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_keys.h>
84 /*---------------------------- Module parameters ----------------------------*/
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
90 static int num_peer_notif = 1;
94 static int use_carrier = 1;
97 static char *primary_reselect;
98 static char *lacp_rate;
100 static char *ad_select;
101 static char *xmit_hash_policy;
102 static int arp_interval;
103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
104 static char *arp_validate;
105 static char *arp_all_targets;
106 static char *fail_over_mac;
107 static int all_slaves_active;
108 static struct bond_params bonding_defaults;
109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
110 static int packets_per_slave = 1;
111 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
119 "failover event (alias of num_unsol_na)");
120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
122 "failover event (alias of num_grat_arp)");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
150 "0 for slow, 1 for fast");
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
153 "0 for stable (default), 1 for bandwidth, "
155 module_param(min_links, int, 0);
156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
158 module_param(xmit_hash_policy, charp, 0);
159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
160 "0 for layer 2 (default), 1 for layer 3+4, "
161 "2 for layer 2+3, 3 for encap layer 2+3, "
162 "4 for encap layer 3+4");
163 module_param(arp_interval, int, 0);
164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165 module_param_array(arp_ip_target, charp, NULL, 0);
166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
167 module_param(arp_validate, charp, 0);
168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets, charp, 0);
172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac, charp, 0);
174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active, int, 0);
178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp, int, 0);
182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
184 module_param(packets_per_slave, int, 0);
185 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
186 "mode; 0 for a random slave, 1 packet per "
187 "slave (default), >1 packets per slave.");
188 module_param(lp_interval, uint, 0);
189 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
190 "the bonding driver sends learning packets to "
191 "each slaves peer switch. The default is 1.");
193 /*----------------------------- Global variables ----------------------------*/
195 #ifdef CONFIG_NET_POLL_CONTROLLER
196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
199 int bond_net_id __read_mostly;
201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
202 static int arp_ip_count;
203 static int bond_mode = BOND_MODE_ROUNDROBIN;
204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
205 static int lacp_fast;
207 /*-------------------------- Forward declarations ---------------------------*/
209 static int bond_init(struct net_device *bond_dev);
210 static void bond_uninit(struct net_device *bond_dev);
212 /*---------------------------- General routines -----------------------------*/
214 const char *bond_mode_name(int mode)
216 static const char *names[] = {
217 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
218 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
219 [BOND_MODE_XOR] = "load balancing (xor)",
220 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
221 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
222 [BOND_MODE_TLB] = "transmit load balancing",
223 [BOND_MODE_ALB] = "adaptive load balancing",
226 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
232 /*---------------------------------- VLAN -----------------------------------*/
235 * bond_dev_queue_xmit - Prepare skb for xmit.
237 * @bond: bond device that got this skb for tx.
238 * @skb: hw accel VLAN tagged skb to transmit
239 * @slave_dev: slave that is supposed to xmit this skbuff
241 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
242 struct net_device *slave_dev)
244 skb->dev = slave_dev;
246 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
247 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
248 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
250 if (unlikely(netpoll_tx_running(bond->dev)))
251 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
257 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
258 * We don't protect the slave list iteration with a lock because:
259 * a. This operation is performed in IOCTL context,
260 * b. The operation is protected by the RTNL semaphore in the 8021q code,
261 * c. Holding a lock with BH disabled while directly calling a base driver
262 * entry point is generally a BAD idea.
264 * The design of synchronization/protection for this operation in the 8021q
265 * module is good for one or more VLAN devices over a single physical device
266 * and cannot be extended for a teaming solution like bonding, so there is a
267 * potential race condition here where a net device from the vlan group might
268 * be referenced (either by a base driver or the 8021q code) while it is being
269 * removed from the system. However, it turns out we're not making matters
270 * worse, and if it works for regular VLAN usage it will work here too.
274 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
275 * @bond_dev: bonding net device that got called
276 * @vid: vlan id being added
278 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
279 __be16 proto, u16 vid)
281 struct bonding *bond = netdev_priv(bond_dev);
282 struct slave *slave, *rollback_slave;
283 struct list_head *iter;
286 bond_for_each_slave(bond, slave, iter) {
287 res = vlan_vid_add(slave->dev, proto, vid);
295 /* unwind to the slave that failed */
296 bond_for_each_slave(bond, rollback_slave, iter) {
297 if (rollback_slave == slave)
300 vlan_vid_del(rollback_slave->dev, proto, vid);
307 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
308 * @bond_dev: bonding net device that got called
309 * @vid: vlan id being removed
311 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
312 __be16 proto, u16 vid)
314 struct bonding *bond = netdev_priv(bond_dev);
315 struct list_head *iter;
318 bond_for_each_slave(bond, slave, iter)
319 vlan_vid_del(slave->dev, proto, vid);
321 if (bond_is_lb(bond))
322 bond_alb_clear_vlan(bond, vid);
327 /*------------------------------- Link status -------------------------------*/
330 * Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
334 * Returns zero if carrier state does not change, nonzero if it does.
336 static int bond_set_carrier(struct bonding *bond)
338 struct list_head *iter;
341 if (!bond_has_slaves(bond))
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
366 * Get link speed and duplex from the slave's base driver
367 * using ethtool. If for some reason the call fails or the
368 * values are invalid, set speed and duplex to -1,
371 static void bond_update_speed_duplex(struct slave *slave)
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_cmd ecmd;
378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
381 res = __ethtool_get_settings(slave_dev, &ecmd);
385 slave_speed = ethtool_cmd_speed(&ecmd);
386 if (slave_speed == 0 || slave_speed == ((__u32) -1))
389 switch (ecmd.duplex) {
397 slave->speed = slave_speed;
398 slave->duplex = ecmd.duplex;
403 const char *bond_slave_link_status(s8 link)
420 * if <dev> supports MII link status reporting, check its link status.
422 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
423 * depending upon the setting of the use_carrier parameter.
425 * Return either BMSR_LSTATUS, meaning that the link is up (or we
426 * can't tell and just pretend it is), or 0, meaning that the link is
429 * If reporting is non-zero, instead of faking link up, return -1 if
430 * both ETHTOOL and MII ioctls fail (meaning the device does not
431 * support them). If use_carrier is set, return whatever it says.
432 * It'd be nice if there was a good way to tell if a driver supports
433 * netif_carrier, but there really isn't.
435 static int bond_check_dev_link(struct bonding *bond,
436 struct net_device *slave_dev, int reporting)
438 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
439 int (*ioctl)(struct net_device *, struct ifreq *, int);
441 struct mii_ioctl_data *mii;
443 if (!reporting && !netif_running(slave_dev))
446 if (bond->params.use_carrier)
447 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
449 /* Try to get link status using Ethtool first. */
450 if (slave_dev->ethtool_ops->get_link)
451 return slave_dev->ethtool_ops->get_link(slave_dev) ?
454 /* Ethtool can't be used, fallback to MII ioctls. */
455 ioctl = slave_ops->ndo_do_ioctl;
457 /* TODO: set pointer to correct ioctl on a per team member */
458 /* bases to make this more efficient. that is, once */
459 /* we determine the correct ioctl, we will always */
460 /* call it and not the others for that team */
464 * We cannot assume that SIOCGMIIPHY will also read a
465 * register; not all network drivers (e.g., e100)
469 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
470 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
472 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
473 mii->reg_num = MII_BMSR;
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
475 return mii->val_out & BMSR_LSTATUS;
480 * If reporting, report that either there's no dev->do_ioctl,
481 * or both SIOCGMIIREG and get_link failed (meaning that we
482 * cannot report link status). If not reporting, pretend
485 return reporting ? -1 : BMSR_LSTATUS;
488 /*----------------------------- Multicast list ------------------------------*/
491 * Push the promiscuity flag down to appropriate slaves
493 static int bond_set_promiscuity(struct bonding *bond, int inc)
495 struct list_head *iter;
498 if (bond_uses_primary(bond)) {
499 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
502 err = dev_set_promiscuity(curr_active->dev, inc);
506 bond_for_each_slave(bond, slave, iter) {
507 err = dev_set_promiscuity(slave->dev, inc);
516 * Push the allmulti flag down to all slaves
518 static int bond_set_allmulti(struct bonding *bond, int inc)
520 struct list_head *iter;
523 if (bond_uses_primary(bond)) {
524 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
527 err = dev_set_allmulti(curr_active->dev, inc);
531 bond_for_each_slave(bond, slave, iter) {
532 err = dev_set_allmulti(slave->dev, inc);
541 * Retrieve the list of registered multicast addresses for the bonding
542 * device and retransmit an IGMP JOIN request to the current active
545 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
547 struct bonding *bond = container_of(work, struct bonding,
550 if (!rtnl_trylock()) {
551 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
554 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
556 if (bond->igmp_retrans > 1) {
557 bond->igmp_retrans--;
558 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
563 /* Flush bond's hardware addresses from slave
565 static void bond_hw_addr_flush(struct net_device *bond_dev,
566 struct net_device *slave_dev)
568 struct bonding *bond = netdev_priv(bond_dev);
570 dev_uc_unsync(slave_dev, bond_dev);
571 dev_mc_unsync(slave_dev, bond_dev);
573 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
574 /* del lacpdu mc addr from mc list */
575 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
577 dev_mc_del(slave_dev, lacpdu_multicast);
581 /*--------------------------- Active slave change ---------------------------*/
583 /* Update the hardware address list and promisc/allmulti for the new and
584 * old active slaves (if any). Modes that are not using primary keep all
585 * slaves up date at all times; only the modes that use primary need to call
586 * this function to swap these settings during a failover.
588 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
589 struct slave *old_active)
594 if (bond->dev->flags & IFF_PROMISC)
595 dev_set_promiscuity(old_active->dev, -1);
597 if (bond->dev->flags & IFF_ALLMULTI)
598 dev_set_allmulti(old_active->dev, -1);
600 bond_hw_addr_flush(bond->dev, old_active->dev);
604 /* FIXME: Signal errors upstream. */
605 if (bond->dev->flags & IFF_PROMISC)
606 dev_set_promiscuity(new_active->dev, 1);
608 if (bond->dev->flags & IFF_ALLMULTI)
609 dev_set_allmulti(new_active->dev, 1);
611 netif_addr_lock_bh(bond->dev);
612 dev_uc_sync(new_active->dev, bond->dev);
613 dev_mc_sync(new_active->dev, bond->dev);
614 netif_addr_unlock_bh(bond->dev);
619 * bond_set_dev_addr - clone slave's address to bond
620 * @bond_dev: bond net device
621 * @slave_dev: slave net device
623 * Should be called with RTNL held.
625 static void bond_set_dev_addr(struct net_device *bond_dev,
626 struct net_device *slave_dev)
628 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
629 bond_dev, slave_dev, slave_dev->addr_len);
630 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
631 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
632 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
636 * bond_do_fail_over_mac
638 * Perform special MAC address swapping for fail_over_mac settings
640 * Called with RTNL, curr_slave_lock for write_bh.
642 static void bond_do_fail_over_mac(struct bonding *bond,
643 struct slave *new_active,
644 struct slave *old_active)
645 __releases(&bond->curr_slave_lock)
646 __acquires(&bond->curr_slave_lock)
648 u8 tmp_mac[ETH_ALEN];
649 struct sockaddr saddr;
652 switch (bond->params.fail_over_mac) {
653 case BOND_FOM_ACTIVE:
655 write_unlock_bh(&bond->curr_slave_lock);
656 bond_set_dev_addr(bond->dev, new_active->dev);
657 write_lock_bh(&bond->curr_slave_lock);
660 case BOND_FOM_FOLLOW:
662 * if new_active && old_active, swap them
663 * if just old_active, do nothing (going to no active slave)
664 * if just new_active, set new_active to bond's MAC
669 write_unlock_bh(&bond->curr_slave_lock);
672 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
673 ether_addr_copy(saddr.sa_data,
674 old_active->dev->dev_addr);
675 saddr.sa_family = new_active->dev->type;
677 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
678 saddr.sa_family = bond->dev->type;
681 rv = dev_set_mac_address(new_active->dev, &saddr);
683 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
684 -rv, new_active->dev->name);
691 ether_addr_copy(saddr.sa_data, tmp_mac);
692 saddr.sa_family = old_active->dev->type;
694 rv = dev_set_mac_address(old_active->dev, &saddr);
696 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
697 -rv, new_active->dev->name);
699 write_lock_bh(&bond->curr_slave_lock);
702 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
703 bond->params.fail_over_mac);
709 static bool bond_should_change_active(struct bonding *bond)
711 struct slave *prim = bond->primary_slave;
712 struct slave *curr = bond_deref_active_protected(bond);
714 if (!prim || !curr || curr->link != BOND_LINK_UP)
716 if (bond->force_primary) {
717 bond->force_primary = false;
720 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
721 (prim->speed < curr->speed ||
722 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
724 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
730 * find_best_interface - select the best available slave to be the active one
731 * @bond: our bonding struct
733 static struct slave *bond_find_best_slave(struct bonding *bond)
735 struct slave *slave, *bestslave = NULL;
736 struct list_head *iter;
737 int mintime = bond->params.updelay;
739 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
740 bond_should_change_active(bond))
741 return bond->primary_slave;
743 bond_for_each_slave(bond, slave, iter) {
744 if (slave->link == BOND_LINK_UP)
746 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
747 slave->delay < mintime) {
748 mintime = slave->delay;
756 static bool bond_should_notify_peers(struct bonding *bond)
761 slave = rcu_dereference(bond->curr_active_slave);
764 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
765 slave ? slave->dev->name : "NULL");
767 if (!slave || !bond->send_peer_notif ||
768 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
775 * change_active_interface - change the active slave into the specified one
776 * @bond: our bonding struct
777 * @new: the new slave to make the active one
779 * Set the new slave to the bond's settings and unset them on the old
781 * Setting include flags, mc-list, promiscuity, allmulti, etc.
783 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
784 * because it is apparently the best available slave we have, even though its
785 * updelay hasn't timed out yet.
787 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
789 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
791 struct slave *old_active;
793 old_active = rcu_dereference_protected(bond->curr_active_slave,
795 lockdep_is_held(&bond->curr_slave_lock));
797 if (old_active == new_active)
801 new_active->last_link_up = jiffies;
803 if (new_active->link == BOND_LINK_BACK) {
804 if (bond_uses_primary(bond)) {
805 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
806 new_active->dev->name,
807 (bond->params.updelay - new_active->delay) * bond->params.miimon);
810 new_active->delay = 0;
811 new_active->link = BOND_LINK_UP;
813 if (BOND_MODE(bond) == BOND_MODE_8023AD)
814 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
816 if (bond_is_lb(bond))
817 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
819 if (bond_uses_primary(bond)) {
820 netdev_info(bond->dev, "making interface %s the new active one\n",
821 new_active->dev->name);
826 if (bond_uses_primary(bond))
827 bond_hw_addr_swap(bond, new_active, old_active);
829 if (bond_is_lb(bond)) {
830 bond_alb_handle_active_change(bond, new_active);
832 bond_set_slave_inactive_flags(old_active,
833 BOND_SLAVE_NOTIFY_NOW);
835 bond_set_slave_active_flags(new_active,
836 BOND_SLAVE_NOTIFY_NOW);
838 rcu_assign_pointer(bond->curr_active_slave, new_active);
841 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
843 bond_set_slave_inactive_flags(old_active,
844 BOND_SLAVE_NOTIFY_NOW);
847 bool should_notify_peers = false;
849 bond_set_slave_active_flags(new_active,
850 BOND_SLAVE_NOTIFY_NOW);
852 if (bond->params.fail_over_mac)
853 bond_do_fail_over_mac(bond, new_active,
856 if (netif_running(bond->dev)) {
857 bond->send_peer_notif =
858 bond->params.num_peer_notif;
859 should_notify_peers =
860 bond_should_notify_peers(bond);
863 write_unlock_bh(&bond->curr_slave_lock);
865 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
866 if (should_notify_peers)
867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
870 write_lock_bh(&bond->curr_slave_lock);
874 /* resend IGMP joins since active slave has changed or
875 * all were sent on curr_active_slave.
876 * resend only if bond is brought up with the affected
877 * bonding modes and the retransmission is enabled */
878 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
879 ((bond_uses_primary(bond) && new_active) ||
880 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
881 bond->igmp_retrans = bond->params.resend_igmp;
882 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
887 * bond_select_active_slave - select a new active slave, if needed
888 * @bond: our bonding struct
890 * This functions should be called when one of the following occurs:
891 * - The old curr_active_slave has been released or lost its link.
892 * - The primary_slave has got its link back.
893 * - A slave has got its link back and there's no old curr_active_slave.
895 * Caller must hold curr_slave_lock for write_bh.
897 void bond_select_active_slave(struct bonding *bond)
899 struct slave *best_slave;
902 best_slave = bond_find_best_slave(bond);
903 if (best_slave != bond_deref_active_protected(bond)) {
904 bond_change_active_slave(bond, best_slave);
905 rv = bond_set_carrier(bond);
909 if (netif_carrier_ok(bond->dev)) {
910 netdev_info(bond->dev, "first active interface up!\n");
912 netdev_info(bond->dev, "now running without any active interface!\n");
917 #ifdef CONFIG_NET_POLL_CONTROLLER
918 static inline int slave_enable_netpoll(struct slave *slave)
923 np = kzalloc(sizeof(*np), GFP_KERNEL);
928 err = __netpoll_setup(np, slave->dev);
937 static inline void slave_disable_netpoll(struct slave *slave)
939 struct netpoll *np = slave->np;
945 __netpoll_free_async(np);
948 static void bond_poll_controller(struct net_device *bond_dev)
952 static void bond_netpoll_cleanup(struct net_device *bond_dev)
954 struct bonding *bond = netdev_priv(bond_dev);
955 struct list_head *iter;
958 bond_for_each_slave(bond, slave, iter)
959 if (bond_slave_is_up(slave))
960 slave_disable_netpoll(slave);
963 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
965 struct bonding *bond = netdev_priv(dev);
966 struct list_head *iter;
970 bond_for_each_slave(bond, slave, iter) {
971 err = slave_enable_netpoll(slave);
973 bond_netpoll_cleanup(dev);
980 static inline int slave_enable_netpoll(struct slave *slave)
984 static inline void slave_disable_netpoll(struct slave *slave)
987 static void bond_netpoll_cleanup(struct net_device *bond_dev)
992 /*---------------------------------- IOCTL ----------------------------------*/
994 static netdev_features_t bond_fix_features(struct net_device *dev,
995 netdev_features_t features)
997 struct bonding *bond = netdev_priv(dev);
998 struct list_head *iter;
999 netdev_features_t mask;
1000 struct slave *slave;
1003 features &= ~NETIF_F_ONE_FOR_ALL;
1004 features |= NETIF_F_ALL_FOR_ALL;
1006 bond_for_each_slave(bond, slave, iter) {
1007 features = netdev_increment_features(features,
1008 slave->dev->features,
1011 features = netdev_add_tso_features(features, mask);
1016 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1017 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1018 NETIF_F_HIGHDMA | NETIF_F_LRO)
1020 #define BOND_ENC_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1021 NETIF_F_TSO | NETIF_F_GSO_UDP_TUNNEL)
1023 static void bond_compute_features(struct bonding *bond)
1025 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1026 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1027 netdev_features_t enc_features = BOND_ENC_FEATURES;
1028 struct net_device *bond_dev = bond->dev;
1029 struct list_head *iter;
1030 struct slave *slave;
1031 unsigned short max_hard_header_len = ETH_HLEN;
1032 unsigned int gso_max_size = GSO_MAX_SIZE;
1033 u16 gso_max_segs = GSO_MAX_SEGS;
1035 if (!bond_has_slaves(bond))
1037 vlan_features &= NETIF_F_ALL_FOR_ALL;
1039 bond_for_each_slave(bond, slave, iter) {
1040 vlan_features = netdev_increment_features(vlan_features,
1041 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1043 enc_features = netdev_increment_features(enc_features,
1044 slave->dev->hw_enc_features,
1046 dst_release_flag &= slave->dev->priv_flags;
1047 if (slave->dev->hard_header_len > max_hard_header_len)
1048 max_hard_header_len = slave->dev->hard_header_len;
1050 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1051 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1055 bond_dev->vlan_features = vlan_features;
1056 bond_dev->hw_enc_features = enc_features;
1057 bond_dev->hard_header_len = max_hard_header_len;
1058 bond_dev->gso_max_segs = gso_max_segs;
1059 netif_set_gso_max_size(bond_dev, gso_max_size);
1061 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1062 bond_dev->priv_flags = flags | dst_release_flag;
1064 netdev_change_features(bond_dev);
1067 static void bond_setup_by_slave(struct net_device *bond_dev,
1068 struct net_device *slave_dev)
1070 bond_dev->header_ops = slave_dev->header_ops;
1072 bond_dev->type = slave_dev->type;
1073 bond_dev->hard_header_len = slave_dev->hard_header_len;
1074 bond_dev->addr_len = slave_dev->addr_len;
1076 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1077 slave_dev->addr_len);
1080 /* On bonding slaves other than the currently active slave, suppress
1081 * duplicates except for alb non-mcast/bcast.
1083 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1084 struct slave *slave,
1085 struct bonding *bond)
1087 if (bond_is_slave_inactive(slave)) {
1088 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1089 skb->pkt_type != PACKET_BROADCAST &&
1090 skb->pkt_type != PACKET_MULTICAST)
1097 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1099 struct sk_buff *skb = *pskb;
1100 struct slave *slave;
1101 struct bonding *bond;
1102 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1104 int ret = RX_HANDLER_ANOTHER;
1106 skb = skb_share_check(skb, GFP_ATOMIC);
1108 return RX_HANDLER_CONSUMED;
1112 slave = bond_slave_get_rcu(skb->dev);
1115 recv_probe = ACCESS_ONCE(bond->recv_probe);
1117 ret = recv_probe(skb, bond, slave);
1118 if (ret == RX_HANDLER_CONSUMED) {
1124 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1125 return RX_HANDLER_EXACT;
1128 skb->dev = bond->dev;
1130 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1131 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1132 skb->pkt_type == PACKET_HOST) {
1134 if (unlikely(skb_cow_head(skb,
1135 skb->data - skb_mac_header(skb)))) {
1137 return RX_HANDLER_CONSUMED;
1139 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1145 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1146 struct net_device *slave_dev,
1147 struct slave *slave)
1151 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1154 slave_dev->flags |= IFF_SLAVE;
1155 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1159 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1160 struct net_device *slave_dev)
1162 netdev_upper_dev_unlink(slave_dev, bond_dev);
1163 slave_dev->flags &= ~IFF_SLAVE;
1164 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1167 static struct slave *bond_alloc_slave(struct bonding *bond)
1169 struct slave *slave = NULL;
1171 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1175 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1176 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1178 if (!SLAVE_AD_INFO(slave)) {
1186 static void bond_free_slave(struct slave *slave)
1188 struct bonding *bond = bond_get_bond_by_slave(slave);
1190 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1191 kfree(SLAVE_AD_INFO(slave));
1196 /* enslave device <slave> to bond device <master> */
1197 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1199 struct bonding *bond = netdev_priv(bond_dev);
1200 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1201 struct slave *new_slave = NULL, *prev_slave;
1202 struct sockaddr addr;
1206 if (!bond->params.use_carrier &&
1207 slave_dev->ethtool_ops->get_link == NULL &&
1208 slave_ops->ndo_do_ioctl == NULL) {
1209 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1213 /* already enslaved */
1214 if (slave_dev->flags & IFF_SLAVE) {
1215 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1219 if (bond_dev == slave_dev) {
1220 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1224 /* vlan challenged mutual exclusion */
1225 /* no need to lock since we're protected by rtnl_lock */
1226 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1227 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1229 if (vlan_uses_dev(bond_dev)) {
1230 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1231 slave_dev->name, bond_dev->name);
1234 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1235 slave_dev->name, slave_dev->name,
1239 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1244 * Old ifenslave binaries are no longer supported. These can
1245 * be identified with moderate accuracy by the state of the slave:
1246 * the current ifenslave will set the interface down prior to
1247 * enslaving it; the old ifenslave will not.
1249 if ((slave_dev->flags & IFF_UP)) {
1250 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1253 goto err_undo_flags;
1256 /* set bonding device ether type by slave - bonding netdevices are
1257 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1258 * there is a need to override some of the type dependent attribs/funcs.
1260 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1261 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1263 if (!bond_has_slaves(bond)) {
1264 if (bond_dev->type != slave_dev->type) {
1265 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1266 bond_dev->type, slave_dev->type);
1268 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1270 res = notifier_to_errno(res);
1272 netdev_err(bond_dev, "refused to change device type\n");
1274 goto err_undo_flags;
1277 /* Flush unicast and multicast addresses */
1278 dev_uc_flush(bond_dev);
1279 dev_mc_flush(bond_dev);
1281 if (slave_dev->type != ARPHRD_ETHER)
1282 bond_setup_by_slave(bond_dev, slave_dev);
1284 ether_setup(bond_dev);
1285 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1288 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1291 } else if (bond_dev->type != slave_dev->type) {
1292 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1293 slave_dev->name, slave_dev->type, bond_dev->type);
1295 goto err_undo_flags;
1298 if (slave_ops->ndo_set_mac_address == NULL) {
1299 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1300 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1301 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1302 if (!bond_has_slaves(bond)) {
1303 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1304 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1306 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");
1308 goto err_undo_flags;
1313 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1315 /* If this is the first slave, then we need to set the master's hardware
1316 * address to be the same as the slave's. */
1317 if (!bond_has_slaves(bond) &&
1318 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1319 bond_set_dev_addr(bond->dev, slave_dev);
1321 new_slave = bond_alloc_slave(bond);
1324 goto err_undo_flags;
1327 new_slave->bond = bond;
1328 new_slave->dev = slave_dev;
1330 * Set the new_slave's queue_id to be zero. Queue ID mapping
1331 * is set via sysfs or module option if desired.
1333 new_slave->queue_id = 0;
1335 /* Save slave's original mtu and then set it to match the bond */
1336 new_slave->original_mtu = slave_dev->mtu;
1337 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1339 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1344 * Save slave's original ("permanent") mac address for modes
1345 * that need it, and for restoring it upon release, and then
1346 * set it to the master's address
1348 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1350 if (!bond->params.fail_over_mac ||
1351 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1353 * Set slave to master's mac address. The application already
1354 * set the master's mac address to that of the first slave
1356 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1357 addr.sa_family = slave_dev->type;
1358 res = dev_set_mac_address(slave_dev, &addr);
1360 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1361 goto err_restore_mtu;
1365 /* open the slave since the application closed it */
1366 res = dev_open(slave_dev);
1368 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1369 goto err_restore_mac;
1372 slave_dev->priv_flags |= IFF_BONDING;
1374 if (bond_is_lb(bond)) {
1375 /* bond_alb_init_slave() must be called before all other stages since
1376 * it might fail and we do not want to have to undo everything
1378 res = bond_alb_init_slave(bond, new_slave);
1383 /* If the mode uses primary, then the following is handled by
1384 * bond_change_active_slave().
1386 if (!bond_uses_primary(bond)) {
1387 /* set promiscuity level to new slave */
1388 if (bond_dev->flags & IFF_PROMISC) {
1389 res = dev_set_promiscuity(slave_dev, 1);
1394 /* set allmulti level to new slave */
1395 if (bond_dev->flags & IFF_ALLMULTI) {
1396 res = dev_set_allmulti(slave_dev, 1);
1401 netif_addr_lock_bh(bond_dev);
1403 dev_mc_sync_multiple(slave_dev, bond_dev);
1404 dev_uc_sync_multiple(slave_dev, bond_dev);
1406 netif_addr_unlock_bh(bond_dev);
1409 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1410 /* add lacpdu mc addr to mc list */
1411 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1413 dev_mc_add(slave_dev, lacpdu_multicast);
1416 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1418 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1423 prev_slave = bond_last_slave(bond);
1425 new_slave->delay = 0;
1426 new_slave->link_failure_count = 0;
1428 bond_update_speed_duplex(new_slave);
1430 new_slave->last_rx = jiffies -
1431 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1432 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1433 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1435 if (bond->params.miimon && !bond->params.use_carrier) {
1436 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1438 if ((link_reporting == -1) && !bond->params.arp_interval) {
1440 * miimon is set but a bonded network driver
1441 * does not support ETHTOOL/MII and
1442 * arp_interval is not set. Note: if
1443 * use_carrier is enabled, we will never go
1444 * here (because netif_carrier is always
1445 * supported); thus, we don't need to change
1446 * the messages for netif_carrier.
1448 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",
1450 } else if (link_reporting == -1) {
1451 /* unable get link status using mii/ethtool */
1452 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",
1457 /* check for initial state */
1458 if (bond->params.miimon) {
1459 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1460 if (bond->params.updelay) {
1461 new_slave->link = BOND_LINK_BACK;
1462 new_slave->delay = bond->params.updelay;
1464 new_slave->link = BOND_LINK_UP;
1467 new_slave->link = BOND_LINK_DOWN;
1469 } else if (bond->params.arp_interval) {
1470 new_slave->link = (netif_carrier_ok(slave_dev) ?
1471 BOND_LINK_UP : BOND_LINK_DOWN);
1473 new_slave->link = BOND_LINK_UP;
1476 if (new_slave->link != BOND_LINK_DOWN)
1477 new_slave->last_link_up = jiffies;
1478 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1479 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1480 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1482 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1483 /* if there is a primary slave, remember it */
1484 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1485 bond->primary_slave = new_slave;
1486 bond->force_primary = true;
1490 switch (BOND_MODE(bond)) {
1491 case BOND_MODE_ACTIVEBACKUP:
1492 bond_set_slave_inactive_flags(new_slave,
1493 BOND_SLAVE_NOTIFY_NOW);
1495 case BOND_MODE_8023AD:
1496 /* in 802.3ad mode, the internal mechanism
1497 * will activate the slaves in the selected
1500 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1501 /* if this is the first slave */
1503 SLAVE_AD_INFO(new_slave)->id = 1;
1504 /* Initialize AD with the number of times that the AD timer is called in 1 second
1505 * can be called only after the mac address of the bond is set
1507 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1509 SLAVE_AD_INFO(new_slave)->id =
1510 SLAVE_AD_INFO(prev_slave)->id + 1;
1513 bond_3ad_bind_slave(new_slave);
1517 bond_set_active_slave(new_slave);
1518 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1521 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1523 /* always active in trunk mode */
1524 bond_set_active_slave(new_slave);
1526 /* In trunking mode there is little meaning to curr_active_slave
1527 * anyway (it holds no special properties of the bond device),
1528 * so we can change it without calling change_active_interface()
1530 if (!rcu_access_pointer(bond->curr_active_slave) &&
1531 new_slave->link == BOND_LINK_UP)
1532 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1535 } /* switch(bond_mode) */
1537 #ifdef CONFIG_NET_POLL_CONTROLLER
1538 slave_dev->npinfo = bond->dev->npinfo;
1539 if (slave_dev->npinfo) {
1540 if (slave_enable_netpoll(new_slave)) {
1541 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1548 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1551 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1555 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1557 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1558 goto err_unregister;
1561 res = bond_sysfs_slave_add(new_slave);
1563 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1564 goto err_upper_unlink;
1568 bond_compute_features(bond);
1569 bond_set_carrier(bond);
1571 if (bond_uses_primary(bond)) {
1573 write_lock_bh(&bond->curr_slave_lock);
1574 bond_select_active_slave(bond);
1575 write_unlock_bh(&bond->curr_slave_lock);
1576 unblock_netpoll_tx();
1579 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1581 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1582 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1584 /* enslave is successful */
1587 /* Undo stages on error */
1589 bond_upper_dev_unlink(bond_dev, slave_dev);
1592 netdev_rx_handler_unregister(slave_dev);
1595 if (!bond_uses_primary(bond))
1596 bond_hw_addr_flush(bond_dev, slave_dev);
1598 vlan_vids_del_by_dev(slave_dev, bond_dev);
1599 if (bond->primary_slave == new_slave)
1600 bond->primary_slave = NULL;
1601 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1603 write_lock_bh(&bond->curr_slave_lock);
1604 bond_change_active_slave(bond, NULL);
1605 bond_select_active_slave(bond);
1606 write_unlock_bh(&bond->curr_slave_lock);
1607 unblock_netpoll_tx();
1609 slave_disable_netpoll(new_slave);
1612 slave_dev->priv_flags &= ~IFF_BONDING;
1613 dev_close(slave_dev);
1616 if (!bond->params.fail_over_mac ||
1617 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1618 /* XXX TODO - fom follow mode needs to change master's
1619 * MAC if this slave's MAC is in use by the bond, or at
1620 * least print a warning.
1622 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1623 addr.sa_family = slave_dev->type;
1624 dev_set_mac_address(slave_dev, &addr);
1628 dev_set_mtu(slave_dev, new_slave->original_mtu);
1631 bond_free_slave(new_slave);
1634 /* Enslave of first slave has failed and we need to fix master's mac */
1635 if (!bond_has_slaves(bond) &&
1636 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1637 eth_hw_addr_random(bond_dev);
1643 * Try to release the slave device <slave> from the bond device <master>
1644 * It is legal to access curr_active_slave without a lock because all the function
1645 * is write-locked. If "all" is true it means that the function is being called
1646 * while destroying a bond interface and all slaves are being released.
1648 * The rules for slave state should be:
1649 * for Active/Backup:
1650 * Active stays on all backups go down
1651 * for Bonded connections:
1652 * The first up interface should be left on and all others downed.
1654 static int __bond_release_one(struct net_device *bond_dev,
1655 struct net_device *slave_dev,
1658 struct bonding *bond = netdev_priv(bond_dev);
1659 struct slave *slave, *oldcurrent;
1660 struct sockaddr addr;
1661 int old_flags = bond_dev->flags;
1662 netdev_features_t old_features = bond_dev->features;
1664 /* slave is not a slave or master is not master of this slave */
1665 if (!(slave_dev->flags & IFF_SLAVE) ||
1666 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1667 netdev_err(bond_dev, "cannot release %s\n",
1674 slave = bond_get_slave_by_dev(bond, slave_dev);
1676 /* not a slave of this bond */
1677 netdev_info(bond_dev, "%s not enslaved\n",
1679 unblock_netpoll_tx();
1683 bond_sysfs_slave_del(slave);
1685 bond_upper_dev_unlink(bond_dev, slave_dev);
1686 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1687 * for this slave anymore.
1689 netdev_rx_handler_unregister(slave_dev);
1690 write_lock_bh(&bond->lock);
1692 /* Inform AD package of unbinding of slave. */
1693 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1694 bond_3ad_unbind_slave(slave);
1696 write_unlock_bh(&bond->lock);
1698 netdev_info(bond_dev, "Releasing %s interface %s\n",
1699 bond_is_active_slave(slave) ? "active" : "backup",
1702 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1704 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1706 if (!all && (!bond->params.fail_over_mac ||
1707 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1708 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1709 bond_has_slaves(bond))
1710 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",
1711 slave_dev->name, slave->perm_hwaddr,
1712 bond_dev->name, slave_dev->name);
1715 if (bond->primary_slave == slave)
1716 bond->primary_slave = NULL;
1718 if (oldcurrent == slave) {
1719 write_lock_bh(&bond->curr_slave_lock);
1720 bond_change_active_slave(bond, NULL);
1721 write_unlock_bh(&bond->curr_slave_lock);
1724 if (bond_is_lb(bond)) {
1725 /* Must be called only after the slave has been
1726 * detached from the list and the curr_active_slave
1727 * has been cleared (if our_slave == old_current),
1728 * but before a new active slave is selected.
1730 bond_alb_deinit_slave(bond, slave);
1734 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1735 } else if (oldcurrent == slave) {
1737 * Note that we hold RTNL over this sequence, so there
1738 * is no concern that another slave add/remove event
1741 write_lock_bh(&bond->curr_slave_lock);
1743 bond_select_active_slave(bond);
1745 write_unlock_bh(&bond->curr_slave_lock);
1748 if (!bond_has_slaves(bond)) {
1749 bond_set_carrier(bond);
1750 eth_hw_addr_random(bond_dev);
1753 unblock_netpoll_tx();
1757 if (!bond_has_slaves(bond)) {
1758 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1759 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1762 bond_compute_features(bond);
1763 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1764 (old_features & NETIF_F_VLAN_CHALLENGED))
1765 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1766 slave_dev->name, bond_dev->name);
1768 /* must do this from outside any spinlocks */
1769 vlan_vids_del_by_dev(slave_dev, bond_dev);
1771 /* If the mode uses primary, then this cases was handled above by
1772 * bond_change_active_slave(..., NULL)
1774 if (!bond_uses_primary(bond)) {
1775 /* unset promiscuity level from slave
1776 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1777 * of the IFF_PROMISC flag in the bond_dev, but we need the
1778 * value of that flag before that change, as that was the value
1779 * when this slave was attached, so we cache at the start of the
1780 * function and use it here. Same goes for ALLMULTI below
1782 if (old_flags & IFF_PROMISC)
1783 dev_set_promiscuity(slave_dev, -1);
1785 /* unset allmulti level from slave */
1786 if (old_flags & IFF_ALLMULTI)
1787 dev_set_allmulti(slave_dev, -1);
1789 bond_hw_addr_flush(bond_dev, slave_dev);
1792 slave_disable_netpoll(slave);
1794 /* close slave before restoring its mac address */
1795 dev_close(slave_dev);
1797 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1798 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1799 /* restore original ("permanent") mac address */
1800 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1801 addr.sa_family = slave_dev->type;
1802 dev_set_mac_address(slave_dev, &addr);
1805 dev_set_mtu(slave_dev, slave->original_mtu);
1807 slave_dev->priv_flags &= ~IFF_BONDING;
1809 bond_free_slave(slave);
1811 return 0; /* deletion OK */
1814 /* A wrapper used because of ndo_del_link */
1815 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1817 return __bond_release_one(bond_dev, slave_dev, false);
1821 * First release a slave and then destroy the bond if no more slaves are left.
1822 * Must be under rtnl_lock when this function is called.
1824 static int bond_release_and_destroy(struct net_device *bond_dev,
1825 struct net_device *slave_dev)
1827 struct bonding *bond = netdev_priv(bond_dev);
1830 ret = bond_release(bond_dev, slave_dev);
1831 if (ret == 0 && !bond_has_slaves(bond)) {
1832 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1833 netdev_info(bond_dev, "Destroying bond %s\n",
1835 unregister_netdevice(bond_dev);
1840 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1842 struct bonding *bond = netdev_priv(bond_dev);
1844 info->bond_mode = BOND_MODE(bond);
1845 info->miimon = bond->params.miimon;
1847 info->num_slaves = bond->slave_cnt;
1852 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1854 struct bonding *bond = netdev_priv(bond_dev);
1855 struct list_head *iter;
1856 int i = 0, res = -ENODEV;
1857 struct slave *slave;
1859 bond_for_each_slave(bond, slave, iter) {
1860 if (i++ == (int)info->slave_id) {
1862 strcpy(info->slave_name, slave->dev->name);
1863 info->link = slave->link;
1864 info->state = bond_slave_state(slave);
1865 info->link_failure_count = slave->link_failure_count;
1873 /*-------------------------------- Monitoring -------------------------------*/
1875 /* called with rcu_read_lock() */
1876 static int bond_miimon_inspect(struct bonding *bond)
1878 int link_state, commit = 0;
1879 struct list_head *iter;
1880 struct slave *slave;
1881 bool ignore_updelay;
1883 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1885 bond_for_each_slave_rcu(bond, slave, iter) {
1886 slave->new_link = BOND_LINK_NOCHANGE;
1888 link_state = bond_check_dev_link(bond, slave->dev, 0);
1890 switch (slave->link) {
1895 slave->link = BOND_LINK_FAIL;
1896 slave->delay = bond->params.downdelay;
1898 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1900 BOND_MODE_ACTIVEBACKUP) ?
1901 (bond_is_active_slave(slave) ?
1902 "active " : "backup ") : "",
1904 bond->params.downdelay * bond->params.miimon);
1907 case BOND_LINK_FAIL:
1910 * recovered before downdelay expired
1912 slave->link = BOND_LINK_UP;
1913 slave->last_link_up = jiffies;
1914 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1915 (bond->params.downdelay - slave->delay) *
1916 bond->params.miimon,
1921 if (slave->delay <= 0) {
1922 slave->new_link = BOND_LINK_DOWN;
1930 case BOND_LINK_DOWN:
1934 slave->link = BOND_LINK_BACK;
1935 slave->delay = bond->params.updelay;
1938 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
1940 ignore_updelay ? 0 :
1941 bond->params.updelay *
1942 bond->params.miimon);
1945 case BOND_LINK_BACK:
1947 slave->link = BOND_LINK_DOWN;
1948 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
1949 (bond->params.updelay - slave->delay) *
1950 bond->params.miimon,
1959 if (slave->delay <= 0) {
1960 slave->new_link = BOND_LINK_UP;
1962 ignore_updelay = false;
1974 static void bond_miimon_commit(struct bonding *bond)
1976 struct list_head *iter;
1977 struct slave *slave;
1979 bond_for_each_slave(bond, slave, iter) {
1980 switch (slave->new_link) {
1981 case BOND_LINK_NOCHANGE:
1985 slave->link = BOND_LINK_UP;
1986 slave->last_link_up = jiffies;
1988 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1989 /* prevent it from being the active one */
1990 bond_set_backup_slave(slave);
1991 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1992 /* make it immediately active */
1993 bond_set_active_slave(slave);
1994 } else if (slave != bond->primary_slave) {
1995 /* prevent it from being the active one */
1996 bond_set_backup_slave(slave);
1999 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2001 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2002 slave->duplex ? "full" : "half");
2004 /* notify ad that the link status has changed */
2005 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2006 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2008 if (bond_is_lb(bond))
2009 bond_alb_handle_link_change(bond, slave,
2012 if (!bond->curr_active_slave ||
2013 (slave == bond->primary_slave))
2018 case BOND_LINK_DOWN:
2019 if (slave->link_failure_count < UINT_MAX)
2020 slave->link_failure_count++;
2022 slave->link = BOND_LINK_DOWN;
2024 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2025 BOND_MODE(bond) == BOND_MODE_8023AD)
2026 bond_set_slave_inactive_flags(slave,
2027 BOND_SLAVE_NOTIFY_NOW);
2029 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2032 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2033 bond_3ad_handle_link_change(slave,
2036 if (bond_is_lb(bond))
2037 bond_alb_handle_link_change(bond, slave,
2040 if (slave == rcu_access_pointer(bond->curr_active_slave))
2046 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2047 slave->new_link, slave->dev->name);
2048 slave->new_link = BOND_LINK_NOCHANGE;
2056 write_lock_bh(&bond->curr_slave_lock);
2057 bond_select_active_slave(bond);
2058 write_unlock_bh(&bond->curr_slave_lock);
2059 unblock_netpoll_tx();
2062 bond_set_carrier(bond);
2068 * Really a wrapper that splits the mii monitor into two phases: an
2069 * inspection, then (if inspection indicates something needs to be done)
2070 * an acquisition of appropriate locks followed by a commit phase to
2071 * implement whatever link state changes are indicated.
2073 static void bond_mii_monitor(struct work_struct *work)
2075 struct bonding *bond = container_of(work, struct bonding,
2077 bool should_notify_peers = false;
2078 unsigned long delay;
2080 delay = msecs_to_jiffies(bond->params.miimon);
2082 if (!bond_has_slaves(bond))
2087 should_notify_peers = bond_should_notify_peers(bond);
2089 if (bond_miimon_inspect(bond)) {
2092 /* Race avoidance with bond_close cancel of workqueue */
2093 if (!rtnl_trylock()) {
2095 should_notify_peers = false;
2099 bond_miimon_commit(bond);
2101 rtnl_unlock(); /* might sleep, hold no other locks */
2106 if (bond->params.miimon)
2107 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2109 if (should_notify_peers) {
2110 if (!rtnl_trylock())
2112 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2117 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2119 struct net_device *upper;
2120 struct list_head *iter;
2123 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2127 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2128 if (ip == bond_confirm_addr(upper, 0, ip)) {
2139 * We go to the (large) trouble of VLAN tagging ARP frames because
2140 * switches in VLAN mode (especially if ports are configured as
2141 * "native" to a VLAN) might not pass non-tagged frames.
2143 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2144 __be32 dest_ip, __be32 src_ip,
2145 struct bond_vlan_tag *tags)
2147 struct sk_buff *skb;
2148 struct bond_vlan_tag *outer_tag = tags;
2150 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2151 arp_op, slave_dev->name, &dest_ip, &src_ip);
2153 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2154 NULL, slave_dev->dev_addr, NULL);
2157 net_err_ratelimited("ARP packet allocation failed\n");
2161 if (!tags || tags->vlan_proto == VLAN_N_VID)
2166 /* Go through all the tags backwards and add them to the packet */
2167 while (tags->vlan_proto != VLAN_N_VID) {
2168 if (!tags->vlan_id) {
2173 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2174 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2175 skb = __vlan_put_tag(skb, tags->vlan_proto,
2178 net_err_ratelimited("failed to insert inner VLAN tag\n");
2184 /* Set the outer tag */
2185 if (outer_tag->vlan_id) {
2186 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2187 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2188 skb = vlan_put_tag(skb, outer_tag->vlan_proto,
2189 outer_tag->vlan_id);
2191 net_err_ratelimited("failed to insert outer VLAN tag\n");
2200 /* Validate the device path between the @start_dev and the @end_dev.
2201 * The path is valid if the @end_dev is reachable through device
2203 * When the path is validated, collect any vlan information in the
2206 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2207 struct net_device *end_dev,
2210 struct bond_vlan_tag *tags;
2211 struct net_device *upper;
2212 struct list_head *iter;
2214 if (start_dev == end_dev) {
2215 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2217 return ERR_PTR(-ENOMEM);
2218 tags[level].vlan_proto = VLAN_N_VID;
2222 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2223 tags = bond_verify_device_path(upper, end_dev, level + 1);
2224 if (IS_ERR_OR_NULL(tags)) {
2229 if (is_vlan_dev(upper)) {
2230 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2231 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2240 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2243 struct bond_vlan_tag *tags;
2244 __be32 *targets = bond->params.arp_targets, addr;
2247 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2248 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2251 /* Find out through which dev should the packet go */
2252 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2255 /* there's no route to target - try to send arp
2256 * probe to generate any traffic (arp_validate=0)
2258 if (bond->params.arp_validate)
2259 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2262 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2267 /* bond device itself */
2268 if (rt->dst.dev == bond->dev)
2272 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2275 if (!IS_ERR_OR_NULL(tags))
2278 /* Not our device - skip */
2279 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2280 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2286 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2288 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2294 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2298 if (!sip || !bond_has_this_ip(bond, tip)) {
2299 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2304 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2306 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2310 slave->last_rx = jiffies;
2311 slave->target_last_arp_rx[i] = jiffies;
2314 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2315 struct slave *slave)
2317 struct arphdr *arp = (struct arphdr *)skb->data;
2318 struct slave *curr_active_slave;
2319 unsigned char *arp_ptr;
2321 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2323 if (!slave_do_arp_validate(bond, slave)) {
2324 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2325 !slave_do_arp_validate_only(bond))
2326 slave->last_rx = jiffies;
2327 return RX_HANDLER_ANOTHER;
2328 } else if (!is_arp) {
2329 return RX_HANDLER_ANOTHER;
2332 alen = arp_hdr_len(bond->dev);
2334 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2337 if (alen > skb_headlen(skb)) {
2338 arp = kmalloc(alen, GFP_ATOMIC);
2341 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2345 if (arp->ar_hln != bond->dev->addr_len ||
2346 skb->pkt_type == PACKET_OTHERHOST ||
2347 skb->pkt_type == PACKET_LOOPBACK ||
2348 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2349 arp->ar_pro != htons(ETH_P_IP) ||
2353 arp_ptr = (unsigned char *)(arp + 1);
2354 arp_ptr += bond->dev->addr_len;
2355 memcpy(&sip, arp_ptr, 4);
2356 arp_ptr += 4 + bond->dev->addr_len;
2357 memcpy(&tip, arp_ptr, 4);
2359 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2360 slave->dev->name, bond_slave_state(slave),
2361 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2364 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2367 * Backup slaves won't see the ARP reply, but do come through
2368 * here for each ARP probe (so we swap the sip/tip to validate
2369 * the probe). In a "redundant switch, common router" type of
2370 * configuration, the ARP probe will (hopefully) travel from
2371 * the active, through one switch, the router, then the other
2372 * switch before reaching the backup.
2374 * We 'trust' the arp requests if there is an active slave and
2375 * it received valid arp reply(s) after it became active. This
2376 * is done to avoid endless looping when we can't reach the
2377 * arp_ip_target and fool ourselves with our own arp requests.
2380 if (bond_is_active_slave(slave))
2381 bond_validate_arp(bond, slave, sip, tip);
2382 else if (curr_active_slave &&
2383 time_after(slave_last_rx(bond, curr_active_slave),
2384 curr_active_slave->last_link_up))
2385 bond_validate_arp(bond, slave, tip, sip);
2388 if (arp != (struct arphdr *)skb->data)
2390 return RX_HANDLER_ANOTHER;
2393 /* function to verify if we're in the arp_interval timeslice, returns true if
2394 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2395 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2397 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2400 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2402 return time_in_range(jiffies,
2403 last_act - delta_in_ticks,
2404 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2408 * this function is called regularly to monitor each slave's link
2409 * ensuring that traffic is being sent and received when arp monitoring
2410 * is used in load-balancing mode. if the adapter has been dormant, then an
2411 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2412 * arp monitoring in active backup mode.
2414 static void bond_loadbalance_arp_mon(struct work_struct *work)
2416 struct bonding *bond = container_of(work, struct bonding,
2418 struct slave *slave, *oldcurrent;
2419 struct list_head *iter;
2420 int do_failover = 0, slave_state_changed = 0;
2422 if (!bond_has_slaves(bond))
2427 oldcurrent = rcu_dereference(bond->curr_active_slave);
2428 /* see if any of the previous devices are up now (i.e. they have
2429 * xmt and rcv traffic). the curr_active_slave does not come into
2430 * the picture unless it is null. also, slave->last_link_up is not
2431 * needed here because we send an arp on each slave and give a slave
2432 * as long as it needs to get the tx/rx within the delta.
2433 * TODO: what about up/down delay in arp mode? it wasn't here before
2436 bond_for_each_slave_rcu(bond, slave, iter) {
2437 unsigned long trans_start = dev_trans_start(slave->dev);
2439 if (slave->link != BOND_LINK_UP) {
2440 if (bond_time_in_interval(bond, trans_start, 1) &&
2441 bond_time_in_interval(bond, slave->last_rx, 1)) {
2443 slave->link = BOND_LINK_UP;
2444 slave_state_changed = 1;
2446 /* primary_slave has no meaning in round-robin
2447 * mode. the window of a slave being up and
2448 * curr_active_slave being null after enslaving
2452 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2456 netdev_info(bond->dev, "interface %s is now up\n",
2461 /* slave->link == BOND_LINK_UP */
2463 /* not all switches will respond to an arp request
2464 * when the source ip is 0, so don't take the link down
2465 * if we don't know our ip yet
2467 if (!bond_time_in_interval(bond, trans_start, 2) ||
2468 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2470 slave->link = BOND_LINK_DOWN;
2471 slave_state_changed = 1;
2473 if (slave->link_failure_count < UINT_MAX)
2474 slave->link_failure_count++;
2476 netdev_info(bond->dev, "interface %s is now down\n",
2479 if (slave == oldcurrent)
2484 /* note: if switch is in round-robin mode, all links
2485 * must tx arp to ensure all links rx an arp - otherwise
2486 * links may oscillate or not come up at all; if switch is
2487 * in something like xor mode, there is nothing we can
2488 * do - all replies will be rx'ed on same link causing slaves
2489 * to be unstable during low/no traffic periods
2491 if (bond_slave_is_up(slave))
2492 bond_arp_send_all(bond, slave);
2497 if (do_failover || slave_state_changed) {
2498 if (!rtnl_trylock())
2501 if (slave_state_changed) {
2502 bond_slave_state_change(bond);
2503 } else if (do_failover) {
2504 /* the bond_select_active_slave must hold RTNL
2505 * and curr_slave_lock for write.
2508 write_lock_bh(&bond->curr_slave_lock);
2510 bond_select_active_slave(bond);
2512 write_unlock_bh(&bond->curr_slave_lock);
2513 unblock_netpoll_tx();
2519 if (bond->params.arp_interval)
2520 queue_delayed_work(bond->wq, &bond->arp_work,
2521 msecs_to_jiffies(bond->params.arp_interval));
2525 * Called to inspect slaves for active-backup mode ARP monitor link state
2526 * changes. Sets new_link in slaves to specify what action should take
2527 * place for the slave. Returns 0 if no changes are found, >0 if changes
2528 * to link states must be committed.
2530 * Called with rcu_read_lock hold.
2532 static int bond_ab_arp_inspect(struct bonding *bond)
2534 unsigned long trans_start, last_rx;
2535 struct list_head *iter;
2536 struct slave *slave;
2539 bond_for_each_slave_rcu(bond, slave, iter) {
2540 slave->new_link = BOND_LINK_NOCHANGE;
2541 last_rx = slave_last_rx(bond, slave);
2543 if (slave->link != BOND_LINK_UP) {
2544 if (bond_time_in_interval(bond, last_rx, 1)) {
2545 slave->new_link = BOND_LINK_UP;
2552 * Give slaves 2*delta after being enslaved or made
2553 * active. This avoids bouncing, as the last receive
2554 * times need a full ARP monitor cycle to be updated.
2556 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2560 * Backup slave is down if:
2561 * - No current_arp_slave AND
2562 * - more than 3*delta since last receive AND
2563 * - the bond has an IP address
2565 * Note: a non-null current_arp_slave indicates
2566 * the curr_active_slave went down and we are
2567 * searching for a new one; under this condition
2568 * we only take the curr_active_slave down - this
2569 * gives each slave a chance to tx/rx traffic
2570 * before being taken out
2572 if (!bond_is_active_slave(slave) &&
2573 !rcu_access_pointer(bond->current_arp_slave) &&
2574 !bond_time_in_interval(bond, last_rx, 3)) {
2575 slave->new_link = BOND_LINK_DOWN;
2580 * Active slave is down if:
2581 * - more than 2*delta since transmitting OR
2582 * - (more than 2*delta since receive AND
2583 * the bond has an IP address)
2585 trans_start = dev_trans_start(slave->dev);
2586 if (bond_is_active_slave(slave) &&
2587 (!bond_time_in_interval(bond, trans_start, 2) ||
2588 !bond_time_in_interval(bond, last_rx, 2))) {
2589 slave->new_link = BOND_LINK_DOWN;
2598 * Called to commit link state changes noted by inspection step of
2599 * active-backup mode ARP monitor.
2601 * Called with RTNL hold.
2603 static void bond_ab_arp_commit(struct bonding *bond)
2605 unsigned long trans_start;
2606 struct list_head *iter;
2607 struct slave *slave;
2609 bond_for_each_slave(bond, slave, iter) {
2610 switch (slave->new_link) {
2611 case BOND_LINK_NOCHANGE:
2615 trans_start = dev_trans_start(slave->dev);
2616 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2617 (!rtnl_dereference(bond->curr_active_slave) &&
2618 bond_time_in_interval(bond, trans_start, 1))) {
2619 struct slave *current_arp_slave;
2621 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2622 slave->link = BOND_LINK_UP;
2623 if (current_arp_slave) {
2624 bond_set_slave_inactive_flags(
2626 BOND_SLAVE_NOTIFY_NOW);
2627 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2630 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2633 if (!rtnl_dereference(bond->curr_active_slave) ||
2634 (slave == bond->primary_slave))
2641 case BOND_LINK_DOWN:
2642 if (slave->link_failure_count < UINT_MAX)
2643 slave->link_failure_count++;
2645 slave->link = BOND_LINK_DOWN;
2646 bond_set_slave_inactive_flags(slave,
2647 BOND_SLAVE_NOTIFY_NOW);
2649 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2652 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2653 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2660 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2661 slave->new_link, slave->dev->name);
2668 write_lock_bh(&bond->curr_slave_lock);
2669 bond_select_active_slave(bond);
2670 write_unlock_bh(&bond->curr_slave_lock);
2671 unblock_netpoll_tx();
2674 bond_set_carrier(bond);
2678 * Send ARP probes for active-backup mode ARP monitor.
2680 * Called with rcu_read_lock hold.
2682 static bool bond_ab_arp_probe(struct bonding *bond)
2684 struct slave *slave, *before = NULL, *new_slave = NULL,
2685 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2686 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2687 struct list_head *iter;
2689 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2691 if (curr_arp_slave && curr_active_slave)
2692 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2693 curr_arp_slave->dev->name,
2694 curr_active_slave->dev->name);
2696 if (curr_active_slave) {
2697 bond_arp_send_all(bond, curr_active_slave);
2698 return should_notify_rtnl;
2701 /* if we don't have a curr_active_slave, search for the next available
2702 * backup slave from the current_arp_slave and make it the candidate
2703 * for becoming the curr_active_slave
2706 if (!curr_arp_slave) {
2707 curr_arp_slave = bond_first_slave_rcu(bond);
2708 if (!curr_arp_slave)
2709 return should_notify_rtnl;
2712 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2714 bond_for_each_slave_rcu(bond, slave, iter) {
2715 if (!found && !before && bond_slave_is_up(slave))
2718 if (found && !new_slave && bond_slave_is_up(slave))
2720 /* if the link state is up at this point, we
2721 * mark it down - this can happen if we have
2722 * simultaneous link failures and
2723 * reselect_active_interface doesn't make this
2724 * one the current slave so it is still marked
2725 * up when it is actually down
2727 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2728 slave->link = BOND_LINK_DOWN;
2729 if (slave->link_failure_count < UINT_MAX)
2730 slave->link_failure_count++;
2732 bond_set_slave_inactive_flags(slave,
2733 BOND_SLAVE_NOTIFY_LATER);
2735 netdev_info(bond->dev, "backup interface %s is now down\n",
2738 if (slave == curr_arp_slave)
2742 if (!new_slave && before)
2748 new_slave->link = BOND_LINK_BACK;
2749 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2750 bond_arp_send_all(bond, new_slave);
2751 new_slave->last_link_up = jiffies;
2752 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2755 bond_for_each_slave_rcu(bond, slave, iter) {
2756 if (slave->should_notify) {
2757 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2761 return should_notify_rtnl;
2764 static void bond_activebackup_arp_mon(struct work_struct *work)
2766 struct bonding *bond = container_of(work, struct bonding,
2768 bool should_notify_peers = false;
2769 bool should_notify_rtnl = false;
2772 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2774 if (!bond_has_slaves(bond))
2779 should_notify_peers = bond_should_notify_peers(bond);
2781 if (bond_ab_arp_inspect(bond)) {
2784 /* Race avoidance with bond_close flush of workqueue */
2785 if (!rtnl_trylock()) {
2787 should_notify_peers = false;
2791 bond_ab_arp_commit(bond);
2797 should_notify_rtnl = bond_ab_arp_probe(bond);
2801 if (bond->params.arp_interval)
2802 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2804 if (should_notify_peers || should_notify_rtnl) {
2805 if (!rtnl_trylock())
2808 if (should_notify_peers)
2809 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2811 if (should_notify_rtnl)
2812 bond_slave_state_notify(bond);
2818 /*-------------------------- netdev event handling --------------------------*/
2821 * Change device name
2823 static int bond_event_changename(struct bonding *bond)
2825 bond_remove_proc_entry(bond);
2826 bond_create_proc_entry(bond);
2828 bond_debug_reregister(bond);
2833 static int bond_master_netdev_event(unsigned long event,
2834 struct net_device *bond_dev)
2836 struct bonding *event_bond = netdev_priv(bond_dev);
2839 case NETDEV_CHANGENAME:
2840 return bond_event_changename(event_bond);
2841 case NETDEV_UNREGISTER:
2842 bond_remove_proc_entry(event_bond);
2844 case NETDEV_REGISTER:
2845 bond_create_proc_entry(event_bond);
2847 case NETDEV_NOTIFY_PEERS:
2848 if (event_bond->send_peer_notif)
2849 event_bond->send_peer_notif--;
2858 static int bond_slave_netdev_event(unsigned long event,
2859 struct net_device *slave_dev)
2861 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2862 struct bonding *bond;
2863 struct net_device *bond_dev;
2867 /* A netdev event can be generated while enslaving a device
2868 * before netdev_rx_handler_register is called in which case
2869 * slave will be NULL
2873 bond_dev = slave->bond->dev;
2877 case NETDEV_UNREGISTER:
2878 if (bond_dev->type != ARPHRD_ETHER)
2879 bond_release_and_destroy(bond_dev, slave_dev);
2881 bond_release(bond_dev, slave_dev);
2885 old_speed = slave->speed;
2886 old_duplex = slave->duplex;
2888 bond_update_speed_duplex(slave);
2890 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2891 if (old_speed != slave->speed)
2892 bond_3ad_adapter_speed_changed(slave);
2893 if (old_duplex != slave->duplex)
2894 bond_3ad_adapter_duplex_changed(slave);
2899 * ... Or is it this?
2902 case NETDEV_CHANGEMTU:
2904 * TODO: Should slaves be allowed to
2905 * independently alter their MTU? For
2906 * an active-backup bond, slaves need
2907 * not be the same type of device, so
2908 * MTUs may vary. For other modes,
2909 * slaves arguably should have the
2910 * same MTUs. To do this, we'd need to
2911 * take over the slave's change_mtu
2912 * function for the duration of their
2916 case NETDEV_CHANGENAME:
2917 /* we don't care if we don't have primary set */
2918 if (!bond_uses_primary(bond) ||
2919 !bond->params.primary[0])
2922 if (slave == bond->primary_slave) {
2923 /* slave's name changed - he's no longer primary */
2924 bond->primary_slave = NULL;
2925 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2926 /* we have a new primary slave */
2927 bond->primary_slave = slave;
2928 } else { /* we didn't change primary - exit */
2932 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2933 bond->primary_slave ? slave_dev->name : "none");
2936 write_lock_bh(&bond->curr_slave_lock);
2937 bond_select_active_slave(bond);
2938 write_unlock_bh(&bond->curr_slave_lock);
2939 unblock_netpoll_tx();
2941 case NETDEV_FEAT_CHANGE:
2942 bond_compute_features(bond);
2944 case NETDEV_RESEND_IGMP:
2945 /* Propagate to master device */
2946 call_netdevice_notifiers(event, slave->bond->dev);
2956 * bond_netdev_event: handle netdev notifier chain events.
2958 * This function receives events for the netdev chain. The caller (an
2959 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2960 * locks for us to safely manipulate the slave devices (RTNL lock,
2963 static int bond_netdev_event(struct notifier_block *this,
2964 unsigned long event, void *ptr)
2966 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2968 netdev_dbg(event_dev, "event: %lx\n", event);
2970 if (!(event_dev->priv_flags & IFF_BONDING))
2973 if (event_dev->flags & IFF_MASTER) {
2974 netdev_dbg(event_dev, "IFF_MASTER\n");
2975 return bond_master_netdev_event(event, event_dev);
2978 if (event_dev->flags & IFF_SLAVE) {
2979 netdev_dbg(event_dev, "IFF_SLAVE\n");
2980 return bond_slave_netdev_event(event, event_dev);
2986 static struct notifier_block bond_netdev_notifier = {
2987 .notifier_call = bond_netdev_event,
2990 /*---------------------------- Hashing Policies -----------------------------*/
2992 /* L2 hash helper */
2993 static inline u32 bond_eth_hash(struct sk_buff *skb)
2995 struct ethhdr *ep, hdr_tmp;
2997 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
2999 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3003 /* Extract the appropriate headers based on bond's xmit policy */
3004 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3005 struct flow_keys *fk)
3007 const struct ipv6hdr *iph6;
3008 const struct iphdr *iph;
3009 int noff, proto = -1;
3011 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3012 return skb_flow_dissect(skb, fk);
3015 noff = skb_network_offset(skb);
3016 if (skb->protocol == htons(ETH_P_IP)) {
3017 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3020 fk->src = iph->saddr;
3021 fk->dst = iph->daddr;
3022 noff += iph->ihl << 2;
3023 if (!ip_is_fragment(iph))
3024 proto = iph->protocol;
3025 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3026 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3028 iph6 = ipv6_hdr(skb);
3029 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3030 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3031 noff += sizeof(*iph6);
3032 proto = iph6->nexthdr;
3036 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3037 fk->ports = skb_flow_get_ports(skb, noff, proto);
3043 * bond_xmit_hash - generate a hash value based on the xmit policy
3044 * @bond: bonding device
3045 * @skb: buffer to use for headers
3047 * This function will extract the necessary headers from the skb buffer and use
3048 * them to generate a hash based on the xmit_policy set in the bonding device
3050 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3052 struct flow_keys flow;
3055 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3056 !bond_flow_dissect(bond, skb, &flow))
3057 return bond_eth_hash(skb);
3059 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3060 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3061 hash = bond_eth_hash(skb);
3063 hash = (__force u32)flow.ports;
3064 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3065 hash ^= (hash >> 16);
3066 hash ^= (hash >> 8);
3071 /*-------------------------- Device entry points ----------------------------*/
3073 static void bond_work_init_all(struct bonding *bond)
3075 INIT_DELAYED_WORK(&bond->mcast_work,
3076 bond_resend_igmp_join_requests_delayed);
3077 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3078 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3079 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3080 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3082 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3083 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3086 static void bond_work_cancel_all(struct bonding *bond)
3088 cancel_delayed_work_sync(&bond->mii_work);
3089 cancel_delayed_work_sync(&bond->arp_work);
3090 cancel_delayed_work_sync(&bond->alb_work);
3091 cancel_delayed_work_sync(&bond->ad_work);
3092 cancel_delayed_work_sync(&bond->mcast_work);
3095 static int bond_open(struct net_device *bond_dev)
3097 struct bonding *bond = netdev_priv(bond_dev);
3098 struct list_head *iter;
3099 struct slave *slave;
3101 /* reset slave->backup and slave->inactive */
3102 read_lock(&bond->lock);
3103 if (bond_has_slaves(bond)) {
3104 read_lock(&bond->curr_slave_lock);
3105 bond_for_each_slave(bond, slave, iter) {
3106 if (bond_uses_primary(bond) &&
3107 slave != rcu_access_pointer(bond->curr_active_slave)) {
3108 bond_set_slave_inactive_flags(slave,
3109 BOND_SLAVE_NOTIFY_NOW);
3111 bond_set_slave_active_flags(slave,
3112 BOND_SLAVE_NOTIFY_NOW);
3115 read_unlock(&bond->curr_slave_lock);
3117 read_unlock(&bond->lock);
3119 bond_work_init_all(bond);
3121 if (bond_is_lb(bond)) {
3122 /* bond_alb_initialize must be called before the timer
3125 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3127 if (bond->params.tlb_dynamic_lb)
3128 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3131 if (bond->params.miimon) /* link check interval, in milliseconds. */
3132 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3134 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3135 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3136 bond->recv_probe = bond_arp_rcv;
3139 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3140 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3141 /* register to receive LACPDUs */
3142 bond->recv_probe = bond_3ad_lacpdu_recv;
3143 bond_3ad_initiate_agg_selection(bond, 1);
3149 static int bond_close(struct net_device *bond_dev)
3151 struct bonding *bond = netdev_priv(bond_dev);
3153 bond_work_cancel_all(bond);
3154 bond->send_peer_notif = 0;
3155 if (bond_is_lb(bond))
3156 bond_alb_deinitialize(bond);
3157 bond->recv_probe = NULL;
3162 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3163 struct rtnl_link_stats64 *stats)
3165 struct bonding *bond = netdev_priv(bond_dev);
3166 struct rtnl_link_stats64 temp;
3167 struct list_head *iter;
3168 struct slave *slave;
3170 memset(stats, 0, sizeof(*stats));
3172 read_lock_bh(&bond->lock);
3173 bond_for_each_slave(bond, slave, iter) {
3174 const struct rtnl_link_stats64 *sstats =
3175 dev_get_stats(slave->dev, &temp);
3177 stats->rx_packets += sstats->rx_packets;
3178 stats->rx_bytes += sstats->rx_bytes;
3179 stats->rx_errors += sstats->rx_errors;
3180 stats->rx_dropped += sstats->rx_dropped;
3182 stats->tx_packets += sstats->tx_packets;
3183 stats->tx_bytes += sstats->tx_bytes;
3184 stats->tx_errors += sstats->tx_errors;
3185 stats->tx_dropped += sstats->tx_dropped;
3187 stats->multicast += sstats->multicast;
3188 stats->collisions += sstats->collisions;
3190 stats->rx_length_errors += sstats->rx_length_errors;
3191 stats->rx_over_errors += sstats->rx_over_errors;
3192 stats->rx_crc_errors += sstats->rx_crc_errors;
3193 stats->rx_frame_errors += sstats->rx_frame_errors;
3194 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3195 stats->rx_missed_errors += sstats->rx_missed_errors;
3197 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3198 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3199 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3200 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3201 stats->tx_window_errors += sstats->tx_window_errors;
3203 read_unlock_bh(&bond->lock);
3208 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3210 struct bonding *bond = netdev_priv(bond_dev);
3211 struct net_device *slave_dev = NULL;
3212 struct ifbond k_binfo;
3213 struct ifbond __user *u_binfo = NULL;
3214 struct ifslave k_sinfo;
3215 struct ifslave __user *u_sinfo = NULL;
3216 struct mii_ioctl_data *mii = NULL;
3217 struct bond_opt_value newval;
3221 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3233 * We do this again just in case we were called by SIOCGMIIREG
3234 * instead of SIOCGMIIPHY.
3241 if (mii->reg_num == 1) {
3243 read_lock(&bond->lock);
3244 read_lock(&bond->curr_slave_lock);
3245 if (netif_carrier_ok(bond->dev))
3246 mii->val_out = BMSR_LSTATUS;
3248 read_unlock(&bond->curr_slave_lock);
3249 read_unlock(&bond->lock);
3253 case BOND_INFO_QUERY_OLD:
3254 case SIOCBONDINFOQUERY:
3255 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3257 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3260 res = bond_info_query(bond_dev, &k_binfo);
3262 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3266 case BOND_SLAVE_INFO_QUERY_OLD:
3267 case SIOCBONDSLAVEINFOQUERY:
3268 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3270 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3273 res = bond_slave_info_query(bond_dev, &k_sinfo);
3275 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3284 net = dev_net(bond_dev);
3286 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3289 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3291 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3296 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3298 case BOND_ENSLAVE_OLD:
3299 case SIOCBONDENSLAVE:
3300 res = bond_enslave(bond_dev, slave_dev);
3302 case BOND_RELEASE_OLD:
3303 case SIOCBONDRELEASE:
3304 res = bond_release(bond_dev, slave_dev);
3306 case BOND_SETHWADDR_OLD:
3307 case SIOCBONDSETHWADDR:
3308 bond_set_dev_addr(bond_dev, slave_dev);
3311 case BOND_CHANGE_ACTIVE_OLD:
3312 case SIOCBONDCHANGEACTIVE:
3313 bond_opt_initstr(&newval, slave_dev->name);
3314 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3323 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3325 struct bonding *bond = netdev_priv(bond_dev);
3327 if (change & IFF_PROMISC)
3328 bond_set_promiscuity(bond,
3329 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3331 if (change & IFF_ALLMULTI)
3332 bond_set_allmulti(bond,
3333 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3336 static void bond_set_rx_mode(struct net_device *bond_dev)
3338 struct bonding *bond = netdev_priv(bond_dev);
3339 struct list_head *iter;
3340 struct slave *slave;
3344 if (bond_uses_primary(bond)) {
3345 slave = rcu_dereference(bond->curr_active_slave);
3347 dev_uc_sync(slave->dev, bond_dev);
3348 dev_mc_sync(slave->dev, bond_dev);
3351 bond_for_each_slave_rcu(bond, slave, iter) {
3352 dev_uc_sync_multiple(slave->dev, bond_dev);
3353 dev_mc_sync_multiple(slave->dev, bond_dev);
3359 static int bond_neigh_init(struct neighbour *n)
3361 struct bonding *bond = netdev_priv(n->dev);
3362 const struct net_device_ops *slave_ops;
3363 struct neigh_parms parms;
3364 struct slave *slave;
3367 slave = bond_first_slave(bond);
3370 slave_ops = slave->dev->netdev_ops;
3371 if (!slave_ops->ndo_neigh_setup)
3374 parms.neigh_setup = NULL;
3375 parms.neigh_cleanup = NULL;
3376 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3381 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3382 * after the last slave has been detached. Assumes that all slaves
3383 * utilize the same neigh_cleanup (true at this writing as only user
3386 n->parms->neigh_cleanup = parms.neigh_cleanup;
3388 if (!parms.neigh_setup)
3391 return parms.neigh_setup(n);
3395 * The bonding ndo_neigh_setup is called at init time beofre any
3396 * slave exists. So we must declare proxy setup function which will
3397 * be used at run time to resolve the actual slave neigh param setup.
3399 * It's also called by master devices (such as vlans) to setup their
3400 * underlying devices. In that case - do nothing, we're already set up from
3403 static int bond_neigh_setup(struct net_device *dev,
3404 struct neigh_parms *parms)
3406 /* modify only our neigh_parms */
3407 if (parms->dev == dev)
3408 parms->neigh_setup = bond_neigh_init;
3414 * Change the MTU of all of a master's slaves to match the master
3416 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3418 struct bonding *bond = netdev_priv(bond_dev);
3419 struct slave *slave, *rollback_slave;
3420 struct list_head *iter;
3423 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3425 /* Can't hold bond->lock with bh disabled here since
3426 * some base drivers panic. On the other hand we can't
3427 * hold bond->lock without bh disabled because we'll
3428 * deadlock. The only solution is to rely on the fact
3429 * that we're under rtnl_lock here, and the slaves
3430 * list won't change. This doesn't solve the problem
3431 * of setting the slave's MTU while it is
3432 * transmitting, but the assumption is that the base
3433 * driver can handle that.
3435 * TODO: figure out a way to safely iterate the slaves
3436 * list, but without holding a lock around the actual
3437 * call to the base driver.
3440 bond_for_each_slave(bond, slave, iter) {
3441 netdev_dbg(bond_dev, "s %p c_m %p\n",
3442 slave, slave->dev->netdev_ops->ndo_change_mtu);
3444 res = dev_set_mtu(slave->dev, new_mtu);
3447 /* If we failed to set the slave's mtu to the new value
3448 * we must abort the operation even in ACTIVE_BACKUP
3449 * mode, because if we allow the backup slaves to have
3450 * different mtu values than the active slave we'll
3451 * need to change their mtu when doing a failover. That
3452 * means changing their mtu from timer context, which
3453 * is probably not a good idea.
3455 netdev_dbg(bond_dev, "err %d %s\n", res,
3461 bond_dev->mtu = new_mtu;
3466 /* unwind from head to the slave that failed */
3467 bond_for_each_slave(bond, rollback_slave, iter) {
3470 if (rollback_slave == slave)
3473 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3475 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3476 tmp_res, rollback_slave->dev->name);
3486 * Note that many devices must be down to change the HW address, and
3487 * downing the master releases all slaves. We can make bonds full of
3488 * bonding devices to test this, however.
3490 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3492 struct bonding *bond = netdev_priv(bond_dev);
3493 struct slave *slave, *rollback_slave;
3494 struct sockaddr *sa = addr, tmp_sa;
3495 struct list_head *iter;
3498 if (BOND_MODE(bond) == BOND_MODE_ALB)
3499 return bond_alb_set_mac_address(bond_dev, addr);
3502 netdev_dbg(bond_dev, "bond=%p\n", bond);
3504 /* If fail_over_mac is enabled, do nothing and return success.
3505 * Returning an error causes ifenslave to fail.
3507 if (bond->params.fail_over_mac &&
3508 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3511 if (!is_valid_ether_addr(sa->sa_data))
3512 return -EADDRNOTAVAIL;
3514 /* Can't hold bond->lock with bh disabled here since
3515 * some base drivers panic. On the other hand we can't
3516 * hold bond->lock without bh disabled because we'll
3517 * deadlock. The only solution is to rely on the fact
3518 * that we're under rtnl_lock here, and the slaves
3519 * list won't change. This doesn't solve the problem
3520 * of setting the slave's hw address while it is
3521 * transmitting, but the assumption is that the base
3522 * driver can handle that.
3524 * TODO: figure out a way to safely iterate the slaves
3525 * list, but without holding a lock around the actual
3526 * call to the base driver.
3529 bond_for_each_slave(bond, slave, iter) {
3530 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3531 res = dev_set_mac_address(slave->dev, addr);
3533 /* TODO: consider downing the slave
3535 * User should expect communications
3536 * breakage anyway until ARP finish
3539 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3545 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3549 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3550 tmp_sa.sa_family = bond_dev->type;
3552 /* unwind from head to the slave that failed */
3553 bond_for_each_slave(bond, rollback_slave, iter) {
3556 if (rollback_slave == slave)
3559 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3561 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3562 tmp_res, rollback_slave->dev->name);
3570 * bond_xmit_slave_id - transmit skb through slave with slave_id
3571 * @bond: bonding device that is transmitting
3572 * @skb: buffer to transmit
3573 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3575 * This function tries to transmit through slave with slave_id but in case
3576 * it fails, it tries to find the first available slave for transmission.
3577 * The skb is consumed in all cases, thus the function is void.
3579 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3581 struct list_head *iter;
3582 struct slave *slave;
3585 /* Here we start from the slave with slave_id */
3586 bond_for_each_slave_rcu(bond, slave, iter) {
3588 if (bond_slave_can_tx(slave)) {
3589 bond_dev_queue_xmit(bond, skb, slave->dev);
3595 /* Here we start from the first slave up to slave_id */
3597 bond_for_each_slave_rcu(bond, slave, iter) {
3600 if (bond_slave_can_tx(slave)) {
3601 bond_dev_queue_xmit(bond, skb, slave->dev);
3605 /* no slave that can tx has been found */
3606 dev_kfree_skb_any(skb);
3610 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3611 * @bond: bonding device to use
3613 * Based on the value of the bonding device's packets_per_slave parameter
3614 * this function generates a slave id, which is usually used as the next
3615 * slave to transmit through.
3617 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3620 struct reciprocal_value reciprocal_packets_per_slave;
3621 int packets_per_slave = bond->params.packets_per_slave;
3623 switch (packets_per_slave) {
3625 slave_id = prandom_u32();
3628 slave_id = bond->rr_tx_counter;
3631 reciprocal_packets_per_slave =
3632 bond->params.reciprocal_packets_per_slave;
3633 slave_id = reciprocal_divide(bond->rr_tx_counter,
3634 reciprocal_packets_per_slave);
3637 bond->rr_tx_counter++;
3642 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3644 struct bonding *bond = netdev_priv(bond_dev);
3645 struct iphdr *iph = ip_hdr(skb);
3646 struct slave *slave;
3649 /* Start with the curr_active_slave that joined the bond as the
3650 * default for sending IGMP traffic. For failover purposes one
3651 * needs to maintain some consistency for the interface that will
3652 * send the join/membership reports. The curr_active_slave found
3653 * will send all of this type of traffic.
3655 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3656 slave = rcu_dereference(bond->curr_active_slave);
3657 if (slave && bond_slave_can_tx(slave))
3658 bond_dev_queue_xmit(bond, skb, slave->dev);
3660 bond_xmit_slave_id(bond, skb, 0);
3662 slave_id = bond_rr_gen_slave_id(bond);
3663 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3666 return NETDEV_TX_OK;
3670 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3671 * the bond has a usable interface.
3673 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3675 struct bonding *bond = netdev_priv(bond_dev);
3676 struct slave *slave;
3678 slave = rcu_dereference(bond->curr_active_slave);
3680 bond_dev_queue_xmit(bond, skb, slave->dev);
3682 dev_kfree_skb_any(skb);
3684 return NETDEV_TX_OK;
3687 /* In bond_xmit_xor() , we determine the output device by using a pre-
3688 * determined xmit_hash_policy(), If the selected device is not enabled,
3689 * find the next active slave.
3691 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3693 struct bonding *bond = netdev_priv(bond_dev);
3695 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb) % bond->slave_cnt);
3697 return NETDEV_TX_OK;
3700 /* in broadcast mode, we send everything to all usable interfaces. */
3701 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3703 struct bonding *bond = netdev_priv(bond_dev);
3704 struct slave *slave = NULL;
3705 struct list_head *iter;
3707 bond_for_each_slave_rcu(bond, slave, iter) {
3708 if (bond_is_last_slave(bond, slave))
3710 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3711 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3714 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3715 bond_dev->name, __func__);
3718 /* bond_dev_queue_xmit always returns 0 */
3719 bond_dev_queue_xmit(bond, skb2, slave->dev);
3722 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3723 bond_dev_queue_xmit(bond, skb, slave->dev);
3725 dev_kfree_skb_any(skb);
3727 return NETDEV_TX_OK;
3730 /*------------------------- Device initialization ---------------------------*/
3733 * Lookup the slave that corresponds to a qid
3735 static inline int bond_slave_override(struct bonding *bond,
3736 struct sk_buff *skb)
3738 struct slave *slave = NULL;
3739 struct list_head *iter;
3741 if (!skb->queue_mapping)
3744 /* Find out if any slaves have the same mapping as this skb. */
3745 bond_for_each_slave_rcu(bond, slave, iter) {
3746 if (slave->queue_id == skb->queue_mapping) {
3747 if (bond_slave_can_tx(slave)) {
3748 bond_dev_queue_xmit(bond, skb, slave->dev);
3751 /* If the slave isn't UP, use default transmit policy. */
3760 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3761 void *accel_priv, select_queue_fallback_t fallback)
3764 * This helper function exists to help dev_pick_tx get the correct
3765 * destination queue. Using a helper function skips a call to
3766 * skb_tx_hash and will put the skbs in the queue we expect on their
3767 * way down to the bonding driver.
3769 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3772 * Save the original txq to restore before passing to the driver
3774 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3776 if (unlikely(txq >= dev->real_num_tx_queues)) {
3778 txq -= dev->real_num_tx_queues;
3779 } while (txq >= dev->real_num_tx_queues);
3784 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3786 struct bonding *bond = netdev_priv(dev);
3788 if (bond_should_override_tx_queue(bond) &&
3789 !bond_slave_override(bond, skb))
3790 return NETDEV_TX_OK;
3792 switch (BOND_MODE(bond)) {
3793 case BOND_MODE_ROUNDROBIN:
3794 return bond_xmit_roundrobin(skb, dev);
3795 case BOND_MODE_ACTIVEBACKUP:
3796 return bond_xmit_activebackup(skb, dev);
3798 return bond_xmit_xor(skb, dev);
3799 case BOND_MODE_BROADCAST:
3800 return bond_xmit_broadcast(skb, dev);
3801 case BOND_MODE_8023AD:
3802 return bond_3ad_xmit_xor(skb, dev);
3804 return bond_alb_xmit(skb, dev);
3806 return bond_tlb_xmit(skb, dev);
3808 /* Should never happen, mode already checked */
3809 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3811 dev_kfree_skb_any(skb);
3812 return NETDEV_TX_OK;
3816 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3818 struct bonding *bond = netdev_priv(dev);
3819 netdev_tx_t ret = NETDEV_TX_OK;
3822 * If we risk deadlock from transmitting this in the
3823 * netpoll path, tell netpoll to queue the frame for later tx
3825 if (unlikely(is_netpoll_tx_blocked(dev)))
3826 return NETDEV_TX_BUSY;
3829 if (bond_has_slaves(bond))
3830 ret = __bond_start_xmit(skb, dev);
3832 dev_kfree_skb_any(skb);
3838 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3839 struct ethtool_cmd *ecmd)
3841 struct bonding *bond = netdev_priv(bond_dev);
3842 unsigned long speed = 0;
3843 struct list_head *iter;
3844 struct slave *slave;
3846 ecmd->duplex = DUPLEX_UNKNOWN;
3847 ecmd->port = PORT_OTHER;
3849 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3850 * do not need to check mode. Though link speed might not represent
3851 * the true receive or transmit bandwidth (not all modes are symmetric)
3852 * this is an accurate maximum.
3854 read_lock(&bond->lock);
3855 bond_for_each_slave(bond, slave, iter) {
3856 if (bond_slave_can_tx(slave)) {
3857 if (slave->speed != SPEED_UNKNOWN)
3858 speed += slave->speed;
3859 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3860 slave->duplex != DUPLEX_UNKNOWN)
3861 ecmd->duplex = slave->duplex;
3864 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3865 read_unlock(&bond->lock);
3870 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3871 struct ethtool_drvinfo *drvinfo)
3873 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3874 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3875 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3879 static const struct ethtool_ops bond_ethtool_ops = {
3880 .get_drvinfo = bond_ethtool_get_drvinfo,
3881 .get_settings = bond_ethtool_get_settings,
3882 .get_link = ethtool_op_get_link,
3885 static const struct net_device_ops bond_netdev_ops = {
3886 .ndo_init = bond_init,
3887 .ndo_uninit = bond_uninit,
3888 .ndo_open = bond_open,
3889 .ndo_stop = bond_close,
3890 .ndo_start_xmit = bond_start_xmit,
3891 .ndo_select_queue = bond_select_queue,
3892 .ndo_get_stats64 = bond_get_stats,
3893 .ndo_do_ioctl = bond_do_ioctl,
3894 .ndo_change_rx_flags = bond_change_rx_flags,
3895 .ndo_set_rx_mode = bond_set_rx_mode,
3896 .ndo_change_mtu = bond_change_mtu,
3897 .ndo_set_mac_address = bond_set_mac_address,
3898 .ndo_neigh_setup = bond_neigh_setup,
3899 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3900 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3901 #ifdef CONFIG_NET_POLL_CONTROLLER
3902 .ndo_netpoll_setup = bond_netpoll_setup,
3903 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3904 .ndo_poll_controller = bond_poll_controller,
3906 .ndo_add_slave = bond_enslave,
3907 .ndo_del_slave = bond_release,
3908 .ndo_fix_features = bond_fix_features,
3911 static const struct device_type bond_type = {
3915 static void bond_destructor(struct net_device *bond_dev)
3917 struct bonding *bond = netdev_priv(bond_dev);
3919 destroy_workqueue(bond->wq);
3920 free_netdev(bond_dev);
3923 void bond_setup(struct net_device *bond_dev)
3925 struct bonding *bond = netdev_priv(bond_dev);
3927 /* initialize rwlocks */
3928 rwlock_init(&bond->lock);
3929 rwlock_init(&bond->curr_slave_lock);
3930 bond->params = bonding_defaults;
3932 /* Initialize pointers */
3933 bond->dev = bond_dev;
3935 /* Initialize the device entry points */
3936 ether_setup(bond_dev);
3937 bond_dev->netdev_ops = &bond_netdev_ops;
3938 bond_dev->ethtool_ops = &bond_ethtool_ops;
3940 bond_dev->destructor = bond_destructor;
3942 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3944 /* Initialize the device options */
3945 bond_dev->tx_queue_len = 0;
3946 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3947 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
3948 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3950 /* don't acquire bond device's netif_tx_lock when
3952 bond_dev->features |= NETIF_F_LLTX;
3954 /* By default, we declare the bond to be fully
3955 * VLAN hardware accelerated capable. Special
3956 * care is taken in the various xmit functions
3957 * when there are slaves that are not hw accel
3961 /* Don't allow bond devices to change network namespaces. */
3962 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3964 bond_dev->hw_features = BOND_VLAN_FEATURES |
3965 NETIF_F_HW_VLAN_CTAG_TX |
3966 NETIF_F_HW_VLAN_CTAG_RX |
3967 NETIF_F_HW_VLAN_CTAG_FILTER;
3969 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3970 bond_dev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3971 bond_dev->features |= bond_dev->hw_features;
3975 * Destroy a bonding device.
3976 * Must be under rtnl_lock when this function is called.
3978 static void bond_uninit(struct net_device *bond_dev)
3980 struct bonding *bond = netdev_priv(bond_dev);
3981 struct list_head *iter;
3982 struct slave *slave;
3984 bond_netpoll_cleanup(bond_dev);
3986 /* Release the bonded slaves */
3987 bond_for_each_slave(bond, slave, iter)
3988 __bond_release_one(bond_dev, slave->dev, true);
3989 netdev_info(bond_dev, "Released all slaves\n");
3991 list_del(&bond->bond_list);
3993 bond_debug_unregister(bond);
3996 /*------------------------- Module initialization ---------------------------*/
3998 static int bond_check_params(struct bond_params *params)
4000 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4001 struct bond_opt_value newval;
4002 const struct bond_opt_value *valptr;
4003 int arp_all_targets_value;
4006 * Convert string parameters.
4009 bond_opt_initstr(&newval, mode);
4010 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4012 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4015 bond_mode = valptr->value;
4018 if (xmit_hash_policy) {
4019 if ((bond_mode != BOND_MODE_XOR) &&
4020 (bond_mode != BOND_MODE_8023AD) &&
4021 (bond_mode != BOND_MODE_TLB)) {
4022 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4023 bond_mode_name(bond_mode));
4025 bond_opt_initstr(&newval, xmit_hash_policy);
4026 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4029 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4033 xmit_hashtype = valptr->value;
4038 if (bond_mode != BOND_MODE_8023AD) {
4039 pr_info("lacp_rate param is irrelevant in mode %s\n",
4040 bond_mode_name(bond_mode));
4042 bond_opt_initstr(&newval, lacp_rate);
4043 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4046 pr_err("Error: Invalid lacp rate \"%s\"\n",
4050 lacp_fast = valptr->value;
4055 bond_opt_initstr(&newval, ad_select);
4056 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4059 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4062 params->ad_select = valptr->value;
4063 if (bond_mode != BOND_MODE_8023AD)
4064 pr_warn("ad_select param only affects 802.3ad mode\n");
4066 params->ad_select = BOND_AD_STABLE;
4069 if (max_bonds < 0) {
4070 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4071 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4072 max_bonds = BOND_DEFAULT_MAX_BONDS;
4076 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4082 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4087 if (downdelay < 0) {
4088 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4089 downdelay, INT_MAX);
4093 if ((use_carrier != 0) && (use_carrier != 1)) {
4094 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4099 if (num_peer_notif < 0 || num_peer_notif > 255) {
4100 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4105 /* reset values for 802.3ad/TLB/ALB */
4106 if (!bond_mode_uses_arp(bond_mode)) {
4108 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");
4109 pr_warn("Forcing miimon to 100msec\n");
4110 miimon = BOND_DEFAULT_MIIMON;
4114 if (tx_queues < 1 || tx_queues > 255) {
4115 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4116 tx_queues, BOND_DEFAULT_TX_QUEUES);
4117 tx_queues = BOND_DEFAULT_TX_QUEUES;
4120 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4121 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4123 all_slaves_active = 0;
4126 if (resend_igmp < 0 || resend_igmp > 255) {
4127 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4128 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4129 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4132 bond_opt_initval(&newval, packets_per_slave);
4133 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4134 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4135 packets_per_slave, USHRT_MAX);
4136 packets_per_slave = 1;
4139 if (bond_mode == BOND_MODE_ALB) {
4140 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",
4145 if (updelay || downdelay) {
4146 /* just warn the user the up/down delay will have
4147 * no effect since miimon is zero...
4149 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",
4150 updelay, downdelay);
4153 /* don't allow arp monitoring */
4155 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4156 miimon, arp_interval);
4160 if ((updelay % miimon) != 0) {
4161 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4162 updelay, miimon, (updelay / miimon) * miimon);
4167 if ((downdelay % miimon) != 0) {
4168 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4170 (downdelay / miimon) * miimon);
4173 downdelay /= miimon;
4176 if (arp_interval < 0) {
4177 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4178 arp_interval, INT_MAX);
4182 for (arp_ip_count = 0, i = 0;
4183 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4184 /* not complete check, but should be good enough to
4187 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4188 !bond_is_ip_target_ok(ip)) {
4189 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4193 if (bond_get_targets_ip(arp_target, ip) == -1)
4194 arp_target[arp_ip_count++] = ip;
4196 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4201 if (arp_interval && !arp_ip_count) {
4202 /* don't allow arping if no arp_ip_target given... */
4203 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4209 if (!arp_interval) {
4210 pr_err("arp_validate requires arp_interval\n");
4214 bond_opt_initstr(&newval, arp_validate);
4215 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4218 pr_err("Error: invalid arp_validate \"%s\"\n",
4222 arp_validate_value = valptr->value;
4224 arp_validate_value = 0;
4227 arp_all_targets_value = 0;
4228 if (arp_all_targets) {
4229 bond_opt_initstr(&newval, arp_all_targets);
4230 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4233 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4235 arp_all_targets_value = 0;
4237 arp_all_targets_value = valptr->value;
4242 pr_info("MII link monitoring set to %d ms\n", miimon);
4243 } else if (arp_interval) {
4244 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4245 arp_validate_value);
4246 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4247 arp_interval, valptr->string, arp_ip_count);
4249 for (i = 0; i < arp_ip_count; i++)
4250 pr_cont(" %s", arp_ip_target[i]);
4254 } else if (max_bonds) {
4255 /* miimon and arp_interval not set, we need one so things
4256 * work as expected, see bonding.txt for details
4258 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");
4261 if (primary && !bond_mode_uses_primary(bond_mode)) {
4262 /* currently, using a primary only makes sense
4263 * in active backup, TLB or ALB modes
4265 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4266 primary, bond_mode_name(bond_mode));
4270 if (primary && primary_reselect) {
4271 bond_opt_initstr(&newval, primary_reselect);
4272 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4275 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4279 primary_reselect_value = valptr->value;
4281 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4284 if (fail_over_mac) {
4285 bond_opt_initstr(&newval, fail_over_mac);
4286 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4289 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4293 fail_over_mac_value = valptr->value;
4294 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4295 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4297 fail_over_mac_value = BOND_FOM_NONE;
4300 if (lp_interval == 0) {
4301 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4302 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4303 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4306 /* fill params struct with the proper values */
4307 params->mode = bond_mode;
4308 params->xmit_policy = xmit_hashtype;
4309 params->miimon = miimon;
4310 params->num_peer_notif = num_peer_notif;
4311 params->arp_interval = arp_interval;
4312 params->arp_validate = arp_validate_value;
4313 params->arp_all_targets = arp_all_targets_value;
4314 params->updelay = updelay;
4315 params->downdelay = downdelay;
4316 params->use_carrier = use_carrier;
4317 params->lacp_fast = lacp_fast;
4318 params->primary[0] = 0;
4319 params->primary_reselect = primary_reselect_value;
4320 params->fail_over_mac = fail_over_mac_value;
4321 params->tx_queues = tx_queues;
4322 params->all_slaves_active = all_slaves_active;
4323 params->resend_igmp = resend_igmp;
4324 params->min_links = min_links;
4325 params->lp_interval = lp_interval;
4326 params->packets_per_slave = packets_per_slave;
4327 params->tlb_dynamic_lb = 1; /* Default value */
4328 if (packets_per_slave > 0) {
4329 params->reciprocal_packets_per_slave =
4330 reciprocal_value(packets_per_slave);
4332 /* reciprocal_packets_per_slave is unused if
4333 * packets_per_slave is 0 or 1, just initialize it
4335 params->reciprocal_packets_per_slave =
4336 (struct reciprocal_value) { 0 };
4340 strncpy(params->primary, primary, IFNAMSIZ);
4341 params->primary[IFNAMSIZ - 1] = 0;
4344 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4349 static struct lock_class_key bonding_netdev_xmit_lock_key;
4350 static struct lock_class_key bonding_netdev_addr_lock_key;
4351 static struct lock_class_key bonding_tx_busylock_key;
4353 static void bond_set_lockdep_class_one(struct net_device *dev,
4354 struct netdev_queue *txq,
4357 lockdep_set_class(&txq->_xmit_lock,
4358 &bonding_netdev_xmit_lock_key);
4361 static void bond_set_lockdep_class(struct net_device *dev)
4363 lockdep_set_class(&dev->addr_list_lock,
4364 &bonding_netdev_addr_lock_key);
4365 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4366 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4370 * Called from registration process
4372 static int bond_init(struct net_device *bond_dev)
4374 struct bonding *bond = netdev_priv(bond_dev);
4375 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4376 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4378 netdev_dbg(bond_dev, "Begin bond_init\n");
4381 * Initialize locks that may be required during
4382 * en/deslave operations. All of the bond_open work
4383 * (of which this is part) should really be moved to
4384 * a phase prior to dev_open
4386 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4387 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4389 bond->wq = create_singlethread_workqueue(bond_dev->name);
4393 bond_set_lockdep_class(bond_dev);
4395 list_add_tail(&bond->bond_list, &bn->dev_list);
4397 bond_prepare_sysfs_group(bond);
4399 bond_debug_register(bond);
4401 /* Ensure valid dev_addr */
4402 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4403 bond_dev->addr_assign_type == NET_ADDR_PERM)
4404 eth_hw_addr_random(bond_dev);
4409 unsigned int bond_get_num_tx_queues(void)
4414 /* Create a new bond based on the specified name and bonding parameters.
4415 * If name is NULL, obtain a suitable "bond%d" name for us.
4416 * Caller must NOT hold rtnl_lock; we need to release it here before we
4417 * set up our sysfs entries.
4419 int bond_create(struct net *net, const char *name)
4421 struct net_device *bond_dev;
4426 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4427 name ? name : "bond%d", NET_NAME_UNKNOWN,
4428 bond_setup, tx_queues);
4430 pr_err("%s: eek! can't alloc netdev!\n", name);
4435 dev_net_set(bond_dev, net);
4436 bond_dev->rtnl_link_ops = &bond_link_ops;
4438 res = register_netdevice(bond_dev);
4440 netif_carrier_off(bond_dev);
4444 bond_destructor(bond_dev);
4448 static int __net_init bond_net_init(struct net *net)
4450 struct bond_net *bn = net_generic(net, bond_net_id);
4453 INIT_LIST_HEAD(&bn->dev_list);
4455 bond_create_proc_dir(bn);
4456 bond_create_sysfs(bn);
4461 static void __net_exit bond_net_exit(struct net *net)
4463 struct bond_net *bn = net_generic(net, bond_net_id);
4464 struct bonding *bond, *tmp_bond;
4467 bond_destroy_sysfs(bn);
4469 /* Kill off any bonds created after unregistering bond rtnl ops */
4471 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4472 unregister_netdevice_queue(bond->dev, &list);
4473 unregister_netdevice_many(&list);
4476 bond_destroy_proc_dir(bn);
4479 static struct pernet_operations bond_net_ops = {
4480 .init = bond_net_init,
4481 .exit = bond_net_exit,
4483 .size = sizeof(struct bond_net),
4486 static int __init bonding_init(void)
4491 pr_info("%s", bond_version);
4493 res = bond_check_params(&bonding_defaults);
4497 res = register_pernet_subsys(&bond_net_ops);
4501 res = bond_netlink_init();
4505 bond_create_debugfs();
4507 for (i = 0; i < max_bonds; i++) {
4508 res = bond_create(&init_net, NULL);
4513 register_netdevice_notifier(&bond_netdev_notifier);
4517 bond_destroy_debugfs();
4518 bond_netlink_fini();
4520 unregister_pernet_subsys(&bond_net_ops);
4525 static void __exit bonding_exit(void)
4527 unregister_netdevice_notifier(&bond_netdev_notifier);
4529 bond_destroy_debugfs();
4531 bond_netlink_fini();
4532 unregister_pernet_subsys(&bond_net_ops);
4534 #ifdef CONFIG_NET_POLL_CONTROLLER
4536 * Make sure we don't have an imbalance on our netpoll blocking
4538 WARN_ON(atomic_read(&netpoll_block_tx));
4542 module_init(bonding_init);
4543 module_exit(bonding_exit);
4544 MODULE_LICENSE("GPL");
4545 MODULE_VERSION(DRV_VERSION);
4546 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4547 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");