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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 int bond_net_id __read_mostly;
203 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
204 static int arp_ip_count;
205 static int bond_mode = BOND_MODE_ROUNDROBIN;
206 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
207 static int lacp_fast;
209 /*-------------------------- Forward declarations ---------------------------*/
211 static int bond_init(struct net_device *bond_dev);
212 static void bond_uninit(struct net_device *bond_dev);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode)
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
246 skb->dev = slave_dev;
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
302 vlan_vid_del(rollback_slave->dev, proto, vid);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
329 /*------------------------------- Link status -------------------------------*/
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
336 * Returns zero if carrier state does not change, nonzero if it does.
338 static int bond_set_carrier(struct bonding *bond)
340 struct list_head *iter;
343 if (!bond_has_slaves(bond))
346 if (bond->params.mode == BOND_MODE_8023AD)
347 return bond_3ad_set_carrier(bond);
349 bond_for_each_slave(bond, slave, iter) {
350 if (slave->link == BOND_LINK_UP) {
351 if (!netif_carrier_ok(bond->dev)) {
352 netif_carrier_on(bond->dev);
360 if (netif_carrier_ok(bond->dev)) {
361 netif_carrier_off(bond->dev);
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
370 * values are invalid, set speed and duplex to -1,
373 static void bond_update_speed_duplex(struct slave *slave)
375 struct net_device *slave_dev = slave->dev;
376 struct ethtool_cmd ecmd;
380 slave->speed = SPEED_UNKNOWN;
381 slave->duplex = DUPLEX_UNKNOWN;
383 res = __ethtool_get_settings(slave_dev, &ecmd);
387 slave_speed = ethtool_cmd_speed(&ecmd);
388 if (slave_speed == 0 || slave_speed == ((__u32) -1))
391 switch (ecmd.duplex) {
399 slave->speed = slave_speed;
400 slave->duplex = ecmd.duplex;
405 const char *bond_slave_link_status(s8 link)
422 * if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
443 struct mii_ioctl_data *mii;
445 if (!reporting && !netif_running(slave_dev))
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
459 /* TODO: set pointer to correct ioctl on a per team member */
460 /* bases to make this more efficient. that is, once */
461 /* we determine the correct ioctl, we will always */
462 /* call it and not the others for that team */
466 * We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
482 * If reporting, report that either there's no dev->do_ioctl,
483 * or both SIOCGMIIREG and get_link failed (meaning that we
484 * cannot report link status). If not reporting, pretend
487 return reporting ? -1 : BMSR_LSTATUS;
490 /*----------------------------- Multicast list ------------------------------*/
493 * Push the promiscuity flag down to appropriate slaves
495 static int bond_set_promiscuity(struct bonding *bond, int inc)
497 struct list_head *iter;
500 if (USES_PRIMARY(bond->params.mode)) {
501 /* write lock already acquired */
502 if (bond->curr_active_slave) {
503 err = dev_set_promiscuity(bond->curr_active_slave->dev,
509 bond_for_each_slave(bond, slave, iter) {
510 err = dev_set_promiscuity(slave->dev, inc);
519 * Push the allmulti flag down to all slaves
521 static int bond_set_allmulti(struct bonding *bond, int inc)
523 struct list_head *iter;
526 if (USES_PRIMARY(bond->params.mode)) {
527 /* write lock already acquired */
528 if (bond->curr_active_slave) {
529 err = dev_set_allmulti(bond->curr_active_slave->dev,
535 bond_for_each_slave(bond, slave, iter) {
536 err = dev_set_allmulti(slave->dev, inc);
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
549 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
551 struct bonding *bond = container_of(work, struct bonding,
554 if (!rtnl_trylock()) {
555 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
558 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
560 if (bond->igmp_retrans > 1) {
561 bond->igmp_retrans--;
562 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
567 /* Flush bond's hardware addresses from slave
569 static void bond_hw_addr_flush(struct net_device *bond_dev,
570 struct net_device *slave_dev)
572 struct bonding *bond = netdev_priv(bond_dev);
574 dev_uc_unsync(slave_dev, bond_dev);
575 dev_mc_unsync(slave_dev, bond_dev);
577 if (bond->params.mode == BOND_MODE_8023AD) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
581 dev_mc_del(slave_dev, lacpdu_multicast);
585 /*--------------------------- Active slave change ---------------------------*/
587 /* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
589 * slaves up date at all times; only the USES_PRIMARY modes need to call
590 * this function to swap these settings during a failover.
592 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
593 struct slave *old_active)
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(old_active->dev, -1);
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(old_active->dev, -1);
604 bond_hw_addr_flush(bond->dev, old_active->dev);
608 /* FIXME: Signal errors upstream. */
609 if (bond->dev->flags & IFF_PROMISC)
610 dev_set_promiscuity(new_active->dev, 1);
612 if (bond->dev->flags & IFF_ALLMULTI)
613 dev_set_allmulti(new_active->dev, 1);
615 netif_addr_lock_bh(bond->dev);
616 dev_uc_sync(new_active->dev, bond->dev);
617 dev_mc_sync(new_active->dev, bond->dev);
618 netif_addr_unlock_bh(bond->dev);
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
627 * Should be called with RTNL held.
629 static void bond_set_dev_addr(struct net_device *bond_dev,
630 struct net_device *slave_dev)
632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev, slave_dev, slave_dev->addr_len);
634 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
635 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
640 * bond_do_fail_over_mac
642 * Perform special MAC address swapping for fail_over_mac settings
644 * Called with RTNL, curr_slave_lock for write_bh.
646 static void bond_do_fail_over_mac(struct bonding *bond,
647 struct slave *new_active,
648 struct slave *old_active)
649 __releases(&bond->curr_slave_lock)
650 __acquires(&bond->curr_slave_lock)
652 u8 tmp_mac[ETH_ALEN];
653 struct sockaddr saddr;
656 switch (bond->params.fail_over_mac) {
657 case BOND_FOM_ACTIVE:
659 write_unlock_bh(&bond->curr_slave_lock);
660 bond_set_dev_addr(bond->dev, new_active->dev);
661 write_lock_bh(&bond->curr_slave_lock);
664 case BOND_FOM_FOLLOW:
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
673 write_unlock_bh(&bond->curr_slave_lock);
676 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
677 memcpy(saddr.sa_data, old_active->dev->dev_addr,
679 saddr.sa_family = new_active->dev->type;
681 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
682 saddr.sa_family = bond->dev->type;
685 rv = dev_set_mac_address(new_active->dev, &saddr);
687 pr_err("%s: Error %d setting MAC of slave %s\n",
688 bond->dev->name, -rv, new_active->dev->name);
695 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
696 saddr.sa_family = old_active->dev->type;
698 rv = dev_set_mac_address(old_active->dev, &saddr);
700 pr_err("%s: Error %d setting MAC of slave %s\n",
701 bond->dev->name, -rv, new_active->dev->name);
703 write_lock_bh(&bond->curr_slave_lock);
706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
707 bond->dev->name, bond->params.fail_over_mac);
713 static bool bond_should_change_active(struct bonding *bond)
715 struct slave *prim = bond->primary_slave;
716 struct slave *curr = bond->curr_active_slave;
718 if (!prim || !curr || curr->link != BOND_LINK_UP)
720 if (bond->force_primary) {
721 bond->force_primary = false;
724 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
725 (prim->speed < curr->speed ||
726 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
728 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
737 static struct slave *bond_find_best_slave(struct bonding *bond)
739 struct slave *slave, *bestslave = NULL;
740 struct list_head *iter;
741 int mintime = bond->params.updelay;
743 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
744 bond_should_change_active(bond))
745 return bond->primary_slave;
747 bond_for_each_slave(bond, slave, iter) {
748 if (slave->link == BOND_LINK_UP)
750 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
751 slave->delay < mintime) {
752 mintime = slave->delay;
760 static bool bond_should_notify_peers(struct bonding *bond)
765 slave = rcu_dereference(bond->curr_active_slave);
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond->dev->name, slave ? slave->dev->name : "NULL");
771 if (!slave || !bond->send_peer_notif ||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
783 * Set the new slave to the bond's settings and unset them on the old
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
793 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
795 struct slave *old_active = bond->curr_active_slave;
797 if (old_active == new_active)
801 new_active->jiffies = jiffies;
803 if (new_active->link == BOND_LINK_BACK) {
804 if (USES_PRIMARY(bond->params.mode)) {
805 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
806 bond->dev->name, 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->params.mode == 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 (USES_PRIMARY(bond->params.mode)) {
820 pr_info("%s: making interface %s the new active one.\n",
821 bond->dev->name, new_active->dev->name);
826 if (USES_PRIMARY(bond->params.mode))
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);
834 bond_set_slave_active_flags(new_active);
836 rcu_assign_pointer(bond->curr_active_slave, new_active);
839 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
841 bond_set_slave_inactive_flags(old_active);
844 bool should_notify_peers = false;
846 bond_set_slave_active_flags(new_active);
848 if (bond->params.fail_over_mac)
849 bond_do_fail_over_mac(bond, new_active,
852 if (netif_running(bond->dev)) {
853 bond->send_peer_notif =
854 bond->params.num_peer_notif;
855 should_notify_peers =
856 bond_should_notify_peers(bond);
859 write_unlock_bh(&bond->curr_slave_lock);
861 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
862 if (should_notify_peers)
863 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
866 write_lock_bh(&bond->curr_slave_lock);
870 /* resend IGMP joins since active slave has changed or
871 * all were sent on curr_active_slave.
872 * resend only if bond is brought up with the affected
873 * bonding modes and the retransmission is enabled */
874 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
875 ((USES_PRIMARY(bond->params.mode) && new_active) ||
876 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
877 bond->igmp_retrans = bond->params.resend_igmp;
878 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
883 * bond_select_active_slave - select a new active slave, if needed
884 * @bond: our bonding struct
886 * This functions should be called when one of the following occurs:
887 * - The old curr_active_slave has been released or lost its link.
888 * - The primary_slave has got its link back.
889 * - A slave has got its link back and there's no old curr_active_slave.
891 * Caller must hold curr_slave_lock for write_bh.
893 void bond_select_active_slave(struct bonding *bond)
895 struct slave *best_slave;
898 best_slave = bond_find_best_slave(bond);
899 if (best_slave != bond->curr_active_slave) {
900 bond_change_active_slave(bond, best_slave);
901 rv = bond_set_carrier(bond);
905 if (netif_carrier_ok(bond->dev)) {
906 pr_info("%s: first active interface up!\n",
909 pr_info("%s: now running without any active interface !\n",
915 #ifdef CONFIG_NET_POLL_CONTROLLER
916 static inline int slave_enable_netpoll(struct slave *slave)
921 np = kzalloc(sizeof(*np), GFP_ATOMIC);
926 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
935 static inline void slave_disable_netpoll(struct slave *slave)
937 struct netpoll *np = slave->np;
943 __netpoll_free_async(np);
945 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
947 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
949 if (!slave_dev->netdev_ops->ndo_poll_controller)
954 static void bond_poll_controller(struct net_device *bond_dev)
958 static void bond_netpoll_cleanup(struct net_device *bond_dev)
960 struct bonding *bond = netdev_priv(bond_dev);
961 struct list_head *iter;
964 bond_for_each_slave(bond, slave, iter)
965 if (IS_UP(slave->dev))
966 slave_disable_netpoll(slave);
969 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
971 struct bonding *bond = netdev_priv(dev);
972 struct list_head *iter;
976 bond_for_each_slave(bond, slave, iter) {
977 err = slave_enable_netpoll(slave);
979 bond_netpoll_cleanup(dev);
986 static inline int slave_enable_netpoll(struct slave *slave)
990 static inline void slave_disable_netpoll(struct slave *slave)
993 static void bond_netpoll_cleanup(struct net_device *bond_dev)
998 /*---------------------------------- IOCTL ----------------------------------*/
1000 static netdev_features_t bond_fix_features(struct net_device *dev,
1001 netdev_features_t features)
1003 struct bonding *bond = netdev_priv(dev);
1004 struct list_head *iter;
1005 netdev_features_t mask;
1006 struct slave *slave;
1008 if (!bond_has_slaves(bond)) {
1009 /* Disable adding VLANs to empty bond. But why? --mq */
1010 features |= NETIF_F_VLAN_CHALLENGED;
1015 features &= ~NETIF_F_ONE_FOR_ALL;
1016 features |= NETIF_F_ALL_FOR_ALL;
1018 bond_for_each_slave(bond, slave, iter) {
1019 features = netdev_increment_features(features,
1020 slave->dev->features,
1023 features = netdev_add_tso_features(features, mask);
1028 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1029 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1030 NETIF_F_HIGHDMA | NETIF_F_LRO)
1032 static void bond_compute_features(struct bonding *bond)
1034 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1035 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1036 struct net_device *bond_dev = bond->dev;
1037 struct list_head *iter;
1038 struct slave *slave;
1039 unsigned short max_hard_header_len = ETH_HLEN;
1040 unsigned int gso_max_size = GSO_MAX_SIZE;
1041 u16 gso_max_segs = GSO_MAX_SEGS;
1043 if (!bond_has_slaves(bond))
1046 bond_for_each_slave(bond, slave, iter) {
1047 vlan_features = netdev_increment_features(vlan_features,
1048 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1050 dst_release_flag &= slave->dev->priv_flags;
1051 if (slave->dev->hard_header_len > max_hard_header_len)
1052 max_hard_header_len = slave->dev->hard_header_len;
1054 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1055 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1059 bond_dev->vlan_features = vlan_features;
1060 bond_dev->hard_header_len = max_hard_header_len;
1061 bond_dev->gso_max_segs = gso_max_segs;
1062 netif_set_gso_max_size(bond_dev, gso_max_size);
1064 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1065 bond_dev->priv_flags = flags | dst_release_flag;
1067 netdev_change_features(bond_dev);
1070 static void bond_setup_by_slave(struct net_device *bond_dev,
1071 struct net_device *slave_dev)
1073 bond_dev->header_ops = slave_dev->header_ops;
1075 bond_dev->type = slave_dev->type;
1076 bond_dev->hard_header_len = slave_dev->hard_header_len;
1077 bond_dev->addr_len = slave_dev->addr_len;
1079 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1080 slave_dev->addr_len);
1083 /* On bonding slaves other than the currently active slave, suppress
1084 * duplicates except for alb non-mcast/bcast.
1086 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1087 struct slave *slave,
1088 struct bonding *bond)
1090 if (bond_is_slave_inactive(slave)) {
1091 if (bond->params.mode == BOND_MODE_ALB &&
1092 skb->pkt_type != PACKET_BROADCAST &&
1093 skb->pkt_type != PACKET_MULTICAST)
1100 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1102 struct sk_buff *skb = *pskb;
1103 struct slave *slave;
1104 struct bonding *bond;
1105 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1107 int ret = RX_HANDLER_ANOTHER;
1109 skb = skb_share_check(skb, GFP_ATOMIC);
1111 return RX_HANDLER_CONSUMED;
1115 slave = bond_slave_get_rcu(skb->dev);
1118 if (bond->params.arp_interval)
1119 slave->dev->last_rx = jiffies;
1121 recv_probe = ACCESS_ONCE(bond->recv_probe);
1123 ret = recv_probe(skb, bond, slave);
1124 if (ret == RX_HANDLER_CONSUMED) {
1130 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1131 return RX_HANDLER_EXACT;
1134 skb->dev = bond->dev;
1136 if (bond->params.mode == BOND_MODE_ALB &&
1137 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1138 skb->pkt_type == PACKET_HOST) {
1140 if (unlikely(skb_cow_head(skb,
1141 skb->data - skb_mac_header(skb)))) {
1143 return RX_HANDLER_CONSUMED;
1145 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1151 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1152 struct net_device *slave_dev,
1153 struct slave *slave)
1157 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1160 slave_dev->flags |= IFF_SLAVE;
1161 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1165 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1166 struct net_device *slave_dev)
1168 netdev_upper_dev_unlink(slave_dev, bond_dev);
1169 slave_dev->flags &= ~IFF_SLAVE;
1170 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1173 /* enslave device <slave> to bond device <master> */
1174 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1176 struct bonding *bond = netdev_priv(bond_dev);
1177 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1178 struct slave *new_slave = NULL, *prev_slave;
1179 struct sockaddr addr;
1183 if (!bond->params.use_carrier &&
1184 slave_dev->ethtool_ops->get_link == NULL &&
1185 slave_ops->ndo_do_ioctl == NULL) {
1186 pr_warning("%s: Warning: no link monitoring support for %s\n",
1187 bond_dev->name, slave_dev->name);
1190 /* already enslaved */
1191 if (slave_dev->flags & IFF_SLAVE) {
1192 pr_debug("Error, Device was already enslaved\n");
1196 /* vlan challenged mutual exclusion */
1197 /* no need to lock since we're protected by rtnl_lock */
1198 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1199 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1200 if (vlan_uses_dev(bond_dev)) {
1201 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1202 bond_dev->name, slave_dev->name, bond_dev->name);
1205 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1206 bond_dev->name, slave_dev->name,
1207 slave_dev->name, bond_dev->name);
1210 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1214 * Old ifenslave binaries are no longer supported. These can
1215 * be identified with moderate accuracy by the state of the slave:
1216 * the current ifenslave will set the interface down prior to
1217 * enslaving it; the old ifenslave will not.
1219 if ((slave_dev->flags & IFF_UP)) {
1220 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1223 goto err_undo_flags;
1226 /* set bonding device ether type by slave - bonding netdevices are
1227 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1228 * there is a need to override some of the type dependent attribs/funcs.
1230 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1231 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1233 if (!bond_has_slaves(bond)) {
1234 if (bond_dev->type != slave_dev->type) {
1235 pr_debug("%s: change device type from %d to %d\n",
1237 bond_dev->type, slave_dev->type);
1239 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1241 res = notifier_to_errno(res);
1243 pr_err("%s: refused to change device type\n",
1246 goto err_undo_flags;
1249 /* Flush unicast and multicast addresses */
1250 dev_uc_flush(bond_dev);
1251 dev_mc_flush(bond_dev);
1253 if (slave_dev->type != ARPHRD_ETHER)
1254 bond_setup_by_slave(bond_dev, slave_dev);
1256 ether_setup(bond_dev);
1257 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1260 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1263 } else if (bond_dev->type != slave_dev->type) {
1264 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1266 slave_dev->type, bond_dev->type);
1268 goto err_undo_flags;
1271 if (slave_ops->ndo_set_mac_address == NULL) {
1272 if (!bond_has_slaves(bond)) {
1273 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address.\n",
1275 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1276 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1277 pr_warn("%s: Setting fail_over_mac to active for active-backup mode.\n",
1280 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1281 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1284 goto err_undo_flags;
1288 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1290 /* If this is the first slave, then we need to set the master's hardware
1291 * address to be the same as the slave's. */
1292 if (!bond_has_slaves(bond) &&
1293 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1294 bond_set_dev_addr(bond->dev, slave_dev);
1296 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1299 goto err_undo_flags;
1302 * Set the new_slave's queue_id to be zero. Queue ID mapping
1303 * is set via sysfs or module option if desired.
1305 new_slave->queue_id = 0;
1307 /* Save slave's original mtu and then set it to match the bond */
1308 new_slave->original_mtu = slave_dev->mtu;
1309 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1311 pr_debug("Error %d calling dev_set_mtu\n", res);
1316 * Save slave's original ("permanent") mac address for modes
1317 * that need it, and for restoring it upon release, and then
1318 * set it to the master's address
1320 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1322 if (!bond->params.fail_over_mac ||
1323 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1325 * Set slave to master's mac address. The application already
1326 * set the master's mac address to that of the first slave
1328 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1329 addr.sa_family = slave_dev->type;
1330 res = dev_set_mac_address(slave_dev, &addr);
1332 pr_debug("Error %d calling set_mac_address\n", res);
1333 goto err_restore_mtu;
1337 /* open the slave since the application closed it */
1338 res = dev_open(slave_dev);
1340 pr_debug("Opening slave %s failed\n", slave_dev->name);
1341 goto err_restore_mac;
1344 new_slave->bond = bond;
1345 new_slave->dev = slave_dev;
1346 slave_dev->priv_flags |= IFF_BONDING;
1348 if (bond_is_lb(bond)) {
1349 /* bond_alb_init_slave() must be called before all other stages since
1350 * it might fail and we do not want to have to undo everything
1352 res = bond_alb_init_slave(bond, new_slave);
1357 /* If the mode USES_PRIMARY, then the following is handled by
1358 * bond_change_active_slave().
1360 if (!USES_PRIMARY(bond->params.mode)) {
1361 /* set promiscuity level to new slave */
1362 if (bond_dev->flags & IFF_PROMISC) {
1363 res = dev_set_promiscuity(slave_dev, 1);
1368 /* set allmulti level to new slave */
1369 if (bond_dev->flags & IFF_ALLMULTI) {
1370 res = dev_set_allmulti(slave_dev, 1);
1375 netif_addr_lock_bh(bond_dev);
1377 dev_mc_sync_multiple(slave_dev, bond_dev);
1378 dev_uc_sync_multiple(slave_dev, bond_dev);
1380 netif_addr_unlock_bh(bond_dev);
1383 if (bond->params.mode == BOND_MODE_8023AD) {
1384 /* add lacpdu mc addr to mc list */
1385 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1387 dev_mc_add(slave_dev, lacpdu_multicast);
1390 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1392 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1393 bond_dev->name, slave_dev->name);
1397 prev_slave = bond_last_slave(bond);
1399 new_slave->delay = 0;
1400 new_slave->link_failure_count = 0;
1402 bond_update_speed_duplex(new_slave);
1404 new_slave->last_arp_rx = jiffies -
1405 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1406 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1407 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1409 if (bond->params.miimon && !bond->params.use_carrier) {
1410 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1412 if ((link_reporting == -1) && !bond->params.arp_interval) {
1414 * miimon is set but a bonded network driver
1415 * does not support ETHTOOL/MII and
1416 * arp_interval is not set. Note: if
1417 * use_carrier is enabled, we will never go
1418 * here (because netif_carrier is always
1419 * supported); thus, we don't need to change
1420 * the messages for netif_carrier.
1422 pr_warning("%s: Warning: 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",
1423 bond_dev->name, slave_dev->name);
1424 } else if (link_reporting == -1) {
1425 /* unable get link status using mii/ethtool */
1426 pr_warning("%s: Warning: 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",
1427 bond_dev->name, slave_dev->name);
1431 /* check for initial state */
1432 if (bond->params.miimon) {
1433 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1434 if (bond->params.updelay) {
1435 new_slave->link = BOND_LINK_BACK;
1436 new_slave->delay = bond->params.updelay;
1438 new_slave->link = BOND_LINK_UP;
1441 new_slave->link = BOND_LINK_DOWN;
1443 } else if (bond->params.arp_interval) {
1444 new_slave->link = (netif_carrier_ok(slave_dev) ?
1445 BOND_LINK_UP : BOND_LINK_DOWN);
1447 new_slave->link = BOND_LINK_UP;
1450 if (new_slave->link != BOND_LINK_DOWN)
1451 new_slave->jiffies = jiffies;
1452 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1453 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1454 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1456 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1457 /* if there is a primary slave, remember it */
1458 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1459 bond->primary_slave = new_slave;
1460 bond->force_primary = true;
1464 switch (bond->params.mode) {
1465 case BOND_MODE_ACTIVEBACKUP:
1466 bond_set_slave_inactive_flags(new_slave);
1468 case BOND_MODE_8023AD:
1469 /* in 802.3ad mode, the internal mechanism
1470 * will activate the slaves in the selected
1473 bond_set_slave_inactive_flags(new_slave);
1474 /* if this is the first slave */
1476 SLAVE_AD_INFO(new_slave).id = 1;
1477 /* Initialize AD with the number of times that the AD timer is called in 1 second
1478 * can be called only after the mac address of the bond is set
1480 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1482 SLAVE_AD_INFO(new_slave).id =
1483 SLAVE_AD_INFO(prev_slave).id + 1;
1486 bond_3ad_bind_slave(new_slave);
1490 bond_set_active_slave(new_slave);
1491 bond_set_slave_inactive_flags(new_slave);
1494 pr_debug("This slave is always active in trunk mode\n");
1496 /* always active in trunk mode */
1497 bond_set_active_slave(new_slave);
1499 /* In trunking mode there is little meaning to curr_active_slave
1500 * anyway (it holds no special properties of the bond device),
1501 * so we can change it without calling change_active_interface()
1503 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1504 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1507 } /* switch(bond_mode) */
1509 #ifdef CONFIG_NET_POLL_CONTROLLER
1510 slave_dev->npinfo = bond->dev->npinfo;
1511 if (slave_dev->npinfo) {
1512 if (slave_enable_netpoll(new_slave)) {
1513 pr_info("Error, %s: master_dev is using netpoll, "
1514 "but new slave device does not support netpoll.\n",
1522 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1525 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1529 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1531 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1532 goto err_unregister;
1535 res = bond_sysfs_slave_add(new_slave);
1537 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1538 goto err_upper_unlink;
1542 bond_compute_features(bond);
1543 bond_set_carrier(bond);
1545 if (USES_PRIMARY(bond->params.mode)) {
1547 write_lock_bh(&bond->curr_slave_lock);
1548 bond_select_active_slave(bond);
1549 write_unlock_bh(&bond->curr_slave_lock);
1550 unblock_netpoll_tx();
1553 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1554 bond_dev->name, slave_dev->name,
1555 bond_is_active_slave(new_slave) ? "n active" : " backup",
1556 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1558 /* enslave is successful */
1561 /* Undo stages on error */
1563 bond_upper_dev_unlink(bond_dev, slave_dev);
1566 netdev_rx_handler_unregister(slave_dev);
1569 if (!USES_PRIMARY(bond->params.mode))
1570 bond_hw_addr_flush(bond_dev, slave_dev);
1572 vlan_vids_del_by_dev(slave_dev, bond_dev);
1573 if (bond->primary_slave == new_slave)
1574 bond->primary_slave = NULL;
1575 if (bond->curr_active_slave == new_slave) {
1577 write_lock_bh(&bond->curr_slave_lock);
1578 bond_change_active_slave(bond, NULL);
1579 bond_select_active_slave(bond);
1580 write_unlock_bh(&bond->curr_slave_lock);
1581 unblock_netpoll_tx();
1583 slave_disable_netpoll(new_slave);
1586 slave_dev->priv_flags &= ~IFF_BONDING;
1587 dev_close(slave_dev);
1590 if (!bond->params.fail_over_mac ||
1591 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1592 /* XXX TODO - fom follow mode needs to change master's
1593 * MAC if this slave's MAC is in use by the bond, or at
1594 * least print a warning.
1596 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1597 addr.sa_family = slave_dev->type;
1598 dev_set_mac_address(slave_dev, &addr);
1602 dev_set_mtu(slave_dev, new_slave->original_mtu);
1608 /* Enslave of first slave has failed and we need to fix master's mac */
1609 if (!bond_has_slaves(bond) &&
1610 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1611 eth_hw_addr_random(bond_dev);
1617 * Try to release the slave device <slave> from the bond device <master>
1618 * It is legal to access curr_active_slave without a lock because all the function
1619 * is write-locked. If "all" is true it means that the function is being called
1620 * while destroying a bond interface and all slaves are being released.
1622 * The rules for slave state should be:
1623 * for Active/Backup:
1624 * Active stays on all backups go down
1625 * for Bonded connections:
1626 * The first up interface should be left on and all others downed.
1628 static int __bond_release_one(struct net_device *bond_dev,
1629 struct net_device *slave_dev,
1632 struct bonding *bond = netdev_priv(bond_dev);
1633 struct slave *slave, *oldcurrent;
1634 struct sockaddr addr;
1635 int old_flags = bond_dev->flags;
1636 netdev_features_t old_features = bond_dev->features;
1638 /* slave is not a slave or master is not master of this slave */
1639 if (!(slave_dev->flags & IFF_SLAVE) ||
1640 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1641 pr_err("%s: Error: cannot release %s.\n",
1642 bond_dev->name, slave_dev->name);
1648 slave = bond_get_slave_by_dev(bond, slave_dev);
1650 /* not a slave of this bond */
1651 pr_info("%s: %s not enslaved\n",
1652 bond_dev->name, slave_dev->name);
1653 unblock_netpoll_tx();
1657 /* release the slave from its bond */
1660 bond_sysfs_slave_del(slave);
1662 bond_upper_dev_unlink(bond_dev, slave_dev);
1663 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1664 * for this slave anymore.
1666 netdev_rx_handler_unregister(slave_dev);
1667 write_lock_bh(&bond->lock);
1669 /* Inform AD package of unbinding of slave. */
1670 if (bond->params.mode == BOND_MODE_8023AD)
1671 bond_3ad_unbind_slave(slave);
1673 write_unlock_bh(&bond->lock);
1675 pr_info("%s: releasing %s interface %s\n",
1677 bond_is_active_slave(slave) ? "active" : "backup",
1680 oldcurrent = bond->curr_active_slave;
1682 bond->current_arp_slave = NULL;
1684 if (!all && (!bond->params.fail_over_mac ||
1685 bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1686 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1687 bond_has_slaves(bond))
1688 pr_warn("%s: Warning: 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",
1689 bond_dev->name, slave_dev->name,
1691 bond_dev->name, slave_dev->name);
1694 if (bond->primary_slave == slave)
1695 bond->primary_slave = NULL;
1697 if (oldcurrent == slave) {
1698 write_lock_bh(&bond->curr_slave_lock);
1699 bond_change_active_slave(bond, NULL);
1700 write_unlock_bh(&bond->curr_slave_lock);
1703 if (bond_is_lb(bond)) {
1704 /* Must be called only after the slave has been
1705 * detached from the list and the curr_active_slave
1706 * has been cleared (if our_slave == old_current),
1707 * but before a new active slave is selected.
1709 bond_alb_deinit_slave(bond, slave);
1713 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1714 } else if (oldcurrent == slave) {
1716 * Note that we hold RTNL over this sequence, so there
1717 * is no concern that another slave add/remove event
1720 write_lock_bh(&bond->curr_slave_lock);
1722 bond_select_active_slave(bond);
1724 write_unlock_bh(&bond->curr_slave_lock);
1727 if (!bond_has_slaves(bond)) {
1728 bond_set_carrier(bond);
1729 eth_hw_addr_random(bond_dev);
1731 if (vlan_uses_dev(bond_dev)) {
1732 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1733 bond_dev->name, bond_dev->name);
1734 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1739 unblock_netpoll_tx();
1742 if (!bond_has_slaves(bond)) {
1743 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1744 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1747 bond_compute_features(bond);
1748 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1749 (old_features & NETIF_F_VLAN_CHALLENGED))
1750 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1751 bond_dev->name, slave_dev->name, bond_dev->name);
1753 /* must do this from outside any spinlocks */
1754 vlan_vids_del_by_dev(slave_dev, bond_dev);
1756 /* If the mode USES_PRIMARY, then this cases was handled above by
1757 * bond_change_active_slave(..., NULL)
1759 if (!USES_PRIMARY(bond->params.mode)) {
1760 /* unset promiscuity level from slave
1761 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1762 * of the IFF_PROMISC flag in the bond_dev, but we need the
1763 * value of that flag before that change, as that was the value
1764 * when this slave was attached, so we cache at the start of the
1765 * function and use it here. Same goes for ALLMULTI below
1767 if (old_flags & IFF_PROMISC)
1768 dev_set_promiscuity(slave_dev, -1);
1770 /* unset allmulti level from slave */
1771 if (old_flags & IFF_ALLMULTI)
1772 dev_set_allmulti(slave_dev, -1);
1774 bond_hw_addr_flush(bond_dev, slave_dev);
1777 slave_disable_netpoll(slave);
1779 /* close slave before restoring its mac address */
1780 dev_close(slave_dev);
1782 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1783 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1784 /* restore original ("permanent") mac address */
1785 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1786 addr.sa_family = slave_dev->type;
1787 dev_set_mac_address(slave_dev, &addr);
1790 dev_set_mtu(slave_dev, slave->original_mtu);
1792 slave_dev->priv_flags &= ~IFF_BONDING;
1796 return 0; /* deletion OK */
1799 /* A wrapper used because of ndo_del_link */
1800 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1802 return __bond_release_one(bond_dev, slave_dev, false);
1806 * First release a slave and then destroy the bond if no more slaves are left.
1807 * Must be under rtnl_lock when this function is called.
1809 static int bond_release_and_destroy(struct net_device *bond_dev,
1810 struct net_device *slave_dev)
1812 struct bonding *bond = netdev_priv(bond_dev);
1815 ret = bond_release(bond_dev, slave_dev);
1816 if (ret == 0 && !bond_has_slaves(bond)) {
1817 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1818 pr_info("%s: destroying bond %s.\n",
1819 bond_dev->name, bond_dev->name);
1820 unregister_netdevice(bond_dev);
1825 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1827 struct bonding *bond = netdev_priv(bond_dev);
1829 info->bond_mode = bond->params.mode;
1830 info->miimon = bond->params.miimon;
1832 read_lock(&bond->lock);
1833 info->num_slaves = bond->slave_cnt;
1834 read_unlock(&bond->lock);
1839 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1841 struct bonding *bond = netdev_priv(bond_dev);
1842 struct list_head *iter;
1843 int i = 0, res = -ENODEV;
1844 struct slave *slave;
1846 read_lock(&bond->lock);
1847 bond_for_each_slave(bond, slave, iter) {
1848 if (i++ == (int)info->slave_id) {
1850 strcpy(info->slave_name, slave->dev->name);
1851 info->link = slave->link;
1852 info->state = bond_slave_state(slave);
1853 info->link_failure_count = slave->link_failure_count;
1857 read_unlock(&bond->lock);
1862 /*-------------------------------- Monitoring -------------------------------*/
1865 static int bond_miimon_inspect(struct bonding *bond)
1867 int link_state, commit = 0;
1868 struct list_head *iter;
1869 struct slave *slave;
1870 bool ignore_updelay;
1872 ignore_updelay = !bond->curr_active_slave ? true : false;
1874 bond_for_each_slave_rcu(bond, slave, iter) {
1875 slave->new_link = BOND_LINK_NOCHANGE;
1877 link_state = bond_check_dev_link(bond, slave->dev, 0);
1879 switch (slave->link) {
1884 slave->link = BOND_LINK_FAIL;
1885 slave->delay = bond->params.downdelay;
1887 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1889 (bond->params.mode ==
1890 BOND_MODE_ACTIVEBACKUP) ?
1891 (bond_is_active_slave(slave) ?
1892 "active " : "backup ") : "",
1894 bond->params.downdelay * bond->params.miimon);
1897 case BOND_LINK_FAIL:
1900 * recovered before downdelay expired
1902 slave->link = BOND_LINK_UP;
1903 slave->jiffies = jiffies;
1904 pr_info("%s: link status up again after %d ms for interface %s.\n",
1906 (bond->params.downdelay - slave->delay) *
1907 bond->params.miimon,
1912 if (slave->delay <= 0) {
1913 slave->new_link = BOND_LINK_DOWN;
1921 case BOND_LINK_DOWN:
1925 slave->link = BOND_LINK_BACK;
1926 slave->delay = bond->params.updelay;
1929 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1930 bond->dev->name, slave->dev->name,
1931 ignore_updelay ? 0 :
1932 bond->params.updelay *
1933 bond->params.miimon);
1936 case BOND_LINK_BACK:
1938 slave->link = BOND_LINK_DOWN;
1939 pr_info("%s: link status down again after %d ms for interface %s.\n",
1941 (bond->params.updelay - slave->delay) *
1942 bond->params.miimon,
1951 if (slave->delay <= 0) {
1952 slave->new_link = BOND_LINK_UP;
1954 ignore_updelay = false;
1966 static void bond_miimon_commit(struct bonding *bond)
1968 struct list_head *iter;
1969 struct slave *slave;
1971 bond_for_each_slave(bond, slave, iter) {
1972 switch (slave->new_link) {
1973 case BOND_LINK_NOCHANGE:
1977 slave->link = BOND_LINK_UP;
1978 slave->jiffies = jiffies;
1980 if (bond->params.mode == BOND_MODE_8023AD) {
1981 /* prevent it from being the active one */
1982 bond_set_backup_slave(slave);
1983 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1984 /* make it immediately active */
1985 bond_set_active_slave(slave);
1986 } else if (slave != bond->primary_slave) {
1987 /* prevent it from being the active one */
1988 bond_set_backup_slave(slave);
1991 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1992 bond->dev->name, slave->dev->name,
1993 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1994 slave->duplex ? "full" : "half");
1996 /* notify ad that the link status has changed */
1997 if (bond->params.mode == BOND_MODE_8023AD)
1998 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2000 if (bond_is_lb(bond))
2001 bond_alb_handle_link_change(bond, slave,
2004 if (!bond->curr_active_slave ||
2005 (slave == bond->primary_slave))
2010 case BOND_LINK_DOWN:
2011 if (slave->link_failure_count < UINT_MAX)
2012 slave->link_failure_count++;
2014 slave->link = BOND_LINK_DOWN;
2016 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2017 bond->params.mode == BOND_MODE_8023AD)
2018 bond_set_slave_inactive_flags(slave);
2020 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2021 bond->dev->name, slave->dev->name);
2023 if (bond->params.mode == BOND_MODE_8023AD)
2024 bond_3ad_handle_link_change(slave,
2027 if (bond_is_lb(bond))
2028 bond_alb_handle_link_change(bond, slave,
2031 if (slave == bond->curr_active_slave)
2037 pr_err("%s: invalid new link %d on slave %s\n",
2038 bond->dev->name, slave->new_link,
2040 slave->new_link = BOND_LINK_NOCHANGE;
2048 write_lock_bh(&bond->curr_slave_lock);
2049 bond_select_active_slave(bond);
2050 write_unlock_bh(&bond->curr_slave_lock);
2051 unblock_netpoll_tx();
2054 bond_set_carrier(bond);
2060 * Really a wrapper that splits the mii monitor into two phases: an
2061 * inspection, then (if inspection indicates something needs to be done)
2062 * an acquisition of appropriate locks followed by a commit phase to
2063 * implement whatever link state changes are indicated.
2065 static void bond_mii_monitor(struct work_struct *work)
2067 struct bonding *bond = container_of(work, struct bonding,
2069 bool should_notify_peers = false;
2070 unsigned long delay;
2072 delay = msecs_to_jiffies(bond->params.miimon);
2074 if (!bond_has_slaves(bond))
2079 should_notify_peers = bond_should_notify_peers(bond);
2081 if (bond_miimon_inspect(bond)) {
2084 /* Race avoidance with bond_close cancel of workqueue */
2085 if (!rtnl_trylock()) {
2087 should_notify_peers = false;
2091 bond_miimon_commit(bond);
2093 rtnl_unlock(); /* might sleep, hold no other locks */
2098 if (bond->params.miimon)
2099 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2101 if (should_notify_peers) {
2102 if (!rtnl_trylock())
2104 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2109 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2111 struct net_device *upper;
2112 struct list_head *iter;
2115 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2119 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2120 if (ip == bond_confirm_addr(upper, 0, ip)) {
2131 * We go to the (large) trouble of VLAN tagging ARP frames because
2132 * switches in VLAN mode (especially if ports are configured as
2133 * "native" to a VLAN) might not pass non-tagged frames.
2135 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2137 struct sk_buff *skb;
2139 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2140 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2142 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2143 NULL, slave_dev->dev_addr, NULL);
2146 pr_err("ARP packet allocation failed\n");
2150 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2152 pr_err("failed to insert VLAN tag\n");
2160 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2162 struct net_device *upper, *vlan_upper;
2163 struct list_head *iter, *vlan_iter;
2165 __be32 *targets = bond->params.arp_targets, addr;
2168 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2169 pr_debug("basa: target %pI4\n", &targets[i]);
2171 /* Find out through which dev should the packet go */
2172 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2175 pr_debug("%s: no route to arp_ip_target %pI4\n",
2176 bond->dev->name, &targets[i]);
2182 /* bond device itself */
2183 if (rt->dst.dev == bond->dev)
2187 /* first we search only for vlan devices. for every vlan
2188 * found we verify its upper dev list, searching for the
2189 * rt->dst.dev. If found we save the tag of the vlan and
2190 * proceed to send the packet.
2194 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2196 if (!is_vlan_dev(vlan_upper))
2198 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2200 if (upper == rt->dst.dev) {
2201 vlan_id = vlan_dev_vlan_id(vlan_upper);
2208 /* if the device we're looking for is not on top of any of
2209 * our upper vlans, then just search for any dev that
2210 * matches, and in case it's a vlan - save the id
2212 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2213 if (upper == rt->dst.dev) {
2214 /* if it's a vlan - get its VID */
2215 if (is_vlan_dev(upper))
2216 vlan_id = vlan_dev_vlan_id(upper);
2224 /* Not our device - skip */
2225 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2226 bond->dev->name, &targets[i],
2227 rt->dst.dev ? rt->dst.dev->name : "NULL");
2233 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2235 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2240 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2244 if (!sip || !bond_has_this_ip(bond, tip)) {
2245 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2249 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2251 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2254 slave->last_arp_rx = jiffies;
2255 slave->target_last_arp_rx[i] = jiffies;
2258 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2259 struct slave *slave)
2261 struct arphdr *arp = (struct arphdr *)skb->data;
2262 unsigned char *arp_ptr;
2266 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2267 return RX_HANDLER_ANOTHER;
2269 read_lock(&bond->lock);
2271 if (!slave_do_arp_validate(bond, slave))
2274 alen = arp_hdr_len(bond->dev);
2276 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2277 bond->dev->name, skb->dev->name);
2279 if (alen > skb_headlen(skb)) {
2280 arp = kmalloc(alen, GFP_ATOMIC);
2283 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2287 if (arp->ar_hln != bond->dev->addr_len ||
2288 skb->pkt_type == PACKET_OTHERHOST ||
2289 skb->pkt_type == PACKET_LOOPBACK ||
2290 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2291 arp->ar_pro != htons(ETH_P_IP) ||
2295 arp_ptr = (unsigned char *)(arp + 1);
2296 arp_ptr += bond->dev->addr_len;
2297 memcpy(&sip, arp_ptr, 4);
2298 arp_ptr += 4 + bond->dev->addr_len;
2299 memcpy(&tip, arp_ptr, 4);
2301 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2302 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2303 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2307 * Backup slaves won't see the ARP reply, but do come through
2308 * here for each ARP probe (so we swap the sip/tip to validate
2309 * the probe). In a "redundant switch, common router" type of
2310 * configuration, the ARP probe will (hopefully) travel from
2311 * the active, through one switch, the router, then the other
2312 * switch before reaching the backup.
2314 * We 'trust' the arp requests if there is an active slave and
2315 * it received valid arp reply(s) after it became active. This
2316 * is done to avoid endless looping when we can't reach the
2317 * arp_ip_target and fool ourselves with our own arp requests.
2319 if (bond_is_active_slave(slave))
2320 bond_validate_arp(bond, slave, sip, tip);
2321 else if (bond->curr_active_slave &&
2322 time_after(slave_last_rx(bond, bond->curr_active_slave),
2323 bond->curr_active_slave->jiffies))
2324 bond_validate_arp(bond, slave, tip, sip);
2327 read_unlock(&bond->lock);
2328 if (arp != (struct arphdr *)skb->data)
2330 return RX_HANDLER_ANOTHER;
2333 /* function to verify if we're in the arp_interval timeslice, returns true if
2334 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2335 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2337 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2340 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2342 return time_in_range(jiffies,
2343 last_act - delta_in_ticks,
2344 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2348 * this function is called regularly to monitor each slave's link
2349 * ensuring that traffic is being sent and received when arp monitoring
2350 * is used in load-balancing mode. if the adapter has been dormant, then an
2351 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2352 * arp monitoring in active backup mode.
2354 static void bond_loadbalance_arp_mon(struct work_struct *work)
2356 struct bonding *bond = container_of(work, struct bonding,
2358 struct slave *slave, *oldcurrent;
2359 struct list_head *iter;
2360 int do_failover = 0, slave_state_changed = 0;
2362 if (!bond_has_slaves(bond))
2367 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2368 /* see if any of the previous devices are up now (i.e. they have
2369 * xmt and rcv traffic). the curr_active_slave does not come into
2370 * the picture unless it is null. also, slave->jiffies is not needed
2371 * here because we send an arp on each slave and give a slave as
2372 * long as it needs to get the tx/rx within the delta.
2373 * TODO: what about up/down delay in arp mode? it wasn't here before
2376 bond_for_each_slave_rcu(bond, slave, iter) {
2377 unsigned long trans_start = dev_trans_start(slave->dev);
2379 if (slave->link != BOND_LINK_UP) {
2380 if (bond_time_in_interval(bond, trans_start, 1) &&
2381 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2383 slave->link = BOND_LINK_UP;
2384 slave_state_changed = 1;
2386 /* primary_slave has no meaning in round-robin
2387 * mode. the window of a slave being up and
2388 * curr_active_slave being null after enslaving
2392 pr_info("%s: link status definitely up for interface %s, ",
2397 pr_info("%s: interface %s is now up\n",
2403 /* slave->link == BOND_LINK_UP */
2405 /* not all switches will respond to an arp request
2406 * when the source ip is 0, so don't take the link down
2407 * if we don't know our ip yet
2409 if (!bond_time_in_interval(bond, trans_start, 2) ||
2410 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2412 slave->link = BOND_LINK_DOWN;
2413 slave_state_changed = 1;
2415 if (slave->link_failure_count < UINT_MAX)
2416 slave->link_failure_count++;
2418 pr_info("%s: interface %s is now down.\n",
2422 if (slave == oldcurrent)
2427 /* note: if switch is in round-robin mode, all links
2428 * must tx arp to ensure all links rx an arp - otherwise
2429 * links may oscillate or not come up at all; if switch is
2430 * in something like xor mode, there is nothing we can
2431 * do - all replies will be rx'ed on same link causing slaves
2432 * to be unstable during low/no traffic periods
2434 if (IS_UP(slave->dev))
2435 bond_arp_send_all(bond, slave);
2440 if (do_failover || slave_state_changed) {
2441 if (!rtnl_trylock())
2444 if (slave_state_changed) {
2445 bond_slave_state_change(bond);
2446 } else if (do_failover) {
2447 /* the bond_select_active_slave must hold RTNL
2448 * and curr_slave_lock for write.
2451 write_lock_bh(&bond->curr_slave_lock);
2453 bond_select_active_slave(bond);
2455 write_unlock_bh(&bond->curr_slave_lock);
2456 unblock_netpoll_tx();
2462 if (bond->params.arp_interval)
2463 queue_delayed_work(bond->wq, &bond->arp_work,
2464 msecs_to_jiffies(bond->params.arp_interval));
2468 * Called to inspect slaves for active-backup mode ARP monitor link state
2469 * changes. Sets new_link in slaves to specify what action should take
2470 * place for the slave. Returns 0 if no changes are found, >0 if changes
2471 * to link states must be committed.
2473 * Called with rcu_read_lock hold.
2475 static int bond_ab_arp_inspect(struct bonding *bond)
2477 unsigned long trans_start, last_rx;
2478 struct list_head *iter;
2479 struct slave *slave;
2482 bond_for_each_slave_rcu(bond, slave, iter) {
2483 slave->new_link = BOND_LINK_NOCHANGE;
2484 last_rx = slave_last_rx(bond, slave);
2486 if (slave->link != BOND_LINK_UP) {
2487 if (bond_time_in_interval(bond, last_rx, 1)) {
2488 slave->new_link = BOND_LINK_UP;
2495 * Give slaves 2*delta after being enslaved or made
2496 * active. This avoids bouncing, as the last receive
2497 * times need a full ARP monitor cycle to be updated.
2499 if (bond_time_in_interval(bond, slave->jiffies, 2))
2503 * Backup slave is down if:
2504 * - No current_arp_slave AND
2505 * - more than 3*delta since last receive AND
2506 * - the bond has an IP address
2508 * Note: a non-null current_arp_slave indicates
2509 * the curr_active_slave went down and we are
2510 * searching for a new one; under this condition
2511 * we only take the curr_active_slave down - this
2512 * gives each slave a chance to tx/rx traffic
2513 * before being taken out
2515 if (!bond_is_active_slave(slave) &&
2516 !bond->current_arp_slave &&
2517 !bond_time_in_interval(bond, last_rx, 3)) {
2518 slave->new_link = BOND_LINK_DOWN;
2523 * Active slave is down if:
2524 * - more than 2*delta since transmitting OR
2525 * - (more than 2*delta since receive AND
2526 * the bond has an IP address)
2528 trans_start = dev_trans_start(slave->dev);
2529 if (bond_is_active_slave(slave) &&
2530 (!bond_time_in_interval(bond, trans_start, 2) ||
2531 !bond_time_in_interval(bond, last_rx, 2))) {
2532 slave->new_link = BOND_LINK_DOWN;
2541 * Called to commit link state changes noted by inspection step of
2542 * active-backup mode ARP monitor.
2544 * Called with RTNL hold.
2546 static void bond_ab_arp_commit(struct bonding *bond)
2548 unsigned long trans_start;
2549 struct list_head *iter;
2550 struct slave *slave;
2552 bond_for_each_slave(bond, slave, iter) {
2553 switch (slave->new_link) {
2554 case BOND_LINK_NOCHANGE:
2558 trans_start = dev_trans_start(slave->dev);
2559 if (bond->curr_active_slave != slave ||
2560 (!bond->curr_active_slave &&
2561 bond_time_in_interval(bond, trans_start, 1))) {
2562 slave->link = BOND_LINK_UP;
2563 if (bond->current_arp_slave) {
2564 bond_set_slave_inactive_flags(
2565 bond->current_arp_slave);
2566 bond->current_arp_slave = NULL;
2569 pr_info("%s: link status definitely up for interface %s.\n",
2570 bond->dev->name, slave->dev->name);
2572 if (!bond->curr_active_slave ||
2573 (slave == bond->primary_slave))
2580 case BOND_LINK_DOWN:
2581 if (slave->link_failure_count < UINT_MAX)
2582 slave->link_failure_count++;
2584 slave->link = BOND_LINK_DOWN;
2585 bond_set_slave_inactive_flags(slave);
2587 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2588 bond->dev->name, slave->dev->name);
2590 if (slave == bond->curr_active_slave) {
2591 bond->current_arp_slave = NULL;
2598 pr_err("%s: impossible: new_link %d on slave %s\n",
2599 bond->dev->name, slave->new_link,
2607 write_lock_bh(&bond->curr_slave_lock);
2608 bond_select_active_slave(bond);
2609 write_unlock_bh(&bond->curr_slave_lock);
2610 unblock_netpoll_tx();
2613 bond_set_carrier(bond);
2617 * Send ARP probes for active-backup mode ARP monitor.
2619 static bool bond_ab_arp_probe(struct bonding *bond)
2621 struct slave *slave, *before = NULL, *new_slave = NULL,
2622 *curr_arp_slave, *curr_active_slave;
2623 struct list_head *iter;
2627 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2628 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2630 if (curr_arp_slave && curr_active_slave)
2631 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2632 curr_arp_slave->dev->name,
2633 curr_active_slave->dev->name);
2635 if (curr_active_slave) {
2636 bond_arp_send_all(bond, curr_active_slave);
2642 /* if we don't have a curr_active_slave, search for the next available
2643 * backup slave from the current_arp_slave and make it the candidate
2644 * for becoming the curr_active_slave
2647 if (!rtnl_trylock())
2649 /* curr_arp_slave might have gone away */
2650 curr_arp_slave = ACCESS_ONCE(bond->current_arp_slave);
2652 if (!curr_arp_slave) {
2653 curr_arp_slave = bond_first_slave(bond);
2654 if (!curr_arp_slave) {
2660 bond_set_slave_inactive_flags(curr_arp_slave);
2662 bond_for_each_slave(bond, slave, iter) {
2663 if (!found && !before && IS_UP(slave->dev))
2666 if (found && !new_slave && IS_UP(slave->dev))
2668 /* if the link state is up at this point, we
2669 * mark it down - this can happen if we have
2670 * simultaneous link failures and
2671 * reselect_active_interface doesn't make this
2672 * one the current slave so it is still marked
2673 * up when it is actually down
2675 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2676 slave->link = BOND_LINK_DOWN;
2677 if (slave->link_failure_count < UINT_MAX)
2678 slave->link_failure_count++;
2680 bond_set_slave_inactive_flags(slave);
2682 pr_info("%s: backup interface %s is now down.\n",
2683 bond->dev->name, slave->dev->name);
2685 if (slave == curr_arp_slave)
2689 if (!new_slave && before)
2697 new_slave->link = BOND_LINK_BACK;
2698 bond_set_slave_active_flags(new_slave);
2699 bond_arp_send_all(bond, new_slave);
2700 new_slave->jiffies = jiffies;
2701 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2707 static void bond_activebackup_arp_mon(struct work_struct *work)
2709 struct bonding *bond = container_of(work, struct bonding,
2711 bool should_notify_peers = false, should_commit = false;
2714 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2716 if (!bond_has_slaves(bond))
2720 should_notify_peers = bond_should_notify_peers(bond);
2721 should_commit = bond_ab_arp_inspect(bond);
2724 if (should_commit) {
2725 /* Race avoidance with bond_close flush of workqueue */
2726 if (!rtnl_trylock()) {
2728 should_notify_peers = false;
2732 bond_ab_arp_commit(bond);
2736 if (!bond_ab_arp_probe(bond)) {
2737 /* rtnl locking failed, re-arm */
2739 should_notify_peers = false;
2743 if (bond->params.arp_interval)
2744 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2746 if (should_notify_peers) {
2747 if (!rtnl_trylock())
2749 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2754 /*-------------------------- netdev event handling --------------------------*/
2757 * Change device name
2759 static int bond_event_changename(struct bonding *bond)
2761 bond_remove_proc_entry(bond);
2762 bond_create_proc_entry(bond);
2764 bond_debug_reregister(bond);
2769 static int bond_master_netdev_event(unsigned long event,
2770 struct net_device *bond_dev)
2772 struct bonding *event_bond = netdev_priv(bond_dev);
2775 case NETDEV_CHANGENAME:
2776 return bond_event_changename(event_bond);
2777 case NETDEV_UNREGISTER:
2778 bond_remove_proc_entry(event_bond);
2780 case NETDEV_REGISTER:
2781 bond_create_proc_entry(event_bond);
2783 case NETDEV_NOTIFY_PEERS:
2784 if (event_bond->send_peer_notif)
2785 event_bond->send_peer_notif--;
2794 static int bond_slave_netdev_event(unsigned long event,
2795 struct net_device *slave_dev)
2797 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2798 struct bonding *bond;
2799 struct net_device *bond_dev;
2803 /* A netdev event can be generated while enslaving a device
2804 * before netdev_rx_handler_register is called in which case
2805 * slave will be NULL
2809 bond_dev = slave->bond->dev;
2813 case NETDEV_UNREGISTER:
2814 if (bond_dev->type != ARPHRD_ETHER)
2815 bond_release_and_destroy(bond_dev, slave_dev);
2817 bond_release(bond_dev, slave_dev);
2821 old_speed = slave->speed;
2822 old_duplex = slave->duplex;
2824 bond_update_speed_duplex(slave);
2826 if (bond->params.mode == BOND_MODE_8023AD) {
2827 if (old_speed != slave->speed)
2828 bond_3ad_adapter_speed_changed(slave);
2829 if (old_duplex != slave->duplex)
2830 bond_3ad_adapter_duplex_changed(slave);
2835 * ... Or is it this?
2838 case NETDEV_CHANGEMTU:
2840 * TODO: Should slaves be allowed to
2841 * independently alter their MTU? For
2842 * an active-backup bond, slaves need
2843 * not be the same type of device, so
2844 * MTUs may vary. For other modes,
2845 * slaves arguably should have the
2846 * same MTUs. To do this, we'd need to
2847 * take over the slave's change_mtu
2848 * function for the duration of their
2852 case NETDEV_CHANGENAME:
2853 /* we don't care if we don't have primary set */
2854 if (!USES_PRIMARY(bond->params.mode) ||
2855 !bond->params.primary[0])
2858 if (slave == bond->primary_slave) {
2859 /* slave's name changed - he's no longer primary */
2860 bond->primary_slave = NULL;
2861 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2862 /* we have a new primary slave */
2863 bond->primary_slave = slave;
2864 } else { /* we didn't change primary - exit */
2868 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2869 bond->dev->name, bond->primary_slave ? slave_dev->name :
2873 write_lock_bh(&bond->curr_slave_lock);
2874 bond_select_active_slave(bond);
2875 write_unlock_bh(&bond->curr_slave_lock);
2876 unblock_netpoll_tx();
2878 case NETDEV_FEAT_CHANGE:
2879 bond_compute_features(bond);
2881 case NETDEV_RESEND_IGMP:
2882 /* Propagate to master device */
2883 call_netdevice_notifiers(event, slave->bond->dev);
2893 * bond_netdev_event: handle netdev notifier chain events.
2895 * This function receives events for the netdev chain. The caller (an
2896 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2897 * locks for us to safely manipulate the slave devices (RTNL lock,
2900 static int bond_netdev_event(struct notifier_block *this,
2901 unsigned long event, void *ptr)
2903 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2905 pr_debug("event_dev: %s, event: %lx\n",
2906 event_dev ? event_dev->name : "None",
2909 if (!(event_dev->priv_flags & IFF_BONDING))
2912 if (event_dev->flags & IFF_MASTER) {
2913 pr_debug("IFF_MASTER\n");
2914 return bond_master_netdev_event(event, event_dev);
2917 if (event_dev->flags & IFF_SLAVE) {
2918 pr_debug("IFF_SLAVE\n");
2919 return bond_slave_netdev_event(event, event_dev);
2925 static struct notifier_block bond_netdev_notifier = {
2926 .notifier_call = bond_netdev_event,
2929 /*---------------------------- Hashing Policies -----------------------------*/
2931 /* L2 hash helper */
2932 static inline u32 bond_eth_hash(struct sk_buff *skb)
2934 struct ethhdr *data = (struct ethhdr *)skb->data;
2936 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2937 return data->h_dest[5] ^ data->h_source[5];
2942 /* Extract the appropriate headers based on bond's xmit policy */
2943 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2944 struct flow_keys *fk)
2946 const struct ipv6hdr *iph6;
2947 const struct iphdr *iph;
2948 int noff, proto = -1;
2950 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2951 return skb_flow_dissect(skb, fk);
2954 noff = skb_network_offset(skb);
2955 if (skb->protocol == htons(ETH_P_IP)) {
2956 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2959 fk->src = iph->saddr;
2960 fk->dst = iph->daddr;
2961 noff += iph->ihl << 2;
2962 if (!ip_is_fragment(iph))
2963 proto = iph->protocol;
2964 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2965 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2967 iph6 = ipv6_hdr(skb);
2968 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2969 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2970 noff += sizeof(*iph6);
2971 proto = iph6->nexthdr;
2975 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2976 fk->ports = skb_flow_get_ports(skb, noff, proto);
2982 * bond_xmit_hash - generate a hash value based on the xmit policy
2983 * @bond: bonding device
2984 * @skb: buffer to use for headers
2985 * @count: modulo value
2987 * This function will extract the necessary headers from the skb buffer and use
2988 * them to generate a hash based on the xmit_policy set in the bonding device
2989 * which will be reduced modulo count before returning.
2991 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2993 struct flow_keys flow;
2996 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2997 !bond_flow_dissect(bond, skb, &flow))
2998 return bond_eth_hash(skb) % count;
3000 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3001 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3002 hash = bond_eth_hash(skb);
3004 hash = (__force u32)flow.ports;
3005 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3006 hash ^= (hash >> 16);
3007 hash ^= (hash >> 8);
3009 return hash % count;
3012 /*-------------------------- Device entry points ----------------------------*/
3014 static void bond_work_init_all(struct bonding *bond)
3016 INIT_DELAYED_WORK(&bond->mcast_work,
3017 bond_resend_igmp_join_requests_delayed);
3018 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3019 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3020 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3021 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3023 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3024 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3027 static void bond_work_cancel_all(struct bonding *bond)
3029 cancel_delayed_work_sync(&bond->mii_work);
3030 cancel_delayed_work_sync(&bond->arp_work);
3031 cancel_delayed_work_sync(&bond->alb_work);
3032 cancel_delayed_work_sync(&bond->ad_work);
3033 cancel_delayed_work_sync(&bond->mcast_work);
3036 static int bond_open(struct net_device *bond_dev)
3038 struct bonding *bond = netdev_priv(bond_dev);
3039 struct list_head *iter;
3040 struct slave *slave;
3042 /* reset slave->backup and slave->inactive */
3043 read_lock(&bond->lock);
3044 if (bond_has_slaves(bond)) {
3045 read_lock(&bond->curr_slave_lock);
3046 bond_for_each_slave(bond, slave, iter) {
3047 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3048 && (slave != bond->curr_active_slave)) {
3049 bond_set_slave_inactive_flags(slave);
3051 bond_set_slave_active_flags(slave);
3054 read_unlock(&bond->curr_slave_lock);
3056 read_unlock(&bond->lock);
3058 bond_work_init_all(bond);
3060 if (bond_is_lb(bond)) {
3061 /* bond_alb_initialize must be called before the timer
3064 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3066 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3069 if (bond->params.miimon) /* link check interval, in milliseconds. */
3070 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3072 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3073 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3074 if (bond->params.arp_validate)
3075 bond->recv_probe = bond_arp_rcv;
3078 if (bond->params.mode == BOND_MODE_8023AD) {
3079 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3080 /* register to receive LACPDUs */
3081 bond->recv_probe = bond_3ad_lacpdu_recv;
3082 bond_3ad_initiate_agg_selection(bond, 1);
3088 static int bond_close(struct net_device *bond_dev)
3090 struct bonding *bond = netdev_priv(bond_dev);
3092 bond_work_cancel_all(bond);
3093 bond->send_peer_notif = 0;
3094 if (bond_is_lb(bond))
3095 bond_alb_deinitialize(bond);
3096 bond->recv_probe = NULL;
3101 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3102 struct rtnl_link_stats64 *stats)
3104 struct bonding *bond = netdev_priv(bond_dev);
3105 struct rtnl_link_stats64 temp;
3106 struct list_head *iter;
3107 struct slave *slave;
3109 memset(stats, 0, sizeof(*stats));
3111 read_lock_bh(&bond->lock);
3112 bond_for_each_slave(bond, slave, iter) {
3113 const struct rtnl_link_stats64 *sstats =
3114 dev_get_stats(slave->dev, &temp);
3116 stats->rx_packets += sstats->rx_packets;
3117 stats->rx_bytes += sstats->rx_bytes;
3118 stats->rx_errors += sstats->rx_errors;
3119 stats->rx_dropped += sstats->rx_dropped;
3121 stats->tx_packets += sstats->tx_packets;
3122 stats->tx_bytes += sstats->tx_bytes;
3123 stats->tx_errors += sstats->tx_errors;
3124 stats->tx_dropped += sstats->tx_dropped;
3126 stats->multicast += sstats->multicast;
3127 stats->collisions += sstats->collisions;
3129 stats->rx_length_errors += sstats->rx_length_errors;
3130 stats->rx_over_errors += sstats->rx_over_errors;
3131 stats->rx_crc_errors += sstats->rx_crc_errors;
3132 stats->rx_frame_errors += sstats->rx_frame_errors;
3133 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3134 stats->rx_missed_errors += sstats->rx_missed_errors;
3136 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3137 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3138 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3139 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3140 stats->tx_window_errors += sstats->tx_window_errors;
3142 read_unlock_bh(&bond->lock);
3147 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3149 struct bonding *bond = netdev_priv(bond_dev);
3150 struct net_device *slave_dev = NULL;
3151 struct ifbond k_binfo;
3152 struct ifbond __user *u_binfo = NULL;
3153 struct ifslave k_sinfo;
3154 struct ifslave __user *u_sinfo = NULL;
3155 struct mii_ioctl_data *mii = NULL;
3156 struct bond_opt_value newval;
3160 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3172 * We do this again just in case we were called by SIOCGMIIREG
3173 * instead of SIOCGMIIPHY.
3180 if (mii->reg_num == 1) {
3182 read_lock(&bond->lock);
3183 read_lock(&bond->curr_slave_lock);
3184 if (netif_carrier_ok(bond->dev))
3185 mii->val_out = BMSR_LSTATUS;
3187 read_unlock(&bond->curr_slave_lock);
3188 read_unlock(&bond->lock);
3192 case BOND_INFO_QUERY_OLD:
3193 case SIOCBONDINFOQUERY:
3194 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3196 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3199 res = bond_info_query(bond_dev, &k_binfo);
3201 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3205 case BOND_SLAVE_INFO_QUERY_OLD:
3206 case SIOCBONDSLAVEINFOQUERY:
3207 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3209 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3212 res = bond_slave_info_query(bond_dev, &k_sinfo);
3214 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3223 net = dev_net(bond_dev);
3225 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3228 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3230 pr_debug("slave_dev=%p:\n", slave_dev);
3235 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3237 case BOND_ENSLAVE_OLD:
3238 case SIOCBONDENSLAVE:
3239 res = bond_enslave(bond_dev, slave_dev);
3241 case BOND_RELEASE_OLD:
3242 case SIOCBONDRELEASE:
3243 res = bond_release(bond_dev, slave_dev);
3245 case BOND_SETHWADDR_OLD:
3246 case SIOCBONDSETHWADDR:
3247 bond_set_dev_addr(bond_dev, slave_dev);
3250 case BOND_CHANGE_ACTIVE_OLD:
3251 case SIOCBONDCHANGEACTIVE:
3252 bond_opt_initstr(&newval, slave_dev->name);
3253 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3262 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3264 struct bonding *bond = netdev_priv(bond_dev);
3266 if (change & IFF_PROMISC)
3267 bond_set_promiscuity(bond,
3268 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3270 if (change & IFF_ALLMULTI)
3271 bond_set_allmulti(bond,
3272 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3275 static void bond_set_rx_mode(struct net_device *bond_dev)
3277 struct bonding *bond = netdev_priv(bond_dev);
3278 struct list_head *iter;
3279 struct slave *slave;
3283 if (USES_PRIMARY(bond->params.mode)) {
3284 slave = rcu_dereference(bond->curr_active_slave);
3286 dev_uc_sync(slave->dev, bond_dev);
3287 dev_mc_sync(slave->dev, bond_dev);
3290 bond_for_each_slave_rcu(bond, slave, iter) {
3291 dev_uc_sync_multiple(slave->dev, bond_dev);
3292 dev_mc_sync_multiple(slave->dev, bond_dev);
3298 static int bond_neigh_init(struct neighbour *n)
3300 struct bonding *bond = netdev_priv(n->dev);
3301 const struct net_device_ops *slave_ops;
3302 struct neigh_parms parms;
3303 struct slave *slave;
3306 slave = bond_first_slave(bond);
3309 slave_ops = slave->dev->netdev_ops;
3310 if (!slave_ops->ndo_neigh_setup)
3313 parms.neigh_setup = NULL;
3314 parms.neigh_cleanup = NULL;
3315 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3320 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3321 * after the last slave has been detached. Assumes that all slaves
3322 * utilize the same neigh_cleanup (true at this writing as only user
3325 n->parms->neigh_cleanup = parms.neigh_cleanup;
3327 if (!parms.neigh_setup)
3330 return parms.neigh_setup(n);
3334 * The bonding ndo_neigh_setup is called at init time beofre any
3335 * slave exists. So we must declare proxy setup function which will
3336 * be used at run time to resolve the actual slave neigh param setup.
3338 * It's also called by master devices (such as vlans) to setup their
3339 * underlying devices. In that case - do nothing, we're already set up from
3342 static int bond_neigh_setup(struct net_device *dev,
3343 struct neigh_parms *parms)
3345 /* modify only our neigh_parms */
3346 if (parms->dev == dev)
3347 parms->neigh_setup = bond_neigh_init;
3353 * Change the MTU of all of a master's slaves to match the master
3355 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3357 struct bonding *bond = netdev_priv(bond_dev);
3358 struct slave *slave, *rollback_slave;
3359 struct list_head *iter;
3362 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3363 (bond_dev ? bond_dev->name : "None"), new_mtu);
3365 /* Can't hold bond->lock with bh disabled here since
3366 * some base drivers panic. On the other hand we can't
3367 * hold bond->lock without bh disabled because we'll
3368 * deadlock. The only solution is to rely on the fact
3369 * that we're under rtnl_lock here, and the slaves
3370 * list won't change. This doesn't solve the problem
3371 * of setting the slave's MTU while it is
3372 * transmitting, but the assumption is that the base
3373 * driver can handle that.
3375 * TODO: figure out a way to safely iterate the slaves
3376 * list, but without holding a lock around the actual
3377 * call to the base driver.
3380 bond_for_each_slave(bond, slave, iter) {
3381 pr_debug("s %p c_m %p\n",
3383 slave->dev->netdev_ops->ndo_change_mtu);
3385 res = dev_set_mtu(slave->dev, new_mtu);
3388 /* If we failed to set the slave's mtu to the new value
3389 * we must abort the operation even in ACTIVE_BACKUP
3390 * mode, because if we allow the backup slaves to have
3391 * different mtu values than the active slave we'll
3392 * need to change their mtu when doing a failover. That
3393 * means changing their mtu from timer context, which
3394 * is probably not a good idea.
3396 pr_debug("err %d %s\n", res, slave->dev->name);
3401 bond_dev->mtu = new_mtu;
3406 /* unwind from head to the slave that failed */
3407 bond_for_each_slave(bond, rollback_slave, iter) {
3410 if (rollback_slave == slave)
3413 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3415 pr_debug("unwind err %d dev %s\n",
3416 tmp_res, rollback_slave->dev->name);
3426 * Note that many devices must be down to change the HW address, and
3427 * downing the master releases all slaves. We can make bonds full of
3428 * bonding devices to test this, however.
3430 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3432 struct bonding *bond = netdev_priv(bond_dev);
3433 struct slave *slave, *rollback_slave;
3434 struct sockaddr *sa = addr, tmp_sa;
3435 struct list_head *iter;
3438 if (bond->params.mode == BOND_MODE_ALB)
3439 return bond_alb_set_mac_address(bond_dev, addr);
3442 pr_debug("bond=%p, name=%s\n",
3443 bond, bond_dev ? bond_dev->name : "None");
3445 /* If fail_over_mac is enabled, do nothing and return success.
3446 * Returning an error causes ifenslave to fail.
3448 if (bond->params.fail_over_mac &&
3449 bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3452 if (!is_valid_ether_addr(sa->sa_data))
3453 return -EADDRNOTAVAIL;
3455 /* Can't hold bond->lock with bh disabled here since
3456 * some base drivers panic. On the other hand we can't
3457 * hold bond->lock without bh disabled because we'll
3458 * deadlock. The only solution is to rely on the fact
3459 * that we're under rtnl_lock here, and the slaves
3460 * list won't change. This doesn't solve the problem
3461 * of setting the slave's hw address while it is
3462 * transmitting, but the assumption is that the base
3463 * driver can handle that.
3465 * TODO: figure out a way to safely iterate the slaves
3466 * list, but without holding a lock around the actual
3467 * call to the base driver.
3470 bond_for_each_slave(bond, slave, iter) {
3471 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3472 pr_debug("slave %p %s\n", slave, slave->dev->name);
3474 if (slave_ops->ndo_set_mac_address == NULL) {
3476 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3480 res = dev_set_mac_address(slave->dev, addr);
3482 /* TODO: consider downing the slave
3484 * User should expect communications
3485 * breakage anyway until ARP finish
3488 pr_debug("err %d %s\n", res, slave->dev->name);
3494 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3498 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3499 tmp_sa.sa_family = bond_dev->type;
3501 /* unwind from head to the slave that failed */
3502 bond_for_each_slave(bond, rollback_slave, iter) {
3505 if (rollback_slave == slave)
3508 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3510 pr_debug("unwind err %d dev %s\n",
3511 tmp_res, rollback_slave->dev->name);
3519 * bond_xmit_slave_id - transmit skb through slave with slave_id
3520 * @bond: bonding device that is transmitting
3521 * @skb: buffer to transmit
3522 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3524 * This function tries to transmit through slave with slave_id but in case
3525 * it fails, it tries to find the first available slave for transmission.
3526 * The skb is consumed in all cases, thus the function is void.
3528 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3530 struct list_head *iter;
3531 struct slave *slave;
3534 /* Here we start from the slave with slave_id */
3535 bond_for_each_slave_rcu(bond, slave, iter) {
3537 if (slave_can_tx(slave)) {
3538 bond_dev_queue_xmit(bond, skb, slave->dev);
3544 /* Here we start from the first slave up to slave_id */
3546 bond_for_each_slave_rcu(bond, slave, iter) {
3549 if (slave_can_tx(slave)) {
3550 bond_dev_queue_xmit(bond, skb, slave->dev);
3554 /* no slave that can tx has been found */
3559 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3560 * @bond: bonding device to use
3562 * Based on the value of the bonding device's packets_per_slave parameter
3563 * this function generates a slave id, which is usually used as the next
3564 * slave to transmit through.
3566 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3569 struct reciprocal_value reciprocal_packets_per_slave;
3570 int packets_per_slave = bond->params.packets_per_slave;
3572 switch (packets_per_slave) {
3574 slave_id = prandom_u32();
3577 slave_id = bond->rr_tx_counter;
3580 reciprocal_packets_per_slave =
3581 bond->params.reciprocal_packets_per_slave;
3582 slave_id = reciprocal_divide(bond->rr_tx_counter,
3583 reciprocal_packets_per_slave);
3586 bond->rr_tx_counter++;
3591 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3593 struct bonding *bond = netdev_priv(bond_dev);
3594 struct iphdr *iph = ip_hdr(skb);
3595 struct slave *slave;
3598 /* Start with the curr_active_slave that joined the bond as the
3599 * default for sending IGMP traffic. For failover purposes one
3600 * needs to maintain some consistency for the interface that will
3601 * send the join/membership reports. The curr_active_slave found
3602 * will send all of this type of traffic.
3604 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3605 slave = rcu_dereference(bond->curr_active_slave);
3606 if (slave && slave_can_tx(slave))
3607 bond_dev_queue_xmit(bond, skb, slave->dev);
3609 bond_xmit_slave_id(bond, skb, 0);
3611 slave_id = bond_rr_gen_slave_id(bond);
3612 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3615 return NETDEV_TX_OK;
3619 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3620 * the bond has a usable interface.
3622 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3624 struct bonding *bond = netdev_priv(bond_dev);
3625 struct slave *slave;
3627 slave = rcu_dereference(bond->curr_active_slave);
3629 bond_dev_queue_xmit(bond, skb, slave->dev);
3633 return NETDEV_TX_OK;
3636 /* In bond_xmit_xor() , we determine the output device by using a pre-
3637 * determined xmit_hash_policy(), If the selected device is not enabled,
3638 * find the next active slave.
3640 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3642 struct bonding *bond = netdev_priv(bond_dev);
3644 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3646 return NETDEV_TX_OK;
3649 /* in broadcast mode, we send everything to all usable interfaces. */
3650 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3652 struct bonding *bond = netdev_priv(bond_dev);
3653 struct slave *slave = NULL;
3654 struct list_head *iter;
3656 bond_for_each_slave_rcu(bond, slave, iter) {
3657 if (bond_is_last_slave(bond, slave))
3659 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3660 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3663 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3667 /* bond_dev_queue_xmit always returns 0 */
3668 bond_dev_queue_xmit(bond, skb2, slave->dev);
3671 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3672 bond_dev_queue_xmit(bond, skb, slave->dev);
3676 return NETDEV_TX_OK;
3679 /*------------------------- Device initialization ---------------------------*/
3682 * Lookup the slave that corresponds to a qid
3684 static inline int bond_slave_override(struct bonding *bond,
3685 struct sk_buff *skb)
3687 struct slave *slave = NULL;
3688 struct list_head *iter;
3690 if (!skb->queue_mapping)
3693 /* Find out if any slaves have the same mapping as this skb. */
3694 bond_for_each_slave_rcu(bond, slave, iter) {
3695 if (slave->queue_id == skb->queue_mapping) {
3696 if (slave_can_tx(slave)) {
3697 bond_dev_queue_xmit(bond, skb, slave->dev);
3700 /* If the slave isn't UP, use default transmit policy. */
3709 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3710 void *accel_priv, select_queue_fallback_t fallback)
3713 * This helper function exists to help dev_pick_tx get the correct
3714 * destination queue. Using a helper function skips a call to
3715 * skb_tx_hash and will put the skbs in the queue we expect on their
3716 * way down to the bonding driver.
3718 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3721 * Save the original txq to restore before passing to the driver
3723 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3725 if (unlikely(txq >= dev->real_num_tx_queues)) {
3727 txq -= dev->real_num_tx_queues;
3728 } while (txq >= dev->real_num_tx_queues);
3733 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3735 struct bonding *bond = netdev_priv(dev);
3737 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3738 if (!bond_slave_override(bond, skb))
3739 return NETDEV_TX_OK;
3742 switch (bond->params.mode) {
3743 case BOND_MODE_ROUNDROBIN:
3744 return bond_xmit_roundrobin(skb, dev);
3745 case BOND_MODE_ACTIVEBACKUP:
3746 return bond_xmit_activebackup(skb, dev);
3748 return bond_xmit_xor(skb, dev);
3749 case BOND_MODE_BROADCAST:
3750 return bond_xmit_broadcast(skb, dev);
3751 case BOND_MODE_8023AD:
3752 return bond_3ad_xmit_xor(skb, dev);
3755 return bond_alb_xmit(skb, dev);
3757 /* Should never happen, mode already checked */
3758 pr_err("%s: Error: Unknown bonding mode %d\n",
3759 dev->name, bond->params.mode);
3762 return NETDEV_TX_OK;
3766 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3768 struct bonding *bond = netdev_priv(dev);
3769 netdev_tx_t ret = NETDEV_TX_OK;
3772 * If we risk deadlock from transmitting this in the
3773 * netpoll path, tell netpoll to queue the frame for later tx
3775 if (is_netpoll_tx_blocked(dev))
3776 return NETDEV_TX_BUSY;
3779 if (bond_has_slaves(bond))
3780 ret = __bond_start_xmit(skb, dev);
3788 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3789 struct ethtool_cmd *ecmd)
3791 struct bonding *bond = netdev_priv(bond_dev);
3792 unsigned long speed = 0;
3793 struct list_head *iter;
3794 struct slave *slave;
3796 ecmd->duplex = DUPLEX_UNKNOWN;
3797 ecmd->port = PORT_OTHER;
3799 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3800 * do not need to check mode. Though link speed might not represent
3801 * the true receive or transmit bandwidth (not all modes are symmetric)
3802 * this is an accurate maximum.
3804 read_lock(&bond->lock);
3805 bond_for_each_slave(bond, slave, iter) {
3806 if (SLAVE_IS_OK(slave)) {
3807 if (slave->speed != SPEED_UNKNOWN)
3808 speed += slave->speed;
3809 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3810 slave->duplex != DUPLEX_UNKNOWN)
3811 ecmd->duplex = slave->duplex;
3814 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3815 read_unlock(&bond->lock);
3820 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3821 struct ethtool_drvinfo *drvinfo)
3823 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3824 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3825 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3829 static const struct ethtool_ops bond_ethtool_ops = {
3830 .get_drvinfo = bond_ethtool_get_drvinfo,
3831 .get_settings = bond_ethtool_get_settings,
3832 .get_link = ethtool_op_get_link,
3835 static const struct net_device_ops bond_netdev_ops = {
3836 .ndo_init = bond_init,
3837 .ndo_uninit = bond_uninit,
3838 .ndo_open = bond_open,
3839 .ndo_stop = bond_close,
3840 .ndo_start_xmit = bond_start_xmit,
3841 .ndo_select_queue = bond_select_queue,
3842 .ndo_get_stats64 = bond_get_stats,
3843 .ndo_do_ioctl = bond_do_ioctl,
3844 .ndo_change_rx_flags = bond_change_rx_flags,
3845 .ndo_set_rx_mode = bond_set_rx_mode,
3846 .ndo_change_mtu = bond_change_mtu,
3847 .ndo_set_mac_address = bond_set_mac_address,
3848 .ndo_neigh_setup = bond_neigh_setup,
3849 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3850 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3851 #ifdef CONFIG_NET_POLL_CONTROLLER
3852 .ndo_netpoll_setup = bond_netpoll_setup,
3853 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3854 .ndo_poll_controller = bond_poll_controller,
3856 .ndo_add_slave = bond_enslave,
3857 .ndo_del_slave = bond_release,
3858 .ndo_fix_features = bond_fix_features,
3861 static const struct device_type bond_type = {
3865 static void bond_destructor(struct net_device *bond_dev)
3867 struct bonding *bond = netdev_priv(bond_dev);
3869 destroy_workqueue(bond->wq);
3870 free_netdev(bond_dev);
3873 void bond_setup(struct net_device *bond_dev)
3875 struct bonding *bond = netdev_priv(bond_dev);
3877 /* initialize rwlocks */
3878 rwlock_init(&bond->lock);
3879 rwlock_init(&bond->curr_slave_lock);
3880 bond->params = bonding_defaults;
3882 /* Initialize pointers */
3883 bond->dev = bond_dev;
3885 /* Initialize the device entry points */
3886 ether_setup(bond_dev);
3887 bond_dev->netdev_ops = &bond_netdev_ops;
3888 bond_dev->ethtool_ops = &bond_ethtool_ops;
3890 bond_dev->destructor = bond_destructor;
3892 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3894 /* Initialize the device options */
3895 bond_dev->tx_queue_len = 0;
3896 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3897 bond_dev->priv_flags |= IFF_BONDING;
3898 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3900 /* At first, we block adding VLANs. That's the only way to
3901 * prevent problems that occur when adding VLANs over an
3902 * empty bond. The block will be removed once non-challenged
3903 * slaves are enslaved.
3905 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3907 /* don't acquire bond device's netif_tx_lock when
3909 bond_dev->features |= NETIF_F_LLTX;
3911 /* By default, we declare the bond to be fully
3912 * VLAN hardware accelerated capable. Special
3913 * care is taken in the various xmit functions
3914 * when there are slaves that are not hw accel
3918 /* Don't allow bond devices to change network namespaces. */
3919 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3921 bond_dev->hw_features = BOND_VLAN_FEATURES |
3922 NETIF_F_HW_VLAN_CTAG_TX |
3923 NETIF_F_HW_VLAN_CTAG_RX |
3924 NETIF_F_HW_VLAN_CTAG_FILTER;
3926 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3927 bond_dev->features |= bond_dev->hw_features;
3931 * Destroy a bonding device.
3932 * Must be under rtnl_lock when this function is called.
3934 static void bond_uninit(struct net_device *bond_dev)
3936 struct bonding *bond = netdev_priv(bond_dev);
3937 struct list_head *iter;
3938 struct slave *slave;
3940 bond_netpoll_cleanup(bond_dev);
3942 /* Release the bonded slaves */
3943 bond_for_each_slave(bond, slave, iter)
3944 __bond_release_one(bond_dev, slave->dev, true);
3945 pr_info("%s: released all slaves\n", bond_dev->name);
3947 list_del(&bond->bond_list);
3949 bond_debug_unregister(bond);
3952 /*------------------------- Module initialization ---------------------------*/
3954 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3958 for (i = 0; tbl[i].modename; i++)
3959 if (mode == tbl[i].mode)
3965 static int bond_parm_tbl_lookup_name(const char *modename,
3966 const struct bond_parm_tbl *tbl)
3970 for (i = 0; tbl[i].modename; i++)
3971 if (strcmp(modename, tbl[i].modename) == 0)
3978 * Convert string input module parms. Accept either the
3979 * number of the mode or its string name. A bit complicated because
3980 * some mode names are substrings of other names, and calls from sysfs
3981 * may have whitespace in the name (trailing newlines, for example).
3983 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3986 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3988 for (p = (char *)buf; *p; p++)
3989 if (!(isdigit(*p) || isspace(*p)))
3992 if (*p && sscanf(buf, "%20s", modestr) != 0)
3993 return bond_parm_tbl_lookup_name(modestr, tbl);
3994 else if (sscanf(buf, "%d", &modeint) != 0)
3995 return bond_parm_tbl_lookup(modeint, tbl);
4000 static int bond_check_params(struct bond_params *params)
4002 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4003 struct bond_opt_value newval, *valptr;
4004 int arp_all_targets_value;
4007 * Convert string parameters.
4010 bond_opt_initstr(&newval, mode);
4011 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4013 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4016 bond_mode = valptr->value;
4019 if (xmit_hash_policy) {
4020 if ((bond_mode != BOND_MODE_XOR) &&
4021 (bond_mode != BOND_MODE_8023AD)) {
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, lacp_rate);
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_warning("ad_select param only affects 802.3ad mode\n");
4066 params->ad_select = BOND_AD_STABLE;
4069 if (max_bonds < 0) {
4070 pr_warning("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_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4082 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4087 if (downdelay < 0) {
4088 pr_warning("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_warning("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_warning("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_NO_USES_ARP(bond_mode)) {
4108 pr_warning("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_warning("Forcing miimon to 100msec\n");
4110 miimon = BOND_DEFAULT_MIIMON;
4114 if (tx_queues < 1 || tx_queues > 255) {
4115 pr_warning("Warning: tx_queues (%d) should be between "
4116 "1 and 255, resetting to %d\n",
4117 tx_queues, BOND_DEFAULT_TX_QUEUES);
4118 tx_queues = BOND_DEFAULT_TX_QUEUES;
4121 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4122 pr_warning("Warning: all_slaves_active module parameter (%d), "
4123 "not of valid value (0/1), so it was set to "
4124 "0\n", all_slaves_active);
4125 all_slaves_active = 0;
4128 if (resend_igmp < 0 || resend_igmp > 255) {
4129 pr_warning("Warning: resend_igmp (%d) should be between "
4130 "0 and 255, resetting to %d\n",
4131 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4132 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4135 bond_opt_initval(&newval, packets_per_slave);
4136 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4137 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4138 packets_per_slave, USHRT_MAX);
4139 packets_per_slave = 1;
4142 if (bond_mode == BOND_MODE_ALB) {
4143 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",
4148 if (updelay || downdelay) {
4149 /* just warn the user the up/down delay will have
4150 * no effect since miimon is zero...
4152 pr_warning("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",
4153 updelay, downdelay);
4156 /* don't allow arp monitoring */
4158 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4159 miimon, arp_interval);
4163 if ((updelay % miimon) != 0) {
4164 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4166 (updelay / miimon) * miimon);
4171 if ((downdelay % miimon) != 0) {
4172 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4174 (downdelay / miimon) * miimon);
4177 downdelay /= miimon;
4180 if (arp_interval < 0) {
4181 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4182 arp_interval, INT_MAX);
4186 for (arp_ip_count = 0, i = 0;
4187 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4188 /* not complete check, but should be good enough to
4191 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4192 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4193 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4197 if (bond_get_targets_ip(arp_target, ip) == -1)
4198 arp_target[arp_ip_count++] = ip;
4200 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4205 if (arp_interval && !arp_ip_count) {
4206 /* don't allow arping if no arp_ip_target given... */
4207 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4213 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4214 pr_err("arp_validate only supported in active-backup mode\n");
4217 if (!arp_interval) {
4218 pr_err("arp_validate requires arp_interval\n");
4222 bond_opt_initstr(&newval, arp_validate);
4223 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4226 pr_err("Error: invalid arp_validate \"%s\"\n",
4230 arp_validate_value = valptr->value;
4232 arp_validate_value = 0;
4235 arp_all_targets_value = 0;
4236 if (arp_all_targets) {
4237 bond_opt_initstr(&newval, arp_all_targets);
4238 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4241 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4243 arp_all_targets_value = 0;
4245 arp_all_targets_value = valptr->value;
4250 pr_info("MII link monitoring set to %d ms\n", miimon);
4251 } else if (arp_interval) {
4252 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4253 arp_validate_value);
4254 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4255 arp_interval, valptr->string, arp_ip_count);
4257 for (i = 0; i < arp_ip_count; i++)
4258 pr_info(" %s", arp_ip_target[i]);
4262 } else if (max_bonds) {
4263 /* miimon and arp_interval not set, we need one so things
4264 * work as expected, see bonding.txt for details
4266 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");
4269 if (primary && !USES_PRIMARY(bond_mode)) {
4270 /* currently, using a primary only makes sense
4271 * in active backup, TLB or ALB modes
4273 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4274 primary, bond_mode_name(bond_mode));
4278 if (primary && primary_reselect) {
4279 bond_opt_initstr(&newval, primary_reselect);
4280 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4283 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4287 primary_reselect_value = valptr->value;
4289 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4292 if (fail_over_mac) {
4293 bond_opt_initstr(&newval, fail_over_mac);
4294 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4297 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4301 fail_over_mac_value = valptr->value;
4302 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4303 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4305 fail_over_mac_value = BOND_FOM_NONE;
4308 if (lp_interval == 0) {
4309 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4310 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4311 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4314 /* fill params struct with the proper values */
4315 params->mode = bond_mode;
4316 params->xmit_policy = xmit_hashtype;
4317 params->miimon = miimon;
4318 params->num_peer_notif = num_peer_notif;
4319 params->arp_interval = arp_interval;
4320 params->arp_validate = arp_validate_value;
4321 params->arp_all_targets = arp_all_targets_value;
4322 params->updelay = updelay;
4323 params->downdelay = downdelay;
4324 params->use_carrier = use_carrier;
4325 params->lacp_fast = lacp_fast;
4326 params->primary[0] = 0;
4327 params->primary_reselect = primary_reselect_value;
4328 params->fail_over_mac = fail_over_mac_value;
4329 params->tx_queues = tx_queues;
4330 params->all_slaves_active = all_slaves_active;
4331 params->resend_igmp = resend_igmp;
4332 params->min_links = min_links;
4333 params->lp_interval = lp_interval;
4334 params->packets_per_slave = packets_per_slave;
4335 if (packets_per_slave > 0) {
4336 params->reciprocal_packets_per_slave =
4337 reciprocal_value(packets_per_slave);
4339 /* reciprocal_packets_per_slave is unused if
4340 * packets_per_slave is 0 or 1, just initialize it
4342 params->reciprocal_packets_per_slave =
4343 (struct reciprocal_value) { 0 };
4347 strncpy(params->primary, primary, IFNAMSIZ);
4348 params->primary[IFNAMSIZ - 1] = 0;
4351 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4356 static struct lock_class_key bonding_netdev_xmit_lock_key;
4357 static struct lock_class_key bonding_netdev_addr_lock_key;
4358 static struct lock_class_key bonding_tx_busylock_key;
4360 static void bond_set_lockdep_class_one(struct net_device *dev,
4361 struct netdev_queue *txq,
4364 lockdep_set_class(&txq->_xmit_lock,
4365 &bonding_netdev_xmit_lock_key);
4368 static void bond_set_lockdep_class(struct net_device *dev)
4370 lockdep_set_class(&dev->addr_list_lock,
4371 &bonding_netdev_addr_lock_key);
4372 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4373 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4377 * Called from registration process
4379 static int bond_init(struct net_device *bond_dev)
4381 struct bonding *bond = netdev_priv(bond_dev);
4382 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4383 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4385 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4388 * Initialize locks that may be required during
4389 * en/deslave operations. All of the bond_open work
4390 * (of which this is part) should really be moved to
4391 * a phase prior to dev_open
4393 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4394 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4396 bond->wq = create_singlethread_workqueue(bond_dev->name);
4400 bond_set_lockdep_class(bond_dev);
4402 list_add_tail(&bond->bond_list, &bn->dev_list);
4404 bond_prepare_sysfs_group(bond);
4406 bond_debug_register(bond);
4408 /* Ensure valid dev_addr */
4409 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4410 bond_dev->addr_assign_type == NET_ADDR_PERM)
4411 eth_hw_addr_random(bond_dev);
4416 unsigned int bond_get_num_tx_queues(void)
4421 /* Create a new bond based on the specified name and bonding parameters.
4422 * If name is NULL, obtain a suitable "bond%d" name for us.
4423 * Caller must NOT hold rtnl_lock; we need to release it here before we
4424 * set up our sysfs entries.
4426 int bond_create(struct net *net, const char *name)
4428 struct net_device *bond_dev;
4433 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4434 name ? name : "bond%d",
4435 bond_setup, tx_queues);
4437 pr_err("%s: eek! can't alloc netdev!\n", name);
4442 dev_net_set(bond_dev, net);
4443 bond_dev->rtnl_link_ops = &bond_link_ops;
4445 res = register_netdevice(bond_dev);
4447 netif_carrier_off(bond_dev);
4451 bond_destructor(bond_dev);
4455 static int __net_init bond_net_init(struct net *net)
4457 struct bond_net *bn = net_generic(net, bond_net_id);
4460 INIT_LIST_HEAD(&bn->dev_list);
4462 bond_create_proc_dir(bn);
4463 bond_create_sysfs(bn);
4468 static void __net_exit bond_net_exit(struct net *net)
4470 struct bond_net *bn = net_generic(net, bond_net_id);
4471 struct bonding *bond, *tmp_bond;
4474 bond_destroy_sysfs(bn);
4475 bond_destroy_proc_dir(bn);
4477 /* Kill off any bonds created after unregistering bond rtnl ops */
4479 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4480 unregister_netdevice_queue(bond->dev, &list);
4481 unregister_netdevice_many(&list);
4485 static struct pernet_operations bond_net_ops = {
4486 .init = bond_net_init,
4487 .exit = bond_net_exit,
4489 .size = sizeof(struct bond_net),
4492 static int __init bonding_init(void)
4497 pr_info("%s", bond_version);
4499 res = bond_check_params(&bonding_defaults);
4503 res = register_pernet_subsys(&bond_net_ops);
4507 res = bond_netlink_init();
4511 bond_create_debugfs();
4513 for (i = 0; i < max_bonds; i++) {
4514 res = bond_create(&init_net, NULL);
4519 register_netdevice_notifier(&bond_netdev_notifier);
4523 bond_netlink_fini();
4525 unregister_pernet_subsys(&bond_net_ops);
4530 static void __exit bonding_exit(void)
4532 unregister_netdevice_notifier(&bond_netdev_notifier);
4534 bond_destroy_debugfs();
4536 bond_netlink_fini();
4537 unregister_pernet_subsys(&bond_net_ops);
4539 #ifdef CONFIG_NET_POLL_CONTROLLER
4541 * Make sure we don't have an imbalance on our netpoll blocking
4543 WARN_ON(atomic_read(&netpoll_block_tx));
4547 module_init(bonding_init);
4548 module_exit(bonding_exit);
4549 MODULE_LICENSE("GPL");
4550 MODULE_VERSION(DRV_VERSION);
4551 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4552 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");