2977 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2978 primary ? slave_dev->name : "none");
2981 bond_select_active_slave(bond);
2982 unblock_netpoll_tx();
2984 case NETDEV_FEAT_CHANGE:
2985 bond_compute_features(bond);
2987 case NETDEV_RESEND_IGMP:
2988 /* Propagate to master device */
2989 call_netdevice_notifiers(event, slave->bond->dev);
2998 /* bond_netdev_event: handle netdev notifier chain events.
3000 * This function receives events for the netdev chain. The caller (an
3001 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3002 * locks for us to safely manipulate the slave devices (RTNL lock,
3005 static int bond_netdev_event(struct notifier_block *this,
3006 unsigned long event, void *ptr)
3008 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3010 netdev_dbg(event_dev, "event: %lx\n", event);
3012 if (!(event_dev->priv_flags & IFF_BONDING))
3015 if (event_dev->flags & IFF_MASTER) {
3016 netdev_dbg(event_dev, "IFF_MASTER\n");
3017 return bond_master_netdev_event(event, event_dev);
3020 if (event_dev->flags & IFF_SLAVE) {
3021 netdev_dbg(event_dev, "IFF_SLAVE\n");
3022 return bond_slave_netdev_event(event, event_dev);
3028 static struct notifier_block bond_netdev_notifier = {
3029 .notifier_call = bond_netdev_event,
3032 /*---------------------------- Hashing Policies -----------------------------*/
3034 /* L2 hash helper */
3035 static inline u32 bond_eth_hash(struct sk_buff *skb)
3037 struct ethhdr *ep, hdr_tmp;
3039 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3041 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3045 /* Extract the appropriate headers based on bond's xmit policy */
3046 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3047 struct flow_keys *fk)
3049 const struct ipv6hdr *iph6;
3050 const struct iphdr *iph;
3051 int noff, proto = -1;
3053 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3054 return skb_flow_dissect_flow_keys(skb, fk);
3056 fk->ports.ports = 0;
3057 noff = skb_network_offset(skb);
3058 if (skb->protocol == htons(ETH_P_IP)) {
3059 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3062 iph_to_flow_copy_v4addrs(fk, iph);
3063 noff += iph->ihl << 2;
3064 if (!ip_is_fragment(iph))
3065 proto = iph->protocol;
3066 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3067 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3069 iph6 = ipv6_hdr(skb);
3070 iph_to_flow_copy_v6addrs(fk, iph6);
3071 noff += sizeof(*iph6);
3072 proto = iph6->nexthdr;
3076 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3077 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3083 * bond_xmit_hash - generate a hash value based on the xmit policy
3084 * @bond: bonding device
3085 * @skb: buffer to use for headers
3087 * This function will extract the necessary headers from the skb buffer and use
3088 * them to generate a hash based on the xmit_policy set in the bonding device
3090 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3092 struct flow_keys flow;
3095 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3096 !bond_flow_dissect(bond, skb, &flow))
3097 return bond_eth_hash(skb);
3099 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3100 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3101 hash = bond_eth_hash(skb);
3103 hash = (__force u32)flow.ports.ports;
3104 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3105 (__force u32)flow_get_u32_src(&flow);
3106 hash ^= (hash >> 16);
3107 hash ^= (hash >> 8);
3112 /*-------------------------- Device entry points ----------------------------*/
3114 static void bond_work_init_all(struct bonding *bond)
3116 INIT_DELAYED_WORK(&bond->mcast_work,
3117 bond_resend_igmp_join_requests_delayed);
3118 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3119 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3120 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3121 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3123 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3124 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3125 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3128 static void bond_work_cancel_all(struct bonding *bond)
3130 cancel_delayed_work_sync(&bond->mii_work);
3131 cancel_delayed_work_sync(&bond->arp_work);
3132 cancel_delayed_work_sync(&bond->alb_work);
3133 cancel_delayed_work_sync(&bond->ad_work);
3134 cancel_delayed_work_sync(&bond->mcast_work);
3135 cancel_delayed_work_sync(&bond->slave_arr_work);
3138 static int bond_open(struct net_device *bond_dev)
3140 struct bonding *bond = netdev_priv(bond_dev);
3141 struct list_head *iter;
3142 struct slave *slave;
3144 /* reset slave->backup and slave->inactive */
3145 if (bond_has_slaves(bond)) {
3146 bond_for_each_slave(bond, slave, iter) {
3147 if (bond_uses_primary(bond) &&
3148 slave != rcu_access_pointer(bond->curr_active_slave)) {
3149 bond_set_slave_inactive_flags(slave,
3150 BOND_SLAVE_NOTIFY_NOW);
3151 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3152 bond_set_slave_active_flags(slave,
3153 BOND_SLAVE_NOTIFY_NOW);
3158 bond_work_init_all(bond);
3160 if (bond_is_lb(bond)) {
3161 /* bond_alb_initialize must be called before the timer
3164 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3166 if (bond->params.tlb_dynamic_lb)
3167 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3170 if (bond->params.miimon) /* link check interval, in milliseconds. */
3171 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3173 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3174 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3175 bond->recv_probe = bond_arp_rcv;
3178 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3179 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3180 /* register to receive LACPDUs */
3181 bond->recv_probe = bond_3ad_lacpdu_recv;
3182 bond_3ad_initiate_agg_selection(bond, 1);
3185 if (bond_mode_uses_xmit_hash(bond))
3186 bond_update_slave_arr(bond, NULL);
3191 static int bond_close(struct net_device *bond_dev)
3193 struct bonding *bond = netdev_priv(bond_dev);
3195 bond_work_cancel_all(bond);
3196 bond->send_peer_notif = 0;
3197 if (bond_is_lb(bond))
3198 bond_alb_deinitialize(bond);
3199 bond->recv_probe = NULL;
3204 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3205 struct rtnl_link_stats64 *stats)
3207 struct bonding *bond = netdev_priv(bond_dev);
3208 struct rtnl_link_stats64 temp;
3209 struct list_head *iter;
3210 struct slave *slave;
3212 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3214 bond_for_each_slave(bond, slave, iter) {
3215 const struct rtnl_link_stats64 *sstats =
3216 dev_get_stats(slave->dev, &temp);
3217 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3219 stats->rx_packets += sstats->rx_packets - pstats->rx_packets;
3220 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3221 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3222 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3224 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3225 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3226 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3227 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3229 stats->multicast += sstats->multicast - pstats->multicast;
3230 stats->collisions += sstats->collisions - pstats->collisions;
3232 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3233 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3234 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3235 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3236 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3237 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3239 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3240 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3241 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3242 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3243 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3245 /* save off the slave stats for the next run */
3246 memcpy(pstats, sstats, sizeof(*sstats));
3248 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3253 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3255 struct bonding *bond = netdev_priv(bond_dev);
3256 struct net_device *slave_dev = NULL;
3257 struct ifbond k_binfo;
3258 struct ifbond __user *u_binfo = NULL;
3259 struct ifslave k_sinfo;
3260 struct ifslave __user *u_sinfo = NULL;
3261 struct mii_ioctl_data *mii = NULL;
3262 struct bond_opt_value newval;
3266 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3277 /* We do this again just in case we were called by SIOCGMIIREG
3278 * instead of SIOCGMIIPHY.
3284 if (mii->reg_num == 1) {
3286 if (netif_carrier_ok(bond->dev))
3287 mii->val_out = BMSR_LSTATUS;
3291 case BOND_INFO_QUERY_OLD:
3292 case SIOCBONDINFOQUERY:
3293 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3295 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3298 res = bond_info_query(bond_dev, &k_binfo);
3300 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3304 case BOND_SLAVE_INFO_QUERY_OLD:
3305 case SIOCBONDSLAVEINFOQUERY:
3306 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3308 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3311 res = bond_slave_info_query(bond_dev, &k_sinfo);
3313 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3321 net = dev_net(bond_dev);
3323 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3326 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3328 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3333 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3335 case BOND_ENSLAVE_OLD:
3336 case SIOCBONDENSLAVE:
3337 res = bond_enslave(bond_dev, slave_dev);
3339 case BOND_RELEASE_OLD:
3340 case SIOCBONDRELEASE:
3341 res = bond_release(bond_dev, slave_dev);
3343 case BOND_SETHWADDR_OLD:
3344 case SIOCBONDSETHWADDR:
3345 bond_set_dev_addr(bond_dev, slave_dev);
3348 case BOND_CHANGE_ACTIVE_OLD:
3349 case SIOCBONDCHANGEACTIVE:
3350 bond_opt_initstr(&newval, slave_dev->name);
3351 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3360 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3362 struct bonding *bond = netdev_priv(bond_dev);
3364 if (change & IFF_PROMISC)
3365 bond_set_promiscuity(bond,
3366 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3368 if (change & IFF_ALLMULTI)
3369 bond_set_allmulti(bond,
3370 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3373 static void bond_set_rx_mode(struct net_device *bond_dev)
3375 struct bonding *bond = netdev_priv(bond_dev);
3376 struct list_head *iter;
3377 struct slave *slave;
3380 if (bond_uses_primary(bond)) {
3381 slave = rcu_dereference(bond->curr_active_slave);
3383 dev_uc_sync(slave->dev, bond_dev);
3384 dev_mc_sync(slave->dev, bond_dev);
3387 bond_for_each_slave_rcu(bond, slave, iter) {
3388 dev_uc_sync_multiple(slave->dev, bond_dev);
3389 dev_mc_sync_multiple(slave->dev, bond_dev);
3395 static int bond_neigh_init(struct neighbour *n)
3397 struct bonding *bond = netdev_priv(n->dev);
3398 const struct net_device_ops *slave_ops;
3399 struct neigh_parms parms;
3400 struct slave *slave;
3403 slave = bond_first_slave(bond);
3406 slave_ops = slave->dev->netdev_ops;
3407 if (!slave_ops->ndo_neigh_setup)
3410 parms.neigh_setup = NULL;
3411 parms.neigh_cleanup = NULL;
3412 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3416 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3417 * after the last slave has been detached. Assumes that all slaves
3418 * utilize the same neigh_cleanup (true at this writing as only user
3421 n->parms->neigh_cleanup = parms.neigh_cleanup;
3423 if (!parms.neigh_setup)
3426 return parms.neigh_setup(n);
3429 /* The bonding ndo_neigh_setup is called at init time beofre any
3430 * slave exists. So we must declare proxy setup function which will
3431 * be used at run time to resolve the actual slave neigh param setup.
3433 * It's also called by master devices (such as vlans) to setup their
3434 * underlying devices. In that case - do nothing, we're already set up from
3437 static int bond_neigh_setup(struct net_device *dev,
3438 struct neigh_parms *parms)
3440 /* modify only our neigh_parms */
3441 if (parms->dev == dev)
3442 parms->neigh_setup = bond_neigh_init;
3447 /* Change the MTU of all of a master's slaves to match the master */
3448 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3450 struct bonding *bond = netdev_priv(bond_dev);
3451 struct slave *slave, *rollback_slave;
3452 struct list_head *iter;
3455 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3457 bond_for_each_slave(bond, slave, iter) {
3458 netdev_dbg(bond_dev, "s %p c_m %p\n",
3459 slave, slave->dev->netdev_ops->ndo_change_mtu);
3461 res = dev_set_mtu(slave->dev, new_mtu);
3464 /* If we failed to set the slave's mtu to the new value
3465 * we must abort the operation even in ACTIVE_BACKUP
3466 * mode, because if we allow the backup slaves to have
3467 * different mtu values than the active slave we'll
3468 * need to change their mtu when doing a failover. That
3469 * means changing their mtu from timer context, which
3470 * is probably not a good idea.
3472 netdev_dbg(bond_dev, "err %d %s\n", res,
3478 bond_dev->mtu = new_mtu;
3483 /* unwind from head to the slave that failed */
3484 bond_for_each_slave(bond, rollback_slave, iter) {
3487 if (rollback_slave == slave)
3490 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3492 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3493 tmp_res, rollback_slave->dev->name);
3500 /* Change HW address
3502 * Note that many devices must be down to change the HW address, and
3503 * downing the master releases all slaves. We can make bonds full of
3504 * bonding devices to test this, however.
3506 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3508 struct bonding *bond = netdev_priv(bond_dev);
3509 struct slave *slave, *rollback_slave;
3510 struct sockaddr *sa = addr, tmp_sa;
3511 struct list_head *iter;
3514 if (BOND_MODE(bond) == BOND_MODE_ALB)
3515 return bond_alb_set_mac_address(bond_dev, addr);
3518 netdev_dbg(bond_dev, "bond=%p\n", bond);
3520 /* If fail_over_mac is enabled, do nothing and return success.
3521 * Returning an error causes ifenslave to fail.
3523 if (bond->params.fail_over_mac &&
3524 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3527 if (!is_valid_ether_addr(sa->sa_data))
3528 return -EADDRNOTAVAIL;
3530 bond_for_each_slave(bond, slave, iter) {
3531 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3532 res = dev_set_mac_address(slave->dev, addr);
3534 /* TODO: consider downing the slave
3536 * User should expect communications
3537 * breakage anyway until ARP finish
3540 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3546 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3550 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3551 tmp_sa.sa_family = bond_dev->type;
3553 /* unwind from head to the slave that failed */
3554 bond_for_each_slave(bond, rollback_slave, iter) {
3557 if (rollback_slave == slave)
3560 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3562 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3563 tmp_res, rollback_slave->dev->name);
3571 * bond_xmit_slave_id - transmit skb through slave with slave_id
3572 * @bond: bonding device that is transmitting
3573 * @skb: buffer to transmit
3574 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3576 * This function tries to transmit through slave with slave_id but in case
3577 * it fails, it tries to find the first available slave for transmission.
3578 * The skb is consumed in all cases, thus the function is void.
3580 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3582 struct list_head *iter;
3583 struct slave *slave;
3586 /* Here we start from the slave with slave_id */
3587 bond_for_each_slave_rcu(bond, slave, iter) {
3589 if (bond_slave_can_tx(slave)) {
3590 bond_dev_queue_xmit(bond, skb, slave->dev);
3596 /* Here we start from the first slave up to slave_id */
3598 bond_for_each_slave_rcu(bond, slave, iter) {
3601 if (bond_slave_can_tx(slave)) {
3602 bond_dev_queue_xmit(bond, skb, slave->dev);
3606 /* no slave that can tx has been found */
3607 bond_tx_drop(bond->dev, skb);
3611 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3612 * @bond: bonding device to use
3614 * Based on the value of the bonding device's packets_per_slave parameter
3615 * this function generates a slave id, which is usually used as the next
3616 * slave to transmit through.
3618 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3621 struct reciprocal_value reciprocal_packets_per_slave;
3622 int packets_per_slave = bond->params.packets_per_slave;
3624 switch (packets_per_slave) {
3626 slave_id = prandom_u32();
3629 slave_id = bond->rr_tx_counter;
3632 reciprocal_packets_per_slave =
3633 bond->params.reciprocal_packets_per_slave;
3634 slave_id = reciprocal_divide(bond->rr_tx_counter,
3635 reciprocal_packets_per_slave);
3638 bond->rr_tx_counter++;
3643 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3645 struct bonding *bond = netdev_priv(bond_dev);
3646 struct iphdr *iph = ip_hdr(skb);
3647 struct slave *slave;
3650 /* Start with the curr_active_slave that joined the bond as the
3651 * default for sending IGMP traffic. For failover purposes one
3652 * needs to maintain some consistency for the interface that will
3653 * send the join/membership reports. The curr_active_slave found
3654 * will send all of this type of traffic.
3656 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3657 slave = rcu_dereference(bond->curr_active_slave);
3659 bond_dev_queue_xmit(bond, skb, slave->dev);
3661 bond_xmit_slave_id(bond, skb, 0);
3663 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3665 if (likely(slave_cnt)) {
3666 slave_id = bond_rr_gen_slave_id(bond);
3667 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3669 bond_tx_drop(bond_dev, skb);
3673 return NETDEV_TX_OK;
3676 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3677 * the bond has a usable interface.
3679 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3681 struct bonding *bond = netdev_priv(bond_dev);
3682 struct slave *slave;
3684 slave = rcu_dereference(bond->curr_active_slave);
3686 bond_dev_queue_xmit(bond, skb, slave->dev);
3688 bond_tx_drop(bond_dev, skb);
3690 return NETDEV_TX_OK;
3693 /* Use this to update slave_array when (a) it's not appropriate to update
3694 * slave_array right away (note that update_slave_array() may sleep)
3695 * and / or (b) RTNL is not held.
3697 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3699 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3702 /* Slave array work handler. Holds only RTNL */
3703 static void bond_slave_arr_handler(struct work_struct *work)
3705 struct bonding *bond = container_of(work, struct bonding,
3706 slave_arr_work.work);
3709 if (!rtnl_trylock())
3712 ret = bond_update_slave_arr(bond, NULL);
3715 pr_warn_ratelimited("Failed to update slave array from WT\n");
3721 bond_slave_arr_work_rearm(bond, 1);
3724 /* Build the usable slaves array in control path for modes that use xmit-hash
3725 * to determine the slave interface -
3726 * (a) BOND_MODE_8023AD
3728 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3730 * The caller is expected to hold RTNL only and NO other lock!
3732 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3734 struct slave *slave;
3735 struct list_head *iter;
3736 struct bond_up_slave *new_arr, *old_arr;
3741 #ifdef CONFIG_LOCKDEP
3742 WARN_ON(lockdep_is_held(&bond->mode_lock));
3745 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3749 pr_err("Failed to build slave-array.\n");
3752 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3753 struct ad_info ad_info;
3755 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3756 pr_debug("bond_3ad_get_active_agg_info failed\n");
3757 kfree_rcu(new_arr, rcu);
3758 /* No active aggragator means it's not safe to use
3759 * the previous array.
3761 old_arr = rtnl_dereference(bond->slave_arr);
3763 RCU_INIT_POINTER(bond->slave_arr, NULL);
3764 kfree_rcu(old_arr, rcu);
3768 slaves_in_agg = ad_info.ports;
3769 agg_id = ad_info.aggregator_id;
3771 bond_for_each_slave(bond, slave, iter) {
3772 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3773 struct aggregator *agg;
3775 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3776 if (!agg || agg->aggregator_identifier != agg_id)
3779 if (!bond_slave_can_tx(slave))
3781 if (skipslave == slave)
3783 new_arr->arr[new_arr->count++] = slave;
3786 old_arr = rtnl_dereference(bond->slave_arr);
3787 rcu_assign_pointer(bond->slave_arr, new_arr);
3789 kfree_rcu(old_arr, rcu);
3791 if (ret != 0 && skipslave) {
3794 /* Rare situation where caller has asked to skip a specific
3795 * slave but allocation failed (most likely!). BTW this is
3796 * only possible when the call is initiated from
3797 * __bond_release_one(). In this situation; overwrite the
3798 * skipslave entry in the array with the last entry from the
3799 * array to avoid a situation where the xmit path may choose
3800 * this to-be-skipped slave to send a packet out.
3802 old_arr = rtnl_dereference(bond->slave_arr);
3803 for (idx = 0; idx < old_arr->count; idx++) {
3804 if (skipslave == old_arr->arr[idx]) {
3806 old_arr->arr[old_arr->count-1];
3815 /* Use this Xmit function for 3AD as well as XOR modes. The current
3816 * usable slave array is formed in the control path. The xmit function
3817 * just calculates hash and sends the packet out.
3819 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3821 struct bonding *bond = netdev_priv(dev);
3822 struct slave *slave;
3823 struct bond_up_slave *slaves;
3826 slaves = rcu_dereference(bond->slave_arr);
3827 count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3828 if (likely(count)) {
3829 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3830 bond_dev_queue_xmit(bond, skb, slave->dev);
3832 bond_tx_drop(dev, skb);
3835 return NETDEV_TX_OK;
3838 /* in broadcast mode, we send everything to all usable interfaces. */
3839 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3841 struct bonding *bond = netdev_priv(bond_dev);
3842 struct slave *slave = NULL;
3843 struct list_head *iter;
3845 bond_for_each_slave_rcu(bond, slave, iter) {
3846 if (bond_is_last_slave(bond, slave))
3848 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3849 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3852 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3853 bond_dev->name, __func__);
3856 bond_dev_queue_xmit(bond, skb2, slave->dev);
3859 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3860 bond_dev_queue_xmit(bond, skb, slave->dev);
3862 bond_tx_drop(bond_dev, skb);
3864 return NETDEV_TX_OK;
3867 /*------------------------- Device initialization ---------------------------*/
3869 /* Lookup the slave that corresponds to a qid */
3870 static inline int bond_slave_override(struct bonding *bond,
3871 struct sk_buff *skb)
3873 struct slave *slave = NULL;
3874 struct list_head *iter;
3876 if (!skb->queue_mapping)
3879 /* Find out if any slaves have the same mapping as this skb. */
3880 bond_for_each_slave_rcu(bond, slave, iter) {
3881 if (slave->queue_id == skb->queue_mapping) {
3882 if (bond_slave_is_up(slave) &&
3883 slave->link == BOND_LINK_UP) {
3884 bond_dev_queue_xmit(bond, skb, slave->dev);
3887 /* If the slave isn't UP, use default transmit policy. */
3896 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3897 void *accel_priv, select_queue_fallback_t fallback)
3899 /* This helper function exists to help dev_pick_tx get the correct
3900 * destination queue. Using a helper function skips a call to
3901 * skb_tx_hash and will put the skbs in the queue we expect on their
3902 * way down to the bonding driver.
3904 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3906 /* Save the original txq to restore before passing to the driver */
3907 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3909 if (unlikely(txq >= dev->real_num_tx_queues)) {
3911 txq -= dev->real_num_tx_queues;
3912 } while (txq >= dev->real_num_tx_queues);
3917 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3919 struct bonding *bond = netdev_priv(dev);
3921 if (bond_should_override_tx_queue(bond) &&
3922 !bond_slave_override(bond, skb))
3923 return NETDEV_TX_OK;
3925 switch (BOND_MODE(bond)) {
3926 case BOND_MODE_ROUNDROBIN:
3927 return bond_xmit_roundrobin(skb, dev);
3928 case BOND_MODE_ACTIVEBACKUP:
3929 return bond_xmit_activebackup(skb, dev);
3930 case BOND_MODE_8023AD:
3932 return bond_3ad_xor_xmit(skb, dev);
3933 case BOND_MODE_BROADCAST:
3934 return bond_xmit_broadcast(skb, dev);
3936 return bond_alb_xmit(skb, dev);
3938 return bond_tlb_xmit(skb, dev);
3940 /* Should never happen, mode already checked */
3941 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3943 bond_tx_drop(dev, skb);
3944 return NETDEV_TX_OK;
3948 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3950 struct bonding *bond = netdev_priv(dev);
3951 netdev_tx_t ret = NETDEV_TX_OK;
3953 /* If we risk deadlock from transmitting this in the
3954 * netpoll path, tell netpoll to queue the frame for later tx
3956 if (unlikely(is_netpoll_tx_blocked(dev)))
3957 return NETDEV_TX_BUSY;
3960 if (bond_has_slaves(bond))
3961 ret = __bond_start_xmit(skb, dev);
3963 bond_tx_drop(dev, skb);
3969 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3970 struct ethtool_cmd *ecmd)
3972 struct bonding *bond = netdev_priv(bond_dev);
3973 unsigned long speed = 0;
3974 struct list_head *iter;
3975 struct slave *slave;
3977 ecmd->duplex = DUPLEX_UNKNOWN;
3978 ecmd->port = PORT_OTHER;
3980 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3981 * do not need to check mode. Though link speed might not represent
3982 * the true receive or transmit bandwidth (not all modes are symmetric)
3983 * this is an accurate maximum.
3985 bond_for_each_slave(bond, slave, iter) {
3986 if (bond_slave_can_tx(slave)) {
3987 if (slave->speed != SPEED_UNKNOWN)
3988 speed += slave->speed;
3989 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3990 slave->duplex != DUPLEX_UNKNOWN)
3991 ecmd->duplex = slave->duplex;
3994 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3999 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4000 struct ethtool_drvinfo *drvinfo)
4002 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4003 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4004 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4008 static const struct ethtool_ops bond_ethtool_ops = {
4009 .get_drvinfo = bond_ethtool_get_drvinfo,
4010 .get_settings = bond_ethtool_get_settings,
4011 .get_link = ethtool_op_get_link,
4014 static const struct net_device_ops bond_netdev_ops = {
4015 .ndo_init = bond_init,
4016 .ndo_uninit = bond_uninit,
4017 .ndo_open = bond_open,
4018 .ndo_stop = bond_close,
4019 .ndo_start_xmit = bond_start_xmit,
4020 .ndo_select_queue = bond_select_queue,
4021 .ndo_get_stats64 = bond_get_stats,
4022 .ndo_do_ioctl = bond_do_ioctl,
4023 .ndo_change_rx_flags = bond_change_rx_flags,
4024 .ndo_set_rx_mode = bond_set_rx_mode,
4025 .ndo_change_mtu = bond_change_mtu,
4026 .ndo_set_mac_address = bond_set_mac_address,
4027 .ndo_neigh_setup = bond_neigh_setup,
4028 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4029 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4030 #ifdef CONFIG_NET_POLL_CONTROLLER
4031 .ndo_netpoll_setup = bond_netpoll_setup,
4032 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4033 .ndo_poll_controller = bond_poll_controller,
4035 .ndo_add_slave = bond_enslave,
4036 .ndo_del_slave = bond_release,
4037 .ndo_fix_features = bond_fix_features,
4038 .ndo_bridge_setlink = switchdev_port_bridge_setlink,
4039 .ndo_bridge_getlink = switchdev_port_bridge_getlink,
4040 .ndo_bridge_dellink = switchdev_port_bridge_dellink,
4041 .ndo_fdb_add = switchdev_port_fdb_add,
4042 .ndo_fdb_del = switchdev_port_fdb_del,
4043 .ndo_fdb_dump = switchdev_port_fdb_dump,
4044 .ndo_features_check = passthru_features_check,
4047 static const struct device_type bond_type = {
4051 static void bond_destructor(struct net_device *bond_dev)
4053 struct bonding *bond = netdev_priv(bond_dev);
4055 destroy_workqueue(bond->wq);
4056 free_netdev(bond_dev);
4059 void bond_setup(struct net_device *bond_dev)
4061 struct bonding *bond = netdev_priv(bond_dev);
4063 spin_lock_init(&bond->mode_lock);
4064 bond->params = bonding_defaults;
4066 /* Initialize pointers */
4067 bond->dev = bond_dev;
4069 /* Initialize the device entry points */
4070 ether_setup(bond_dev);
4071 bond_dev->netdev_ops = &bond_netdev_ops;
4072 bond_dev->ethtool_ops = &bond_ethtool_ops;
4074 bond_dev->destructor = bond_destructor;
4076 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4078 /* Initialize the device options */
4079 bond_dev->tx_queue_len = 0;
4080 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4081 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4082 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4084 /* don't acquire bond device's netif_tx_lock when transmitting */
4085 bond_dev->features |= NETIF_F_LLTX;
4087 /* By default, we declare the bond to be fully
4088 * VLAN hardware accelerated capable. Special
4089 * care is taken in the various xmit functions
4090 * when there are slaves that are not hw accel
4094 /* Don't allow bond devices to change network namespaces. */
4095 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4097 bond_dev->hw_features = BOND_VLAN_FEATURES |
4098 NETIF_F_HW_VLAN_CTAG_TX |
4099 NETIF_F_HW_VLAN_CTAG_RX |
4100 NETIF_F_HW_VLAN_CTAG_FILTER;
4102 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4103 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4104 bond_dev->features |= bond_dev->hw_features;
4107 /* Destroy a bonding device.
4108 * Must be under rtnl_lock when this function is called.
4110 static void bond_uninit(struct net_device *bond_dev)
4112 struct bonding *bond = netdev_priv(bond_dev);
4113 struct list_head *iter;
4114 struct slave *slave;
4115 struct bond_up_slave *arr;
4117 bond_netpoll_cleanup(bond_dev);
4119 /* Release the bonded slaves */
4120 bond_for_each_slave(bond, slave, iter)
4121 __bond_release_one(bond_dev, slave->dev, true);
4122 netdev_info(bond_dev, "Released all slaves\n");
4124 arr = rtnl_dereference(bond->slave_arr);
4126 RCU_INIT_POINTER(bond->slave_arr, NULL);
4127 kfree_rcu(arr, rcu);
4130 list_del(&bond->bond_list);
4132 bond_debug_unregister(bond);
4135 /*------------------------- Module initialization ---------------------------*/
4137 static int bond_check_params(struct bond_params *params)
4139 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4140 struct bond_opt_value newval;
4141 const struct bond_opt_value *valptr;
4142 int arp_all_targets_value;
4143 u16 ad_actor_sys_prio = 0;
4144 u16 ad_user_port_key = 0;
4146 /* Convert string parameters. */
4148 bond_opt_initstr(&newval, mode);
4149 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4151 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4154 bond_mode = valptr->value;
4157 if (xmit_hash_policy) {
4158 if ((bond_mode != BOND_MODE_XOR) &&
4159 (bond_mode != BOND_MODE_8023AD) &&
4160 (bond_mode != BOND_MODE_TLB)) {
4161 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4162 bond_mode_name(bond_mode));
4164 bond_opt_initstr(&newval, xmit_hash_policy);
4165 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4168 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4172 xmit_hashtype = valptr->value;
4177 if (bond_mode != BOND_MODE_8023AD) {
4178 pr_info("lacp_rate param is irrelevant in mode %s\n",
4179 bond_mode_name(bond_mode));
4181 bond_opt_initstr(&newval, lacp_rate);
4182 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4185 pr_err("Error: Invalid lacp rate \"%s\"\n",
4189 lacp_fast = valptr->value;
4194 bond_opt_initstr(&newval, ad_select);
4195 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4198 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4201 params->ad_select = valptr->value;
4202 if (bond_mode != BOND_MODE_8023AD)
4203 pr_warn("ad_select param only affects 802.3ad mode\n");
4205 params->ad_select = BOND_AD_STABLE;
4208 if (max_bonds < 0) {
4209 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4210 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4211 max_bonds = BOND_DEFAULT_MAX_BONDS;
4215 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4221 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4226 if (downdelay < 0) {
4227 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4228 downdelay, INT_MAX);
4232 if ((use_carrier != 0) && (use_carrier != 1)) {
4233 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4238 if (num_peer_notif < 0 || num_peer_notif > 255) {
4239 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4244 /* reset values for 802.3ad/TLB/ALB */
4245 if (!bond_mode_uses_arp(bond_mode)) {
4247 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");
4248 pr_warn("Forcing miimon to 100msec\n");
4249 miimon = BOND_DEFAULT_MIIMON;
4253 if (tx_queues < 1 || tx_queues > 255) {
4254 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4255 tx_queues, BOND_DEFAULT_TX_QUEUES);
4256 tx_queues = BOND_DEFAULT_TX_QUEUES;
4259 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4260 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4262 all_slaves_active = 0;
4265 if (resend_igmp < 0 || resend_igmp > 255) {
4266 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4267 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4268 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4271 bond_opt_initval(&newval, packets_per_slave);
4272 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4273 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4274 packets_per_slave, USHRT_MAX);
4275 packets_per_slave = 1;
4278 if (bond_mode == BOND_MODE_ALB) {
4279 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",
4284 if (updelay || downdelay) {
4285 /* just warn the user the up/down delay will have
4286 * no effect since miimon is zero...
4288 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",
4289 updelay, downdelay);
4292 /* don't allow arp monitoring */
4294 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4295 miimon, arp_interval);
4299 if ((updelay % miimon) != 0) {
4300 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4301 updelay, miimon, (updelay / miimon) * miimon);
4306 if ((downdelay % miimon) != 0) {
4307 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4309 (downdelay / miimon) * miimon);
4312 downdelay /= miimon;
4315 if (arp_interval < 0) {
4316 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4317 arp_interval, INT_MAX);
4321 for (arp_ip_count = 0, i = 0;
4322 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4325 /* not a complete check, but good enough to catch mistakes */
4326 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4327 !bond_is_ip_target_ok(ip)) {
4328 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4332 if (bond_get_targets_ip(arp_target, ip) == -1)
4333 arp_target[arp_ip_count++] = ip;
4335 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4340 if (arp_interval && !arp_ip_count) {
4341 /* don't allow arping if no arp_ip_target given... */
4342 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4348 if (!arp_interval) {
4349 pr_err("arp_validate requires arp_interval\n");
4353 bond_opt_initstr(&newval, arp_validate);
4354 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4357 pr_err("Error: invalid arp_validate \"%s\"\n",
4361 arp_validate_value = valptr->value;
4363 arp_validate_value = 0;
4366 arp_all_targets_value = 0;
4367 if (arp_all_targets) {
4368 bond_opt_initstr(&newval, arp_all_targets);
4369 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4372 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4374 arp_all_targets_value = 0;
4376 arp_all_targets_value = valptr->value;
4381 pr_info("MII link monitoring set to %d ms\n", miimon);
4382 } else if (arp_interval) {
4383 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4384 arp_validate_value);
4385 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4386 arp_interval, valptr->string, arp_ip_count);
4388 for (i = 0; i < arp_ip_count; i++)
4389 pr_cont(" %s", arp_ip_target[i]);
4393 } else if (max_bonds) {
4394 /* miimon and arp_interval not set, we need one so things
4395 * work as expected, see bonding.txt for details
4397 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");
4400 if (primary && !bond_mode_uses_primary(bond_mode)) {
4401 /* currently, using a primary only makes sense
4402 * in active backup, TLB or ALB modes
4404 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4405 primary, bond_mode_name(bond_mode));
4409 if (primary && primary_reselect) {
4410 bond_opt_initstr(&newval, primary_reselect);
4411 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4414 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4418 primary_reselect_value = valptr->value;
4420 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4423 if (fail_over_mac) {
4424 bond_opt_initstr(&newval, fail_over_mac);
4425 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4428 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4432 fail_over_mac_value = valptr->value;
4433 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4434 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4436 fail_over_mac_value = BOND_FOM_NONE;
4439 bond_opt_initstr(&newval, "default");
4440 valptr = bond_opt_parse(
4441 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4444 pr_err("Error: No ad_actor_sys_prio default value");
4447 ad_actor_sys_prio = valptr->value;
4449 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4452 pr_err("Error: No ad_user_port_key default value");
4455 ad_user_port_key = valptr->value;
4457 if (lp_interval == 0) {
4458 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4459 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4460 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4463 /* fill params struct with the proper values */
4464 params->mode = bond_mode;
4465 params->xmit_policy = xmit_hashtype;
4466 params->miimon = miimon;
4467 params->num_peer_notif = num_peer_notif;
4468 params->arp_interval = arp_interval;
4469 params->arp_validate = arp_validate_value;
4470 params->arp_all_targets = arp_all_targets_value;
4471 params->updelay = updelay;
4472 params->downdelay = downdelay;
4473 params->use_carrier = use_carrier;
4474 params->lacp_fast = lacp_fast;
4475 params->primary[0] = 0;
4476 params->primary_reselect = primary_reselect_value;
4477 params->fail_over_mac = fail_over_mac_value;
4478 params->tx_queues = tx_queues;
4479 params->all_slaves_active = all_slaves_active;
4480 params->resend_igmp = resend_igmp;
4481 params->min_links = min_links;
4482 params->lp_interval = lp_interval;
4483 params->packets_per_slave = packets_per_slave;
4484 params->tlb_dynamic_lb = 1; /* Default value */
4485 params->ad_actor_sys_prio = ad_actor_sys_prio;
4486 eth_zero_addr(params->ad_actor_system);
4487 params->ad_user_port_key = ad_user_port_key;
4488 if (packets_per_slave > 0) {
4489 params->reciprocal_packets_per_slave =
4490 reciprocal_value(packets_per_slave);
4492 /* reciprocal_packets_per_slave is unused if
4493 * packets_per_slave is 0 or 1, just initialize it
4495 params->reciprocal_packets_per_slave =
4496 (struct reciprocal_value) { 0 };
4500 strncpy(params->primary, primary, IFNAMSIZ);
4501 params->primary[IFNAMSIZ - 1] = 0;
4504 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4509 static struct lock_class_key bonding_netdev_xmit_lock_key;
4510 static struct lock_class_key bonding_netdev_addr_lock_key;
4511 static struct lock_class_key bonding_tx_busylock_key;
4513 static void bond_set_lockdep_class_one(struct net_device *dev,
4514 struct netdev_queue *txq,
4517 lockdep_set_class(&txq->_xmit_lock,
4518 &bonding_netdev_xmit_lock_key);
4521 static void bond_set_lockdep_class(struct net_device *dev)
4523 lockdep_set_class(&dev->addr_list_lock,
4524 &bonding_netdev_addr_lock_key);
4525 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4526 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4529 /* Called from registration process */
4530 static int bond_init(struct net_device *bond_dev)
4532 struct bonding *bond = netdev_priv(bond_dev);
4533 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4535 netdev_dbg(bond_dev, "Begin bond_init\n");
4537 bond->wq = create_singlethread_workqueue(bond_dev->name);
4541 bond_set_lockdep_class(bond_dev);
4543 list_add_tail(&bond->bond_list, &bn->dev_list);
4545 bond_prepare_sysfs_group(bond);
4547 bond_debug_register(bond);
4549 /* Ensure valid dev_addr */
4550 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4551 bond_dev->addr_assign_type == NET_ADDR_PERM)
4552 eth_hw_addr_random(bond_dev);
4557 unsigned int bond_get_num_tx_queues(void)
4562 /* Create a new bond based on the specified name and bonding parameters.
4563 * If name is NULL, obtain a suitable "bond%d" name for us.
4564 * Caller must NOT hold rtnl_lock; we need to release it here before we
4565 * set up our sysfs entries.
4567 int bond_create(struct net *net, const char *name)
4569 struct net_device *bond_dev;
4570 struct bonding *bond;
4571 struct alb_bond_info *bond_info;
4576 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4577 name ? name : "bond%d", NET_NAME_UNKNOWN,
4578 bond_setup, tx_queues);
4580 pr_err("%s: eek! can't alloc netdev!\n", name);
4586 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4587 * It is set to 0 by default which is wrong.
4589 bond = netdev_priv(bond_dev);
4590 bond_info = &(BOND_ALB_INFO(bond));
4591 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4593 dev_net_set(bond_dev, net);
4594 bond_dev->rtnl_link_ops = &bond_link_ops;
4596 res = register_netdevice(bond_dev);
4598 netif_carrier_off(bond_dev);
4602 bond_destructor(bond_dev);
4606 static int __net_init bond_net_init(struct net *net)
4608 struct bond_net *bn = net_generic(net, bond_net_id);
4611 INIT_LIST_HEAD(&bn->dev_list);
4613 bond_create_proc_dir(bn);
4614 bond_create_sysfs(bn);
4619 static void __net_exit bond_net_exit(struct net *net)
4621 struct bond_net *bn = net_generic(net, bond_net_id);
4622 struct bonding *bond, *tmp_bond;
4625 bond_destroy_sysfs(bn);
4627 /* Kill off any bonds created after unregistering bond rtnl ops */
4629 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4630 unregister_netdevice_queue(bond->dev, &list);
4631 unregister_netdevice_many(&list);
4634 bond_destroy_proc_dir(bn);
4637 static struct pernet_operations bond_net_ops = {
4638 .init = bond_net_init,
4639 .exit = bond_net_exit,
4641 .size = sizeof(struct bond_net),
4644 static int __init bonding_init(void)
4649 pr_info("%s", bond_version);
4651 res = bond_check_params(&bonding_defaults);
4655 res = register_pernet_subsys(&bond_net_ops);
4659 res = bond_netlink_init();
4663 bond_create_debugfs();
4665 for (i = 0; i < max_bonds; i++) {
4666 res = bond_create(&init_net, NULL);
4671 register_netdevice_notifier(&bond_netdev_notifier);
4675 bond_destroy_debugfs();
4676 bond_netlink_fini();
4678 unregister_pernet_subsys(&bond_net_ops);
4683 static void __exit bonding_exit(void)
4685 unregister_netdevice_notifier(&bond_netdev_notifier);
4687 bond_destroy_debugfs();
4689 bond_netlink_fini();
4690 unregister_pernet_subsys(&bond_net_ops);
4692 #ifdef CONFIG_NET_POLL_CONTROLLER
4693 /* Make sure we don't have an imbalance on our netpoll blocking */
4694 WARN_ON(atomic_read(&netpoll_block_tx));
4698 module_init(bonding_init);
4699 module_exit(bonding_exit);
4700 MODULE_LICENSE("GPL");
4701 MODULE_VERSION(DRV_VERSION);
4702 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4703 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");