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_MODE(bond) == 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 (bond_uses_primary(bond)) {
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 (bond_uses_primary(bond)) {
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_MODE(bond) == 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 not using primary keep all
589 * slaves up date at all times; only the modes that use primary 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 ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
677 ether_addr_copy(saddr.sa_data,
678 old_active->dev->dev_addr);
679 saddr.sa_family = new_active->dev->type;
681 ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
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 ether_addr_copy(saddr.sa_data, tmp_mac);
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 && bond_slave_is_up(slave) &&
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->last_link_up = jiffies;
803 if (new_active->link == BOND_LINK_BACK) {
804 if (bond_uses_primary(bond)) {
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_MODE(bond) == BOND_MODE_8023AD)
814 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
816 if (bond_is_lb(bond))
817 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
819 if (bond_uses_primary(bond)) {
820 pr_info("%s: making interface %s the new active one\n",
821 bond->dev->name, new_active->dev->name);
826 if (bond_uses_primary(bond))
827 bond_hw_addr_swap(bond, new_active, old_active);
829 if (bond_is_lb(bond)) {
830 bond_alb_handle_active_change(bond, new_active);
832 bond_set_slave_inactive_flags(old_active,
833 BOND_SLAVE_NOTIFY_NOW);
835 bond_set_slave_active_flags(new_active,
836 BOND_SLAVE_NOTIFY_NOW);
838 rcu_assign_pointer(bond->curr_active_slave, new_active);
841 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
843 bond_set_slave_inactive_flags(old_active,
844 BOND_SLAVE_NOTIFY_NOW);
847 bool should_notify_peers = false;
849 bond_set_slave_active_flags(new_active,
850 BOND_SLAVE_NOTIFY_NOW);
852 if (bond->params.fail_over_mac)
853 bond_do_fail_over_mac(bond, new_active,
856 if (netif_running(bond->dev)) {
857 bond->send_peer_notif =
858 bond->params.num_peer_notif;
859 should_notify_peers =
860 bond_should_notify_peers(bond);
863 write_unlock_bh(&bond->curr_slave_lock);
865 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
866 if (should_notify_peers)
867 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
870 write_lock_bh(&bond->curr_slave_lock);
874 /* resend IGMP joins since active slave has changed or
875 * all were sent on curr_active_slave.
876 * resend only if bond is brought up with the affected
877 * bonding modes and the retransmission is enabled */
878 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
879 ((bond_uses_primary(bond) && new_active) ||
880 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
881 bond->igmp_retrans = bond->params.resend_igmp;
882 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
887 * bond_select_active_slave - select a new active slave, if needed
888 * @bond: our bonding struct
890 * This functions should be called when one of the following occurs:
891 * - The old curr_active_slave has been released or lost its link.
892 * - The primary_slave has got its link back.
893 * - A slave has got its link back and there's no old curr_active_slave.
895 * Caller must hold curr_slave_lock for write_bh.
897 void bond_select_active_slave(struct bonding *bond)
899 struct slave *best_slave;
902 best_slave = bond_find_best_slave(bond);
903 if (best_slave != bond->curr_active_slave) {
904 bond_change_active_slave(bond, best_slave);
905 rv = bond_set_carrier(bond);
909 if (netif_carrier_ok(bond->dev)) {
910 pr_info("%s: first active interface up!\n",
913 pr_info("%s: now running without any active interface!\n",
919 #ifdef CONFIG_NET_POLL_CONTROLLER
920 static inline int slave_enable_netpoll(struct slave *slave)
925 np = kzalloc(sizeof(*np), GFP_KERNEL);
930 err = __netpoll_setup(np, slave->dev);
939 static inline void slave_disable_netpoll(struct slave *slave)
941 struct netpoll *np = slave->np;
947 __netpoll_free_async(np);
950 static void bond_poll_controller(struct net_device *bond_dev)
954 static void bond_netpoll_cleanup(struct net_device *bond_dev)
956 struct bonding *bond = netdev_priv(bond_dev);
957 struct list_head *iter;
960 bond_for_each_slave(bond, slave, iter)
961 if (bond_slave_is_up(slave))
962 slave_disable_netpoll(slave);
965 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
967 struct bonding *bond = netdev_priv(dev);
968 struct list_head *iter;
972 bond_for_each_slave(bond, slave, iter) {
973 err = slave_enable_netpoll(slave);
975 bond_netpoll_cleanup(dev);
982 static inline int slave_enable_netpoll(struct slave *slave)
986 static inline void slave_disable_netpoll(struct slave *slave)
989 static void bond_netpoll_cleanup(struct net_device *bond_dev)
994 /*---------------------------------- IOCTL ----------------------------------*/
996 static netdev_features_t bond_fix_features(struct net_device *dev,
997 netdev_features_t features)
999 struct bonding *bond = netdev_priv(dev);
1000 struct list_head *iter;
1001 netdev_features_t mask;
1002 struct slave *slave;
1004 if (!bond_has_slaves(bond)) {
1005 /* Disable adding VLANs to empty bond. But why? --mq */
1006 features |= NETIF_F_VLAN_CHALLENGED;
1011 features &= ~NETIF_F_ONE_FOR_ALL;
1012 features |= NETIF_F_ALL_FOR_ALL;
1014 bond_for_each_slave(bond, slave, iter) {
1015 features = netdev_increment_features(features,
1016 slave->dev->features,
1019 features = netdev_add_tso_features(features, mask);
1024 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1025 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1026 NETIF_F_HIGHDMA | NETIF_F_LRO)
1028 static void bond_compute_features(struct bonding *bond)
1030 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1031 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1032 struct net_device *bond_dev = bond->dev;
1033 struct list_head *iter;
1034 struct slave *slave;
1035 unsigned short max_hard_header_len = ETH_HLEN;
1036 unsigned int gso_max_size = GSO_MAX_SIZE;
1037 u16 gso_max_segs = GSO_MAX_SEGS;
1039 if (!bond_has_slaves(bond))
1042 bond_for_each_slave(bond, slave, iter) {
1043 vlan_features = netdev_increment_features(vlan_features,
1044 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1046 dst_release_flag &= slave->dev->priv_flags;
1047 if (slave->dev->hard_header_len > max_hard_header_len)
1048 max_hard_header_len = slave->dev->hard_header_len;
1050 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1051 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1055 bond_dev->vlan_features = vlan_features;
1056 bond_dev->hard_header_len = max_hard_header_len;
1057 bond_dev->gso_max_segs = gso_max_segs;
1058 netif_set_gso_max_size(bond_dev, gso_max_size);
1060 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1061 bond_dev->priv_flags = flags | dst_release_flag;
1063 netdev_change_features(bond_dev);
1066 static void bond_setup_by_slave(struct net_device *bond_dev,
1067 struct net_device *slave_dev)
1069 bond_dev->header_ops = slave_dev->header_ops;
1071 bond_dev->type = slave_dev->type;
1072 bond_dev->hard_header_len = slave_dev->hard_header_len;
1073 bond_dev->addr_len = slave_dev->addr_len;
1075 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1076 slave_dev->addr_len);
1079 /* On bonding slaves other than the currently active slave, suppress
1080 * duplicates except for alb non-mcast/bcast.
1082 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1083 struct slave *slave,
1084 struct bonding *bond)
1086 if (bond_is_slave_inactive(slave)) {
1087 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1088 skb->pkt_type != PACKET_BROADCAST &&
1089 skb->pkt_type != PACKET_MULTICAST)
1096 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1098 struct sk_buff *skb = *pskb;
1099 struct slave *slave;
1100 struct bonding *bond;
1101 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1103 int ret = RX_HANDLER_ANOTHER;
1105 skb = skb_share_check(skb, GFP_ATOMIC);
1107 return RX_HANDLER_CONSUMED;
1111 slave = bond_slave_get_rcu(skb->dev);
1114 recv_probe = ACCESS_ONCE(bond->recv_probe);
1116 ret = recv_probe(skb, bond, slave);
1117 if (ret == RX_HANDLER_CONSUMED) {
1123 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1124 return RX_HANDLER_EXACT;
1127 skb->dev = bond->dev;
1129 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1130 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1131 skb->pkt_type == PACKET_HOST) {
1133 if (unlikely(skb_cow_head(skb,
1134 skb->data - skb_mac_header(skb)))) {
1136 return RX_HANDLER_CONSUMED;
1138 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1144 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1145 struct net_device *slave_dev,
1146 struct slave *slave)
1150 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1153 slave_dev->flags |= IFF_SLAVE;
1154 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1158 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1159 struct net_device *slave_dev)
1161 netdev_upper_dev_unlink(slave_dev, bond_dev);
1162 slave_dev->flags &= ~IFF_SLAVE;
1163 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1166 static struct slave *bond_alloc_slave(struct bonding *bond)
1168 struct slave *slave = NULL;
1170 slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1174 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1175 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1177 if (!SLAVE_AD_INFO(slave)) {
1185 static void bond_free_slave(struct slave *slave)
1187 struct bonding *bond = bond_get_bond_by_slave(slave);
1189 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1190 kfree(SLAVE_AD_INFO(slave));
1195 /* enslave device <slave> to bond device <master> */
1196 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1198 struct bonding *bond = netdev_priv(bond_dev);
1199 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1200 struct slave *new_slave = NULL, *prev_slave;
1201 struct sockaddr addr;
1205 if (!bond->params.use_carrier &&
1206 slave_dev->ethtool_ops->get_link == NULL &&
1207 slave_ops->ndo_do_ioctl == NULL) {
1208 pr_warn("%s: Warning: no link monitoring support for %s\n",
1209 bond_dev->name, slave_dev->name);
1212 /* already enslaved */
1213 if (slave_dev->flags & IFF_SLAVE) {
1214 pr_debug("Error: Device was already enslaved\n");
1218 if (bond_dev == slave_dev) {
1219 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1223 /* vlan challenged mutual exclusion */
1224 /* no need to lock since we're protected by rtnl_lock */
1225 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1226 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1227 if (vlan_uses_dev(bond_dev)) {
1228 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1229 bond_dev->name, slave_dev->name, bond_dev->name);
1232 pr_warn("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1233 bond_dev->name, slave_dev->name,
1234 slave_dev->name, bond_dev->name);
1237 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1241 * Old ifenslave binaries are no longer supported. These can
1242 * be identified with moderate accuracy by the state of the slave:
1243 * the current ifenslave will set the interface down prior to
1244 * enslaving it; the old ifenslave will not.
1246 if ((slave_dev->flags & IFF_UP)) {
1247 pr_err("%s is up - this may be due to an out of date ifenslave\n",
1250 goto err_undo_flags;
1253 /* set bonding device ether type by slave - bonding netdevices are
1254 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1255 * there is a need to override some of the type dependent attribs/funcs.
1257 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1258 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1260 if (!bond_has_slaves(bond)) {
1261 if (bond_dev->type != slave_dev->type) {
1262 pr_debug("%s: change device type from %d to %d\n",
1264 bond_dev->type, slave_dev->type);
1266 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1268 res = notifier_to_errno(res);
1270 pr_err("%s: refused to change device type\n",
1273 goto err_undo_flags;
1276 /* Flush unicast and multicast addresses */
1277 dev_uc_flush(bond_dev);
1278 dev_mc_flush(bond_dev);
1280 if (slave_dev->type != ARPHRD_ETHER)
1281 bond_setup_by_slave(bond_dev, slave_dev);
1283 ether_setup(bond_dev);
1284 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1287 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1290 } else if (bond_dev->type != slave_dev->type) {
1291 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1292 slave_dev->name, slave_dev->type, bond_dev->type);
1294 goto err_undo_flags;
1297 if (slave_ops->ndo_set_mac_address == NULL) {
1298 if (!bond_has_slaves(bond)) {
1299 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address\n",
1301 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1302 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1303 pr_warn("%s: Setting fail_over_mac to active for active-backup mode\n",
1306 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1307 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",
1310 goto err_undo_flags;
1314 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1316 /* If this is the first slave, then we need to set the master's hardware
1317 * address to be the same as the slave's. */
1318 if (!bond_has_slaves(bond) &&
1319 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1320 bond_set_dev_addr(bond->dev, slave_dev);
1322 new_slave = bond_alloc_slave(bond);
1325 goto err_undo_flags;
1329 * Set the new_slave's queue_id to be zero. Queue ID mapping
1330 * is set via sysfs or module option if desired.
1332 new_slave->queue_id = 0;
1334 /* Save slave's original mtu and then set it to match the bond */
1335 new_slave->original_mtu = slave_dev->mtu;
1336 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1338 pr_debug("Error %d calling dev_set_mtu\n", res);
1343 * Save slave's original ("permanent") mac address for modes
1344 * that need it, and for restoring it upon release, and then
1345 * set it to the master's address
1347 ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1349 if (!bond->params.fail_over_mac ||
1350 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1352 * Set slave to master's mac address. The application already
1353 * set the master's mac address to that of the first slave
1355 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1356 addr.sa_family = slave_dev->type;
1357 res = dev_set_mac_address(slave_dev, &addr);
1359 pr_debug("Error %d calling set_mac_address\n", res);
1360 goto err_restore_mtu;
1364 /* open the slave since the application closed it */
1365 res = dev_open(slave_dev);
1367 pr_debug("Opening slave %s failed\n", slave_dev->name);
1368 goto err_restore_mac;
1371 new_slave->bond = bond;
1372 new_slave->dev = slave_dev;
1373 slave_dev->priv_flags |= IFF_BONDING;
1375 if (bond_is_lb(bond)) {
1376 /* bond_alb_init_slave() must be called before all other stages since
1377 * it might fail and we do not want to have to undo everything
1379 res = bond_alb_init_slave(bond, new_slave);
1384 /* If the mode uses primary, then the following is handled by
1385 * bond_change_active_slave().
1387 if (!bond_uses_primary(bond)) {
1388 /* set promiscuity level to new slave */
1389 if (bond_dev->flags & IFF_PROMISC) {
1390 res = dev_set_promiscuity(slave_dev, 1);
1395 /* set allmulti level to new slave */
1396 if (bond_dev->flags & IFF_ALLMULTI) {
1397 res = dev_set_allmulti(slave_dev, 1);
1402 netif_addr_lock_bh(bond_dev);
1404 dev_mc_sync_multiple(slave_dev, bond_dev);
1405 dev_uc_sync_multiple(slave_dev, bond_dev);
1407 netif_addr_unlock_bh(bond_dev);
1410 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1411 /* add lacpdu mc addr to mc list */
1412 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1414 dev_mc_add(slave_dev, lacpdu_multicast);
1417 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1419 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1420 bond_dev->name, slave_dev->name);
1424 prev_slave = bond_last_slave(bond);
1426 new_slave->delay = 0;
1427 new_slave->link_failure_count = 0;
1429 bond_update_speed_duplex(new_slave);
1431 new_slave->last_rx = jiffies -
1432 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1433 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1434 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1436 if (bond->params.miimon && !bond->params.use_carrier) {
1437 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1439 if ((link_reporting == -1) && !bond->params.arp_interval) {
1441 * miimon is set but a bonded network driver
1442 * does not support ETHTOOL/MII and
1443 * arp_interval is not set. Note: if
1444 * use_carrier is enabled, we will never go
1445 * here (because netif_carrier is always
1446 * supported); thus, we don't need to change
1447 * the messages for netif_carrier.
1449 pr_warn("%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",
1450 bond_dev->name, slave_dev->name);
1451 } else if (link_reporting == -1) {
1452 /* unable get link status using mii/ethtool */
1453 pr_warn("%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",
1454 bond_dev->name, slave_dev->name);
1458 /* check for initial state */
1459 if (bond->params.miimon) {
1460 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1461 if (bond->params.updelay) {
1462 new_slave->link = BOND_LINK_BACK;
1463 new_slave->delay = bond->params.updelay;
1465 new_slave->link = BOND_LINK_UP;
1468 new_slave->link = BOND_LINK_DOWN;
1470 } else if (bond->params.arp_interval) {
1471 new_slave->link = (netif_carrier_ok(slave_dev) ?
1472 BOND_LINK_UP : BOND_LINK_DOWN);
1474 new_slave->link = BOND_LINK_UP;
1477 if (new_slave->link != BOND_LINK_DOWN)
1478 new_slave->last_link_up = jiffies;
1479 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1480 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1481 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1483 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1484 /* if there is a primary slave, remember it */
1485 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1486 bond->primary_slave = new_slave;
1487 bond->force_primary = true;
1491 switch (BOND_MODE(bond)) {
1492 case BOND_MODE_ACTIVEBACKUP:
1493 bond_set_slave_inactive_flags(new_slave,
1494 BOND_SLAVE_NOTIFY_NOW);
1496 case BOND_MODE_8023AD:
1497 /* in 802.3ad mode, the internal mechanism
1498 * will activate the slaves in the selected
1501 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1502 /* if this is the first slave */
1504 SLAVE_AD_INFO(new_slave)->id = 1;
1505 /* Initialize AD with the number of times that the AD timer is called in 1 second
1506 * can be called only after the mac address of the bond is set
1508 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1510 SLAVE_AD_INFO(new_slave)->id =
1511 SLAVE_AD_INFO(prev_slave)->id + 1;
1514 bond_3ad_bind_slave(new_slave);
1518 bond_set_active_slave(new_slave);
1519 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1522 pr_debug("This slave is always active in trunk mode\n");
1524 /* always active in trunk mode */
1525 bond_set_active_slave(new_slave);
1527 /* In trunking mode there is little meaning to curr_active_slave
1528 * anyway (it holds no special properties of the bond device),
1529 * so we can change it without calling change_active_interface()
1531 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1532 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1535 } /* switch(bond_mode) */
1537 #ifdef CONFIG_NET_POLL_CONTROLLER
1538 slave_dev->npinfo = bond->dev->npinfo;
1539 if (slave_dev->npinfo) {
1540 if (slave_enable_netpoll(new_slave)) {
1541 pr_info("Error, %s: master_dev is using netpoll, but new slave device does not support netpoll\n",
1549 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1552 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1556 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1558 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1559 goto err_unregister;
1562 res = bond_sysfs_slave_add(new_slave);
1564 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1565 goto err_upper_unlink;
1569 bond_compute_features(bond);
1570 bond_set_carrier(bond);
1572 if (bond_uses_primary(bond)) {
1574 write_lock_bh(&bond->curr_slave_lock);
1575 bond_select_active_slave(bond);
1576 write_unlock_bh(&bond->curr_slave_lock);
1577 unblock_netpoll_tx();
1580 pr_info("%s: Enslaving %s as %s interface with %s link\n",
1581 bond_dev->name, slave_dev->name,
1582 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1583 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1585 /* enslave is successful */
1588 /* Undo stages on error */
1590 bond_upper_dev_unlink(bond_dev, slave_dev);
1593 netdev_rx_handler_unregister(slave_dev);
1596 if (!bond_uses_primary(bond))
1597 bond_hw_addr_flush(bond_dev, slave_dev);
1599 vlan_vids_del_by_dev(slave_dev, bond_dev);
1600 if (bond->primary_slave == new_slave)
1601 bond->primary_slave = NULL;
1602 if (bond->curr_active_slave == new_slave) {
1604 write_lock_bh(&bond->curr_slave_lock);
1605 bond_change_active_slave(bond, NULL);
1606 bond_select_active_slave(bond);
1607 write_unlock_bh(&bond->curr_slave_lock);
1608 unblock_netpoll_tx();
1610 slave_disable_netpoll(new_slave);
1613 slave_dev->priv_flags &= ~IFF_BONDING;
1614 dev_close(slave_dev);
1617 if (!bond->params.fail_over_mac ||
1618 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1619 /* XXX TODO - fom follow mode needs to change master's
1620 * MAC if this slave's MAC is in use by the bond, or at
1621 * least print a warning.
1623 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1624 addr.sa_family = slave_dev->type;
1625 dev_set_mac_address(slave_dev, &addr);
1629 dev_set_mtu(slave_dev, new_slave->original_mtu);
1632 bond_free_slave(new_slave);
1635 /* Enslave of first slave has failed and we need to fix master's mac */
1636 if (!bond_has_slaves(bond) &&
1637 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1638 eth_hw_addr_random(bond_dev);
1644 * Try to release the slave device <slave> from the bond device <master>
1645 * It is legal to access curr_active_slave without a lock because all the function
1646 * is write-locked. If "all" is true it means that the function is being called
1647 * while destroying a bond interface and all slaves are being released.
1649 * The rules for slave state should be:
1650 * for Active/Backup:
1651 * Active stays on all backups go down
1652 * for Bonded connections:
1653 * The first up interface should be left on and all others downed.
1655 static int __bond_release_one(struct net_device *bond_dev,
1656 struct net_device *slave_dev,
1659 struct bonding *bond = netdev_priv(bond_dev);
1660 struct slave *slave, *oldcurrent;
1661 struct sockaddr addr;
1662 int old_flags = bond_dev->flags;
1663 netdev_features_t old_features = bond_dev->features;
1665 /* slave is not a slave or master is not master of this slave */
1666 if (!(slave_dev->flags & IFF_SLAVE) ||
1667 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1668 pr_err("%s: Error: cannot release %s\n",
1669 bond_dev->name, slave_dev->name);
1675 slave = bond_get_slave_by_dev(bond, slave_dev);
1677 /* not a slave of this bond */
1678 pr_info("%s: %s not enslaved\n",
1679 bond_dev->name, slave_dev->name);
1680 unblock_netpoll_tx();
1684 bond_sysfs_slave_del(slave);
1686 bond_upper_dev_unlink(bond_dev, slave_dev);
1687 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1688 * for this slave anymore.
1690 netdev_rx_handler_unregister(slave_dev);
1691 write_lock_bh(&bond->lock);
1693 /* Inform AD package of unbinding of slave. */
1694 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1695 bond_3ad_unbind_slave(slave);
1697 write_unlock_bh(&bond->lock);
1699 pr_info("%s: Releasing %s interface %s\n",
1701 bond_is_active_slave(slave) ? "active" : "backup",
1704 oldcurrent = bond->curr_active_slave;
1706 bond->current_arp_slave = NULL;
1708 if (!all && (!bond->params.fail_over_mac ||
1709 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1710 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1711 bond_has_slaves(bond))
1712 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",
1713 bond_dev->name, slave_dev->name,
1715 bond_dev->name, slave_dev->name);
1718 if (bond->primary_slave == slave)
1719 bond->primary_slave = NULL;
1721 if (oldcurrent == slave) {
1722 write_lock_bh(&bond->curr_slave_lock);
1723 bond_change_active_slave(bond, NULL);
1724 write_unlock_bh(&bond->curr_slave_lock);
1727 if (bond_is_lb(bond)) {
1728 /* Must be called only after the slave has been
1729 * detached from the list and the curr_active_slave
1730 * has been cleared (if our_slave == old_current),
1731 * but before a new active slave is selected.
1733 bond_alb_deinit_slave(bond, slave);
1737 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1738 } else if (oldcurrent == slave) {
1740 * Note that we hold RTNL over this sequence, so there
1741 * is no concern that another slave add/remove event
1744 write_lock_bh(&bond->curr_slave_lock);
1746 bond_select_active_slave(bond);
1748 write_unlock_bh(&bond->curr_slave_lock);
1751 if (!bond_has_slaves(bond)) {
1752 bond_set_carrier(bond);
1753 eth_hw_addr_random(bond_dev);
1755 if (vlan_uses_dev(bond_dev)) {
1756 pr_warn("%s: Warning: clearing HW address of %s while it still has VLANs\n",
1757 bond_dev->name, bond_dev->name);
1758 pr_warn("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs\n",
1763 unblock_netpoll_tx();
1767 if (!bond_has_slaves(bond)) {
1768 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1769 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1772 bond_compute_features(bond);
1773 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1774 (old_features & NETIF_F_VLAN_CHALLENGED))
1775 pr_info("%s: last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1776 bond_dev->name, slave_dev->name, bond_dev->name);
1778 /* must do this from outside any spinlocks */
1779 vlan_vids_del_by_dev(slave_dev, bond_dev);
1781 /* If the mode uses primary, then this cases was handled above by
1782 * bond_change_active_slave(..., NULL)
1784 if (!bond_uses_primary(bond)) {
1785 /* unset promiscuity level from slave
1786 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1787 * of the IFF_PROMISC flag in the bond_dev, but we need the
1788 * value of that flag before that change, as that was the value
1789 * when this slave was attached, so we cache at the start of the
1790 * function and use it here. Same goes for ALLMULTI below
1792 if (old_flags & IFF_PROMISC)
1793 dev_set_promiscuity(slave_dev, -1);
1795 /* unset allmulti level from slave */
1796 if (old_flags & IFF_ALLMULTI)
1797 dev_set_allmulti(slave_dev, -1);
1799 bond_hw_addr_flush(bond_dev, slave_dev);
1802 slave_disable_netpoll(slave);
1804 /* close slave before restoring its mac address */
1805 dev_close(slave_dev);
1807 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1808 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1809 /* restore original ("permanent") mac address */
1810 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1811 addr.sa_family = slave_dev->type;
1812 dev_set_mac_address(slave_dev, &addr);
1815 dev_set_mtu(slave_dev, slave->original_mtu);
1817 slave_dev->priv_flags &= ~IFF_BONDING;
1819 bond_free_slave(slave);
1821 return 0; /* deletion OK */
1824 /* A wrapper used because of ndo_del_link */
1825 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1827 return __bond_release_one(bond_dev, slave_dev, false);
1831 * First release a slave and then destroy the bond if no more slaves are left.
1832 * Must be under rtnl_lock when this function is called.
1834 static int bond_release_and_destroy(struct net_device *bond_dev,
1835 struct net_device *slave_dev)
1837 struct bonding *bond = netdev_priv(bond_dev);
1840 ret = bond_release(bond_dev, slave_dev);
1841 if (ret == 0 && !bond_has_slaves(bond)) {
1842 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1843 pr_info("%s: Destroying bond %s\n",
1844 bond_dev->name, bond_dev->name);
1845 unregister_netdevice(bond_dev);
1850 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1852 struct bonding *bond = netdev_priv(bond_dev);
1854 info->bond_mode = BOND_MODE(bond);
1855 info->miimon = bond->params.miimon;
1857 info->num_slaves = bond->slave_cnt;
1862 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1864 struct bonding *bond = netdev_priv(bond_dev);
1865 struct list_head *iter;
1866 int i = 0, res = -ENODEV;
1867 struct slave *slave;
1869 bond_for_each_slave(bond, slave, iter) {
1870 if (i++ == (int)info->slave_id) {
1872 strcpy(info->slave_name, slave->dev->name);
1873 info->link = slave->link;
1874 info->state = bond_slave_state(slave);
1875 info->link_failure_count = slave->link_failure_count;
1883 /*-------------------------------- Monitoring -------------------------------*/
1886 static int bond_miimon_inspect(struct bonding *bond)
1888 int link_state, commit = 0;
1889 struct list_head *iter;
1890 struct slave *slave;
1891 bool ignore_updelay;
1893 ignore_updelay = !bond->curr_active_slave ? true : false;
1895 bond_for_each_slave_rcu(bond, slave, iter) {
1896 slave->new_link = BOND_LINK_NOCHANGE;
1898 link_state = bond_check_dev_link(bond, slave->dev, 0);
1900 switch (slave->link) {
1905 slave->link = BOND_LINK_FAIL;
1906 slave->delay = bond->params.downdelay;
1908 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms\n",
1911 BOND_MODE_ACTIVEBACKUP) ?
1912 (bond_is_active_slave(slave) ?
1913 "active " : "backup ") : "",
1915 bond->params.downdelay * bond->params.miimon);
1918 case BOND_LINK_FAIL:
1921 * recovered before downdelay expired
1923 slave->link = BOND_LINK_UP;
1924 slave->last_link_up = jiffies;
1925 pr_info("%s: link status up again after %d ms for interface %s\n",
1927 (bond->params.downdelay - slave->delay) *
1928 bond->params.miimon,
1933 if (slave->delay <= 0) {
1934 slave->new_link = BOND_LINK_DOWN;
1942 case BOND_LINK_DOWN:
1946 slave->link = BOND_LINK_BACK;
1947 slave->delay = bond->params.updelay;
1950 pr_info("%s: link status up for interface %s, enabling it in %d ms\n",
1951 bond->dev->name, slave->dev->name,
1952 ignore_updelay ? 0 :
1953 bond->params.updelay *
1954 bond->params.miimon);
1957 case BOND_LINK_BACK:
1959 slave->link = BOND_LINK_DOWN;
1960 pr_info("%s: link status down again after %d ms for interface %s\n",
1962 (bond->params.updelay - slave->delay) *
1963 bond->params.miimon,
1972 if (slave->delay <= 0) {
1973 slave->new_link = BOND_LINK_UP;
1975 ignore_updelay = false;
1987 static void bond_miimon_commit(struct bonding *bond)
1989 struct list_head *iter;
1990 struct slave *slave;
1992 bond_for_each_slave(bond, slave, iter) {
1993 switch (slave->new_link) {
1994 case BOND_LINK_NOCHANGE:
1998 slave->link = BOND_LINK_UP;
1999 slave->last_link_up = jiffies;
2001 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2002 /* prevent it from being the active one */
2003 bond_set_backup_slave(slave);
2004 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2005 /* make it immediately active */
2006 bond_set_active_slave(slave);
2007 } else if (slave != bond->primary_slave) {
2008 /* prevent it from being the active one */
2009 bond_set_backup_slave(slave);
2012 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex\n",
2013 bond->dev->name, slave->dev->name,
2014 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2015 slave->duplex ? "full" : "half");
2017 /* notify ad that the link status has changed */
2018 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2019 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2021 if (bond_is_lb(bond))
2022 bond_alb_handle_link_change(bond, slave,
2025 if (!bond->curr_active_slave ||
2026 (slave == bond->primary_slave))
2031 case BOND_LINK_DOWN:
2032 if (slave->link_failure_count < UINT_MAX)
2033 slave->link_failure_count++;
2035 slave->link = BOND_LINK_DOWN;
2037 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2038 BOND_MODE(bond) == BOND_MODE_8023AD)
2039 bond_set_slave_inactive_flags(slave,
2040 BOND_SLAVE_NOTIFY_NOW);
2042 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2043 bond->dev->name, slave->dev->name);
2045 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2046 bond_3ad_handle_link_change(slave,
2049 if (bond_is_lb(bond))
2050 bond_alb_handle_link_change(bond, slave,
2053 if (slave == bond->curr_active_slave)
2059 pr_err("%s: invalid new link %d on slave %s\n",
2060 bond->dev->name, slave->new_link,
2062 slave->new_link = BOND_LINK_NOCHANGE;
2070 write_lock_bh(&bond->curr_slave_lock);
2071 bond_select_active_slave(bond);
2072 write_unlock_bh(&bond->curr_slave_lock);
2073 unblock_netpoll_tx();
2076 bond_set_carrier(bond);
2082 * Really a wrapper that splits the mii monitor into two phases: an
2083 * inspection, then (if inspection indicates something needs to be done)
2084 * an acquisition of appropriate locks followed by a commit phase to
2085 * implement whatever link state changes are indicated.
2087 static void bond_mii_monitor(struct work_struct *work)
2089 struct bonding *bond = container_of(work, struct bonding,
2091 bool should_notify_peers = false;
2092 unsigned long delay;
2094 delay = msecs_to_jiffies(bond->params.miimon);
2096 if (!bond_has_slaves(bond))
2101 should_notify_peers = bond_should_notify_peers(bond);
2103 if (bond_miimon_inspect(bond)) {
2106 /* Race avoidance with bond_close cancel of workqueue */
2107 if (!rtnl_trylock()) {
2109 should_notify_peers = false;
2113 bond_miimon_commit(bond);
2115 rtnl_unlock(); /* might sleep, hold no other locks */
2120 if (bond->params.miimon)
2121 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2123 if (should_notify_peers) {
2124 if (!rtnl_trylock())
2126 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2131 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2133 struct net_device *upper;
2134 struct list_head *iter;
2137 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2141 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2142 if (ip == bond_confirm_addr(upper, 0, ip)) {
2153 * We go to the (large) trouble of VLAN tagging ARP frames because
2154 * switches in VLAN mode (especially if ports are configured as
2155 * "native" to a VLAN) might not pass non-tagged frames.
2157 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2158 __be32 dest_ip, __be32 src_ip,
2159 struct bond_vlan_tag *inner,
2160 struct bond_vlan_tag *outer)
2162 struct sk_buff *skb;
2164 pr_debug("arp %d on slave %s: dst %pI4 src %pI4\n",
2165 arp_op, slave_dev->name, &dest_ip, &src_ip);
2167 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2168 NULL, slave_dev->dev_addr, NULL);
2171 net_err_ratelimited("ARP packet allocation failed\n");
2174 if (outer->vlan_id) {
2175 if (inner->vlan_id) {
2176 pr_debug("inner tag: proto %X vid %X\n",
2177 ntohs(inner->vlan_proto), inner->vlan_id);
2178 skb = __vlan_put_tag(skb, inner->vlan_proto,
2181 net_err_ratelimited("failed to insert inner VLAN tag\n");
2186 pr_debug("outer reg: proto %X vid %X\n",
2187 ntohs(outer->vlan_proto), outer->vlan_id);
2188 skb = vlan_put_tag(skb, outer->vlan_proto, outer->vlan_id);
2190 net_err_ratelimited("failed to insert outer VLAN tag\n");
2198 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2200 struct net_device *upper, *vlan_upper;
2201 struct list_head *iter, *vlan_iter;
2203 struct bond_vlan_tag inner, outer;
2204 __be32 *targets = bond->params.arp_targets, addr;
2207 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2208 pr_debug("basa: target %pI4\n", &targets[i]);
2209 inner.vlan_proto = 0;
2211 outer.vlan_proto = 0;
2214 /* Find out through which dev should the packet go */
2215 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2218 /* there's no route to target - try to send arp
2219 * probe to generate any traffic (arp_validate=0)
2221 if (bond->params.arp_validate)
2222 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2225 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i], 0, &inner, &outer);
2229 /* bond device itself */
2230 if (rt->dst.dev == bond->dev)
2234 /* first we search only for vlan devices. for every vlan
2235 * found we verify its upper dev list, searching for the
2236 * rt->dst.dev. If found we save the tag of the vlan and
2237 * proceed to send the packet.
2239 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2241 if (!is_vlan_dev(vlan_upper))
2244 if (vlan_upper == rt->dst.dev) {
2245 outer.vlan_proto = vlan_dev_vlan_proto(vlan_upper);
2246 outer.vlan_id = vlan_dev_vlan_id(vlan_upper);
2250 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2252 if (upper == rt->dst.dev) {
2253 /* If the upper dev is a vlan dev too,
2254 * set the vlan tag to inner tag.
2256 if (is_vlan_dev(upper)) {
2257 inner.vlan_proto = vlan_dev_vlan_proto(upper);
2258 inner.vlan_id = vlan_dev_vlan_id(upper);
2260 outer.vlan_proto = vlan_dev_vlan_proto(vlan_upper);
2261 outer.vlan_id = vlan_dev_vlan_id(vlan_upper);
2268 /* if the device we're looking for is not on top of any of
2269 * our upper vlans, then just search for any dev that
2270 * matches, and in case it's a vlan - save the id
2272 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2273 if (upper == rt->dst.dev) {
2280 /* Not our device - skip */
2281 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2282 bond->dev->name, &targets[i],
2283 rt->dst.dev ? rt->dst.dev->name : "NULL");
2289 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2291 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2292 addr, &inner, &outer);
2296 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2300 if (!sip || !bond_has_this_ip(bond, tip)) {
2301 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2305 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2307 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2310 slave->last_rx = jiffies;
2311 slave->target_last_arp_rx[i] = jiffies;
2314 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2315 struct slave *slave)
2317 struct arphdr *arp = (struct arphdr *)skb->data;
2318 struct slave *curr_active_slave;
2319 unsigned char *arp_ptr;
2321 int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2323 if (!slave_do_arp_validate(bond, slave)) {
2324 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2325 !slave_do_arp_validate_only(bond))
2326 slave->last_rx = jiffies;
2327 return RX_HANDLER_ANOTHER;
2328 } else if (!is_arp) {
2329 return RX_HANDLER_ANOTHER;
2332 alen = arp_hdr_len(bond->dev);
2334 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2335 bond->dev->name, skb->dev->name);
2337 if (alen > skb_headlen(skb)) {
2338 arp = kmalloc(alen, GFP_ATOMIC);
2341 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2345 if (arp->ar_hln != bond->dev->addr_len ||
2346 skb->pkt_type == PACKET_OTHERHOST ||
2347 skb->pkt_type == PACKET_LOOPBACK ||
2348 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2349 arp->ar_pro != htons(ETH_P_IP) ||
2353 arp_ptr = (unsigned char *)(arp + 1);
2354 arp_ptr += bond->dev->addr_len;
2355 memcpy(&sip, arp_ptr, 4);
2356 arp_ptr += 4 + bond->dev->addr_len;
2357 memcpy(&tip, arp_ptr, 4);
2359 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2360 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2361 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2364 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2367 * Backup slaves won't see the ARP reply, but do come through
2368 * here for each ARP probe (so we swap the sip/tip to validate
2369 * the probe). In a "redundant switch, common router" type of
2370 * configuration, the ARP probe will (hopefully) travel from
2371 * the active, through one switch, the router, then the other
2372 * switch before reaching the backup.
2374 * We 'trust' the arp requests if there is an active slave and
2375 * it received valid arp reply(s) after it became active. This
2376 * is done to avoid endless looping when we can't reach the
2377 * arp_ip_target and fool ourselves with our own arp requests.
2380 if (bond_is_active_slave(slave))
2381 bond_validate_arp(bond, slave, sip, tip);
2382 else if (curr_active_slave &&
2383 time_after(slave_last_rx(bond, curr_active_slave),
2384 curr_active_slave->last_link_up))
2385 bond_validate_arp(bond, slave, tip, sip);
2388 if (arp != (struct arphdr *)skb->data)
2390 return RX_HANDLER_ANOTHER;
2393 /* function to verify if we're in the arp_interval timeslice, returns true if
2394 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2395 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2397 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2400 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2402 return time_in_range(jiffies,
2403 last_act - delta_in_ticks,
2404 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2408 * this function is called regularly to monitor each slave's link
2409 * ensuring that traffic is being sent and received when arp monitoring
2410 * is used in load-balancing mode. if the adapter has been dormant, then an
2411 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2412 * arp monitoring in active backup mode.
2414 static void bond_loadbalance_arp_mon(struct work_struct *work)
2416 struct bonding *bond = container_of(work, struct bonding,
2418 struct slave *slave, *oldcurrent;
2419 struct list_head *iter;
2420 int do_failover = 0, slave_state_changed = 0;
2422 if (!bond_has_slaves(bond))
2427 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2428 /* see if any of the previous devices are up now (i.e. they have
2429 * xmt and rcv traffic). the curr_active_slave does not come into
2430 * the picture unless it is null. also, slave->last_link_up is not
2431 * needed here because we send an arp on each slave and give a slave
2432 * as long as it needs to get the tx/rx within the delta.
2433 * TODO: what about up/down delay in arp mode? it wasn't here before
2436 bond_for_each_slave_rcu(bond, slave, iter) {
2437 unsigned long trans_start = dev_trans_start(slave->dev);
2439 if (slave->link != BOND_LINK_UP) {
2440 if (bond_time_in_interval(bond, trans_start, 1) &&
2441 bond_time_in_interval(bond, slave->last_rx, 1)) {
2443 slave->link = BOND_LINK_UP;
2444 slave_state_changed = 1;
2446 /* primary_slave has no meaning in round-robin
2447 * mode. the window of a slave being up and
2448 * curr_active_slave being null after enslaving
2452 pr_info("%s: link status definitely up for interface %s\n",
2457 pr_info("%s: interface %s is now up\n",
2463 /* slave->link == BOND_LINK_UP */
2465 /* not all switches will respond to an arp request
2466 * when the source ip is 0, so don't take the link down
2467 * if we don't know our ip yet
2469 if (!bond_time_in_interval(bond, trans_start, 2) ||
2470 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2472 slave->link = BOND_LINK_DOWN;
2473 slave_state_changed = 1;
2475 if (slave->link_failure_count < UINT_MAX)
2476 slave->link_failure_count++;
2478 pr_info("%s: interface %s is now down\n",
2479 bond->dev->name, slave->dev->name);
2481 if (slave == oldcurrent)
2486 /* note: if switch is in round-robin mode, all links
2487 * must tx arp to ensure all links rx an arp - otherwise
2488 * links may oscillate or not come up at all; if switch is
2489 * in something like xor mode, there is nothing we can
2490 * do - all replies will be rx'ed on same link causing slaves
2491 * to be unstable during low/no traffic periods
2493 if (bond_slave_is_up(slave))
2494 bond_arp_send_all(bond, slave);
2499 if (do_failover || slave_state_changed) {
2500 if (!rtnl_trylock())
2503 if (slave_state_changed) {
2504 bond_slave_state_change(bond);
2505 } else if (do_failover) {
2506 /* the bond_select_active_slave must hold RTNL
2507 * and curr_slave_lock for write.
2510 write_lock_bh(&bond->curr_slave_lock);
2512 bond_select_active_slave(bond);
2514 write_unlock_bh(&bond->curr_slave_lock);
2515 unblock_netpoll_tx();
2521 if (bond->params.arp_interval)
2522 queue_delayed_work(bond->wq, &bond->arp_work,
2523 msecs_to_jiffies(bond->params.arp_interval));
2527 * Called to inspect slaves for active-backup mode ARP monitor link state
2528 * changes. Sets new_link in slaves to specify what action should take
2529 * place for the slave. Returns 0 if no changes are found, >0 if changes
2530 * to link states must be committed.
2532 * Called with rcu_read_lock hold.
2534 static int bond_ab_arp_inspect(struct bonding *bond)
2536 unsigned long trans_start, last_rx;
2537 struct list_head *iter;
2538 struct slave *slave;
2541 bond_for_each_slave_rcu(bond, slave, iter) {
2542 slave->new_link = BOND_LINK_NOCHANGE;
2543 last_rx = slave_last_rx(bond, slave);
2545 if (slave->link != BOND_LINK_UP) {
2546 if (bond_time_in_interval(bond, last_rx, 1)) {
2547 slave->new_link = BOND_LINK_UP;
2554 * Give slaves 2*delta after being enslaved or made
2555 * active. This avoids bouncing, as the last receive
2556 * times need a full ARP monitor cycle to be updated.
2558 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2562 * Backup slave is down if:
2563 * - No current_arp_slave AND
2564 * - more than 3*delta since last receive AND
2565 * - the bond has an IP address
2567 * Note: a non-null current_arp_slave indicates
2568 * the curr_active_slave went down and we are
2569 * searching for a new one; under this condition
2570 * we only take the curr_active_slave down - this
2571 * gives each slave a chance to tx/rx traffic
2572 * before being taken out
2574 if (!bond_is_active_slave(slave) &&
2575 !bond->current_arp_slave &&
2576 !bond_time_in_interval(bond, last_rx, 3)) {
2577 slave->new_link = BOND_LINK_DOWN;
2582 * Active slave is down if:
2583 * - more than 2*delta since transmitting OR
2584 * - (more than 2*delta since receive AND
2585 * the bond has an IP address)
2587 trans_start = dev_trans_start(slave->dev);
2588 if (bond_is_active_slave(slave) &&
2589 (!bond_time_in_interval(bond, trans_start, 2) ||
2590 !bond_time_in_interval(bond, last_rx, 2))) {
2591 slave->new_link = BOND_LINK_DOWN;
2600 * Called to commit link state changes noted by inspection step of
2601 * active-backup mode ARP monitor.
2603 * Called with RTNL hold.
2605 static void bond_ab_arp_commit(struct bonding *bond)
2607 unsigned long trans_start;
2608 struct list_head *iter;
2609 struct slave *slave;
2611 bond_for_each_slave(bond, slave, iter) {
2612 switch (slave->new_link) {
2613 case BOND_LINK_NOCHANGE:
2617 trans_start = dev_trans_start(slave->dev);
2618 if (bond->curr_active_slave != slave ||
2619 (!bond->curr_active_slave &&
2620 bond_time_in_interval(bond, trans_start, 1))) {
2621 slave->link = BOND_LINK_UP;
2622 if (bond->current_arp_slave) {
2623 bond_set_slave_inactive_flags(
2624 bond->current_arp_slave,
2625 BOND_SLAVE_NOTIFY_NOW);
2626 bond->current_arp_slave = NULL;
2629 pr_info("%s: link status definitely up for interface %s\n",
2630 bond->dev->name, slave->dev->name);
2632 if (!bond->curr_active_slave ||
2633 (slave == bond->primary_slave))
2640 case BOND_LINK_DOWN:
2641 if (slave->link_failure_count < UINT_MAX)
2642 slave->link_failure_count++;
2644 slave->link = BOND_LINK_DOWN;
2645 bond_set_slave_inactive_flags(slave,
2646 BOND_SLAVE_NOTIFY_NOW);
2648 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2649 bond->dev->name, slave->dev->name);
2651 if (slave == bond->curr_active_slave) {
2652 bond->current_arp_slave = NULL;
2659 pr_err("%s: impossible: new_link %d on slave %s\n",
2660 bond->dev->name, slave->new_link,
2668 write_lock_bh(&bond->curr_slave_lock);
2669 bond_select_active_slave(bond);
2670 write_unlock_bh(&bond->curr_slave_lock);
2671 unblock_netpoll_tx();
2674 bond_set_carrier(bond);
2678 * Send ARP probes for active-backup mode ARP monitor.
2680 * Called with rcu_read_lock hold.
2682 static bool bond_ab_arp_probe(struct bonding *bond)
2684 struct slave *slave, *before = NULL, *new_slave = NULL,
2685 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2686 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2687 struct list_head *iter;
2689 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2691 if (curr_arp_slave && curr_active_slave)
2692 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2693 curr_arp_slave->dev->name,
2694 curr_active_slave->dev->name);
2696 if (curr_active_slave) {
2697 bond_arp_send_all(bond, curr_active_slave);
2698 return should_notify_rtnl;
2701 /* if we don't have a curr_active_slave, search for the next available
2702 * backup slave from the current_arp_slave and make it the candidate
2703 * for becoming the curr_active_slave
2706 if (!curr_arp_slave) {
2707 curr_arp_slave = bond_first_slave_rcu(bond);
2708 if (!curr_arp_slave)
2709 return should_notify_rtnl;
2712 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2714 bond_for_each_slave_rcu(bond, slave, iter) {
2715 if (!found && !before && bond_slave_is_up(slave))
2718 if (found && !new_slave && bond_slave_is_up(slave))
2720 /* if the link state is up at this point, we
2721 * mark it down - this can happen if we have
2722 * simultaneous link failures and
2723 * reselect_active_interface doesn't make this
2724 * one the current slave so it is still marked
2725 * up when it is actually down
2727 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2728 slave->link = BOND_LINK_DOWN;
2729 if (slave->link_failure_count < UINT_MAX)
2730 slave->link_failure_count++;
2732 bond_set_slave_inactive_flags(slave,
2733 BOND_SLAVE_NOTIFY_LATER);
2735 pr_info("%s: backup interface %s is now down\n",
2736 bond->dev->name, slave->dev->name);
2738 if (slave == curr_arp_slave)
2742 if (!new_slave && before)
2748 new_slave->link = BOND_LINK_BACK;
2749 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2750 bond_arp_send_all(bond, new_slave);
2751 new_slave->last_link_up = jiffies;
2752 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2755 bond_for_each_slave_rcu(bond, slave, iter) {
2756 if (slave->should_notify) {
2757 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2761 return should_notify_rtnl;
2764 static void bond_activebackup_arp_mon(struct work_struct *work)
2766 struct bonding *bond = container_of(work, struct bonding,
2768 bool should_notify_peers = false;
2769 bool should_notify_rtnl = false;
2772 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2774 if (!bond_has_slaves(bond))
2779 should_notify_peers = bond_should_notify_peers(bond);
2781 if (bond_ab_arp_inspect(bond)) {
2784 /* Race avoidance with bond_close flush of workqueue */
2785 if (!rtnl_trylock()) {
2787 should_notify_peers = false;
2791 bond_ab_arp_commit(bond);
2797 should_notify_rtnl = bond_ab_arp_probe(bond);
2801 if (bond->params.arp_interval)
2802 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2804 if (should_notify_peers || should_notify_rtnl) {
2805 if (!rtnl_trylock())
2808 if (should_notify_peers)
2809 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2811 if (should_notify_rtnl)
2812 bond_slave_state_notify(bond);
2818 /*-------------------------- netdev event handling --------------------------*/
2821 * Change device name
2823 static int bond_event_changename(struct bonding *bond)
2825 bond_remove_proc_entry(bond);
2826 bond_create_proc_entry(bond);
2828 bond_debug_reregister(bond);
2833 static int bond_master_netdev_event(unsigned long event,
2834 struct net_device *bond_dev)
2836 struct bonding *event_bond = netdev_priv(bond_dev);
2839 case NETDEV_CHANGENAME:
2840 return bond_event_changename(event_bond);
2841 case NETDEV_UNREGISTER:
2842 bond_remove_proc_entry(event_bond);
2844 case NETDEV_REGISTER:
2845 bond_create_proc_entry(event_bond);
2847 case NETDEV_NOTIFY_PEERS:
2848 if (event_bond->send_peer_notif)
2849 event_bond->send_peer_notif--;
2858 static int bond_slave_netdev_event(unsigned long event,
2859 struct net_device *slave_dev)
2861 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2862 struct bonding *bond;
2863 struct net_device *bond_dev;
2867 /* A netdev event can be generated while enslaving a device
2868 * before netdev_rx_handler_register is called in which case
2869 * slave will be NULL
2873 bond_dev = slave->bond->dev;
2877 case NETDEV_UNREGISTER:
2878 if (bond_dev->type != ARPHRD_ETHER)
2879 bond_release_and_destroy(bond_dev, slave_dev);
2881 bond_release(bond_dev, slave_dev);
2885 old_speed = slave->speed;
2886 old_duplex = slave->duplex;
2888 bond_update_speed_duplex(slave);
2890 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2891 if (old_speed != slave->speed)
2892 bond_3ad_adapter_speed_changed(slave);
2893 if (old_duplex != slave->duplex)
2894 bond_3ad_adapter_duplex_changed(slave);
2899 * ... Or is it this?
2902 case NETDEV_CHANGEMTU:
2904 * TODO: Should slaves be allowed to
2905 * independently alter their MTU? For
2906 * an active-backup bond, slaves need
2907 * not be the same type of device, so
2908 * MTUs may vary. For other modes,
2909 * slaves arguably should have the
2910 * same MTUs. To do this, we'd need to
2911 * take over the slave's change_mtu
2912 * function for the duration of their
2916 case NETDEV_CHANGENAME:
2917 /* we don't care if we don't have primary set */
2918 if (!bond_uses_primary(bond) ||
2919 !bond->params.primary[0])
2922 if (slave == bond->primary_slave) {
2923 /* slave's name changed - he's no longer primary */
2924 bond->primary_slave = NULL;
2925 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2926 /* we have a new primary slave */
2927 bond->primary_slave = slave;
2928 } else { /* we didn't change primary - exit */
2932 pr_info("%s: Primary slave changed to %s, reselecting active slave\n",
2934 bond->primary_slave ? slave_dev->name : "none");
2937 write_lock_bh(&bond->curr_slave_lock);
2938 bond_select_active_slave(bond);
2939 write_unlock_bh(&bond->curr_slave_lock);
2940 unblock_netpoll_tx();
2942 case NETDEV_FEAT_CHANGE:
2943 bond_compute_features(bond);
2945 case NETDEV_RESEND_IGMP:
2946 /* Propagate to master device */
2947 call_netdevice_notifiers(event, slave->bond->dev);
2957 * bond_netdev_event: handle netdev notifier chain events.
2959 * This function receives events for the netdev chain. The caller (an
2960 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2961 * locks for us to safely manipulate the slave devices (RTNL lock,
2964 static int bond_netdev_event(struct notifier_block *this,
2965 unsigned long event, void *ptr)
2967 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2969 pr_debug("event_dev: %s, event: %lx\n",
2970 event_dev ? event_dev->name : "None", event);
2972 if (!(event_dev->priv_flags & IFF_BONDING))
2975 if (event_dev->flags & IFF_MASTER) {
2976 pr_debug("IFF_MASTER\n");
2977 return bond_master_netdev_event(event, event_dev);
2980 if (event_dev->flags & IFF_SLAVE) {
2981 pr_debug("IFF_SLAVE\n");
2982 return bond_slave_netdev_event(event, event_dev);
2988 static struct notifier_block bond_netdev_notifier = {
2989 .notifier_call = bond_netdev_event,
2992 /*---------------------------- Hashing Policies -----------------------------*/
2994 /* L2 hash helper */
2995 static inline u32 bond_eth_hash(struct sk_buff *skb)
2997 struct ethhdr *data = (struct ethhdr *)skb->data;
2999 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3000 return data->h_dest[5] ^ data->h_source[5];
3005 /* Extract the appropriate headers based on bond's xmit policy */
3006 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3007 struct flow_keys *fk)
3009 const struct ipv6hdr *iph6;
3010 const struct iphdr *iph;
3011 int noff, proto = -1;
3013 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3014 return skb_flow_dissect(skb, fk);
3017 noff = skb_network_offset(skb);
3018 if (skb->protocol == htons(ETH_P_IP)) {
3019 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3022 fk->src = iph->saddr;
3023 fk->dst = iph->daddr;
3024 noff += iph->ihl << 2;
3025 if (!ip_is_fragment(iph))
3026 proto = iph->protocol;
3027 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3028 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3030 iph6 = ipv6_hdr(skb);
3031 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
3032 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
3033 noff += sizeof(*iph6);
3034 proto = iph6->nexthdr;
3038 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3039 fk->ports = skb_flow_get_ports(skb, noff, proto);
3045 * bond_xmit_hash - generate a hash value based on the xmit policy
3046 * @bond: bonding device
3047 * @skb: buffer to use for headers
3049 * This function will extract the necessary headers from the skb buffer and use
3050 * them to generate a hash based on the xmit_policy set in the bonding device
3052 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3054 struct flow_keys flow;
3057 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3058 !bond_flow_dissect(bond, skb, &flow))
3059 return bond_eth_hash(skb);
3061 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3062 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3063 hash = bond_eth_hash(skb);
3065 hash = (__force u32)flow.ports;
3066 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3067 hash ^= (hash >> 16);
3068 hash ^= (hash >> 8);
3073 /*-------------------------- Device entry points ----------------------------*/
3075 static void bond_work_init_all(struct bonding *bond)
3077 INIT_DELAYED_WORK(&bond->mcast_work,
3078 bond_resend_igmp_join_requests_delayed);
3079 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3080 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3081 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3082 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3084 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3085 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3088 static void bond_work_cancel_all(struct bonding *bond)
3090 cancel_delayed_work_sync(&bond->mii_work);
3091 cancel_delayed_work_sync(&bond->arp_work);
3092 cancel_delayed_work_sync(&bond->alb_work);
3093 cancel_delayed_work_sync(&bond->ad_work);
3094 cancel_delayed_work_sync(&bond->mcast_work);
3097 static int bond_open(struct net_device *bond_dev)
3099 struct bonding *bond = netdev_priv(bond_dev);
3100 struct list_head *iter;
3101 struct slave *slave;
3103 /* reset slave->backup and slave->inactive */
3104 read_lock(&bond->lock);
3105 if (bond_has_slaves(bond)) {
3106 read_lock(&bond->curr_slave_lock);
3107 bond_for_each_slave(bond, slave, iter) {
3108 if (bond_uses_primary(bond)
3109 && (slave != bond->curr_active_slave)) {
3110 bond_set_slave_inactive_flags(slave,
3111 BOND_SLAVE_NOTIFY_NOW);
3113 bond_set_slave_active_flags(slave,
3114 BOND_SLAVE_NOTIFY_NOW);
3117 read_unlock(&bond->curr_slave_lock);
3119 read_unlock(&bond->lock);
3121 bond_work_init_all(bond);
3123 if (bond_is_lb(bond)) {
3124 /* bond_alb_initialize must be called before the timer
3127 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3129 if (bond->params.tlb_dynamic_lb)
3130 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3133 if (bond->params.miimon) /* link check interval, in milliseconds. */
3134 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3136 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3137 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3138 bond->recv_probe = bond_arp_rcv;
3141 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3142 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3143 /* register to receive LACPDUs */
3144 bond->recv_probe = bond_3ad_lacpdu_recv;
3145 bond_3ad_initiate_agg_selection(bond, 1);
3151 static int bond_close(struct net_device *bond_dev)
3153 struct bonding *bond = netdev_priv(bond_dev);
3155 bond_work_cancel_all(bond);
3156 bond->send_peer_notif = 0;
3157 if (bond_is_lb(bond))
3158 bond_alb_deinitialize(bond);
3159 bond->recv_probe = NULL;
3164 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3165 struct rtnl_link_stats64 *stats)
3167 struct bonding *bond = netdev_priv(bond_dev);
3168 struct rtnl_link_stats64 temp;
3169 struct list_head *iter;
3170 struct slave *slave;
3172 memset(stats, 0, sizeof(*stats));
3174 read_lock_bh(&bond->lock);
3175 bond_for_each_slave(bond, slave, iter) {
3176 const struct rtnl_link_stats64 *sstats =
3177 dev_get_stats(slave->dev, &temp);
3179 stats->rx_packets += sstats->rx_packets;
3180 stats->rx_bytes += sstats->rx_bytes;
3181 stats->rx_errors += sstats->rx_errors;
3182 stats->rx_dropped += sstats->rx_dropped;
3184 stats->tx_packets += sstats->tx_packets;
3185 stats->tx_bytes += sstats->tx_bytes;
3186 stats->tx_errors += sstats->tx_errors;
3187 stats->tx_dropped += sstats->tx_dropped;
3189 stats->multicast += sstats->multicast;
3190 stats->collisions += sstats->collisions;
3192 stats->rx_length_errors += sstats->rx_length_errors;
3193 stats->rx_over_errors += sstats->rx_over_errors;
3194 stats->rx_crc_errors += sstats->rx_crc_errors;
3195 stats->rx_frame_errors += sstats->rx_frame_errors;
3196 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3197 stats->rx_missed_errors += sstats->rx_missed_errors;
3199 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3200 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3201 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3202 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3203 stats->tx_window_errors += sstats->tx_window_errors;
3205 read_unlock_bh(&bond->lock);
3210 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3212 struct bonding *bond = netdev_priv(bond_dev);
3213 struct net_device *slave_dev = NULL;
3214 struct ifbond k_binfo;
3215 struct ifbond __user *u_binfo = NULL;
3216 struct ifslave k_sinfo;
3217 struct ifslave __user *u_sinfo = NULL;
3218 struct mii_ioctl_data *mii = NULL;
3219 struct bond_opt_value newval;
3223 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3235 * We do this again just in case we were called by SIOCGMIIREG
3236 * instead of SIOCGMIIPHY.
3243 if (mii->reg_num == 1) {
3245 read_lock(&bond->lock);
3246 read_lock(&bond->curr_slave_lock);
3247 if (netif_carrier_ok(bond->dev))
3248 mii->val_out = BMSR_LSTATUS;
3250 read_unlock(&bond->curr_slave_lock);
3251 read_unlock(&bond->lock);
3255 case BOND_INFO_QUERY_OLD:
3256 case SIOCBONDINFOQUERY:
3257 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3259 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3262 res = bond_info_query(bond_dev, &k_binfo);
3264 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3268 case BOND_SLAVE_INFO_QUERY_OLD:
3269 case SIOCBONDSLAVEINFOQUERY:
3270 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3272 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3275 res = bond_slave_info_query(bond_dev, &k_sinfo);
3277 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3286 net = dev_net(bond_dev);
3288 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3291 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3293 pr_debug("slave_dev=%p:\n", slave_dev);
3298 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3300 case BOND_ENSLAVE_OLD:
3301 case SIOCBONDENSLAVE:
3302 res = bond_enslave(bond_dev, slave_dev);
3304 case BOND_RELEASE_OLD:
3305 case SIOCBONDRELEASE:
3306 res = bond_release(bond_dev, slave_dev);
3308 case BOND_SETHWADDR_OLD:
3309 case SIOCBONDSETHWADDR:
3310 bond_set_dev_addr(bond_dev, slave_dev);
3313 case BOND_CHANGE_ACTIVE_OLD:
3314 case SIOCBONDCHANGEACTIVE:
3315 bond_opt_initstr(&newval, slave_dev->name);
3316 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3325 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3327 struct bonding *bond = netdev_priv(bond_dev);
3329 if (change & IFF_PROMISC)
3330 bond_set_promiscuity(bond,
3331 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3333 if (change & IFF_ALLMULTI)
3334 bond_set_allmulti(bond,
3335 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3338 static void bond_set_rx_mode(struct net_device *bond_dev)
3340 struct bonding *bond = netdev_priv(bond_dev);
3341 struct list_head *iter;
3342 struct slave *slave;
3346 if (bond_uses_primary(bond)) {
3347 slave = rcu_dereference(bond->curr_active_slave);
3349 dev_uc_sync(slave->dev, bond_dev);
3350 dev_mc_sync(slave->dev, bond_dev);
3353 bond_for_each_slave_rcu(bond, slave, iter) {
3354 dev_uc_sync_multiple(slave->dev, bond_dev);
3355 dev_mc_sync_multiple(slave->dev, bond_dev);
3361 static int bond_neigh_init(struct neighbour *n)
3363 struct bonding *bond = netdev_priv(n->dev);
3364 const struct net_device_ops *slave_ops;
3365 struct neigh_parms parms;
3366 struct slave *slave;
3369 slave = bond_first_slave(bond);
3372 slave_ops = slave->dev->netdev_ops;
3373 if (!slave_ops->ndo_neigh_setup)
3376 parms.neigh_setup = NULL;
3377 parms.neigh_cleanup = NULL;
3378 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3383 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3384 * after the last slave has been detached. Assumes that all slaves
3385 * utilize the same neigh_cleanup (true at this writing as only user
3388 n->parms->neigh_cleanup = parms.neigh_cleanup;
3390 if (!parms.neigh_setup)
3393 return parms.neigh_setup(n);
3397 * The bonding ndo_neigh_setup is called at init time beofre any
3398 * slave exists. So we must declare proxy setup function which will
3399 * be used at run time to resolve the actual slave neigh param setup.
3401 * It's also called by master devices (such as vlans) to setup their
3402 * underlying devices. In that case - do nothing, we're already set up from
3405 static int bond_neigh_setup(struct net_device *dev,
3406 struct neigh_parms *parms)
3408 /* modify only our neigh_parms */
3409 if (parms->dev == dev)
3410 parms->neigh_setup = bond_neigh_init;
3416 * Change the MTU of all of a master's slaves to match the master
3418 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3420 struct bonding *bond = netdev_priv(bond_dev);
3421 struct slave *slave, *rollback_slave;
3422 struct list_head *iter;
3425 pr_debug("bond=%p, name=%s, new_mtu=%d\n",
3426 bond, bond_dev ? bond_dev->name : "None", new_mtu);
3428 /* Can't hold bond->lock with bh disabled here since
3429 * some base drivers panic. On the other hand we can't
3430 * hold bond->lock without bh disabled because we'll
3431 * deadlock. The only solution is to rely on the fact
3432 * that we're under rtnl_lock here, and the slaves
3433 * list won't change. This doesn't solve the problem
3434 * of setting the slave's MTU while it is
3435 * transmitting, but the assumption is that the base
3436 * driver can handle that.
3438 * TODO: figure out a way to safely iterate the slaves
3439 * list, but without holding a lock around the actual
3440 * call to the base driver.
3443 bond_for_each_slave(bond, slave, iter) {
3444 pr_debug("s %p c_m %p\n",
3445 slave, slave->dev->netdev_ops->ndo_change_mtu);
3447 res = dev_set_mtu(slave->dev, new_mtu);
3450 /* If we failed to set the slave's mtu to the new value
3451 * we must abort the operation even in ACTIVE_BACKUP
3452 * mode, because if we allow the backup slaves to have
3453 * different mtu values than the active slave we'll
3454 * need to change their mtu when doing a failover. That
3455 * means changing their mtu from timer context, which
3456 * is probably not a good idea.
3458 pr_debug("err %d %s\n", res, slave->dev->name);
3463 bond_dev->mtu = new_mtu;
3468 /* unwind from head to the slave that failed */
3469 bond_for_each_slave(bond, rollback_slave, iter) {
3472 if (rollback_slave == slave)
3475 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3477 pr_debug("unwind err %d dev %s\n",
3478 tmp_res, rollback_slave->dev->name);
3488 * Note that many devices must be down to change the HW address, and
3489 * downing the master releases all slaves. We can make bonds full of
3490 * bonding devices to test this, however.
3492 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3494 struct bonding *bond = netdev_priv(bond_dev);
3495 struct slave *slave, *rollback_slave;
3496 struct sockaddr *sa = addr, tmp_sa;
3497 struct list_head *iter;
3500 if (BOND_MODE(bond) == BOND_MODE_ALB)
3501 return bond_alb_set_mac_address(bond_dev, addr);
3504 pr_debug("bond=%p, name=%s\n",
3505 bond, bond_dev ? bond_dev->name : "None");
3507 /* If fail_over_mac is enabled, do nothing and return success.
3508 * Returning an error causes ifenslave to fail.
3510 if (bond->params.fail_over_mac &&
3511 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3514 if (!is_valid_ether_addr(sa->sa_data))
3515 return -EADDRNOTAVAIL;
3517 /* Can't hold bond->lock with bh disabled here since
3518 * some base drivers panic. On the other hand we can't
3519 * hold bond->lock without bh disabled because we'll
3520 * deadlock. The only solution is to rely on the fact
3521 * that we're under rtnl_lock here, and the slaves
3522 * list won't change. This doesn't solve the problem
3523 * of setting the slave's hw address while it is
3524 * transmitting, but the assumption is that the base
3525 * driver can handle that.
3527 * TODO: figure out a way to safely iterate the slaves
3528 * list, but without holding a lock around the actual
3529 * call to the base driver.
3532 bond_for_each_slave(bond, slave, iter) {
3533 pr_debug("slave %p %s\n", slave, slave->dev->name);
3534 res = dev_set_mac_address(slave->dev, addr);
3536 /* TODO: consider downing the slave
3538 * User should expect communications
3539 * breakage anyway until ARP finish
3542 pr_debug("err %d %s\n", res, slave->dev->name);
3548 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3552 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3553 tmp_sa.sa_family = bond_dev->type;
3555 /* unwind from head to the slave that failed */
3556 bond_for_each_slave(bond, rollback_slave, iter) {
3559 if (rollback_slave == slave)
3562 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3564 pr_debug("unwind err %d dev %s\n",
3565 tmp_res, rollback_slave->dev->name);
3573 * bond_xmit_slave_id - transmit skb through slave with slave_id
3574 * @bond: bonding device that is transmitting
3575 * @skb: buffer to transmit
3576 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3578 * This function tries to transmit through slave with slave_id but in case
3579 * it fails, it tries to find the first available slave for transmission.
3580 * The skb is consumed in all cases, thus the function is void.
3582 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3584 struct list_head *iter;
3585 struct slave *slave;
3588 /* Here we start from the slave with slave_id */
3589 bond_for_each_slave_rcu(bond, slave, iter) {
3591 if (bond_slave_can_tx(slave)) {
3592 bond_dev_queue_xmit(bond, skb, slave->dev);
3598 /* Here we start from the first slave up to slave_id */
3600 bond_for_each_slave_rcu(bond, slave, iter) {
3603 if (bond_slave_can_tx(slave)) {
3604 bond_dev_queue_xmit(bond, skb, slave->dev);
3608 /* no slave that can tx has been found */
3609 dev_kfree_skb_any(skb);
3613 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3614 * @bond: bonding device to use
3616 * Based on the value of the bonding device's packets_per_slave parameter
3617 * this function generates a slave id, which is usually used as the next
3618 * slave to transmit through.
3620 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3623 struct reciprocal_value reciprocal_packets_per_slave;
3624 int packets_per_slave = bond->params.packets_per_slave;
3626 switch (packets_per_slave) {
3628 slave_id = prandom_u32();
3631 slave_id = bond->rr_tx_counter;
3634 reciprocal_packets_per_slave =
3635 bond->params.reciprocal_packets_per_slave;
3636 slave_id = reciprocal_divide(bond->rr_tx_counter,
3637 reciprocal_packets_per_slave);
3640 bond->rr_tx_counter++;
3645 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3647 struct bonding *bond = netdev_priv(bond_dev);
3648 struct iphdr *iph = ip_hdr(skb);
3649 struct slave *slave;
3652 /* Start with the curr_active_slave that joined the bond as the
3653 * default for sending IGMP traffic. For failover purposes one
3654 * needs to maintain some consistency for the interface that will
3655 * send the join/membership reports. The curr_active_slave found
3656 * will send all of this type of traffic.
3658 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3659 slave = rcu_dereference(bond->curr_active_slave);
3660 if (slave && bond_slave_can_tx(slave))
3661 bond_dev_queue_xmit(bond, skb, slave->dev);
3663 bond_xmit_slave_id(bond, skb, 0);
3665 slave_id = bond_rr_gen_slave_id(bond);
3666 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3669 return NETDEV_TX_OK;
3673 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3674 * the bond has a usable interface.
3676 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3678 struct bonding *bond = netdev_priv(bond_dev);
3679 struct slave *slave;
3681 slave = rcu_dereference(bond->curr_active_slave);
3683 bond_dev_queue_xmit(bond, skb, slave->dev);
3685 dev_kfree_skb_any(skb);
3687 return NETDEV_TX_OK;
3690 /* In bond_xmit_xor() , we determine the output device by using a pre-
3691 * determined xmit_hash_policy(), If the selected device is not enabled,
3692 * find the next active slave.
3694 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3696 struct bonding *bond = netdev_priv(bond_dev);
3698 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb) % bond->slave_cnt);
3700 return NETDEV_TX_OK;
3703 /* in broadcast mode, we send everything to all usable interfaces. */
3704 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3706 struct bonding *bond = netdev_priv(bond_dev);
3707 struct slave *slave = NULL;
3708 struct list_head *iter;
3710 bond_for_each_slave_rcu(bond, slave, iter) {
3711 if (bond_is_last_slave(bond, slave))
3713 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3714 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3717 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3718 bond_dev->name, __func__);
3721 /* bond_dev_queue_xmit always returns 0 */
3722 bond_dev_queue_xmit(bond, skb2, slave->dev);
3725 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3726 bond_dev_queue_xmit(bond, skb, slave->dev);
3728 dev_kfree_skb_any(skb);
3730 return NETDEV_TX_OK;
3733 /*------------------------- Device initialization ---------------------------*/
3736 * Lookup the slave that corresponds to a qid
3738 static inline int bond_slave_override(struct bonding *bond,
3739 struct sk_buff *skb)
3741 struct slave *slave = NULL;
3742 struct list_head *iter;
3744 if (!skb->queue_mapping)
3747 /* Find out if any slaves have the same mapping as this skb. */
3748 bond_for_each_slave_rcu(bond, slave, iter) {
3749 if (slave->queue_id == skb->queue_mapping) {
3750 if (bond_slave_can_tx(slave)) {
3751 bond_dev_queue_xmit(bond, skb, slave->dev);
3754 /* If the slave isn't UP, use default transmit policy. */
3763 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3764 void *accel_priv, select_queue_fallback_t fallback)
3767 * This helper function exists to help dev_pick_tx get the correct
3768 * destination queue. Using a helper function skips a call to
3769 * skb_tx_hash and will put the skbs in the queue we expect on their
3770 * way down to the bonding driver.
3772 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3775 * Save the original txq to restore before passing to the driver
3777 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3779 if (unlikely(txq >= dev->real_num_tx_queues)) {
3781 txq -= dev->real_num_tx_queues;
3782 } while (txq >= dev->real_num_tx_queues);
3787 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3789 struct bonding *bond = netdev_priv(dev);
3791 if (bond_should_override_tx_queue(bond) &&
3792 !bond_slave_override(bond, skb))
3793 return NETDEV_TX_OK;
3795 switch (BOND_MODE(bond)) {
3796 case BOND_MODE_ROUNDROBIN:
3797 return bond_xmit_roundrobin(skb, dev);
3798 case BOND_MODE_ACTIVEBACKUP:
3799 return bond_xmit_activebackup(skb, dev);
3801 return bond_xmit_xor(skb, dev);
3802 case BOND_MODE_BROADCAST:
3803 return bond_xmit_broadcast(skb, dev);
3804 case BOND_MODE_8023AD:
3805 return bond_3ad_xmit_xor(skb, dev);
3807 return bond_alb_xmit(skb, dev);
3809 return bond_tlb_xmit(skb, dev);
3811 /* Should never happen, mode already checked */
3812 pr_err("%s: Error: Unknown bonding mode %d\n",
3813 dev->name, BOND_MODE(bond));
3815 dev_kfree_skb_any(skb);
3816 return NETDEV_TX_OK;
3820 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3822 struct bonding *bond = netdev_priv(dev);
3823 netdev_tx_t ret = NETDEV_TX_OK;
3826 * If we risk deadlock from transmitting this in the
3827 * netpoll path, tell netpoll to queue the frame for later tx
3829 if (unlikely(is_netpoll_tx_blocked(dev)))
3830 return NETDEV_TX_BUSY;
3833 if (bond_has_slaves(bond))
3834 ret = __bond_start_xmit(skb, dev);
3836 dev_kfree_skb_any(skb);
3842 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3843 struct ethtool_cmd *ecmd)
3845 struct bonding *bond = netdev_priv(bond_dev);
3846 unsigned long speed = 0;
3847 struct list_head *iter;
3848 struct slave *slave;
3850 ecmd->duplex = DUPLEX_UNKNOWN;
3851 ecmd->port = PORT_OTHER;
3853 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3854 * do not need to check mode. Though link speed might not represent
3855 * the true receive or transmit bandwidth (not all modes are symmetric)
3856 * this is an accurate maximum.
3858 read_lock(&bond->lock);
3859 bond_for_each_slave(bond, slave, iter) {
3860 if (bond_slave_can_tx(slave)) {
3861 if (slave->speed != SPEED_UNKNOWN)
3862 speed += slave->speed;
3863 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3864 slave->duplex != DUPLEX_UNKNOWN)
3865 ecmd->duplex = slave->duplex;
3868 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3869 read_unlock(&bond->lock);
3874 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3875 struct ethtool_drvinfo *drvinfo)
3877 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3878 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3879 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3883 static const struct ethtool_ops bond_ethtool_ops = {
3884 .get_drvinfo = bond_ethtool_get_drvinfo,
3885 .get_settings = bond_ethtool_get_settings,
3886 .get_link = ethtool_op_get_link,
3889 static const struct net_device_ops bond_netdev_ops = {
3890 .ndo_init = bond_init,
3891 .ndo_uninit = bond_uninit,
3892 .ndo_open = bond_open,
3893 .ndo_stop = bond_close,
3894 .ndo_start_xmit = bond_start_xmit,
3895 .ndo_select_queue = bond_select_queue,
3896 .ndo_get_stats64 = bond_get_stats,
3897 .ndo_do_ioctl = bond_do_ioctl,
3898 .ndo_change_rx_flags = bond_change_rx_flags,
3899 .ndo_set_rx_mode = bond_set_rx_mode,
3900 .ndo_change_mtu = bond_change_mtu,
3901 .ndo_set_mac_address = bond_set_mac_address,
3902 .ndo_neigh_setup = bond_neigh_setup,
3903 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3904 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3905 #ifdef CONFIG_NET_POLL_CONTROLLER
3906 .ndo_netpoll_setup = bond_netpoll_setup,
3907 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3908 .ndo_poll_controller = bond_poll_controller,
3910 .ndo_add_slave = bond_enslave,
3911 .ndo_del_slave = bond_release,
3912 .ndo_fix_features = bond_fix_features,
3915 static const struct device_type bond_type = {
3919 static void bond_destructor(struct net_device *bond_dev)
3921 struct bonding *bond = netdev_priv(bond_dev);
3923 destroy_workqueue(bond->wq);
3924 free_netdev(bond_dev);
3927 void bond_setup(struct net_device *bond_dev)
3929 struct bonding *bond = netdev_priv(bond_dev);
3931 /* initialize rwlocks */
3932 rwlock_init(&bond->lock);
3933 rwlock_init(&bond->curr_slave_lock);
3934 bond->params = bonding_defaults;
3936 /* Initialize pointers */
3937 bond->dev = bond_dev;
3939 /* Initialize the device entry points */
3940 ether_setup(bond_dev);
3941 bond_dev->netdev_ops = &bond_netdev_ops;
3942 bond_dev->ethtool_ops = &bond_ethtool_ops;
3944 bond_dev->destructor = bond_destructor;
3946 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3948 /* Initialize the device options */
3949 bond_dev->tx_queue_len = 0;
3950 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3951 bond_dev->priv_flags |= IFF_BONDING;
3952 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3954 /* At first, we block adding VLANs. That's the only way to
3955 * prevent problems that occur when adding VLANs over an
3956 * empty bond. The block will be removed once non-challenged
3957 * slaves are enslaved.
3959 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3961 /* don't acquire bond device's netif_tx_lock when
3963 bond_dev->features |= NETIF_F_LLTX;
3965 /* By default, we declare the bond to be fully
3966 * VLAN hardware accelerated capable. Special
3967 * care is taken in the various xmit functions
3968 * when there are slaves that are not hw accel
3972 /* Don't allow bond devices to change network namespaces. */
3973 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3975 bond_dev->hw_features = BOND_VLAN_FEATURES |
3976 NETIF_F_HW_VLAN_CTAG_TX |
3977 NETIF_F_HW_VLAN_CTAG_RX |
3978 NETIF_F_HW_VLAN_CTAG_FILTER;
3980 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3981 bond_dev->features |= bond_dev->hw_features;
3985 * Destroy a bonding device.
3986 * Must be under rtnl_lock when this function is called.
3988 static void bond_uninit(struct net_device *bond_dev)
3990 struct bonding *bond = netdev_priv(bond_dev);
3991 struct list_head *iter;
3992 struct slave *slave;
3994 bond_netpoll_cleanup(bond_dev);
3996 /* Release the bonded slaves */
3997 bond_for_each_slave(bond, slave, iter)
3998 __bond_release_one(bond_dev, slave->dev, true);
3999 pr_info("%s: Released all slaves\n", bond_dev->name);
4001 list_del(&bond->bond_list);
4003 bond_debug_unregister(bond);
4006 /*------------------------- Module initialization ---------------------------*/
4008 static int bond_check_params(struct bond_params *params)
4010 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4011 struct bond_opt_value newval;
4012 const struct bond_opt_value *valptr;
4013 int arp_all_targets_value;
4016 * Convert string parameters.
4019 bond_opt_initstr(&newval, mode);
4020 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4022 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4025 bond_mode = valptr->value;
4028 if (xmit_hash_policy) {
4029 if ((bond_mode != BOND_MODE_XOR) &&
4030 (bond_mode != BOND_MODE_8023AD) &&
4031 (bond_mode != BOND_MODE_TLB)) {
4032 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4033 bond_mode_name(bond_mode));
4035 bond_opt_initstr(&newval, xmit_hash_policy);
4036 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4039 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4043 xmit_hashtype = valptr->value;
4048 if (bond_mode != BOND_MODE_8023AD) {
4049 pr_info("lacp_rate param is irrelevant in mode %s\n",
4050 bond_mode_name(bond_mode));
4052 bond_opt_initstr(&newval, lacp_rate);
4053 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4056 pr_err("Error: Invalid lacp rate \"%s\"\n",
4060 lacp_fast = valptr->value;
4065 bond_opt_initstr(&newval, lacp_rate);
4066 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4069 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4072 params->ad_select = valptr->value;
4073 if (bond_mode != BOND_MODE_8023AD)
4074 pr_warn("ad_select param only affects 802.3ad mode\n");
4076 params->ad_select = BOND_AD_STABLE;
4079 if (max_bonds < 0) {
4080 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4081 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4082 max_bonds = BOND_DEFAULT_MAX_BONDS;
4086 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4092 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4097 if (downdelay < 0) {
4098 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4099 downdelay, INT_MAX);
4103 if ((use_carrier != 0) && (use_carrier != 1)) {
4104 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4109 if (num_peer_notif < 0 || num_peer_notif > 255) {
4110 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4115 /* reset values for 802.3ad/TLB/ALB */
4116 if (!bond_mode_uses_arp(bond_mode)) {
4118 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4119 pr_warn("Forcing miimon to 100msec\n");
4120 miimon = BOND_DEFAULT_MIIMON;
4124 if (tx_queues < 1 || tx_queues > 255) {
4125 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4126 tx_queues, BOND_DEFAULT_TX_QUEUES);
4127 tx_queues = BOND_DEFAULT_TX_QUEUES;
4130 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4131 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4133 all_slaves_active = 0;
4136 if (resend_igmp < 0 || resend_igmp > 255) {
4137 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4138 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4139 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4142 bond_opt_initval(&newval, packets_per_slave);
4143 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4144 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4145 packets_per_slave, USHRT_MAX);
4146 packets_per_slave = 1;
4149 if (bond_mode == BOND_MODE_ALB) {
4150 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",
4155 if (updelay || downdelay) {
4156 /* just warn the user the up/down delay will have
4157 * no effect since miimon is zero...
4159 pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4160 updelay, downdelay);
4163 /* don't allow arp monitoring */
4165 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4166 miimon, arp_interval);
4170 if ((updelay % miimon) != 0) {
4171 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4172 updelay, miimon, (updelay / miimon) * miimon);
4177 if ((downdelay % miimon) != 0) {
4178 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4180 (downdelay / miimon) * miimon);
4183 downdelay /= miimon;
4186 if (arp_interval < 0) {
4187 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4188 arp_interval, INT_MAX);
4192 for (arp_ip_count = 0, i = 0;
4193 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4194 /* not complete check, but should be good enough to
4197 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4198 !bond_is_ip_target_ok(ip)) {
4199 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4203 if (bond_get_targets_ip(arp_target, ip) == -1)
4204 arp_target[arp_ip_count++] = ip;
4206 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4211 if (arp_interval && !arp_ip_count) {
4212 /* don't allow arping if no arp_ip_target given... */
4213 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4219 if (!arp_interval) {
4220 pr_err("arp_validate requires arp_interval\n");
4224 bond_opt_initstr(&newval, arp_validate);
4225 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4228 pr_err("Error: invalid arp_validate \"%s\"\n",
4232 arp_validate_value = valptr->value;
4234 arp_validate_value = 0;
4237 arp_all_targets_value = 0;
4238 if (arp_all_targets) {
4239 bond_opt_initstr(&newval, arp_all_targets);
4240 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4243 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4245 arp_all_targets_value = 0;
4247 arp_all_targets_value = valptr->value;
4252 pr_info("MII link monitoring set to %d ms\n", miimon);
4253 } else if (arp_interval) {
4254 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4255 arp_validate_value);
4256 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4257 arp_interval, valptr->string, arp_ip_count);
4259 for (i = 0; i < arp_ip_count; i++)
4260 pr_cont(" %s", arp_ip_target[i]);
4264 } else if (max_bonds) {
4265 /* miimon and arp_interval not set, we need one so things
4266 * work as expected, see bonding.txt for details
4268 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");
4271 if (primary && !bond_mode_uses_primary(bond_mode)) {
4272 /* currently, using a primary only makes sense
4273 * in active backup, TLB or ALB modes
4275 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4276 primary, bond_mode_name(bond_mode));
4280 if (primary && primary_reselect) {
4281 bond_opt_initstr(&newval, primary_reselect);
4282 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4285 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4289 primary_reselect_value = valptr->value;
4291 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4294 if (fail_over_mac) {
4295 bond_opt_initstr(&newval, fail_over_mac);
4296 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4299 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4303 fail_over_mac_value = valptr->value;
4304 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4305 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4307 fail_over_mac_value = BOND_FOM_NONE;
4310 if (lp_interval == 0) {
4311 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4312 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4313 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4316 /* fill params struct with the proper values */
4317 params->mode = bond_mode;
4318 params->xmit_policy = xmit_hashtype;
4319 params->miimon = miimon;
4320 params->num_peer_notif = num_peer_notif;
4321 params->arp_interval = arp_interval;
4322 params->arp_validate = arp_validate_value;
4323 params->arp_all_targets = arp_all_targets_value;
4324 params->updelay = updelay;
4325 params->downdelay = downdelay;
4326 params->use_carrier = use_carrier;
4327 params->lacp_fast = lacp_fast;
4328 params->primary[0] = 0;
4329 params->primary_reselect = primary_reselect_value;
4330 params->fail_over_mac = fail_over_mac_value;
4331 params->tx_queues = tx_queues;
4332 params->all_slaves_active = all_slaves_active;
4333 params->resend_igmp = resend_igmp;
4334 params->min_links = min_links;
4335 params->lp_interval = lp_interval;
4336 params->packets_per_slave = packets_per_slave;
4337 params->tlb_dynamic_lb = 1; /* Default value */
4338 if (packets_per_slave > 0) {
4339 params->reciprocal_packets_per_slave =
4340 reciprocal_value(packets_per_slave);
4342 /* reciprocal_packets_per_slave is unused if
4343 * packets_per_slave is 0 or 1, just initialize it
4345 params->reciprocal_packets_per_slave =
4346 (struct reciprocal_value) { 0 };
4350 strncpy(params->primary, primary, IFNAMSIZ);
4351 params->primary[IFNAMSIZ - 1] = 0;
4354 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4359 static struct lock_class_key bonding_netdev_xmit_lock_key;
4360 static struct lock_class_key bonding_netdev_addr_lock_key;
4361 static struct lock_class_key bonding_tx_busylock_key;
4363 static void bond_set_lockdep_class_one(struct net_device *dev,
4364 struct netdev_queue *txq,
4367 lockdep_set_class(&txq->_xmit_lock,
4368 &bonding_netdev_xmit_lock_key);
4371 static void bond_set_lockdep_class(struct net_device *dev)
4373 lockdep_set_class(&dev->addr_list_lock,
4374 &bonding_netdev_addr_lock_key);
4375 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4376 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4380 * Called from registration process
4382 static int bond_init(struct net_device *bond_dev)
4384 struct bonding *bond = netdev_priv(bond_dev);
4385 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4386 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4388 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4391 * Initialize locks that may be required during
4392 * en/deslave operations. All of the bond_open work
4393 * (of which this is part) should really be moved to
4394 * a phase prior to dev_open
4396 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4397 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4399 bond->wq = create_singlethread_workqueue(bond_dev->name);
4403 bond_set_lockdep_class(bond_dev);
4405 list_add_tail(&bond->bond_list, &bn->dev_list);
4407 bond_prepare_sysfs_group(bond);
4409 bond_debug_register(bond);
4411 /* Ensure valid dev_addr */
4412 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4413 bond_dev->addr_assign_type == NET_ADDR_PERM)
4414 eth_hw_addr_random(bond_dev);
4419 unsigned int bond_get_num_tx_queues(void)
4424 /* Create a new bond based on the specified name and bonding parameters.
4425 * If name is NULL, obtain a suitable "bond%d" name for us.
4426 * Caller must NOT hold rtnl_lock; we need to release it here before we
4427 * set up our sysfs entries.
4429 int bond_create(struct net *net, const char *name)
4431 struct net_device *bond_dev;
4436 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4437 name ? name : "bond%d",
4438 bond_setup, tx_queues);
4440 pr_err("%s: eek! can't alloc netdev!\n", name);
4445 dev_net_set(bond_dev, net);
4446 bond_dev->rtnl_link_ops = &bond_link_ops;
4448 res = register_netdevice(bond_dev);
4450 netif_carrier_off(bond_dev);
4454 bond_destructor(bond_dev);
4458 static int __net_init bond_net_init(struct net *net)
4460 struct bond_net *bn = net_generic(net, bond_net_id);
4463 INIT_LIST_HEAD(&bn->dev_list);
4465 bond_create_proc_dir(bn);
4466 bond_create_sysfs(bn);
4471 static void __net_exit bond_net_exit(struct net *net)
4473 struct bond_net *bn = net_generic(net, bond_net_id);
4474 struct bonding *bond, *tmp_bond;
4477 bond_destroy_sysfs(bn);
4478 bond_destroy_proc_dir(bn);
4480 /* Kill off any bonds created after unregistering bond rtnl ops */
4482 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4483 unregister_netdevice_queue(bond->dev, &list);
4484 unregister_netdevice_many(&list);
4488 static struct pernet_operations bond_net_ops = {
4489 .init = bond_net_init,
4490 .exit = bond_net_exit,
4492 .size = sizeof(struct bond_net),
4495 static int __init bonding_init(void)
4500 pr_info("%s", bond_version);
4502 res = bond_check_params(&bonding_defaults);
4506 res = register_pernet_subsys(&bond_net_ops);
4510 res = bond_netlink_init();
4514 bond_create_debugfs();
4516 for (i = 0; i < max_bonds; i++) {
4517 res = bond_create(&init_net, NULL);
4522 register_netdevice_notifier(&bond_netdev_notifier);
4526 bond_destroy_debugfs();
4527 bond_netlink_fini();
4529 unregister_pernet_subsys(&bond_net_ops);
4534 static void __exit bonding_exit(void)
4536 unregister_netdevice_notifier(&bond_netdev_notifier);
4538 bond_destroy_debugfs();
4540 bond_netlink_fini();
4541 unregister_pernet_subsys(&bond_net_ops);
4543 #ifdef CONFIG_NET_POLL_CONTROLLER
4545 * Make sure we don't have an imbalance on our netpoll blocking
4547 WARN_ON(atomic_read(&netpoll_block_tx));
4551 module_init(bonding_init);
4552 module_exit(bonding_exit);
4553 MODULE_LICENSE("GPL");
4554 MODULE_VERSION(DRV_VERSION);
4555 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4556 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");