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 read_unlock(&bond->lock);
1514 pr_info("Error, %s: master_dev is using netpoll, "
1515 "but new slave device does not support netpoll.\n",
1523 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1526 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1530 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1532 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1533 goto err_unregister;
1536 res = bond_sysfs_slave_add(new_slave);
1538 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1539 goto err_upper_unlink;
1543 bond_compute_features(bond);
1544 bond_set_carrier(bond);
1546 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);
1552 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1553 bond_dev->name, slave_dev->name,
1554 bond_is_active_slave(new_slave) ? "n active" : " backup",
1555 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1557 /* enslave is successful */
1560 /* Undo stages on error */
1562 bond_upper_dev_unlink(bond_dev, slave_dev);
1565 netdev_rx_handler_unregister(slave_dev);
1568 if (!USES_PRIMARY(bond->params.mode))
1569 bond_hw_addr_flush(bond_dev, slave_dev);
1571 vlan_vids_del_by_dev(slave_dev, bond_dev);
1572 if (bond->primary_slave == new_slave)
1573 bond->primary_slave = NULL;
1574 if (bond->curr_active_slave == new_slave) {
1575 write_lock_bh(&bond->curr_slave_lock);
1576 bond_change_active_slave(bond, NULL);
1577 bond_select_active_slave(bond);
1578 write_unlock_bh(&bond->curr_slave_lock);
1580 slave_disable_netpoll(new_slave);
1583 slave_dev->priv_flags &= ~IFF_BONDING;
1584 dev_close(slave_dev);
1587 if (!bond->params.fail_over_mac ||
1588 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1589 /* XXX TODO - fom follow mode needs to change master's
1590 * MAC if this slave's MAC is in use by the bond, or at
1591 * least print a warning.
1593 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1594 addr.sa_family = slave_dev->type;
1595 dev_set_mac_address(slave_dev, &addr);
1599 dev_set_mtu(slave_dev, new_slave->original_mtu);
1605 /* Enslave of first slave has failed and we need to fix master's mac */
1606 if (!bond_has_slaves(bond) &&
1607 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1608 eth_hw_addr_random(bond_dev);
1614 * Try to release the slave device <slave> from the bond device <master>
1615 * It is legal to access curr_active_slave without a lock because all the function
1616 * is write-locked. If "all" is true it means that the function is being called
1617 * while destroying a bond interface and all slaves are being released.
1619 * The rules for slave state should be:
1620 * for Active/Backup:
1621 * Active stays on all backups go down
1622 * for Bonded connections:
1623 * The first up interface should be left on and all others downed.
1625 static int __bond_release_one(struct net_device *bond_dev,
1626 struct net_device *slave_dev,
1629 struct bonding *bond = netdev_priv(bond_dev);
1630 struct slave *slave, *oldcurrent;
1631 struct sockaddr addr;
1632 int old_flags = bond_dev->flags;
1633 netdev_features_t old_features = bond_dev->features;
1635 /* slave is not a slave or master is not master of this slave */
1636 if (!(slave_dev->flags & IFF_SLAVE) ||
1637 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1638 pr_err("%s: Error: cannot release %s.\n",
1639 bond_dev->name, slave_dev->name);
1645 slave = bond_get_slave_by_dev(bond, slave_dev);
1647 /* not a slave of this bond */
1648 pr_info("%s: %s not enslaved\n",
1649 bond_dev->name, slave_dev->name);
1650 unblock_netpoll_tx();
1654 /* release the slave from its bond */
1657 bond_sysfs_slave_del(slave);
1659 bond_upper_dev_unlink(bond_dev, slave_dev);
1660 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1661 * for this slave anymore.
1663 netdev_rx_handler_unregister(slave_dev);
1664 write_lock_bh(&bond->lock);
1666 /* Inform AD package of unbinding of slave. */
1667 if (bond->params.mode == BOND_MODE_8023AD)
1668 bond_3ad_unbind_slave(slave);
1670 write_unlock_bh(&bond->lock);
1672 pr_info("%s: releasing %s interface %s\n",
1674 bond_is_active_slave(slave) ? "active" : "backup",
1677 oldcurrent = bond->curr_active_slave;
1679 bond->current_arp_slave = NULL;
1681 if (!all && (!bond->params.fail_over_mac ||
1682 bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1683 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1684 bond_has_slaves(bond))
1685 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",
1686 bond_dev->name, slave_dev->name,
1688 bond_dev->name, slave_dev->name);
1691 if (bond->primary_slave == slave)
1692 bond->primary_slave = NULL;
1694 if (oldcurrent == slave) {
1695 write_lock_bh(&bond->curr_slave_lock);
1696 bond_change_active_slave(bond, NULL);
1697 write_unlock_bh(&bond->curr_slave_lock);
1700 if (bond_is_lb(bond)) {
1701 /* Must be called only after the slave has been
1702 * detached from the list and the curr_active_slave
1703 * has been cleared (if our_slave == old_current),
1704 * but before a new active slave is selected.
1706 bond_alb_deinit_slave(bond, slave);
1710 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1711 } else if (oldcurrent == slave) {
1713 * Note that we hold RTNL over this sequence, so there
1714 * is no concern that another slave add/remove event
1717 write_lock_bh(&bond->curr_slave_lock);
1719 bond_select_active_slave(bond);
1721 write_unlock_bh(&bond->curr_slave_lock);
1724 if (!bond_has_slaves(bond)) {
1725 bond_set_carrier(bond);
1726 eth_hw_addr_random(bond_dev);
1728 if (vlan_uses_dev(bond_dev)) {
1729 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1730 bond_dev->name, bond_dev->name);
1731 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1736 unblock_netpoll_tx();
1739 if (!bond_has_slaves(bond)) {
1740 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1741 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1744 bond_compute_features(bond);
1745 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1746 (old_features & NETIF_F_VLAN_CHALLENGED))
1747 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1748 bond_dev->name, slave_dev->name, bond_dev->name);
1750 /* must do this from outside any spinlocks */
1751 vlan_vids_del_by_dev(slave_dev, bond_dev);
1753 /* If the mode USES_PRIMARY, then this cases was handled above by
1754 * bond_change_active_slave(..., NULL)
1756 if (!USES_PRIMARY(bond->params.mode)) {
1757 /* unset promiscuity level from slave
1758 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1759 * of the IFF_PROMISC flag in the bond_dev, but we need the
1760 * value of that flag before that change, as that was the value
1761 * when this slave was attached, so we cache at the start of the
1762 * function and use it here. Same goes for ALLMULTI below
1764 if (old_flags & IFF_PROMISC)
1765 dev_set_promiscuity(slave_dev, -1);
1767 /* unset allmulti level from slave */
1768 if (old_flags & IFF_ALLMULTI)
1769 dev_set_allmulti(slave_dev, -1);
1771 bond_hw_addr_flush(bond_dev, slave_dev);
1774 slave_disable_netpoll(slave);
1776 /* close slave before restoring its mac address */
1777 dev_close(slave_dev);
1779 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1780 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1781 /* restore original ("permanent") mac address */
1782 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1783 addr.sa_family = slave_dev->type;
1784 dev_set_mac_address(slave_dev, &addr);
1787 dev_set_mtu(slave_dev, slave->original_mtu);
1789 slave_dev->priv_flags &= ~IFF_BONDING;
1793 return 0; /* deletion OK */
1796 /* A wrapper used because of ndo_del_link */
1797 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1799 return __bond_release_one(bond_dev, slave_dev, false);
1803 * First release a slave and then destroy the bond if no more slaves are left.
1804 * Must be under rtnl_lock when this function is called.
1806 static int bond_release_and_destroy(struct net_device *bond_dev,
1807 struct net_device *slave_dev)
1809 struct bonding *bond = netdev_priv(bond_dev);
1812 ret = bond_release(bond_dev, slave_dev);
1813 if (ret == 0 && !bond_has_slaves(bond)) {
1814 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1815 pr_info("%s: destroying bond %s.\n",
1816 bond_dev->name, bond_dev->name);
1817 unregister_netdevice(bond_dev);
1822 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1824 struct bonding *bond = netdev_priv(bond_dev);
1826 info->bond_mode = bond->params.mode;
1827 info->miimon = bond->params.miimon;
1829 read_lock(&bond->lock);
1830 info->num_slaves = bond->slave_cnt;
1831 read_unlock(&bond->lock);
1836 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1838 struct bonding *bond = netdev_priv(bond_dev);
1839 struct list_head *iter;
1840 int i = 0, res = -ENODEV;
1841 struct slave *slave;
1843 read_lock(&bond->lock);
1844 bond_for_each_slave(bond, slave, iter) {
1845 if (i++ == (int)info->slave_id) {
1847 strcpy(info->slave_name, slave->dev->name);
1848 info->link = slave->link;
1849 info->state = bond_slave_state(slave);
1850 info->link_failure_count = slave->link_failure_count;
1854 read_unlock(&bond->lock);
1859 /*-------------------------------- Monitoring -------------------------------*/
1862 static int bond_miimon_inspect(struct bonding *bond)
1864 int link_state, commit = 0;
1865 struct list_head *iter;
1866 struct slave *slave;
1867 bool ignore_updelay;
1869 ignore_updelay = !bond->curr_active_slave ? true : false;
1871 bond_for_each_slave_rcu(bond, slave, iter) {
1872 slave->new_link = BOND_LINK_NOCHANGE;
1874 link_state = bond_check_dev_link(bond, slave->dev, 0);
1876 switch (slave->link) {
1881 slave->link = BOND_LINK_FAIL;
1882 slave->delay = bond->params.downdelay;
1884 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1886 (bond->params.mode ==
1887 BOND_MODE_ACTIVEBACKUP) ?
1888 (bond_is_active_slave(slave) ?
1889 "active " : "backup ") : "",
1891 bond->params.downdelay * bond->params.miimon);
1894 case BOND_LINK_FAIL:
1897 * recovered before downdelay expired
1899 slave->link = BOND_LINK_UP;
1900 slave->jiffies = jiffies;
1901 pr_info("%s: link status up again after %d ms for interface %s.\n",
1903 (bond->params.downdelay - slave->delay) *
1904 bond->params.miimon,
1909 if (slave->delay <= 0) {
1910 slave->new_link = BOND_LINK_DOWN;
1918 case BOND_LINK_DOWN:
1922 slave->link = BOND_LINK_BACK;
1923 slave->delay = bond->params.updelay;
1926 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1927 bond->dev->name, slave->dev->name,
1928 ignore_updelay ? 0 :
1929 bond->params.updelay *
1930 bond->params.miimon);
1933 case BOND_LINK_BACK:
1935 slave->link = BOND_LINK_DOWN;
1936 pr_info("%s: link status down again after %d ms for interface %s.\n",
1938 (bond->params.updelay - slave->delay) *
1939 bond->params.miimon,
1948 if (slave->delay <= 0) {
1949 slave->new_link = BOND_LINK_UP;
1951 ignore_updelay = false;
1963 static void bond_miimon_commit(struct bonding *bond)
1965 struct list_head *iter;
1966 struct slave *slave;
1968 bond_for_each_slave(bond, slave, iter) {
1969 switch (slave->new_link) {
1970 case BOND_LINK_NOCHANGE:
1974 slave->link = BOND_LINK_UP;
1975 slave->jiffies = jiffies;
1977 if (bond->params.mode == BOND_MODE_8023AD) {
1978 /* prevent it from being the active one */
1979 bond_set_backup_slave(slave);
1980 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1981 /* make it immediately active */
1982 bond_set_active_slave(slave);
1983 } else if (slave != bond->primary_slave) {
1984 /* prevent it from being the active one */
1985 bond_set_backup_slave(slave);
1988 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
1989 bond->dev->name, slave->dev->name,
1990 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
1991 slave->duplex ? "full" : "half");
1993 /* notify ad that the link status has changed */
1994 if (bond->params.mode == BOND_MODE_8023AD)
1995 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
1997 if (bond_is_lb(bond))
1998 bond_alb_handle_link_change(bond, slave,
2001 if (!bond->curr_active_slave ||
2002 (slave == bond->primary_slave))
2007 case BOND_LINK_DOWN:
2008 if (slave->link_failure_count < UINT_MAX)
2009 slave->link_failure_count++;
2011 slave->link = BOND_LINK_DOWN;
2013 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2014 bond->params.mode == BOND_MODE_8023AD)
2015 bond_set_slave_inactive_flags(slave);
2017 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2018 bond->dev->name, slave->dev->name);
2020 if (bond->params.mode == BOND_MODE_8023AD)
2021 bond_3ad_handle_link_change(slave,
2024 if (bond_is_lb(bond))
2025 bond_alb_handle_link_change(bond, slave,
2028 if (slave == bond->curr_active_slave)
2034 pr_err("%s: invalid new link %d on slave %s\n",
2035 bond->dev->name, slave->new_link,
2037 slave->new_link = BOND_LINK_NOCHANGE;
2045 write_lock_bh(&bond->curr_slave_lock);
2046 bond_select_active_slave(bond);
2047 write_unlock_bh(&bond->curr_slave_lock);
2048 unblock_netpoll_tx();
2051 bond_set_carrier(bond);
2057 * Really a wrapper that splits the mii monitor into two phases: an
2058 * inspection, then (if inspection indicates something needs to be done)
2059 * an acquisition of appropriate locks followed by a commit phase to
2060 * implement whatever link state changes are indicated.
2062 static void bond_mii_monitor(struct work_struct *work)
2064 struct bonding *bond = container_of(work, struct bonding,
2066 bool should_notify_peers = false;
2067 unsigned long delay;
2069 delay = msecs_to_jiffies(bond->params.miimon);
2071 if (!bond_has_slaves(bond))
2076 should_notify_peers = bond_should_notify_peers(bond);
2078 if (bond_miimon_inspect(bond)) {
2081 /* Race avoidance with bond_close cancel of workqueue */
2082 if (!rtnl_trylock()) {
2084 should_notify_peers = false;
2088 bond_miimon_commit(bond);
2090 rtnl_unlock(); /* might sleep, hold no other locks */
2095 if (bond->params.miimon)
2096 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2098 if (should_notify_peers) {
2099 if (!rtnl_trylock())
2101 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2106 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2108 struct net_device *upper;
2109 struct list_head *iter;
2112 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2116 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2117 if (ip == bond_confirm_addr(upper, 0, ip)) {
2128 * We go to the (large) trouble of VLAN tagging ARP frames because
2129 * switches in VLAN mode (especially if ports are configured as
2130 * "native" to a VLAN) might not pass non-tagged frames.
2132 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2134 struct sk_buff *skb;
2136 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2137 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2139 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2140 NULL, slave_dev->dev_addr, NULL);
2143 pr_err("ARP packet allocation failed\n");
2147 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2149 pr_err("failed to insert VLAN tag\n");
2157 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2159 struct net_device *upper, *vlan_upper;
2160 struct list_head *iter, *vlan_iter;
2162 __be32 *targets = bond->params.arp_targets, addr;
2165 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2166 pr_debug("basa: target %pI4\n", &targets[i]);
2168 /* Find out through which dev should the packet go */
2169 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2172 pr_debug("%s: no route to arp_ip_target %pI4\n",
2173 bond->dev->name, &targets[i]);
2179 /* bond device itself */
2180 if (rt->dst.dev == bond->dev)
2184 /* first we search only for vlan devices. for every vlan
2185 * found we verify its upper dev list, searching for the
2186 * rt->dst.dev. If found we save the tag of the vlan and
2187 * proceed to send the packet.
2191 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2193 if (!is_vlan_dev(vlan_upper))
2195 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2197 if (upper == rt->dst.dev) {
2198 vlan_id = vlan_dev_vlan_id(vlan_upper);
2205 /* if the device we're looking for is not on top of any of
2206 * our upper vlans, then just search for any dev that
2207 * matches, and in case it's a vlan - save the id
2209 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2210 if (upper == rt->dst.dev) {
2211 /* if it's a vlan - get its VID */
2212 if (is_vlan_dev(upper))
2213 vlan_id = vlan_dev_vlan_id(upper);
2221 /* Not our device - skip */
2222 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2223 bond->dev->name, &targets[i],
2224 rt->dst.dev ? rt->dst.dev->name : "NULL");
2230 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2232 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2237 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2241 if (!sip || !bond_has_this_ip(bond, tip)) {
2242 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2246 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2248 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2251 slave->last_arp_rx = jiffies;
2252 slave->target_last_arp_rx[i] = jiffies;
2255 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2256 struct slave *slave)
2258 struct arphdr *arp = (struct arphdr *)skb->data;
2259 unsigned char *arp_ptr;
2263 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2264 return RX_HANDLER_ANOTHER;
2266 read_lock(&bond->lock);
2268 if (!slave_do_arp_validate(bond, slave))
2271 alen = arp_hdr_len(bond->dev);
2273 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2274 bond->dev->name, skb->dev->name);
2276 if (alen > skb_headlen(skb)) {
2277 arp = kmalloc(alen, GFP_ATOMIC);
2280 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2284 if (arp->ar_hln != bond->dev->addr_len ||
2285 skb->pkt_type == PACKET_OTHERHOST ||
2286 skb->pkt_type == PACKET_LOOPBACK ||
2287 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2288 arp->ar_pro != htons(ETH_P_IP) ||
2292 arp_ptr = (unsigned char *)(arp + 1);
2293 arp_ptr += bond->dev->addr_len;
2294 memcpy(&sip, arp_ptr, 4);
2295 arp_ptr += 4 + bond->dev->addr_len;
2296 memcpy(&tip, arp_ptr, 4);
2298 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2299 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2300 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2304 * Backup slaves won't see the ARP reply, but do come through
2305 * here for each ARP probe (so we swap the sip/tip to validate
2306 * the probe). In a "redundant switch, common router" type of
2307 * configuration, the ARP probe will (hopefully) travel from
2308 * the active, through one switch, the router, then the other
2309 * switch before reaching the backup.
2311 * We 'trust' the arp requests if there is an active slave and
2312 * it received valid arp reply(s) after it became active. This
2313 * is done to avoid endless looping when we can't reach the
2314 * arp_ip_target and fool ourselves with our own arp requests.
2316 if (bond_is_active_slave(slave))
2317 bond_validate_arp(bond, slave, sip, tip);
2318 else if (bond->curr_active_slave &&
2319 time_after(slave_last_rx(bond, bond->curr_active_slave),
2320 bond->curr_active_slave->jiffies))
2321 bond_validate_arp(bond, slave, tip, sip);
2324 read_unlock(&bond->lock);
2325 if (arp != (struct arphdr *)skb->data)
2327 return RX_HANDLER_ANOTHER;
2330 /* function to verify if we're in the arp_interval timeslice, returns true if
2331 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2332 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2334 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2337 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2339 return time_in_range(jiffies,
2340 last_act - delta_in_ticks,
2341 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2345 * this function is called regularly to monitor each slave's link
2346 * ensuring that traffic is being sent and received when arp monitoring
2347 * is used in load-balancing mode. if the adapter has been dormant, then an
2348 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2349 * arp monitoring in active backup mode.
2351 static void bond_loadbalance_arp_mon(struct work_struct *work)
2353 struct bonding *bond = container_of(work, struct bonding,
2355 struct slave *slave, *oldcurrent;
2356 struct list_head *iter;
2357 int do_failover = 0, slave_state_changed = 0;
2359 if (!bond_has_slaves(bond))
2364 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2365 /* see if any of the previous devices are up now (i.e. they have
2366 * xmt and rcv traffic). the curr_active_slave does not come into
2367 * the picture unless it is null. also, slave->jiffies is not needed
2368 * here because we send an arp on each slave and give a slave as
2369 * long as it needs to get the tx/rx within the delta.
2370 * TODO: what about up/down delay in arp mode? it wasn't here before
2373 bond_for_each_slave_rcu(bond, slave, iter) {
2374 unsigned long trans_start = dev_trans_start(slave->dev);
2376 if (slave->link != BOND_LINK_UP) {
2377 if (bond_time_in_interval(bond, trans_start, 1) &&
2378 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2380 slave->link = BOND_LINK_UP;
2381 slave_state_changed = 1;
2383 /* primary_slave has no meaning in round-robin
2384 * mode. the window of a slave being up and
2385 * curr_active_slave being null after enslaving
2389 pr_info("%s: link status definitely up for interface %s, ",
2394 pr_info("%s: interface %s is now up\n",
2400 /* slave->link == BOND_LINK_UP */
2402 /* not all switches will respond to an arp request
2403 * when the source ip is 0, so don't take the link down
2404 * if we don't know our ip yet
2406 if (!bond_time_in_interval(bond, trans_start, 2) ||
2407 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2409 slave->link = BOND_LINK_DOWN;
2410 slave_state_changed = 1;
2412 if (slave->link_failure_count < UINT_MAX)
2413 slave->link_failure_count++;
2415 pr_info("%s: interface %s is now down.\n",
2419 if (slave == oldcurrent)
2424 /* note: if switch is in round-robin mode, all links
2425 * must tx arp to ensure all links rx an arp - otherwise
2426 * links may oscillate or not come up at all; if switch is
2427 * in something like xor mode, there is nothing we can
2428 * do - all replies will be rx'ed on same link causing slaves
2429 * to be unstable during low/no traffic periods
2431 if (IS_UP(slave->dev))
2432 bond_arp_send_all(bond, slave);
2437 if (do_failover || slave_state_changed) {
2438 if (!rtnl_trylock())
2441 if (slave_state_changed) {
2442 bond_slave_state_change(bond);
2443 } else if (do_failover) {
2444 /* the bond_select_active_slave must hold RTNL
2445 * and curr_slave_lock for write.
2448 write_lock_bh(&bond->curr_slave_lock);
2450 bond_select_active_slave(bond);
2452 write_unlock_bh(&bond->curr_slave_lock);
2453 unblock_netpoll_tx();
2459 if (bond->params.arp_interval)
2460 queue_delayed_work(bond->wq, &bond->arp_work,
2461 msecs_to_jiffies(bond->params.arp_interval));
2465 * Called to inspect slaves for active-backup mode ARP monitor link state
2466 * changes. Sets new_link in slaves to specify what action should take
2467 * place for the slave. Returns 0 if no changes are found, >0 if changes
2468 * to link states must be committed.
2470 * Called with rcu_read_lock hold.
2472 static int bond_ab_arp_inspect(struct bonding *bond)
2474 unsigned long trans_start, last_rx;
2475 struct list_head *iter;
2476 struct slave *slave;
2479 bond_for_each_slave_rcu(bond, slave, iter) {
2480 slave->new_link = BOND_LINK_NOCHANGE;
2481 last_rx = slave_last_rx(bond, slave);
2483 if (slave->link != BOND_LINK_UP) {
2484 if (bond_time_in_interval(bond, last_rx, 1)) {
2485 slave->new_link = BOND_LINK_UP;
2492 * Give slaves 2*delta after being enslaved or made
2493 * active. This avoids bouncing, as the last receive
2494 * times need a full ARP monitor cycle to be updated.
2496 if (bond_time_in_interval(bond, slave->jiffies, 2))
2500 * Backup slave is down if:
2501 * - No current_arp_slave AND
2502 * - more than 3*delta since last receive AND
2503 * - the bond has an IP address
2505 * Note: a non-null current_arp_slave indicates
2506 * the curr_active_slave went down and we are
2507 * searching for a new one; under this condition
2508 * we only take the curr_active_slave down - this
2509 * gives each slave a chance to tx/rx traffic
2510 * before being taken out
2512 if (!bond_is_active_slave(slave) &&
2513 !bond->current_arp_slave &&
2514 !bond_time_in_interval(bond, last_rx, 3)) {
2515 slave->new_link = BOND_LINK_DOWN;
2520 * Active slave is down if:
2521 * - more than 2*delta since transmitting OR
2522 * - (more than 2*delta since receive AND
2523 * the bond has an IP address)
2525 trans_start = dev_trans_start(slave->dev);
2526 if (bond_is_active_slave(slave) &&
2527 (!bond_time_in_interval(bond, trans_start, 2) ||
2528 !bond_time_in_interval(bond, last_rx, 2))) {
2529 slave->new_link = BOND_LINK_DOWN;
2538 * Called to commit link state changes noted by inspection step of
2539 * active-backup mode ARP monitor.
2541 * Called with RTNL hold.
2543 static void bond_ab_arp_commit(struct bonding *bond)
2545 unsigned long trans_start;
2546 struct list_head *iter;
2547 struct slave *slave;
2549 bond_for_each_slave(bond, slave, iter) {
2550 switch (slave->new_link) {
2551 case BOND_LINK_NOCHANGE:
2555 trans_start = dev_trans_start(slave->dev);
2556 if (bond->curr_active_slave != slave ||
2557 (!bond->curr_active_slave &&
2558 bond_time_in_interval(bond, trans_start, 1))) {
2559 slave->link = BOND_LINK_UP;
2560 if (bond->current_arp_slave) {
2561 bond_set_slave_inactive_flags(
2562 bond->current_arp_slave);
2563 bond->current_arp_slave = NULL;
2566 pr_info("%s: link status definitely up for interface %s.\n",
2567 bond->dev->name, slave->dev->name);
2569 if (!bond->curr_active_slave ||
2570 (slave == bond->primary_slave))
2577 case BOND_LINK_DOWN:
2578 if (slave->link_failure_count < UINT_MAX)
2579 slave->link_failure_count++;
2581 slave->link = BOND_LINK_DOWN;
2582 bond_set_slave_inactive_flags(slave);
2584 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2585 bond->dev->name, slave->dev->name);
2587 if (slave == bond->curr_active_slave) {
2588 bond->current_arp_slave = NULL;
2595 pr_err("%s: impossible: new_link %d on slave %s\n",
2596 bond->dev->name, slave->new_link,
2604 write_lock_bh(&bond->curr_slave_lock);
2605 bond_select_active_slave(bond);
2606 write_unlock_bh(&bond->curr_slave_lock);
2607 unblock_netpoll_tx();
2610 bond_set_carrier(bond);
2614 * Send ARP probes for active-backup mode ARP monitor.
2616 static bool bond_ab_arp_probe(struct bonding *bond)
2618 struct slave *slave, *before = NULL, *new_slave = NULL,
2619 *curr_arp_slave, *curr_active_slave;
2620 struct list_head *iter;
2624 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2625 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2627 if (curr_arp_slave && curr_active_slave)
2628 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2629 curr_arp_slave->dev->name,
2630 curr_active_slave->dev->name);
2632 if (curr_active_slave) {
2633 bond_arp_send_all(bond, curr_active_slave);
2639 /* if we don't have a curr_active_slave, search for the next available
2640 * backup slave from the current_arp_slave and make it the candidate
2641 * for becoming the curr_active_slave
2644 if (!rtnl_trylock())
2646 /* curr_arp_slave might have gone away */
2647 curr_arp_slave = ACCESS_ONCE(bond->current_arp_slave);
2649 if (!curr_arp_slave) {
2650 curr_arp_slave = bond_first_slave(bond);
2651 if (!curr_arp_slave) {
2657 bond_set_slave_inactive_flags(curr_arp_slave);
2659 bond_for_each_slave(bond, slave, iter) {
2660 if (!found && !before && IS_UP(slave->dev))
2663 if (found && !new_slave && IS_UP(slave->dev))
2665 /* if the link state is up at this point, we
2666 * mark it down - this can happen if we have
2667 * simultaneous link failures and
2668 * reselect_active_interface doesn't make this
2669 * one the current slave so it is still marked
2670 * up when it is actually down
2672 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2673 slave->link = BOND_LINK_DOWN;
2674 if (slave->link_failure_count < UINT_MAX)
2675 slave->link_failure_count++;
2677 bond_set_slave_inactive_flags(slave);
2679 pr_info("%s: backup interface %s is now down.\n",
2680 bond->dev->name, slave->dev->name);
2682 if (slave == curr_arp_slave)
2686 if (!new_slave && before)
2694 new_slave->link = BOND_LINK_BACK;
2695 bond_set_slave_active_flags(new_slave);
2696 bond_arp_send_all(bond, new_slave);
2697 new_slave->jiffies = jiffies;
2698 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2704 static void bond_activebackup_arp_mon(struct work_struct *work)
2706 struct bonding *bond = container_of(work, struct bonding,
2708 bool should_notify_peers = false, should_commit = false;
2711 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2713 if (!bond_has_slaves(bond))
2717 should_notify_peers = bond_should_notify_peers(bond);
2718 should_commit = bond_ab_arp_inspect(bond);
2721 if (should_commit) {
2722 /* Race avoidance with bond_close flush of workqueue */
2723 if (!rtnl_trylock()) {
2725 should_notify_peers = false;
2729 bond_ab_arp_commit(bond);
2733 if (!bond_ab_arp_probe(bond)) {
2734 /* rtnl locking failed, re-arm */
2736 should_notify_peers = false;
2740 if (bond->params.arp_interval)
2741 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2743 if (should_notify_peers) {
2744 if (!rtnl_trylock())
2746 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2751 /*-------------------------- netdev event handling --------------------------*/
2754 * Change device name
2756 static int bond_event_changename(struct bonding *bond)
2758 bond_remove_proc_entry(bond);
2759 bond_create_proc_entry(bond);
2761 bond_debug_reregister(bond);
2766 static int bond_master_netdev_event(unsigned long event,
2767 struct net_device *bond_dev)
2769 struct bonding *event_bond = netdev_priv(bond_dev);
2772 case NETDEV_CHANGENAME:
2773 return bond_event_changename(event_bond);
2774 case NETDEV_UNREGISTER:
2775 bond_remove_proc_entry(event_bond);
2777 case NETDEV_REGISTER:
2778 bond_create_proc_entry(event_bond);
2780 case NETDEV_NOTIFY_PEERS:
2781 if (event_bond->send_peer_notif)
2782 event_bond->send_peer_notif--;
2791 static int bond_slave_netdev_event(unsigned long event,
2792 struct net_device *slave_dev)
2794 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2795 struct bonding *bond;
2796 struct net_device *bond_dev;
2800 /* A netdev event can be generated while enslaving a device
2801 * before netdev_rx_handler_register is called in which case
2802 * slave will be NULL
2806 bond_dev = slave->bond->dev;
2810 case NETDEV_UNREGISTER:
2811 if (bond_dev->type != ARPHRD_ETHER)
2812 bond_release_and_destroy(bond_dev, slave_dev);
2814 bond_release(bond_dev, slave_dev);
2818 old_speed = slave->speed;
2819 old_duplex = slave->duplex;
2821 bond_update_speed_duplex(slave);
2823 if (bond->params.mode == BOND_MODE_8023AD) {
2824 if (old_speed != slave->speed)
2825 bond_3ad_adapter_speed_changed(slave);
2826 if (old_duplex != slave->duplex)
2827 bond_3ad_adapter_duplex_changed(slave);
2832 * ... Or is it this?
2835 case NETDEV_CHANGEMTU:
2837 * TODO: Should slaves be allowed to
2838 * independently alter their MTU? For
2839 * an active-backup bond, slaves need
2840 * not be the same type of device, so
2841 * MTUs may vary. For other modes,
2842 * slaves arguably should have the
2843 * same MTUs. To do this, we'd need to
2844 * take over the slave's change_mtu
2845 * function for the duration of their
2849 case NETDEV_CHANGENAME:
2850 /* we don't care if we don't have primary set */
2851 if (!USES_PRIMARY(bond->params.mode) ||
2852 !bond->params.primary[0])
2855 if (slave == bond->primary_slave) {
2856 /* slave's name changed - he's no longer primary */
2857 bond->primary_slave = NULL;
2858 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2859 /* we have a new primary slave */
2860 bond->primary_slave = slave;
2861 } else { /* we didn't change primary - exit */
2865 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2866 bond->dev->name, bond->primary_slave ? slave_dev->name :
2868 write_lock_bh(&bond->curr_slave_lock);
2869 bond_select_active_slave(bond);
2870 write_unlock_bh(&bond->curr_slave_lock);
2872 case NETDEV_FEAT_CHANGE:
2873 bond_compute_features(bond);
2875 case NETDEV_RESEND_IGMP:
2876 /* Propagate to master device */
2877 call_netdevice_notifiers(event, slave->bond->dev);
2887 * bond_netdev_event: handle netdev notifier chain events.
2889 * This function receives events for the netdev chain. The caller (an
2890 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2891 * locks for us to safely manipulate the slave devices (RTNL lock,
2894 static int bond_netdev_event(struct notifier_block *this,
2895 unsigned long event, void *ptr)
2897 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2899 pr_debug("event_dev: %s, event: %lx\n",
2900 event_dev ? event_dev->name : "None",
2903 if (!(event_dev->priv_flags & IFF_BONDING))
2906 if (event_dev->flags & IFF_MASTER) {
2907 pr_debug("IFF_MASTER\n");
2908 return bond_master_netdev_event(event, event_dev);
2911 if (event_dev->flags & IFF_SLAVE) {
2912 pr_debug("IFF_SLAVE\n");
2913 return bond_slave_netdev_event(event, event_dev);
2919 static struct notifier_block bond_netdev_notifier = {
2920 .notifier_call = bond_netdev_event,
2923 /*---------------------------- Hashing Policies -----------------------------*/
2925 /* L2 hash helper */
2926 static inline u32 bond_eth_hash(struct sk_buff *skb)
2928 struct ethhdr *data = (struct ethhdr *)skb->data;
2930 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2931 return data->h_dest[5] ^ data->h_source[5];
2936 /* Extract the appropriate headers based on bond's xmit policy */
2937 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2938 struct flow_keys *fk)
2940 const struct ipv6hdr *iph6;
2941 const struct iphdr *iph;
2942 int noff, proto = -1;
2944 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2945 return skb_flow_dissect(skb, fk);
2948 noff = skb_network_offset(skb);
2949 if (skb->protocol == htons(ETH_P_IP)) {
2950 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2953 fk->src = iph->saddr;
2954 fk->dst = iph->daddr;
2955 noff += iph->ihl << 2;
2956 if (!ip_is_fragment(iph))
2957 proto = iph->protocol;
2958 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2959 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2961 iph6 = ipv6_hdr(skb);
2962 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2963 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2964 noff += sizeof(*iph6);
2965 proto = iph6->nexthdr;
2969 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2970 fk->ports = skb_flow_get_ports(skb, noff, proto);
2976 * bond_xmit_hash - generate a hash value based on the xmit policy
2977 * @bond: bonding device
2978 * @skb: buffer to use for headers
2979 * @count: modulo value
2981 * This function will extract the necessary headers from the skb buffer and use
2982 * them to generate a hash based on the xmit_policy set in the bonding device
2983 * which will be reduced modulo count before returning.
2985 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
2987 struct flow_keys flow;
2990 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
2991 !bond_flow_dissect(bond, skb, &flow))
2992 return bond_eth_hash(skb) % count;
2994 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
2995 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
2996 hash = bond_eth_hash(skb);
2998 hash = (__force u32)flow.ports;
2999 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3000 hash ^= (hash >> 16);
3001 hash ^= (hash >> 8);
3003 return hash % count;
3006 /*-------------------------- Device entry points ----------------------------*/
3008 static void bond_work_init_all(struct bonding *bond)
3010 INIT_DELAYED_WORK(&bond->mcast_work,
3011 bond_resend_igmp_join_requests_delayed);
3012 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3013 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3014 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3015 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3017 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3018 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3021 static void bond_work_cancel_all(struct bonding *bond)
3023 cancel_delayed_work_sync(&bond->mii_work);
3024 cancel_delayed_work_sync(&bond->arp_work);
3025 cancel_delayed_work_sync(&bond->alb_work);
3026 cancel_delayed_work_sync(&bond->ad_work);
3027 cancel_delayed_work_sync(&bond->mcast_work);
3030 static int bond_open(struct net_device *bond_dev)
3032 struct bonding *bond = netdev_priv(bond_dev);
3033 struct list_head *iter;
3034 struct slave *slave;
3036 /* reset slave->backup and slave->inactive */
3037 read_lock(&bond->lock);
3038 if (bond_has_slaves(bond)) {
3039 read_lock(&bond->curr_slave_lock);
3040 bond_for_each_slave(bond, slave, iter) {
3041 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3042 && (slave != bond->curr_active_slave)) {
3043 bond_set_slave_inactive_flags(slave);
3045 bond_set_slave_active_flags(slave);
3048 read_unlock(&bond->curr_slave_lock);
3050 read_unlock(&bond->lock);
3052 bond_work_init_all(bond);
3054 if (bond_is_lb(bond)) {
3055 /* bond_alb_initialize must be called before the timer
3058 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3060 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3063 if (bond->params.miimon) /* link check interval, in milliseconds. */
3064 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3066 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3067 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3068 if (bond->params.arp_validate)
3069 bond->recv_probe = bond_arp_rcv;
3072 if (bond->params.mode == BOND_MODE_8023AD) {
3073 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3074 /* register to receive LACPDUs */
3075 bond->recv_probe = bond_3ad_lacpdu_recv;
3076 bond_3ad_initiate_agg_selection(bond, 1);
3082 static int bond_close(struct net_device *bond_dev)
3084 struct bonding *bond = netdev_priv(bond_dev);
3086 bond_work_cancel_all(bond);
3087 bond->send_peer_notif = 0;
3088 if (bond_is_lb(bond))
3089 bond_alb_deinitialize(bond);
3090 bond->recv_probe = NULL;
3095 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3096 struct rtnl_link_stats64 *stats)
3098 struct bonding *bond = netdev_priv(bond_dev);
3099 struct rtnl_link_stats64 temp;
3100 struct list_head *iter;
3101 struct slave *slave;
3103 memset(stats, 0, sizeof(*stats));
3105 read_lock_bh(&bond->lock);
3106 bond_for_each_slave(bond, slave, iter) {
3107 const struct rtnl_link_stats64 *sstats =
3108 dev_get_stats(slave->dev, &temp);
3110 stats->rx_packets += sstats->rx_packets;
3111 stats->rx_bytes += sstats->rx_bytes;
3112 stats->rx_errors += sstats->rx_errors;
3113 stats->rx_dropped += sstats->rx_dropped;
3115 stats->tx_packets += sstats->tx_packets;
3116 stats->tx_bytes += sstats->tx_bytes;
3117 stats->tx_errors += sstats->tx_errors;
3118 stats->tx_dropped += sstats->tx_dropped;
3120 stats->multicast += sstats->multicast;
3121 stats->collisions += sstats->collisions;
3123 stats->rx_length_errors += sstats->rx_length_errors;
3124 stats->rx_over_errors += sstats->rx_over_errors;
3125 stats->rx_crc_errors += sstats->rx_crc_errors;
3126 stats->rx_frame_errors += sstats->rx_frame_errors;
3127 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3128 stats->rx_missed_errors += sstats->rx_missed_errors;
3130 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3131 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3132 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3133 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3134 stats->tx_window_errors += sstats->tx_window_errors;
3136 read_unlock_bh(&bond->lock);
3141 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3143 struct bonding *bond = netdev_priv(bond_dev);
3144 struct net_device *slave_dev = NULL;
3145 struct ifbond k_binfo;
3146 struct ifbond __user *u_binfo = NULL;
3147 struct ifslave k_sinfo;
3148 struct ifslave __user *u_sinfo = NULL;
3149 struct mii_ioctl_data *mii = NULL;
3150 struct bond_opt_value newval;
3154 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3166 * We do this again just in case we were called by SIOCGMIIREG
3167 * instead of SIOCGMIIPHY.
3174 if (mii->reg_num == 1) {
3176 read_lock(&bond->lock);
3177 read_lock(&bond->curr_slave_lock);
3178 if (netif_carrier_ok(bond->dev))
3179 mii->val_out = BMSR_LSTATUS;
3181 read_unlock(&bond->curr_slave_lock);
3182 read_unlock(&bond->lock);
3186 case BOND_INFO_QUERY_OLD:
3187 case SIOCBONDINFOQUERY:
3188 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3190 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3193 res = bond_info_query(bond_dev, &k_binfo);
3195 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3199 case BOND_SLAVE_INFO_QUERY_OLD:
3200 case SIOCBONDSLAVEINFOQUERY:
3201 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3203 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3206 res = bond_slave_info_query(bond_dev, &k_sinfo);
3208 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3217 net = dev_net(bond_dev);
3219 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3222 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3224 pr_debug("slave_dev=%p:\n", slave_dev);
3229 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3231 case BOND_ENSLAVE_OLD:
3232 case SIOCBONDENSLAVE:
3233 res = bond_enslave(bond_dev, slave_dev);
3235 case BOND_RELEASE_OLD:
3236 case SIOCBONDRELEASE:
3237 res = bond_release(bond_dev, slave_dev);
3239 case BOND_SETHWADDR_OLD:
3240 case SIOCBONDSETHWADDR:
3241 bond_set_dev_addr(bond_dev, slave_dev);
3244 case BOND_CHANGE_ACTIVE_OLD:
3245 case SIOCBONDCHANGEACTIVE:
3246 bond_opt_initstr(&newval, slave_dev->name);
3247 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3256 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3258 struct bonding *bond = netdev_priv(bond_dev);
3260 if (change & IFF_PROMISC)
3261 bond_set_promiscuity(bond,
3262 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3264 if (change & IFF_ALLMULTI)
3265 bond_set_allmulti(bond,
3266 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3269 static void bond_set_rx_mode(struct net_device *bond_dev)
3271 struct bonding *bond = netdev_priv(bond_dev);
3272 struct list_head *iter;
3273 struct slave *slave;
3277 if (USES_PRIMARY(bond->params.mode)) {
3278 slave = rcu_dereference(bond->curr_active_slave);
3280 dev_uc_sync(slave->dev, bond_dev);
3281 dev_mc_sync(slave->dev, bond_dev);
3284 bond_for_each_slave_rcu(bond, slave, iter) {
3285 dev_uc_sync_multiple(slave->dev, bond_dev);
3286 dev_mc_sync_multiple(slave->dev, bond_dev);
3292 static int bond_neigh_init(struct neighbour *n)
3294 struct bonding *bond = netdev_priv(n->dev);
3295 const struct net_device_ops *slave_ops;
3296 struct neigh_parms parms;
3297 struct slave *slave;
3300 slave = bond_first_slave(bond);
3303 slave_ops = slave->dev->netdev_ops;
3304 if (!slave_ops->ndo_neigh_setup)
3307 parms.neigh_setup = NULL;
3308 parms.neigh_cleanup = NULL;
3309 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3314 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3315 * after the last slave has been detached. Assumes that all slaves
3316 * utilize the same neigh_cleanup (true at this writing as only user
3319 n->parms->neigh_cleanup = parms.neigh_cleanup;
3321 if (!parms.neigh_setup)
3324 return parms.neigh_setup(n);
3328 * The bonding ndo_neigh_setup is called at init time beofre any
3329 * slave exists. So we must declare proxy setup function which will
3330 * be used at run time to resolve the actual slave neigh param setup.
3332 * It's also called by master devices (such as vlans) to setup their
3333 * underlying devices. In that case - do nothing, we're already set up from
3336 static int bond_neigh_setup(struct net_device *dev,
3337 struct neigh_parms *parms)
3339 /* modify only our neigh_parms */
3340 if (parms->dev == dev)
3341 parms->neigh_setup = bond_neigh_init;
3347 * Change the MTU of all of a master's slaves to match the master
3349 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3351 struct bonding *bond = netdev_priv(bond_dev);
3352 struct slave *slave, *rollback_slave;
3353 struct list_head *iter;
3356 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3357 (bond_dev ? bond_dev->name : "None"), new_mtu);
3359 /* Can't hold bond->lock with bh disabled here since
3360 * some base drivers panic. On the other hand we can't
3361 * hold bond->lock without bh disabled because we'll
3362 * deadlock. The only solution is to rely on the fact
3363 * that we're under rtnl_lock here, and the slaves
3364 * list won't change. This doesn't solve the problem
3365 * of setting the slave's MTU while it is
3366 * transmitting, but the assumption is that the base
3367 * driver can handle that.
3369 * TODO: figure out a way to safely iterate the slaves
3370 * list, but without holding a lock around the actual
3371 * call to the base driver.
3374 bond_for_each_slave(bond, slave, iter) {
3375 pr_debug("s %p c_m %p\n",
3377 slave->dev->netdev_ops->ndo_change_mtu);
3379 res = dev_set_mtu(slave->dev, new_mtu);
3382 /* If we failed to set the slave's mtu to the new value
3383 * we must abort the operation even in ACTIVE_BACKUP
3384 * mode, because if we allow the backup slaves to have
3385 * different mtu values than the active slave we'll
3386 * need to change their mtu when doing a failover. That
3387 * means changing their mtu from timer context, which
3388 * is probably not a good idea.
3390 pr_debug("err %d %s\n", res, slave->dev->name);
3395 bond_dev->mtu = new_mtu;
3400 /* unwind from head to the slave that failed */
3401 bond_for_each_slave(bond, rollback_slave, iter) {
3404 if (rollback_slave == slave)
3407 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3409 pr_debug("unwind err %d dev %s\n",
3410 tmp_res, rollback_slave->dev->name);
3420 * Note that many devices must be down to change the HW address, and
3421 * downing the master releases all slaves. We can make bonds full of
3422 * bonding devices to test this, however.
3424 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3426 struct bonding *bond = netdev_priv(bond_dev);
3427 struct slave *slave, *rollback_slave;
3428 struct sockaddr *sa = addr, tmp_sa;
3429 struct list_head *iter;
3432 if (bond->params.mode == BOND_MODE_ALB)
3433 return bond_alb_set_mac_address(bond_dev, addr);
3436 pr_debug("bond=%p, name=%s\n",
3437 bond, bond_dev ? bond_dev->name : "None");
3439 /* If fail_over_mac is enabled, do nothing and return success.
3440 * Returning an error causes ifenslave to fail.
3442 if (bond->params.fail_over_mac &&
3443 bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3446 if (!is_valid_ether_addr(sa->sa_data))
3447 return -EADDRNOTAVAIL;
3449 /* Can't hold bond->lock with bh disabled here since
3450 * some base drivers panic. On the other hand we can't
3451 * hold bond->lock without bh disabled because we'll
3452 * deadlock. The only solution is to rely on the fact
3453 * that we're under rtnl_lock here, and the slaves
3454 * list won't change. This doesn't solve the problem
3455 * of setting the slave's hw address while it is
3456 * transmitting, but the assumption is that the base
3457 * driver can handle that.
3459 * TODO: figure out a way to safely iterate the slaves
3460 * list, but without holding a lock around the actual
3461 * call to the base driver.
3464 bond_for_each_slave(bond, slave, iter) {
3465 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3466 pr_debug("slave %p %s\n", slave, slave->dev->name);
3468 if (slave_ops->ndo_set_mac_address == NULL) {
3470 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3474 res = dev_set_mac_address(slave->dev, addr);
3476 /* TODO: consider downing the slave
3478 * User should expect communications
3479 * breakage anyway until ARP finish
3482 pr_debug("err %d %s\n", res, slave->dev->name);
3488 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3492 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3493 tmp_sa.sa_family = bond_dev->type;
3495 /* unwind from head to the slave that failed */
3496 bond_for_each_slave(bond, rollback_slave, iter) {
3499 if (rollback_slave == slave)
3502 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3504 pr_debug("unwind err %d dev %s\n",
3505 tmp_res, rollback_slave->dev->name);
3513 * bond_xmit_slave_id - transmit skb through slave with slave_id
3514 * @bond: bonding device that is transmitting
3515 * @skb: buffer to transmit
3516 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3518 * This function tries to transmit through slave with slave_id but in case
3519 * it fails, it tries to find the first available slave for transmission.
3520 * The skb is consumed in all cases, thus the function is void.
3522 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3524 struct list_head *iter;
3525 struct slave *slave;
3528 /* Here we start from the slave with slave_id */
3529 bond_for_each_slave_rcu(bond, slave, iter) {
3531 if (slave_can_tx(slave)) {
3532 bond_dev_queue_xmit(bond, skb, slave->dev);
3538 /* Here we start from the first slave up to slave_id */
3540 bond_for_each_slave_rcu(bond, slave, iter) {
3543 if (slave_can_tx(slave)) {
3544 bond_dev_queue_xmit(bond, skb, slave->dev);
3548 /* no slave that can tx has been found */
3553 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3554 * @bond: bonding device to use
3556 * Based on the value of the bonding device's packets_per_slave parameter
3557 * this function generates a slave id, which is usually used as the next
3558 * slave to transmit through.
3560 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3563 struct reciprocal_value reciprocal_packets_per_slave;
3564 int packets_per_slave = bond->params.packets_per_slave;
3566 switch (packets_per_slave) {
3568 slave_id = prandom_u32();
3571 slave_id = bond->rr_tx_counter;
3574 reciprocal_packets_per_slave =
3575 bond->params.reciprocal_packets_per_slave;
3576 slave_id = reciprocal_divide(bond->rr_tx_counter,
3577 reciprocal_packets_per_slave);
3580 bond->rr_tx_counter++;
3585 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3587 struct bonding *bond = netdev_priv(bond_dev);
3588 struct iphdr *iph = ip_hdr(skb);
3589 struct slave *slave;
3592 /* Start with the curr_active_slave that joined the bond as the
3593 * default for sending IGMP traffic. For failover purposes one
3594 * needs to maintain some consistency for the interface that will
3595 * send the join/membership reports. The curr_active_slave found
3596 * will send all of this type of traffic.
3598 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3599 slave = rcu_dereference(bond->curr_active_slave);
3600 if (slave && slave_can_tx(slave))
3601 bond_dev_queue_xmit(bond, skb, slave->dev);
3603 bond_xmit_slave_id(bond, skb, 0);
3605 slave_id = bond_rr_gen_slave_id(bond);
3606 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3609 return NETDEV_TX_OK;
3613 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3614 * the bond has a usable interface.
3616 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3618 struct bonding *bond = netdev_priv(bond_dev);
3619 struct slave *slave;
3621 slave = rcu_dereference(bond->curr_active_slave);
3623 bond_dev_queue_xmit(bond, skb, slave->dev);
3627 return NETDEV_TX_OK;
3630 /* In bond_xmit_xor() , we determine the output device by using a pre-
3631 * determined xmit_hash_policy(), If the selected device is not enabled,
3632 * find the next active slave.
3634 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3636 struct bonding *bond = netdev_priv(bond_dev);
3638 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3640 return NETDEV_TX_OK;
3643 /* in broadcast mode, we send everything to all usable interfaces. */
3644 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3646 struct bonding *bond = netdev_priv(bond_dev);
3647 struct slave *slave = NULL;
3648 struct list_head *iter;
3650 bond_for_each_slave_rcu(bond, slave, iter) {
3651 if (bond_is_last_slave(bond, slave))
3653 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3654 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3657 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3661 /* bond_dev_queue_xmit always returns 0 */
3662 bond_dev_queue_xmit(bond, skb2, slave->dev);
3665 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3666 bond_dev_queue_xmit(bond, skb, slave->dev);
3670 return NETDEV_TX_OK;
3673 /*------------------------- Device initialization ---------------------------*/
3676 * Lookup the slave that corresponds to a qid
3678 static inline int bond_slave_override(struct bonding *bond,
3679 struct sk_buff *skb)
3681 struct slave *slave = NULL;
3682 struct list_head *iter;
3684 if (!skb->queue_mapping)
3687 /* Find out if any slaves have the same mapping as this skb. */
3688 bond_for_each_slave_rcu(bond, slave, iter) {
3689 if (slave->queue_id == skb->queue_mapping) {
3690 if (slave_can_tx(slave)) {
3691 bond_dev_queue_xmit(bond, skb, slave->dev);
3694 /* If the slave isn't UP, use default transmit policy. */
3703 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3707 * This helper function exists to help dev_pick_tx get the correct
3708 * destination queue. Using a helper function skips a call to
3709 * skb_tx_hash and will put the skbs in the queue we expect on their
3710 * way down to the bonding driver.
3712 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3715 * Save the original txq to restore before passing to the driver
3717 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3719 if (unlikely(txq >= dev->real_num_tx_queues)) {
3721 txq -= dev->real_num_tx_queues;
3722 } while (txq >= dev->real_num_tx_queues);
3727 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3729 struct bonding *bond = netdev_priv(dev);
3731 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3732 if (!bond_slave_override(bond, skb))
3733 return NETDEV_TX_OK;
3736 switch (bond->params.mode) {
3737 case BOND_MODE_ROUNDROBIN:
3738 return bond_xmit_roundrobin(skb, dev);
3739 case BOND_MODE_ACTIVEBACKUP:
3740 return bond_xmit_activebackup(skb, dev);
3742 return bond_xmit_xor(skb, dev);
3743 case BOND_MODE_BROADCAST:
3744 return bond_xmit_broadcast(skb, dev);
3745 case BOND_MODE_8023AD:
3746 return bond_3ad_xmit_xor(skb, dev);
3749 return bond_alb_xmit(skb, dev);
3751 /* Should never happen, mode already checked */
3752 pr_err("%s: Error: Unknown bonding mode %d\n",
3753 dev->name, bond->params.mode);
3756 return NETDEV_TX_OK;
3760 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3762 struct bonding *bond = netdev_priv(dev);
3763 netdev_tx_t ret = NETDEV_TX_OK;
3766 * If we risk deadlock from transmitting this in the
3767 * netpoll path, tell netpoll to queue the frame for later tx
3769 if (is_netpoll_tx_blocked(dev))
3770 return NETDEV_TX_BUSY;
3773 if (bond_has_slaves(bond))
3774 ret = __bond_start_xmit(skb, dev);
3782 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3783 struct ethtool_cmd *ecmd)
3785 struct bonding *bond = netdev_priv(bond_dev);
3786 unsigned long speed = 0;
3787 struct list_head *iter;
3788 struct slave *slave;
3790 ecmd->duplex = DUPLEX_UNKNOWN;
3791 ecmd->port = PORT_OTHER;
3793 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3794 * do not need to check mode. Though link speed might not represent
3795 * the true receive or transmit bandwidth (not all modes are symmetric)
3796 * this is an accurate maximum.
3798 read_lock(&bond->lock);
3799 bond_for_each_slave(bond, slave, iter) {
3800 if (SLAVE_IS_OK(slave)) {
3801 if (slave->speed != SPEED_UNKNOWN)
3802 speed += slave->speed;
3803 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3804 slave->duplex != DUPLEX_UNKNOWN)
3805 ecmd->duplex = slave->duplex;
3808 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3809 read_unlock(&bond->lock);
3814 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3815 struct ethtool_drvinfo *drvinfo)
3817 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3818 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3819 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3823 static const struct ethtool_ops bond_ethtool_ops = {
3824 .get_drvinfo = bond_ethtool_get_drvinfo,
3825 .get_settings = bond_ethtool_get_settings,
3826 .get_link = ethtool_op_get_link,
3829 static const struct net_device_ops bond_netdev_ops = {
3830 .ndo_init = bond_init,
3831 .ndo_uninit = bond_uninit,
3832 .ndo_open = bond_open,
3833 .ndo_stop = bond_close,
3834 .ndo_start_xmit = bond_start_xmit,
3835 .ndo_select_queue = bond_select_queue,
3836 .ndo_get_stats64 = bond_get_stats,
3837 .ndo_do_ioctl = bond_do_ioctl,
3838 .ndo_change_rx_flags = bond_change_rx_flags,
3839 .ndo_set_rx_mode = bond_set_rx_mode,
3840 .ndo_change_mtu = bond_change_mtu,
3841 .ndo_set_mac_address = bond_set_mac_address,
3842 .ndo_neigh_setup = bond_neigh_setup,
3843 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3844 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3845 #ifdef CONFIG_NET_POLL_CONTROLLER
3846 .ndo_netpoll_setup = bond_netpoll_setup,
3847 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3848 .ndo_poll_controller = bond_poll_controller,
3850 .ndo_add_slave = bond_enslave,
3851 .ndo_del_slave = bond_release,
3852 .ndo_fix_features = bond_fix_features,
3855 static const struct device_type bond_type = {
3859 static void bond_destructor(struct net_device *bond_dev)
3861 struct bonding *bond = netdev_priv(bond_dev);
3863 destroy_workqueue(bond->wq);
3864 free_netdev(bond_dev);
3867 void bond_setup(struct net_device *bond_dev)
3869 struct bonding *bond = netdev_priv(bond_dev);
3871 /* initialize rwlocks */
3872 rwlock_init(&bond->lock);
3873 rwlock_init(&bond->curr_slave_lock);
3874 bond->params = bonding_defaults;
3876 /* Initialize pointers */
3877 bond->dev = bond_dev;
3879 /* Initialize the device entry points */
3880 ether_setup(bond_dev);
3881 bond_dev->netdev_ops = &bond_netdev_ops;
3882 bond_dev->ethtool_ops = &bond_ethtool_ops;
3884 bond_dev->destructor = bond_destructor;
3886 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3888 /* Initialize the device options */
3889 bond_dev->tx_queue_len = 0;
3890 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3891 bond_dev->priv_flags |= IFF_BONDING;
3892 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3894 /* At first, we block adding VLANs. That's the only way to
3895 * prevent problems that occur when adding VLANs over an
3896 * empty bond. The block will be removed once non-challenged
3897 * slaves are enslaved.
3899 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3901 /* don't acquire bond device's netif_tx_lock when
3903 bond_dev->features |= NETIF_F_LLTX;
3905 /* By default, we declare the bond to be fully
3906 * VLAN hardware accelerated capable. Special
3907 * care is taken in the various xmit functions
3908 * when there are slaves that are not hw accel
3912 /* Don't allow bond devices to change network namespaces. */
3913 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3915 bond_dev->hw_features = BOND_VLAN_FEATURES |
3916 NETIF_F_HW_VLAN_CTAG_TX |
3917 NETIF_F_HW_VLAN_CTAG_RX |
3918 NETIF_F_HW_VLAN_CTAG_FILTER;
3920 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3921 bond_dev->features |= bond_dev->hw_features;
3925 * Destroy a bonding device.
3926 * Must be under rtnl_lock when this function is called.
3928 static void bond_uninit(struct net_device *bond_dev)
3930 struct bonding *bond = netdev_priv(bond_dev);
3931 struct list_head *iter;
3932 struct slave *slave;
3934 bond_netpoll_cleanup(bond_dev);
3936 /* Release the bonded slaves */
3937 bond_for_each_slave(bond, slave, iter)
3938 __bond_release_one(bond_dev, slave->dev, true);
3939 pr_info("%s: released all slaves\n", bond_dev->name);
3941 list_del(&bond->bond_list);
3943 bond_debug_unregister(bond);
3946 /*------------------------- Module initialization ---------------------------*/
3948 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3952 for (i = 0; tbl[i].modename; i++)
3953 if (mode == tbl[i].mode)
3959 static int bond_parm_tbl_lookup_name(const char *modename,
3960 const struct bond_parm_tbl *tbl)
3964 for (i = 0; tbl[i].modename; i++)
3965 if (strcmp(modename, tbl[i].modename) == 0)
3972 * Convert string input module parms. Accept either the
3973 * number of the mode or its string name. A bit complicated because
3974 * some mode names are substrings of other names, and calls from sysfs
3975 * may have whitespace in the name (trailing newlines, for example).
3977 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3980 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
3982 for (p = (char *)buf; *p; p++)
3983 if (!(isdigit(*p) || isspace(*p)))
3986 if (*p && sscanf(buf, "%20s", modestr) != 0)
3987 return bond_parm_tbl_lookup_name(modestr, tbl);
3988 else if (sscanf(buf, "%d", &modeint) != 0)
3989 return bond_parm_tbl_lookup(modeint, tbl);
3994 static int bond_check_params(struct bond_params *params)
3996 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
3997 struct bond_opt_value newval, *valptr;
3998 int arp_all_targets_value;
4001 * Convert string parameters.
4004 bond_opt_initstr(&newval, mode);
4005 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4007 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4010 bond_mode = valptr->value;
4013 if (xmit_hash_policy) {
4014 if ((bond_mode != BOND_MODE_XOR) &&
4015 (bond_mode != BOND_MODE_8023AD)) {
4016 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4017 bond_mode_name(bond_mode));
4019 bond_opt_initstr(&newval, xmit_hash_policy);
4020 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4023 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4027 xmit_hashtype = valptr->value;
4032 if (bond_mode != BOND_MODE_8023AD) {
4033 pr_info("lacp_rate param is irrelevant in mode %s\n",
4034 bond_mode_name(bond_mode));
4036 bond_opt_initstr(&newval, lacp_rate);
4037 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4040 pr_err("Error: Invalid lacp rate \"%s\"\n",
4044 lacp_fast = valptr->value;
4049 bond_opt_initstr(&newval, lacp_rate);
4050 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4053 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4056 params->ad_select = valptr->value;
4057 if (bond_mode != BOND_MODE_8023AD)
4058 pr_warning("ad_select param only affects 802.3ad mode\n");
4060 params->ad_select = BOND_AD_STABLE;
4063 if (max_bonds < 0) {
4064 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4065 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4066 max_bonds = BOND_DEFAULT_MAX_BONDS;
4070 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4076 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4081 if (downdelay < 0) {
4082 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4083 downdelay, INT_MAX);
4087 if ((use_carrier != 0) && (use_carrier != 1)) {
4088 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4093 if (num_peer_notif < 0 || num_peer_notif > 255) {
4094 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4099 /* reset values for 802.3ad/TLB/ALB */
4100 if (BOND_NO_USES_ARP(bond_mode)) {
4102 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");
4103 pr_warning("Forcing miimon to 100msec\n");
4104 miimon = BOND_DEFAULT_MIIMON;
4108 if (tx_queues < 1 || tx_queues > 255) {
4109 pr_warning("Warning: tx_queues (%d) should be between "
4110 "1 and 255, resetting to %d\n",
4111 tx_queues, BOND_DEFAULT_TX_QUEUES);
4112 tx_queues = BOND_DEFAULT_TX_QUEUES;
4115 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4116 pr_warning("Warning: all_slaves_active module parameter (%d), "
4117 "not of valid value (0/1), so it was set to "
4118 "0\n", all_slaves_active);
4119 all_slaves_active = 0;
4122 if (resend_igmp < 0 || resend_igmp > 255) {
4123 pr_warning("Warning: resend_igmp (%d) should be between "
4124 "0 and 255, resetting to %d\n",
4125 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4126 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4129 bond_opt_initval(&newval, packets_per_slave);
4130 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4131 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4132 packets_per_slave, USHRT_MAX);
4133 packets_per_slave = 1;
4136 if (bond_mode == BOND_MODE_ALB) {
4137 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",
4142 if (updelay || downdelay) {
4143 /* just warn the user the up/down delay will have
4144 * no effect since miimon is zero...
4146 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",
4147 updelay, downdelay);
4150 /* don't allow arp monitoring */
4152 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4153 miimon, arp_interval);
4157 if ((updelay % miimon) != 0) {
4158 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4160 (updelay / miimon) * miimon);
4165 if ((downdelay % miimon) != 0) {
4166 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4168 (downdelay / miimon) * miimon);
4171 downdelay /= miimon;
4174 if (arp_interval < 0) {
4175 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4176 arp_interval, INT_MAX);
4180 for (arp_ip_count = 0, i = 0;
4181 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4182 /* not complete check, but should be good enough to
4185 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4186 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4187 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4191 if (bond_get_targets_ip(arp_target, ip) == -1)
4192 arp_target[arp_ip_count++] = ip;
4194 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4199 if (arp_interval && !arp_ip_count) {
4200 /* don't allow arping if no arp_ip_target given... */
4201 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4207 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4208 pr_err("arp_validate only supported in active-backup mode\n");
4211 if (!arp_interval) {
4212 pr_err("arp_validate requires arp_interval\n");
4216 bond_opt_initstr(&newval, arp_validate);
4217 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4220 pr_err("Error: invalid arp_validate \"%s\"\n",
4224 arp_validate_value = valptr->value;
4226 arp_validate_value = 0;
4229 arp_all_targets_value = 0;
4230 if (arp_all_targets) {
4231 bond_opt_initstr(&newval, arp_all_targets);
4232 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4235 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4237 arp_all_targets_value = 0;
4239 arp_all_targets_value = valptr->value;
4244 pr_info("MII link monitoring set to %d ms\n", miimon);
4245 } else if (arp_interval) {
4246 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4247 arp_validate_value);
4248 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4249 arp_interval, valptr->string, arp_ip_count);
4251 for (i = 0; i < arp_ip_count; i++)
4252 pr_info(" %s", arp_ip_target[i]);
4256 } else if (max_bonds) {
4257 /* miimon and arp_interval not set, we need one so things
4258 * work as expected, see bonding.txt for details
4260 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");
4263 if (primary && !USES_PRIMARY(bond_mode)) {
4264 /* currently, using a primary only makes sense
4265 * in active backup, TLB or ALB modes
4267 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4268 primary, bond_mode_name(bond_mode));
4272 if (primary && primary_reselect) {
4273 bond_opt_initstr(&newval, primary_reselect);
4274 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4277 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4281 primary_reselect_value = valptr->value;
4283 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4286 if (fail_over_mac) {
4287 bond_opt_initstr(&newval, fail_over_mac);
4288 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4291 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4295 fail_over_mac_value = valptr->value;
4296 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4297 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4299 fail_over_mac_value = BOND_FOM_NONE;
4302 if (lp_interval == 0) {
4303 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4304 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4305 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4308 /* fill params struct with the proper values */
4309 params->mode = bond_mode;
4310 params->xmit_policy = xmit_hashtype;
4311 params->miimon = miimon;
4312 params->num_peer_notif = num_peer_notif;
4313 params->arp_interval = arp_interval;
4314 params->arp_validate = arp_validate_value;
4315 params->arp_all_targets = arp_all_targets_value;
4316 params->updelay = updelay;
4317 params->downdelay = downdelay;
4318 params->use_carrier = use_carrier;
4319 params->lacp_fast = lacp_fast;
4320 params->primary[0] = 0;
4321 params->primary_reselect = primary_reselect_value;
4322 params->fail_over_mac = fail_over_mac_value;
4323 params->tx_queues = tx_queues;
4324 params->all_slaves_active = all_slaves_active;
4325 params->resend_igmp = resend_igmp;
4326 params->min_links = min_links;
4327 params->lp_interval = lp_interval;
4328 params->packets_per_slave = packets_per_slave;
4329 if (packets_per_slave > 0) {
4330 params->reciprocal_packets_per_slave =
4331 reciprocal_value(packets_per_slave);
4333 /* reciprocal_packets_per_slave is unused if
4334 * packets_per_slave is 0 or 1, just initialize it
4336 params->reciprocal_packets_per_slave =
4337 (struct reciprocal_value) { 0 };
4341 strncpy(params->primary, primary, IFNAMSIZ);
4342 params->primary[IFNAMSIZ - 1] = 0;
4345 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4350 static struct lock_class_key bonding_netdev_xmit_lock_key;
4351 static struct lock_class_key bonding_netdev_addr_lock_key;
4352 static struct lock_class_key bonding_tx_busylock_key;
4354 static void bond_set_lockdep_class_one(struct net_device *dev,
4355 struct netdev_queue *txq,
4358 lockdep_set_class(&txq->_xmit_lock,
4359 &bonding_netdev_xmit_lock_key);
4362 static void bond_set_lockdep_class(struct net_device *dev)
4364 lockdep_set_class(&dev->addr_list_lock,
4365 &bonding_netdev_addr_lock_key);
4366 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4367 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4371 * Called from registration process
4373 static int bond_init(struct net_device *bond_dev)
4375 struct bonding *bond = netdev_priv(bond_dev);
4376 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4377 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4379 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4382 * Initialize locks that may be required during
4383 * en/deslave operations. All of the bond_open work
4384 * (of which this is part) should really be moved to
4385 * a phase prior to dev_open
4387 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4388 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4390 bond->wq = create_singlethread_workqueue(bond_dev->name);
4394 bond_set_lockdep_class(bond_dev);
4396 list_add_tail(&bond->bond_list, &bn->dev_list);
4398 bond_prepare_sysfs_group(bond);
4400 bond_debug_register(bond);
4402 /* Ensure valid dev_addr */
4403 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4404 bond_dev->addr_assign_type == NET_ADDR_PERM)
4405 eth_hw_addr_random(bond_dev);
4410 unsigned int bond_get_num_tx_queues(void)
4415 /* Create a new bond based on the specified name and bonding parameters.
4416 * If name is NULL, obtain a suitable "bond%d" name for us.
4417 * Caller must NOT hold rtnl_lock; we need to release it here before we
4418 * set up our sysfs entries.
4420 int bond_create(struct net *net, const char *name)
4422 struct net_device *bond_dev;
4427 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4428 name ? name : "bond%d",
4429 bond_setup, tx_queues);
4431 pr_err("%s: eek! can't alloc netdev!\n", name);
4436 dev_net_set(bond_dev, net);
4437 bond_dev->rtnl_link_ops = &bond_link_ops;
4439 res = register_netdevice(bond_dev);
4441 netif_carrier_off(bond_dev);
4445 bond_destructor(bond_dev);
4449 static int __net_init bond_net_init(struct net *net)
4451 struct bond_net *bn = net_generic(net, bond_net_id);
4454 INIT_LIST_HEAD(&bn->dev_list);
4456 bond_create_proc_dir(bn);
4457 bond_create_sysfs(bn);
4462 static void __net_exit bond_net_exit(struct net *net)
4464 struct bond_net *bn = net_generic(net, bond_net_id);
4465 struct bonding *bond, *tmp_bond;
4468 bond_destroy_sysfs(bn);
4469 bond_destroy_proc_dir(bn);
4471 /* Kill off any bonds created after unregistering bond rtnl ops */
4473 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4474 unregister_netdevice_queue(bond->dev, &list);
4475 unregister_netdevice_many(&list);
4479 static struct pernet_operations bond_net_ops = {
4480 .init = bond_net_init,
4481 .exit = bond_net_exit,
4483 .size = sizeof(struct bond_net),
4486 static int __init bonding_init(void)
4491 pr_info("%s", bond_version);
4493 res = bond_check_params(&bonding_defaults);
4497 res = register_pernet_subsys(&bond_net_ops);
4501 res = bond_netlink_init();
4505 bond_create_debugfs();
4507 for (i = 0; i < max_bonds; i++) {
4508 res = bond_create(&init_net, NULL);
4513 register_netdevice_notifier(&bond_netdev_notifier);
4517 bond_netlink_fini();
4519 unregister_pernet_subsys(&bond_net_ops);
4524 static void __exit bonding_exit(void)
4526 unregister_netdevice_notifier(&bond_netdev_notifier);
4528 bond_destroy_debugfs();
4530 bond_netlink_fini();
4531 unregister_pernet_subsys(&bond_net_ops);
4533 #ifdef CONFIG_NET_POLL_CONTROLLER
4535 * Make sure we don't have an imbalance on our netpoll blocking
4537 WARN_ON(atomic_read(&netpoll_block_tx));
4541 module_init(bonding_init);
4542 module_exit(bonding_exit);
4543 MODULE_LICENSE("GPL");
4544 MODULE_VERSION(DRV_VERSION);
4545 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4546 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");