2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
96 static int use_carrier = 1;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *arp_all_targets;
108 static char *fail_over_mac;
109 static int all_slaves_active;
110 static struct bond_params bonding_defaults;
111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
112 static int packets_per_slave = 1;
113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
121 "failover event (alias of num_unsol_na)");
122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
124 "failover event (alias of num_grat_arp)");
125 module_param(miimon, int, 0);
126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
127 module_param(updelay, int, 0);
128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
129 module_param(downdelay, int, 0);
130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 module_param(use_carrier, int, 0);
133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
134 "0 for off, 1 for on (default)");
135 module_param(mode, charp, 0);
136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
137 "1 for active-backup, 2 for balance-xor, "
138 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
139 "6 for balance-alb");
140 module_param(primary, charp, 0);
141 MODULE_PARM_DESC(primary, "Primary network device to use");
142 module_param(primary_reselect, charp, 0);
143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "0 for always (default), "
146 "1 for only if speed of primary is "
148 "2 for only on active slave "
150 module_param(lacp_rate, charp, 0);
151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
152 "0 for slow, 1 for fast");
153 module_param(ad_select, charp, 0);
154 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
155 "0 for stable (default), 1 for bandwidth, "
157 module_param(min_links, int, 0);
158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
160 module_param(xmit_hash_policy, charp, 0);
161 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
162 "0 for layer 2 (default), 1 for layer 3+4, "
163 "2 for layer 2+3, 3 for encap layer 2+3, "
164 "4 for encap layer 3+4");
165 module_param(arp_interval, int, 0);
166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
167 module_param_array(arp_ip_target, charp, NULL, 0);
168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
169 module_param(arp_validate, charp, 0);
170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
171 "0 for none (default), 1 for active, "
172 "2 for backup, 3 for all");
173 module_param(arp_all_targets, charp, 0);
174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
175 module_param(fail_over_mac, charp, 0);
176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
177 "the same MAC; 0 for none (default), "
178 "1 for active, 2 for follow");
179 module_param(all_slaves_active, int, 0);
180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
181 "by setting active flag for all slaves; "
182 "0 for never (default), 1 for always.");
183 module_param(resend_igmp, int, 0);
184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 module_param(packets_per_slave, int, 0);
187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
188 "mode; 0 for a random slave, 1 packet per "
189 "slave (default), >1 packets per slave.");
190 module_param(lp_interval, uint, 0);
191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
192 "the bonding driver sends learning packets to "
193 "each slaves peer switch. The default is 1.");
195 /*----------------------------- Global variables ----------------------------*/
197 #ifdef CONFIG_NET_POLL_CONTROLLER
198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
201 int bond_net_id __read_mostly;
203 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
204 static int arp_ip_count;
205 static int bond_mode = BOND_MODE_ROUNDROBIN;
206 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
207 static int lacp_fast;
209 /*-------------------------- Forward declarations ---------------------------*/
211 static int bond_init(struct net_device *bond_dev);
212 static void bond_uninit(struct net_device *bond_dev);
214 /*---------------------------- General routines -----------------------------*/
216 const char *bond_mode_name(int mode)
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
234 /*---------------------------------- VLAN -----------------------------------*/
237 * bond_dev_queue_xmit - Prepare skb for xmit.
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
246 skb->dev = slave_dev;
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
259 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
302 vlan_vid_del(rollback_slave->dev, proto, vid);
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
329 /*------------------------------- Link status -------------------------------*/
332 * Set the carrier state for the master according to the state of its
333 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
334 * do special 802.3ad magic.
336 * Returns zero if carrier state does not change, nonzero if it does.
338 static int bond_set_carrier(struct bonding *bond)
340 struct list_head *iter;
343 if (!bond_has_slaves(bond))
346 if (bond->params.mode == BOND_MODE_8023AD)
347 return bond_3ad_set_carrier(bond);
349 bond_for_each_slave(bond, slave, iter) {
350 if (slave->link == BOND_LINK_UP) {
351 if (!netif_carrier_ok(bond->dev)) {
352 netif_carrier_on(bond->dev);
360 if (netif_carrier_ok(bond->dev)) {
361 netif_carrier_off(bond->dev);
368 * Get link speed and duplex from the slave's base driver
369 * using ethtool. If for some reason the call fails or the
370 * values are invalid, set speed and duplex to -1,
373 static void bond_update_speed_duplex(struct slave *slave)
375 struct net_device *slave_dev = slave->dev;
376 struct ethtool_cmd ecmd;
380 slave->speed = SPEED_UNKNOWN;
381 slave->duplex = DUPLEX_UNKNOWN;
383 res = __ethtool_get_settings(slave_dev, &ecmd);
387 slave_speed = ethtool_cmd_speed(&ecmd);
388 if (slave_speed == 0 || slave_speed == ((__u32) -1))
391 switch (ecmd.duplex) {
399 slave->speed = slave_speed;
400 slave->duplex = ecmd.duplex;
405 const char *bond_slave_link_status(s8 link)
422 * if <dev> supports MII link status reporting, check its link status.
424 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
425 * depending upon the setting of the use_carrier parameter.
427 * Return either BMSR_LSTATUS, meaning that the link is up (or we
428 * can't tell and just pretend it is), or 0, meaning that the link is
431 * If reporting is non-zero, instead of faking link up, return -1 if
432 * both ETHTOOL and MII ioctls fail (meaning the device does not
433 * support them). If use_carrier is set, return whatever it says.
434 * It'd be nice if there was a good way to tell if a driver supports
435 * netif_carrier, but there really isn't.
437 static int bond_check_dev_link(struct bonding *bond,
438 struct net_device *slave_dev, int reporting)
440 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
441 int (*ioctl)(struct net_device *, struct ifreq *, int);
443 struct mii_ioctl_data *mii;
445 if (!reporting && !netif_running(slave_dev))
448 if (bond->params.use_carrier)
449 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
451 /* Try to get link status using Ethtool first. */
452 if (slave_dev->ethtool_ops->get_link)
453 return slave_dev->ethtool_ops->get_link(slave_dev) ?
456 /* Ethtool can't be used, fallback to MII ioctls. */
457 ioctl = slave_ops->ndo_do_ioctl;
459 /* TODO: set pointer to correct ioctl on a per team member */
460 /* bases to make this more efficient. that is, once */
461 /* we determine the correct ioctl, we will always */
462 /* call it and not the others for that team */
466 * We cannot assume that SIOCGMIIPHY will also read a
467 * register; not all network drivers (e.g., e100)
471 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
472 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
474 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
475 mii->reg_num = MII_BMSR;
476 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
477 return mii->val_out & BMSR_LSTATUS;
482 * If reporting, report that either there's no dev->do_ioctl,
483 * or both SIOCGMIIREG and get_link failed (meaning that we
484 * cannot report link status). If not reporting, pretend
487 return reporting ? -1 : BMSR_LSTATUS;
490 /*----------------------------- Multicast list ------------------------------*/
493 * Push the promiscuity flag down to appropriate slaves
495 static int bond_set_promiscuity(struct bonding *bond, int inc)
497 struct list_head *iter;
500 if (USES_PRIMARY(bond->params.mode)) {
501 /* write lock already acquired */
502 if (bond->curr_active_slave) {
503 err = dev_set_promiscuity(bond->curr_active_slave->dev,
509 bond_for_each_slave(bond, slave, iter) {
510 err = dev_set_promiscuity(slave->dev, inc);
519 * Push the allmulti flag down to all slaves
521 static int bond_set_allmulti(struct bonding *bond, int inc)
523 struct list_head *iter;
526 if (USES_PRIMARY(bond->params.mode)) {
527 /* write lock already acquired */
528 if (bond->curr_active_slave) {
529 err = dev_set_allmulti(bond->curr_active_slave->dev,
535 bond_for_each_slave(bond, slave, iter) {
536 err = dev_set_allmulti(slave->dev, inc);
545 * Retrieve the list of registered multicast addresses for the bonding
546 * device and retransmit an IGMP JOIN request to the current active
549 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
551 struct bonding *bond = container_of(work, struct bonding,
554 if (!rtnl_trylock()) {
555 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
558 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
560 if (bond->igmp_retrans > 1) {
561 bond->igmp_retrans--;
562 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
567 /* Flush bond's hardware addresses from slave
569 static void bond_hw_addr_flush(struct net_device *bond_dev,
570 struct net_device *slave_dev)
572 struct bonding *bond = netdev_priv(bond_dev);
574 dev_uc_unsync(slave_dev, bond_dev);
575 dev_mc_unsync(slave_dev, bond_dev);
577 if (bond->params.mode == BOND_MODE_8023AD) {
578 /* del lacpdu mc addr from mc list */
579 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
581 dev_mc_del(slave_dev, lacpdu_multicast);
585 /*--------------------------- Active slave change ---------------------------*/
587 /* Update the hardware address list and promisc/allmulti for the new and
588 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
589 * slaves up date at all times; only the USES_PRIMARY modes need to call
590 * this function to swap these settings during a failover.
592 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
593 struct slave *old_active)
598 if (bond->dev->flags & IFF_PROMISC)
599 dev_set_promiscuity(old_active->dev, -1);
601 if (bond->dev->flags & IFF_ALLMULTI)
602 dev_set_allmulti(old_active->dev, -1);
604 bond_hw_addr_flush(bond->dev, old_active->dev);
608 /* FIXME: Signal errors upstream. */
609 if (bond->dev->flags & IFF_PROMISC)
610 dev_set_promiscuity(new_active->dev, 1);
612 if (bond->dev->flags & IFF_ALLMULTI)
613 dev_set_allmulti(new_active->dev, 1);
615 netif_addr_lock_bh(bond->dev);
616 dev_uc_sync(new_active->dev, bond->dev);
617 dev_mc_sync(new_active->dev, bond->dev);
618 netif_addr_unlock_bh(bond->dev);
623 * bond_set_dev_addr - clone slave's address to bond
624 * @bond_dev: bond net device
625 * @slave_dev: slave net device
627 * Should be called with RTNL held.
629 static void bond_set_dev_addr(struct net_device *bond_dev,
630 struct net_device *slave_dev)
632 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
633 bond_dev, slave_dev, slave_dev->addr_len);
634 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
635 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
636 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
640 * bond_do_fail_over_mac
642 * Perform special MAC address swapping for fail_over_mac settings
644 * Called with RTNL, curr_slave_lock for write_bh.
646 static void bond_do_fail_over_mac(struct bonding *bond,
647 struct slave *new_active,
648 struct slave *old_active)
649 __releases(&bond->curr_slave_lock)
650 __acquires(&bond->curr_slave_lock)
652 u8 tmp_mac[ETH_ALEN];
653 struct sockaddr saddr;
656 switch (bond->params.fail_over_mac) {
657 case BOND_FOM_ACTIVE:
659 write_unlock_bh(&bond->curr_slave_lock);
660 bond_set_dev_addr(bond->dev, new_active->dev);
661 write_lock_bh(&bond->curr_slave_lock);
664 case BOND_FOM_FOLLOW:
666 * if new_active && old_active, swap them
667 * if just old_active, do nothing (going to no active slave)
668 * if just new_active, set new_active to bond's MAC
673 write_unlock_bh(&bond->curr_slave_lock);
676 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
677 memcpy(saddr.sa_data, old_active->dev->dev_addr,
679 saddr.sa_family = new_active->dev->type;
681 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
682 saddr.sa_family = bond->dev->type;
685 rv = dev_set_mac_address(new_active->dev, &saddr);
687 pr_err("%s: Error %d setting MAC of slave %s\n",
688 bond->dev->name, -rv, new_active->dev->name);
695 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
696 saddr.sa_family = old_active->dev->type;
698 rv = dev_set_mac_address(old_active->dev, &saddr);
700 pr_err("%s: Error %d setting MAC of slave %s\n",
701 bond->dev->name, -rv, new_active->dev->name);
703 write_lock_bh(&bond->curr_slave_lock);
706 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
707 bond->dev->name, bond->params.fail_over_mac);
713 static bool bond_should_change_active(struct bonding *bond)
715 struct slave *prim = bond->primary_slave;
716 struct slave *curr = bond->curr_active_slave;
718 if (!prim || !curr || curr->link != BOND_LINK_UP)
720 if (bond->force_primary) {
721 bond->force_primary = false;
724 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
725 (prim->speed < curr->speed ||
726 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
728 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
734 * find_best_interface - select the best available slave to be the active one
735 * @bond: our bonding struct
737 static struct slave *bond_find_best_slave(struct bonding *bond)
739 struct slave *slave, *bestslave = NULL;
740 struct list_head *iter;
741 int mintime = bond->params.updelay;
743 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
744 bond_should_change_active(bond))
745 return bond->primary_slave;
747 bond_for_each_slave(bond, slave, iter) {
748 if (slave->link == BOND_LINK_UP)
750 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
751 slave->delay < mintime) {
752 mintime = slave->delay;
760 static bool bond_should_notify_peers(struct bonding *bond)
765 slave = rcu_dereference(bond->curr_active_slave);
768 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
769 bond->dev->name, slave ? slave->dev->name : "NULL");
771 if (!slave || !bond->send_peer_notif ||
772 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
779 * change_active_interface - change the active slave into the specified one
780 * @bond: our bonding struct
781 * @new: the new slave to make the active one
783 * Set the new slave to the bond's settings and unset them on the old
785 * Setting include flags, mc-list, promiscuity, allmulti, etc.
787 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
788 * because it is apparently the best available slave we have, even though its
789 * updelay hasn't timed out yet.
791 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
793 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
795 struct slave *old_active = bond->curr_active_slave;
797 if (old_active == new_active)
801 new_active->jiffies = jiffies;
803 if (new_active->link == BOND_LINK_BACK) {
804 if (USES_PRIMARY(bond->params.mode)) {
805 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
806 bond->dev->name, new_active->dev->name,
807 (bond->params.updelay - new_active->delay) * bond->params.miimon);
810 new_active->delay = 0;
811 new_active->link = BOND_LINK_UP;
813 if (bond->params.mode == BOND_MODE_8023AD)
814 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
816 if (bond_is_lb(bond))
817 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
819 if (USES_PRIMARY(bond->params.mode)) {
820 pr_info("%s: making interface %s the new active one.\n",
821 bond->dev->name, new_active->dev->name);
826 if (USES_PRIMARY(bond->params.mode))
827 bond_hw_addr_swap(bond, new_active, old_active);
829 if (bond_is_lb(bond)) {
830 bond_alb_handle_active_change(bond, new_active);
832 bond_set_slave_inactive_flags(old_active,
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->params.mode == 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 ((USES_PRIMARY(bond->params.mode) && new_active) ||
880 bond->params.mode == 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_ATOMIC);
930 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
939 static inline void slave_disable_netpoll(struct slave *slave)
941 struct netpoll *np = slave->np;
947 __netpoll_free_async(np);
949 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
951 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
953 if (!slave_dev->netdev_ops->ndo_poll_controller)
958 static void bond_poll_controller(struct net_device *bond_dev)
962 static void bond_netpoll_cleanup(struct net_device *bond_dev)
964 struct bonding *bond = netdev_priv(bond_dev);
965 struct list_head *iter;
968 bond_for_each_slave(bond, slave, iter)
969 if (IS_UP(slave->dev))
970 slave_disable_netpoll(slave);
973 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
975 struct bonding *bond = netdev_priv(dev);
976 struct list_head *iter;
980 bond_for_each_slave(bond, slave, iter) {
981 err = slave_enable_netpoll(slave);
983 bond_netpoll_cleanup(dev);
990 static inline int slave_enable_netpoll(struct slave *slave)
994 static inline void slave_disable_netpoll(struct slave *slave)
997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1002 /*---------------------------------- IOCTL ----------------------------------*/
1004 static netdev_features_t bond_fix_features(struct net_device *dev,
1005 netdev_features_t features)
1007 struct bonding *bond = netdev_priv(dev);
1008 struct list_head *iter;
1009 netdev_features_t mask;
1010 struct slave *slave;
1012 if (!bond_has_slaves(bond)) {
1013 /* Disable adding VLANs to empty bond. But why? --mq */
1014 features |= NETIF_F_VLAN_CHALLENGED;
1019 features &= ~NETIF_F_ONE_FOR_ALL;
1020 features |= NETIF_F_ALL_FOR_ALL;
1022 bond_for_each_slave(bond, slave, iter) {
1023 features = netdev_increment_features(features,
1024 slave->dev->features,
1027 features = netdev_add_tso_features(features, mask);
1032 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1033 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1034 NETIF_F_HIGHDMA | NETIF_F_LRO)
1036 static void bond_compute_features(struct bonding *bond)
1038 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1039 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1040 struct net_device *bond_dev = bond->dev;
1041 struct list_head *iter;
1042 struct slave *slave;
1043 unsigned short max_hard_header_len = ETH_HLEN;
1044 unsigned int gso_max_size = GSO_MAX_SIZE;
1045 u16 gso_max_segs = GSO_MAX_SEGS;
1047 if (!bond_has_slaves(bond))
1050 bond_for_each_slave(bond, slave, iter) {
1051 vlan_features = netdev_increment_features(vlan_features,
1052 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1054 dst_release_flag &= slave->dev->priv_flags;
1055 if (slave->dev->hard_header_len > max_hard_header_len)
1056 max_hard_header_len = slave->dev->hard_header_len;
1058 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1059 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1063 bond_dev->vlan_features = vlan_features;
1064 bond_dev->hard_header_len = max_hard_header_len;
1065 bond_dev->gso_max_segs = gso_max_segs;
1066 netif_set_gso_max_size(bond_dev, gso_max_size);
1068 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1069 bond_dev->priv_flags = flags | dst_release_flag;
1071 netdev_change_features(bond_dev);
1074 static void bond_setup_by_slave(struct net_device *bond_dev,
1075 struct net_device *slave_dev)
1077 bond_dev->header_ops = slave_dev->header_ops;
1079 bond_dev->type = slave_dev->type;
1080 bond_dev->hard_header_len = slave_dev->hard_header_len;
1081 bond_dev->addr_len = slave_dev->addr_len;
1083 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1084 slave_dev->addr_len);
1087 /* On bonding slaves other than the currently active slave, suppress
1088 * duplicates except for alb non-mcast/bcast.
1090 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1091 struct slave *slave,
1092 struct bonding *bond)
1094 if (bond_is_slave_inactive(slave)) {
1095 if (bond->params.mode == BOND_MODE_ALB &&
1096 skb->pkt_type != PACKET_BROADCAST &&
1097 skb->pkt_type != PACKET_MULTICAST)
1104 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1106 struct sk_buff *skb = *pskb;
1107 struct slave *slave;
1108 struct bonding *bond;
1109 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1111 int ret = RX_HANDLER_ANOTHER;
1113 skb = skb_share_check(skb, GFP_ATOMIC);
1115 return RX_HANDLER_CONSUMED;
1119 slave = bond_slave_get_rcu(skb->dev);
1122 if (bond->params.arp_interval)
1123 slave->dev->last_rx = jiffies;
1125 recv_probe = ACCESS_ONCE(bond->recv_probe);
1127 ret = recv_probe(skb, bond, slave);
1128 if (ret == RX_HANDLER_CONSUMED) {
1134 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1135 return RX_HANDLER_EXACT;
1138 skb->dev = bond->dev;
1140 if (bond->params.mode == BOND_MODE_ALB &&
1141 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1142 skb->pkt_type == PACKET_HOST) {
1144 if (unlikely(skb_cow_head(skb,
1145 skb->data - skb_mac_header(skb)))) {
1147 return RX_HANDLER_CONSUMED;
1149 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1155 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1156 struct net_device *slave_dev,
1157 struct slave *slave)
1161 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1164 slave_dev->flags |= IFF_SLAVE;
1165 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1169 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1170 struct net_device *slave_dev)
1172 netdev_upper_dev_unlink(slave_dev, bond_dev);
1173 slave_dev->flags &= ~IFF_SLAVE;
1174 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1177 /* enslave device <slave> to bond device <master> */
1178 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1180 struct bonding *bond = netdev_priv(bond_dev);
1181 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1182 struct slave *new_slave = NULL, *prev_slave;
1183 struct sockaddr addr;
1187 if (!bond->params.use_carrier &&
1188 slave_dev->ethtool_ops->get_link == NULL &&
1189 slave_ops->ndo_do_ioctl == NULL) {
1190 pr_warning("%s: Warning: no link monitoring support for %s\n",
1191 bond_dev->name, slave_dev->name);
1194 /* already enslaved */
1195 if (slave_dev->flags & IFF_SLAVE) {
1196 pr_debug("Error, Device was already enslaved\n");
1200 if (bond_dev == slave_dev) {
1201 pr_err("%s: cannot enslave bond to itself.\n", bond_dev->name);
1205 /* vlan challenged mutual exclusion */
1206 /* no need to lock since we're protected by rtnl_lock */
1207 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1208 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1209 if (vlan_uses_dev(bond_dev)) {
1210 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1211 bond_dev->name, slave_dev->name, bond_dev->name);
1214 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1215 bond_dev->name, slave_dev->name,
1216 slave_dev->name, bond_dev->name);
1219 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1223 * Old ifenslave binaries are no longer supported. These can
1224 * be identified with moderate accuracy by the state of the slave:
1225 * the current ifenslave will set the interface down prior to
1226 * enslaving it; the old ifenslave will not.
1228 if ((slave_dev->flags & IFF_UP)) {
1229 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1232 goto err_undo_flags;
1235 /* set bonding device ether type by slave - bonding netdevices are
1236 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1237 * there is a need to override some of the type dependent attribs/funcs.
1239 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1240 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1242 if (!bond_has_slaves(bond)) {
1243 if (bond_dev->type != slave_dev->type) {
1244 pr_debug("%s: change device type from %d to %d\n",
1246 bond_dev->type, slave_dev->type);
1248 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1250 res = notifier_to_errno(res);
1252 pr_err("%s: refused to change device type\n",
1255 goto err_undo_flags;
1258 /* Flush unicast and multicast addresses */
1259 dev_uc_flush(bond_dev);
1260 dev_mc_flush(bond_dev);
1262 if (slave_dev->type != ARPHRD_ETHER)
1263 bond_setup_by_slave(bond_dev, slave_dev);
1265 ether_setup(bond_dev);
1266 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1269 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1272 } else if (bond_dev->type != slave_dev->type) {
1273 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1275 slave_dev->type, bond_dev->type);
1277 goto err_undo_flags;
1280 if (slave_ops->ndo_set_mac_address == NULL) {
1281 if (!bond_has_slaves(bond)) {
1282 pr_warn("%s: Warning: The first slave device specified does not support setting the MAC address.\n",
1284 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1285 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1286 pr_warn("%s: Setting fail_over_mac to active for active-backup mode.\n",
1289 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1290 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",
1293 goto err_undo_flags;
1297 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1299 /* If this is the first slave, then we need to set the master's hardware
1300 * address to be the same as the slave's. */
1301 if (!bond_has_slaves(bond) &&
1302 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1303 bond_set_dev_addr(bond->dev, slave_dev);
1305 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1308 goto err_undo_flags;
1311 * Set the new_slave's queue_id to be zero. Queue ID mapping
1312 * is set via sysfs or module option if desired.
1314 new_slave->queue_id = 0;
1316 /* Save slave's original mtu and then set it to match the bond */
1317 new_slave->original_mtu = slave_dev->mtu;
1318 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1320 pr_debug("Error %d calling dev_set_mtu\n", res);
1325 * Save slave's original ("permanent") mac address for modes
1326 * that need it, and for restoring it upon release, and then
1327 * set it to the master's address
1329 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1331 if (!bond->params.fail_over_mac ||
1332 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1334 * Set slave to master's mac address. The application already
1335 * set the master's mac address to that of the first slave
1337 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1338 addr.sa_family = slave_dev->type;
1339 res = dev_set_mac_address(slave_dev, &addr);
1341 pr_debug("Error %d calling set_mac_address\n", res);
1342 goto err_restore_mtu;
1346 /* open the slave since the application closed it */
1347 res = dev_open(slave_dev);
1349 pr_debug("Opening slave %s failed\n", slave_dev->name);
1350 goto err_restore_mac;
1353 new_slave->bond = bond;
1354 new_slave->dev = slave_dev;
1355 slave_dev->priv_flags |= IFF_BONDING;
1357 if (bond_is_lb(bond)) {
1358 /* bond_alb_init_slave() must be called before all other stages since
1359 * it might fail and we do not want to have to undo everything
1361 res = bond_alb_init_slave(bond, new_slave);
1366 /* If the mode USES_PRIMARY, then the following is handled by
1367 * bond_change_active_slave().
1369 if (!USES_PRIMARY(bond->params.mode)) {
1370 /* set promiscuity level to new slave */
1371 if (bond_dev->flags & IFF_PROMISC) {
1372 res = dev_set_promiscuity(slave_dev, 1);
1377 /* set allmulti level to new slave */
1378 if (bond_dev->flags & IFF_ALLMULTI) {
1379 res = dev_set_allmulti(slave_dev, 1);
1384 netif_addr_lock_bh(bond_dev);
1386 dev_mc_sync_multiple(slave_dev, bond_dev);
1387 dev_uc_sync_multiple(slave_dev, bond_dev);
1389 netif_addr_unlock_bh(bond_dev);
1392 if (bond->params.mode == BOND_MODE_8023AD) {
1393 /* add lacpdu mc addr to mc list */
1394 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1396 dev_mc_add(slave_dev, lacpdu_multicast);
1399 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1401 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1402 bond_dev->name, slave_dev->name);
1406 prev_slave = bond_last_slave(bond);
1408 new_slave->delay = 0;
1409 new_slave->link_failure_count = 0;
1411 bond_update_speed_duplex(new_slave);
1413 new_slave->last_arp_rx = jiffies -
1414 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1415 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1416 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1418 if (bond->params.miimon && !bond->params.use_carrier) {
1419 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1421 if ((link_reporting == -1) && !bond->params.arp_interval) {
1423 * miimon is set but a bonded network driver
1424 * does not support ETHTOOL/MII and
1425 * arp_interval is not set. Note: if
1426 * use_carrier is enabled, we will never go
1427 * here (because netif_carrier is always
1428 * supported); thus, we don't need to change
1429 * the messages for netif_carrier.
1431 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1432 bond_dev->name, slave_dev->name);
1433 } else if (link_reporting == -1) {
1434 /* unable get link status using mii/ethtool */
1435 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1436 bond_dev->name, slave_dev->name);
1440 /* check for initial state */
1441 if (bond->params.miimon) {
1442 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1443 if (bond->params.updelay) {
1444 new_slave->link = BOND_LINK_BACK;
1445 new_slave->delay = bond->params.updelay;
1447 new_slave->link = BOND_LINK_UP;
1450 new_slave->link = BOND_LINK_DOWN;
1452 } else if (bond->params.arp_interval) {
1453 new_slave->link = (netif_carrier_ok(slave_dev) ?
1454 BOND_LINK_UP : BOND_LINK_DOWN);
1456 new_slave->link = BOND_LINK_UP;
1459 if (new_slave->link != BOND_LINK_DOWN)
1460 new_slave->jiffies = jiffies;
1461 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1462 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1463 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1465 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1466 /* if there is a primary slave, remember it */
1467 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1468 bond->primary_slave = new_slave;
1469 bond->force_primary = true;
1473 switch (bond->params.mode) {
1474 case BOND_MODE_ACTIVEBACKUP:
1475 bond_set_slave_inactive_flags(new_slave,
1476 BOND_SLAVE_NOTIFY_NOW);
1478 case BOND_MODE_8023AD:
1479 /* in 802.3ad mode, the internal mechanism
1480 * will activate the slaves in the selected
1483 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1484 /* if this is the first slave */
1486 SLAVE_AD_INFO(new_slave).id = 1;
1487 /* Initialize AD with the number of times that the AD timer is called in 1 second
1488 * can be called only after the mac address of the bond is set
1490 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1492 SLAVE_AD_INFO(new_slave).id =
1493 SLAVE_AD_INFO(prev_slave).id + 1;
1496 bond_3ad_bind_slave(new_slave);
1500 bond_set_active_slave(new_slave);
1501 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1504 pr_debug("This slave is always active in trunk mode\n");
1506 /* always active in trunk mode */
1507 bond_set_active_slave(new_slave);
1509 /* In trunking mode there is little meaning to curr_active_slave
1510 * anyway (it holds no special properties of the bond device),
1511 * so we can change it without calling change_active_interface()
1513 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1514 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1517 } /* switch(bond_mode) */
1519 #ifdef CONFIG_NET_POLL_CONTROLLER
1520 slave_dev->npinfo = bond->dev->npinfo;
1521 if (slave_dev->npinfo) {
1522 if (slave_enable_netpoll(new_slave)) {
1523 pr_info("Error, %s: master_dev is using netpoll, "
1524 "but new slave device does not support netpoll.\n",
1532 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1535 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1539 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1541 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1542 goto err_unregister;
1545 res = bond_sysfs_slave_add(new_slave);
1547 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1548 goto err_upper_unlink;
1552 bond_compute_features(bond);
1553 bond_set_carrier(bond);
1555 if (USES_PRIMARY(bond->params.mode)) {
1557 write_lock_bh(&bond->curr_slave_lock);
1558 bond_select_active_slave(bond);
1559 write_unlock_bh(&bond->curr_slave_lock);
1560 unblock_netpoll_tx();
1563 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1564 bond_dev->name, slave_dev->name,
1565 bond_is_active_slave(new_slave) ? "n active" : " backup",
1566 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1568 /* enslave is successful */
1571 /* Undo stages on error */
1573 bond_upper_dev_unlink(bond_dev, slave_dev);
1576 netdev_rx_handler_unregister(slave_dev);
1579 if (!USES_PRIMARY(bond->params.mode))
1580 bond_hw_addr_flush(bond_dev, slave_dev);
1582 vlan_vids_del_by_dev(slave_dev, bond_dev);
1583 if (bond->primary_slave == new_slave)
1584 bond->primary_slave = NULL;
1585 if (bond->curr_active_slave == new_slave) {
1587 write_lock_bh(&bond->curr_slave_lock);
1588 bond_change_active_slave(bond, NULL);
1589 bond_select_active_slave(bond);
1590 write_unlock_bh(&bond->curr_slave_lock);
1591 unblock_netpoll_tx();
1593 slave_disable_netpoll(new_slave);
1596 slave_dev->priv_flags &= ~IFF_BONDING;
1597 dev_close(slave_dev);
1600 if (!bond->params.fail_over_mac ||
1601 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1602 /* XXX TODO - fom follow mode needs to change master's
1603 * MAC if this slave's MAC is in use by the bond, or at
1604 * least print a warning.
1606 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1607 addr.sa_family = slave_dev->type;
1608 dev_set_mac_address(slave_dev, &addr);
1612 dev_set_mtu(slave_dev, new_slave->original_mtu);
1618 /* Enslave of first slave has failed and we need to fix master's mac */
1619 if (!bond_has_slaves(bond) &&
1620 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1621 eth_hw_addr_random(bond_dev);
1627 * Try to release the slave device <slave> from the bond device <master>
1628 * It is legal to access curr_active_slave without a lock because all the function
1629 * is write-locked. If "all" is true it means that the function is being called
1630 * while destroying a bond interface and all slaves are being released.
1632 * The rules for slave state should be:
1633 * for Active/Backup:
1634 * Active stays on all backups go down
1635 * for Bonded connections:
1636 * The first up interface should be left on and all others downed.
1638 static int __bond_release_one(struct net_device *bond_dev,
1639 struct net_device *slave_dev,
1642 struct bonding *bond = netdev_priv(bond_dev);
1643 struct slave *slave, *oldcurrent;
1644 struct sockaddr addr;
1645 int old_flags = bond_dev->flags;
1646 netdev_features_t old_features = bond_dev->features;
1648 /* slave is not a slave or master is not master of this slave */
1649 if (!(slave_dev->flags & IFF_SLAVE) ||
1650 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1651 pr_err("%s: Error: cannot release %s.\n",
1652 bond_dev->name, slave_dev->name);
1658 slave = bond_get_slave_by_dev(bond, slave_dev);
1660 /* not a slave of this bond */
1661 pr_info("%s: %s not enslaved\n",
1662 bond_dev->name, slave_dev->name);
1663 unblock_netpoll_tx();
1667 bond_sysfs_slave_del(slave);
1669 bond_upper_dev_unlink(bond_dev, slave_dev);
1670 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1671 * for this slave anymore.
1673 netdev_rx_handler_unregister(slave_dev);
1674 write_lock_bh(&bond->lock);
1676 /* Inform AD package of unbinding of slave. */
1677 if (bond->params.mode == BOND_MODE_8023AD)
1678 bond_3ad_unbind_slave(slave);
1680 write_unlock_bh(&bond->lock);
1682 pr_info("%s: releasing %s interface %s\n",
1684 bond_is_active_slave(slave) ? "active" : "backup",
1687 oldcurrent = bond->curr_active_slave;
1689 bond->current_arp_slave = NULL;
1691 if (!all && (!bond->params.fail_over_mac ||
1692 bond->params.mode != BOND_MODE_ACTIVEBACKUP)) {
1693 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1694 bond_has_slaves(bond))
1695 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",
1696 bond_dev->name, slave_dev->name,
1698 bond_dev->name, slave_dev->name);
1701 if (bond->primary_slave == slave)
1702 bond->primary_slave = NULL;
1704 if (oldcurrent == slave) {
1705 write_lock_bh(&bond->curr_slave_lock);
1706 bond_change_active_slave(bond, NULL);
1707 write_unlock_bh(&bond->curr_slave_lock);
1710 if (bond_is_lb(bond)) {
1711 /* Must be called only after the slave has been
1712 * detached from the list and the curr_active_slave
1713 * has been cleared (if our_slave == old_current),
1714 * but before a new active slave is selected.
1716 bond_alb_deinit_slave(bond, slave);
1720 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1721 } else if (oldcurrent == slave) {
1723 * Note that we hold RTNL over this sequence, so there
1724 * is no concern that another slave add/remove event
1727 write_lock_bh(&bond->curr_slave_lock);
1729 bond_select_active_slave(bond);
1731 write_unlock_bh(&bond->curr_slave_lock);
1734 if (!bond_has_slaves(bond)) {
1735 bond_set_carrier(bond);
1736 eth_hw_addr_random(bond_dev);
1738 if (vlan_uses_dev(bond_dev)) {
1739 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1740 bond_dev->name, bond_dev->name);
1741 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1746 unblock_netpoll_tx();
1750 if (!bond_has_slaves(bond)) {
1751 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1752 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1755 bond_compute_features(bond);
1756 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1757 (old_features & NETIF_F_VLAN_CHALLENGED))
1758 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1759 bond_dev->name, slave_dev->name, bond_dev->name);
1761 /* must do this from outside any spinlocks */
1762 vlan_vids_del_by_dev(slave_dev, bond_dev);
1764 /* If the mode USES_PRIMARY, then this cases was handled above by
1765 * bond_change_active_slave(..., NULL)
1767 if (!USES_PRIMARY(bond->params.mode)) {
1768 /* unset promiscuity level from slave
1769 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1770 * of the IFF_PROMISC flag in the bond_dev, but we need the
1771 * value of that flag before that change, as that was the value
1772 * when this slave was attached, so we cache at the start of the
1773 * function and use it here. Same goes for ALLMULTI below
1775 if (old_flags & IFF_PROMISC)
1776 dev_set_promiscuity(slave_dev, -1);
1778 /* unset allmulti level from slave */
1779 if (old_flags & IFF_ALLMULTI)
1780 dev_set_allmulti(slave_dev, -1);
1782 bond_hw_addr_flush(bond_dev, slave_dev);
1785 slave_disable_netpoll(slave);
1787 /* close slave before restoring its mac address */
1788 dev_close(slave_dev);
1790 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1791 bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1792 /* restore original ("permanent") mac address */
1793 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1794 addr.sa_family = slave_dev->type;
1795 dev_set_mac_address(slave_dev, &addr);
1798 dev_set_mtu(slave_dev, slave->original_mtu);
1800 slave_dev->priv_flags &= ~IFF_BONDING;
1804 return 0; /* deletion OK */
1807 /* A wrapper used because of ndo_del_link */
1808 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1810 return __bond_release_one(bond_dev, slave_dev, false);
1814 * First release a slave and then destroy the bond if no more slaves are left.
1815 * Must be under rtnl_lock when this function is called.
1817 static int bond_release_and_destroy(struct net_device *bond_dev,
1818 struct net_device *slave_dev)
1820 struct bonding *bond = netdev_priv(bond_dev);
1823 ret = bond_release(bond_dev, slave_dev);
1824 if (ret == 0 && !bond_has_slaves(bond)) {
1825 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1826 pr_info("%s: destroying bond %s.\n",
1827 bond_dev->name, bond_dev->name);
1828 unregister_netdevice(bond_dev);
1833 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1835 struct bonding *bond = netdev_priv(bond_dev);
1837 info->bond_mode = bond->params.mode;
1838 info->miimon = bond->params.miimon;
1840 read_lock(&bond->lock);
1841 info->num_slaves = bond->slave_cnt;
1842 read_unlock(&bond->lock);
1847 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1849 struct bonding *bond = netdev_priv(bond_dev);
1850 struct list_head *iter;
1851 int i = 0, res = -ENODEV;
1852 struct slave *slave;
1854 read_lock(&bond->lock);
1855 bond_for_each_slave(bond, slave, iter) {
1856 if (i++ == (int)info->slave_id) {
1858 strcpy(info->slave_name, slave->dev->name);
1859 info->link = slave->link;
1860 info->state = bond_slave_state(slave);
1861 info->link_failure_count = slave->link_failure_count;
1865 read_unlock(&bond->lock);
1870 /*-------------------------------- Monitoring -------------------------------*/
1873 static int bond_miimon_inspect(struct bonding *bond)
1875 int link_state, commit = 0;
1876 struct list_head *iter;
1877 struct slave *slave;
1878 bool ignore_updelay;
1880 ignore_updelay = !bond->curr_active_slave ? true : false;
1882 bond_for_each_slave_rcu(bond, slave, iter) {
1883 slave->new_link = BOND_LINK_NOCHANGE;
1885 link_state = bond_check_dev_link(bond, slave->dev, 0);
1887 switch (slave->link) {
1892 slave->link = BOND_LINK_FAIL;
1893 slave->delay = bond->params.downdelay;
1895 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1897 (bond->params.mode ==
1898 BOND_MODE_ACTIVEBACKUP) ?
1899 (bond_is_active_slave(slave) ?
1900 "active " : "backup ") : "",
1902 bond->params.downdelay * bond->params.miimon);
1905 case BOND_LINK_FAIL:
1908 * recovered before downdelay expired
1910 slave->link = BOND_LINK_UP;
1911 slave->jiffies = jiffies;
1912 pr_info("%s: link status up again after %d ms for interface %s.\n",
1914 (bond->params.downdelay - slave->delay) *
1915 bond->params.miimon,
1920 if (slave->delay <= 0) {
1921 slave->new_link = BOND_LINK_DOWN;
1929 case BOND_LINK_DOWN:
1933 slave->link = BOND_LINK_BACK;
1934 slave->delay = bond->params.updelay;
1937 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1938 bond->dev->name, slave->dev->name,
1939 ignore_updelay ? 0 :
1940 bond->params.updelay *
1941 bond->params.miimon);
1944 case BOND_LINK_BACK:
1946 slave->link = BOND_LINK_DOWN;
1947 pr_info("%s: link status down again after %d ms for interface %s.\n",
1949 (bond->params.updelay - slave->delay) *
1950 bond->params.miimon,
1959 if (slave->delay <= 0) {
1960 slave->new_link = BOND_LINK_UP;
1962 ignore_updelay = false;
1974 static void bond_miimon_commit(struct bonding *bond)
1976 struct list_head *iter;
1977 struct slave *slave;
1979 bond_for_each_slave(bond, slave, iter) {
1980 switch (slave->new_link) {
1981 case BOND_LINK_NOCHANGE:
1985 slave->link = BOND_LINK_UP;
1986 slave->jiffies = jiffies;
1988 if (bond->params.mode == BOND_MODE_8023AD) {
1989 /* prevent it from being the active one */
1990 bond_set_backup_slave(slave);
1991 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
1992 /* make it immediately active */
1993 bond_set_active_slave(slave);
1994 } else if (slave != bond->primary_slave) {
1995 /* prevent it from being the active one */
1996 bond_set_backup_slave(slave);
1999 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2000 bond->dev->name, slave->dev->name,
2001 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2002 slave->duplex ? "full" : "half");
2004 /* notify ad that the link status has changed */
2005 if (bond->params.mode == BOND_MODE_8023AD)
2006 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2008 if (bond_is_lb(bond))
2009 bond_alb_handle_link_change(bond, slave,
2012 if (!bond->curr_active_slave ||
2013 (slave == bond->primary_slave))
2018 case BOND_LINK_DOWN:
2019 if (slave->link_failure_count < UINT_MAX)
2020 slave->link_failure_count++;
2022 slave->link = BOND_LINK_DOWN;
2024 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2025 bond->params.mode == BOND_MODE_8023AD)
2026 bond_set_slave_inactive_flags(slave,
2027 BOND_SLAVE_NOTIFY_NOW);
2029 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2030 bond->dev->name, slave->dev->name);
2032 if (bond->params.mode == BOND_MODE_8023AD)
2033 bond_3ad_handle_link_change(slave,
2036 if (bond_is_lb(bond))
2037 bond_alb_handle_link_change(bond, slave,
2040 if (slave == bond->curr_active_slave)
2046 pr_err("%s: invalid new link %d on slave %s\n",
2047 bond->dev->name, slave->new_link,
2049 slave->new_link = BOND_LINK_NOCHANGE;
2057 write_lock_bh(&bond->curr_slave_lock);
2058 bond_select_active_slave(bond);
2059 write_unlock_bh(&bond->curr_slave_lock);
2060 unblock_netpoll_tx();
2063 bond_set_carrier(bond);
2069 * Really a wrapper that splits the mii monitor into two phases: an
2070 * inspection, then (if inspection indicates something needs to be done)
2071 * an acquisition of appropriate locks followed by a commit phase to
2072 * implement whatever link state changes are indicated.
2074 static void bond_mii_monitor(struct work_struct *work)
2076 struct bonding *bond = container_of(work, struct bonding,
2078 bool should_notify_peers = false;
2079 unsigned long delay;
2081 delay = msecs_to_jiffies(bond->params.miimon);
2083 if (!bond_has_slaves(bond))
2088 should_notify_peers = bond_should_notify_peers(bond);
2090 if (bond_miimon_inspect(bond)) {
2093 /* Race avoidance with bond_close cancel of workqueue */
2094 if (!rtnl_trylock()) {
2096 should_notify_peers = false;
2100 bond_miimon_commit(bond);
2102 rtnl_unlock(); /* might sleep, hold no other locks */
2107 if (bond->params.miimon)
2108 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2110 if (should_notify_peers) {
2111 if (!rtnl_trylock())
2113 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2118 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2120 struct net_device *upper;
2121 struct list_head *iter;
2124 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2128 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2129 if (ip == bond_confirm_addr(upper, 0, ip)) {
2140 * We go to the (large) trouble of VLAN tagging ARP frames because
2141 * switches in VLAN mode (especially if ports are configured as
2142 * "native" to a VLAN) might not pass non-tagged frames.
2144 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2146 struct sk_buff *skb;
2148 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2149 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2151 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2152 NULL, slave_dev->dev_addr, NULL);
2155 pr_err("ARP packet allocation failed\n");
2159 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2161 pr_err("failed to insert VLAN tag\n");
2169 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2171 struct net_device *upper, *vlan_upper;
2172 struct list_head *iter, *vlan_iter;
2174 __be32 *targets = bond->params.arp_targets, addr;
2177 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2178 pr_debug("basa: target %pI4\n", &targets[i]);
2180 /* Find out through which dev should the packet go */
2181 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2184 pr_debug("%s: no route to arp_ip_target %pI4\n",
2185 bond->dev->name, &targets[i]);
2191 /* bond device itself */
2192 if (rt->dst.dev == bond->dev)
2196 /* first we search only for vlan devices. for every vlan
2197 * found we verify its upper dev list, searching for the
2198 * rt->dst.dev. If found we save the tag of the vlan and
2199 * proceed to send the packet.
2203 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2205 if (!is_vlan_dev(vlan_upper))
2207 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2209 if (upper == rt->dst.dev) {
2210 vlan_id = vlan_dev_vlan_id(vlan_upper);
2217 /* if the device we're looking for is not on top of any of
2218 * our upper vlans, then just search for any dev that
2219 * matches, and in case it's a vlan - save the id
2221 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2222 if (upper == rt->dst.dev) {
2223 /* if it's a vlan - get its VID */
2224 if (is_vlan_dev(upper))
2225 vlan_id = vlan_dev_vlan_id(upper);
2233 /* Not our device - skip */
2234 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2235 bond->dev->name, &targets[i],
2236 rt->dst.dev ? rt->dst.dev->name : "NULL");
2242 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2244 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2249 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2253 if (!sip || !bond_has_this_ip(bond, tip)) {
2254 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2258 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2260 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2263 slave->last_arp_rx = jiffies;
2264 slave->target_last_arp_rx[i] = jiffies;
2267 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2268 struct slave *slave)
2270 struct arphdr *arp = (struct arphdr *)skb->data;
2271 unsigned char *arp_ptr;
2275 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2276 return RX_HANDLER_ANOTHER;
2278 read_lock(&bond->lock);
2280 if (!slave_do_arp_validate(bond, slave))
2283 alen = arp_hdr_len(bond->dev);
2285 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2286 bond->dev->name, skb->dev->name);
2288 if (alen > skb_headlen(skb)) {
2289 arp = kmalloc(alen, GFP_ATOMIC);
2292 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2296 if (arp->ar_hln != bond->dev->addr_len ||
2297 skb->pkt_type == PACKET_OTHERHOST ||
2298 skb->pkt_type == PACKET_LOOPBACK ||
2299 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2300 arp->ar_pro != htons(ETH_P_IP) ||
2304 arp_ptr = (unsigned char *)(arp + 1);
2305 arp_ptr += bond->dev->addr_len;
2306 memcpy(&sip, arp_ptr, 4);
2307 arp_ptr += 4 + bond->dev->addr_len;
2308 memcpy(&tip, arp_ptr, 4);
2310 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2311 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2312 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2316 * Backup slaves won't see the ARP reply, but do come through
2317 * here for each ARP probe (so we swap the sip/tip to validate
2318 * the probe). In a "redundant switch, common router" type of
2319 * configuration, the ARP probe will (hopefully) travel from
2320 * the active, through one switch, the router, then the other
2321 * switch before reaching the backup.
2323 * We 'trust' the arp requests if there is an active slave and
2324 * it received valid arp reply(s) after it became active. This
2325 * is done to avoid endless looping when we can't reach the
2326 * arp_ip_target and fool ourselves with our own arp requests.
2328 if (bond_is_active_slave(slave))
2329 bond_validate_arp(bond, slave, sip, tip);
2330 else if (bond->curr_active_slave &&
2331 time_after(slave_last_rx(bond, bond->curr_active_slave),
2332 bond->curr_active_slave->jiffies))
2333 bond_validate_arp(bond, slave, tip, sip);
2336 read_unlock(&bond->lock);
2337 if (arp != (struct arphdr *)skb->data)
2339 return RX_HANDLER_ANOTHER;
2342 /* function to verify if we're in the arp_interval timeslice, returns true if
2343 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2344 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2346 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2349 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2351 return time_in_range(jiffies,
2352 last_act - delta_in_ticks,
2353 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2357 * this function is called regularly to monitor each slave's link
2358 * ensuring that traffic is being sent and received when arp monitoring
2359 * is used in load-balancing mode. if the adapter has been dormant, then an
2360 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2361 * arp monitoring in active backup mode.
2363 static void bond_loadbalance_arp_mon(struct work_struct *work)
2365 struct bonding *bond = container_of(work, struct bonding,
2367 struct slave *slave, *oldcurrent;
2368 struct list_head *iter;
2369 int do_failover = 0, slave_state_changed = 0;
2371 if (!bond_has_slaves(bond))
2376 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2377 /* see if any of the previous devices are up now (i.e. they have
2378 * xmt and rcv traffic). the curr_active_slave does not come into
2379 * the picture unless it is null. also, slave->jiffies is not needed
2380 * here because we send an arp on each slave and give a slave as
2381 * long as it needs to get the tx/rx within the delta.
2382 * TODO: what about up/down delay in arp mode? it wasn't here before
2385 bond_for_each_slave_rcu(bond, slave, iter) {
2386 unsigned long trans_start = dev_trans_start(slave->dev);
2388 if (slave->link != BOND_LINK_UP) {
2389 if (bond_time_in_interval(bond, trans_start, 1) &&
2390 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2392 slave->link = BOND_LINK_UP;
2393 slave_state_changed = 1;
2395 /* primary_slave has no meaning in round-robin
2396 * mode. the window of a slave being up and
2397 * curr_active_slave being null after enslaving
2401 pr_info("%s: link status definitely up for interface %s, ",
2406 pr_info("%s: interface %s is now up\n",
2412 /* slave->link == BOND_LINK_UP */
2414 /* not all switches will respond to an arp request
2415 * when the source ip is 0, so don't take the link down
2416 * if we don't know our ip yet
2418 if (!bond_time_in_interval(bond, trans_start, 2) ||
2419 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2421 slave->link = BOND_LINK_DOWN;
2422 slave_state_changed = 1;
2424 if (slave->link_failure_count < UINT_MAX)
2425 slave->link_failure_count++;
2427 pr_info("%s: interface %s is now down.\n",
2431 if (slave == oldcurrent)
2436 /* note: if switch is in round-robin mode, all links
2437 * must tx arp to ensure all links rx an arp - otherwise
2438 * links may oscillate or not come up at all; if switch is
2439 * in something like xor mode, there is nothing we can
2440 * do - all replies will be rx'ed on same link causing slaves
2441 * to be unstable during low/no traffic periods
2443 if (IS_UP(slave->dev))
2444 bond_arp_send_all(bond, slave);
2449 if (do_failover || slave_state_changed) {
2450 if (!rtnl_trylock())
2453 if (slave_state_changed) {
2454 bond_slave_state_change(bond);
2455 } else if (do_failover) {
2456 /* the bond_select_active_slave must hold RTNL
2457 * and curr_slave_lock for write.
2460 write_lock_bh(&bond->curr_slave_lock);
2462 bond_select_active_slave(bond);
2464 write_unlock_bh(&bond->curr_slave_lock);
2465 unblock_netpoll_tx();
2471 if (bond->params.arp_interval)
2472 queue_delayed_work(bond->wq, &bond->arp_work,
2473 msecs_to_jiffies(bond->params.arp_interval));
2477 * Called to inspect slaves for active-backup mode ARP monitor link state
2478 * changes. Sets new_link in slaves to specify what action should take
2479 * place for the slave. Returns 0 if no changes are found, >0 if changes
2480 * to link states must be committed.
2482 * Called with rcu_read_lock hold.
2484 static int bond_ab_arp_inspect(struct bonding *bond)
2486 unsigned long trans_start, last_rx;
2487 struct list_head *iter;
2488 struct slave *slave;
2491 bond_for_each_slave_rcu(bond, slave, iter) {
2492 slave->new_link = BOND_LINK_NOCHANGE;
2493 last_rx = slave_last_rx(bond, slave);
2495 if (slave->link != BOND_LINK_UP) {
2496 if (bond_time_in_interval(bond, last_rx, 1)) {
2497 slave->new_link = BOND_LINK_UP;
2504 * Give slaves 2*delta after being enslaved or made
2505 * active. This avoids bouncing, as the last receive
2506 * times need a full ARP monitor cycle to be updated.
2508 if (bond_time_in_interval(bond, slave->jiffies, 2))
2512 * Backup slave is down if:
2513 * - No current_arp_slave AND
2514 * - more than 3*delta since last receive AND
2515 * - the bond has an IP address
2517 * Note: a non-null current_arp_slave indicates
2518 * the curr_active_slave went down and we are
2519 * searching for a new one; under this condition
2520 * we only take the curr_active_slave down - this
2521 * gives each slave a chance to tx/rx traffic
2522 * before being taken out
2524 if (!bond_is_active_slave(slave) &&
2525 !bond->current_arp_slave &&
2526 !bond_time_in_interval(bond, last_rx, 3)) {
2527 slave->new_link = BOND_LINK_DOWN;
2532 * Active slave is down if:
2533 * - more than 2*delta since transmitting OR
2534 * - (more than 2*delta since receive AND
2535 * the bond has an IP address)
2537 trans_start = dev_trans_start(slave->dev);
2538 if (bond_is_active_slave(slave) &&
2539 (!bond_time_in_interval(bond, trans_start, 2) ||
2540 !bond_time_in_interval(bond, last_rx, 2))) {
2541 slave->new_link = BOND_LINK_DOWN;
2550 * Called to commit link state changes noted by inspection step of
2551 * active-backup mode ARP monitor.
2553 * Called with RTNL hold.
2555 static void bond_ab_arp_commit(struct bonding *bond)
2557 unsigned long trans_start;
2558 struct list_head *iter;
2559 struct slave *slave;
2561 bond_for_each_slave(bond, slave, iter) {
2562 switch (slave->new_link) {
2563 case BOND_LINK_NOCHANGE:
2567 trans_start = dev_trans_start(slave->dev);
2568 if (bond->curr_active_slave != slave ||
2569 (!bond->curr_active_slave &&
2570 bond_time_in_interval(bond, trans_start, 1))) {
2571 slave->link = BOND_LINK_UP;
2572 if (bond->current_arp_slave) {
2573 bond_set_slave_inactive_flags(
2574 bond->current_arp_slave,
2575 BOND_SLAVE_NOTIFY_NOW);
2576 bond->current_arp_slave = NULL;
2579 pr_info("%s: link status definitely up for interface %s.\n",
2580 bond->dev->name, slave->dev->name);
2582 if (!bond->curr_active_slave ||
2583 (slave == bond->primary_slave))
2590 case BOND_LINK_DOWN:
2591 if (slave->link_failure_count < UINT_MAX)
2592 slave->link_failure_count++;
2594 slave->link = BOND_LINK_DOWN;
2595 bond_set_slave_inactive_flags(slave,
2596 BOND_SLAVE_NOTIFY_NOW);
2598 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2599 bond->dev->name, slave->dev->name);
2601 if (slave == bond->curr_active_slave) {
2602 bond->current_arp_slave = NULL;
2609 pr_err("%s: impossible: new_link %d on slave %s\n",
2610 bond->dev->name, slave->new_link,
2618 write_lock_bh(&bond->curr_slave_lock);
2619 bond_select_active_slave(bond);
2620 write_unlock_bh(&bond->curr_slave_lock);
2621 unblock_netpoll_tx();
2624 bond_set_carrier(bond);
2628 * Send ARP probes for active-backup mode ARP monitor.
2630 * Called with rcu_read_lock hold.
2632 static bool bond_ab_arp_probe(struct bonding *bond)
2634 struct slave *slave, *before = NULL, *new_slave = NULL,
2635 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2636 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2637 struct list_head *iter;
2639 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2641 if (curr_arp_slave && curr_active_slave)
2642 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2643 curr_arp_slave->dev->name,
2644 curr_active_slave->dev->name);
2646 if (curr_active_slave) {
2647 bond_arp_send_all(bond, curr_active_slave);
2648 return should_notify_rtnl;
2651 /* if we don't have a curr_active_slave, search for the next available
2652 * backup slave from the current_arp_slave and make it the candidate
2653 * for becoming the curr_active_slave
2656 if (!curr_arp_slave) {
2657 curr_arp_slave = bond_first_slave_rcu(bond);
2658 if (!curr_arp_slave)
2659 return should_notify_rtnl;
2662 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2664 bond_for_each_slave_rcu(bond, slave, iter) {
2665 if (!found && !before && IS_UP(slave->dev))
2668 if (found && !new_slave && IS_UP(slave->dev))
2670 /* if the link state is up at this point, we
2671 * mark it down - this can happen if we have
2672 * simultaneous link failures and
2673 * reselect_active_interface doesn't make this
2674 * one the current slave so it is still marked
2675 * up when it is actually down
2677 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2678 slave->link = BOND_LINK_DOWN;
2679 if (slave->link_failure_count < UINT_MAX)
2680 slave->link_failure_count++;
2682 bond_set_slave_inactive_flags(slave,
2683 BOND_SLAVE_NOTIFY_LATER);
2685 pr_info("%s: backup interface %s is now down.\n",
2686 bond->dev->name, slave->dev->name);
2688 if (slave == curr_arp_slave)
2692 if (!new_slave && before)
2698 new_slave->link = BOND_LINK_BACK;
2699 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2700 bond_arp_send_all(bond, new_slave);
2701 new_slave->jiffies = jiffies;
2702 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2705 bond_for_each_slave_rcu(bond, slave, iter) {
2706 if (slave->should_notify) {
2707 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2711 return should_notify_rtnl;
2714 static void bond_activebackup_arp_mon(struct work_struct *work)
2716 struct bonding *bond = container_of(work, struct bonding,
2718 bool should_notify_peers = false;
2719 bool should_notify_rtnl = false;
2722 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2724 if (!bond_has_slaves(bond))
2729 should_notify_peers = bond_should_notify_peers(bond);
2731 if (bond_ab_arp_inspect(bond)) {
2734 /* Race avoidance with bond_close flush of workqueue */
2735 if (!rtnl_trylock()) {
2737 should_notify_peers = false;
2741 bond_ab_arp_commit(bond);
2747 should_notify_rtnl = bond_ab_arp_probe(bond);
2751 if (bond->params.arp_interval)
2752 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2754 if (should_notify_peers || should_notify_rtnl) {
2755 if (!rtnl_trylock())
2758 if (should_notify_peers)
2759 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2761 if (should_notify_rtnl)
2762 bond_slave_state_notify(bond);
2768 /*-------------------------- netdev event handling --------------------------*/
2771 * Change device name
2773 static int bond_event_changename(struct bonding *bond)
2775 bond_remove_proc_entry(bond);
2776 bond_create_proc_entry(bond);
2778 bond_debug_reregister(bond);
2783 static int bond_master_netdev_event(unsigned long event,
2784 struct net_device *bond_dev)
2786 struct bonding *event_bond = netdev_priv(bond_dev);
2789 case NETDEV_CHANGENAME:
2790 return bond_event_changename(event_bond);
2791 case NETDEV_UNREGISTER:
2792 bond_remove_proc_entry(event_bond);
2794 case NETDEV_REGISTER:
2795 bond_create_proc_entry(event_bond);
2797 case NETDEV_NOTIFY_PEERS:
2798 if (event_bond->send_peer_notif)
2799 event_bond->send_peer_notif--;
2808 static int bond_slave_netdev_event(unsigned long event,
2809 struct net_device *slave_dev)
2811 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2812 struct bonding *bond;
2813 struct net_device *bond_dev;
2817 /* A netdev event can be generated while enslaving a device
2818 * before netdev_rx_handler_register is called in which case
2819 * slave will be NULL
2823 bond_dev = slave->bond->dev;
2827 case NETDEV_UNREGISTER:
2828 if (bond_dev->type != ARPHRD_ETHER)
2829 bond_release_and_destroy(bond_dev, slave_dev);
2831 bond_release(bond_dev, slave_dev);
2835 old_speed = slave->speed;
2836 old_duplex = slave->duplex;
2838 bond_update_speed_duplex(slave);
2840 if (bond->params.mode == BOND_MODE_8023AD) {
2841 if (old_speed != slave->speed)
2842 bond_3ad_adapter_speed_changed(slave);
2843 if (old_duplex != slave->duplex)
2844 bond_3ad_adapter_duplex_changed(slave);
2849 * ... Or is it this?
2852 case NETDEV_CHANGEMTU:
2854 * TODO: Should slaves be allowed to
2855 * independently alter their MTU? For
2856 * an active-backup bond, slaves need
2857 * not be the same type of device, so
2858 * MTUs may vary. For other modes,
2859 * slaves arguably should have the
2860 * same MTUs. To do this, we'd need to
2861 * take over the slave's change_mtu
2862 * function for the duration of their
2866 case NETDEV_CHANGENAME:
2867 /* we don't care if we don't have primary set */
2868 if (!USES_PRIMARY(bond->params.mode) ||
2869 !bond->params.primary[0])
2872 if (slave == bond->primary_slave) {
2873 /* slave's name changed - he's no longer primary */
2874 bond->primary_slave = NULL;
2875 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2876 /* we have a new primary slave */
2877 bond->primary_slave = slave;
2878 } else { /* we didn't change primary - exit */
2882 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2883 bond->dev->name, bond->primary_slave ? slave_dev->name :
2887 write_lock_bh(&bond->curr_slave_lock);
2888 bond_select_active_slave(bond);
2889 write_unlock_bh(&bond->curr_slave_lock);
2890 unblock_netpoll_tx();
2892 case NETDEV_FEAT_CHANGE:
2893 bond_compute_features(bond);
2895 case NETDEV_RESEND_IGMP:
2896 /* Propagate to master device */
2897 call_netdevice_notifiers(event, slave->bond->dev);
2907 * bond_netdev_event: handle netdev notifier chain events.
2909 * This function receives events for the netdev chain. The caller (an
2910 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2911 * locks for us to safely manipulate the slave devices (RTNL lock,
2914 static int bond_netdev_event(struct notifier_block *this,
2915 unsigned long event, void *ptr)
2917 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2919 pr_debug("event_dev: %s, event: %lx\n",
2920 event_dev ? event_dev->name : "None",
2923 if (!(event_dev->priv_flags & IFF_BONDING))
2926 if (event_dev->flags & IFF_MASTER) {
2927 pr_debug("IFF_MASTER\n");
2928 return bond_master_netdev_event(event, event_dev);
2931 if (event_dev->flags & IFF_SLAVE) {
2932 pr_debug("IFF_SLAVE\n");
2933 return bond_slave_netdev_event(event, event_dev);
2939 static struct notifier_block bond_netdev_notifier = {
2940 .notifier_call = bond_netdev_event,
2943 /*---------------------------- Hashing Policies -----------------------------*/
2945 /* L2 hash helper */
2946 static inline u32 bond_eth_hash(struct sk_buff *skb)
2948 struct ethhdr *data = (struct ethhdr *)skb->data;
2950 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2951 return data->h_dest[5] ^ data->h_source[5];
2956 /* Extract the appropriate headers based on bond's xmit policy */
2957 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2958 struct flow_keys *fk)
2960 const struct ipv6hdr *iph6;
2961 const struct iphdr *iph;
2962 int noff, proto = -1;
2964 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2965 return skb_flow_dissect(skb, fk);
2968 noff = skb_network_offset(skb);
2969 if (skb->protocol == htons(ETH_P_IP)) {
2970 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2973 fk->src = iph->saddr;
2974 fk->dst = iph->daddr;
2975 noff += iph->ihl << 2;
2976 if (!ip_is_fragment(iph))
2977 proto = iph->protocol;
2978 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2979 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2981 iph6 = ipv6_hdr(skb);
2982 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2983 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2984 noff += sizeof(*iph6);
2985 proto = iph6->nexthdr;
2989 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2990 fk->ports = skb_flow_get_ports(skb, noff, proto);
2996 * bond_xmit_hash - generate a hash value based on the xmit policy
2997 * @bond: bonding device
2998 * @skb: buffer to use for headers
2999 * @count: modulo value
3001 * This function will extract the necessary headers from the skb buffer and use
3002 * them to generate a hash based on the xmit_policy set in the bonding device
3003 * which will be reduced modulo count before returning.
3005 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3007 struct flow_keys flow;
3010 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3011 !bond_flow_dissect(bond, skb, &flow))
3012 return bond_eth_hash(skb) % count;
3014 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3015 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3016 hash = bond_eth_hash(skb);
3018 hash = (__force u32)flow.ports;
3019 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3020 hash ^= (hash >> 16);
3021 hash ^= (hash >> 8);
3023 return hash % count;
3026 /*-------------------------- Device entry points ----------------------------*/
3028 static void bond_work_init_all(struct bonding *bond)
3030 INIT_DELAYED_WORK(&bond->mcast_work,
3031 bond_resend_igmp_join_requests_delayed);
3032 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3033 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3034 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3035 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3037 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3038 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3041 static void bond_work_cancel_all(struct bonding *bond)
3043 cancel_delayed_work_sync(&bond->mii_work);
3044 cancel_delayed_work_sync(&bond->arp_work);
3045 cancel_delayed_work_sync(&bond->alb_work);
3046 cancel_delayed_work_sync(&bond->ad_work);
3047 cancel_delayed_work_sync(&bond->mcast_work);
3050 static int bond_open(struct net_device *bond_dev)
3052 struct bonding *bond = netdev_priv(bond_dev);
3053 struct list_head *iter;
3054 struct slave *slave;
3056 /* reset slave->backup and slave->inactive */
3057 read_lock(&bond->lock);
3058 if (bond_has_slaves(bond)) {
3059 read_lock(&bond->curr_slave_lock);
3060 bond_for_each_slave(bond, slave, iter) {
3061 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3062 && (slave != bond->curr_active_slave)) {
3063 bond_set_slave_inactive_flags(slave,
3064 BOND_SLAVE_NOTIFY_NOW);
3066 bond_set_slave_active_flags(slave,
3067 BOND_SLAVE_NOTIFY_NOW);
3070 read_unlock(&bond->curr_slave_lock);
3072 read_unlock(&bond->lock);
3074 bond_work_init_all(bond);
3076 if (bond_is_lb(bond)) {
3077 /* bond_alb_initialize must be called before the timer
3080 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3082 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3085 if (bond->params.miimon) /* link check interval, in milliseconds. */
3086 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3088 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3089 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3090 if (bond->params.arp_validate)
3091 bond->recv_probe = bond_arp_rcv;
3094 if (bond->params.mode == BOND_MODE_8023AD) {
3095 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3096 /* register to receive LACPDUs */
3097 bond->recv_probe = bond_3ad_lacpdu_recv;
3098 bond_3ad_initiate_agg_selection(bond, 1);
3104 static int bond_close(struct net_device *bond_dev)
3106 struct bonding *bond = netdev_priv(bond_dev);
3108 bond_work_cancel_all(bond);
3109 bond->send_peer_notif = 0;
3110 if (bond_is_lb(bond))
3111 bond_alb_deinitialize(bond);
3112 bond->recv_probe = NULL;
3117 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3118 struct rtnl_link_stats64 *stats)
3120 struct bonding *bond = netdev_priv(bond_dev);
3121 struct rtnl_link_stats64 temp;
3122 struct list_head *iter;
3123 struct slave *slave;
3125 memset(stats, 0, sizeof(*stats));
3127 read_lock_bh(&bond->lock);
3128 bond_for_each_slave(bond, slave, iter) {
3129 const struct rtnl_link_stats64 *sstats =
3130 dev_get_stats(slave->dev, &temp);
3132 stats->rx_packets += sstats->rx_packets;
3133 stats->rx_bytes += sstats->rx_bytes;
3134 stats->rx_errors += sstats->rx_errors;
3135 stats->rx_dropped += sstats->rx_dropped;
3137 stats->tx_packets += sstats->tx_packets;
3138 stats->tx_bytes += sstats->tx_bytes;
3139 stats->tx_errors += sstats->tx_errors;
3140 stats->tx_dropped += sstats->tx_dropped;
3142 stats->multicast += sstats->multicast;
3143 stats->collisions += sstats->collisions;
3145 stats->rx_length_errors += sstats->rx_length_errors;
3146 stats->rx_over_errors += sstats->rx_over_errors;
3147 stats->rx_crc_errors += sstats->rx_crc_errors;
3148 stats->rx_frame_errors += sstats->rx_frame_errors;
3149 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3150 stats->rx_missed_errors += sstats->rx_missed_errors;
3152 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3153 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3154 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3155 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3156 stats->tx_window_errors += sstats->tx_window_errors;
3158 read_unlock_bh(&bond->lock);
3163 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3165 struct bonding *bond = netdev_priv(bond_dev);
3166 struct net_device *slave_dev = NULL;
3167 struct ifbond k_binfo;
3168 struct ifbond __user *u_binfo = NULL;
3169 struct ifslave k_sinfo;
3170 struct ifslave __user *u_sinfo = NULL;
3171 struct mii_ioctl_data *mii = NULL;
3172 struct bond_opt_value newval;
3176 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3188 * We do this again just in case we were called by SIOCGMIIREG
3189 * instead of SIOCGMIIPHY.
3196 if (mii->reg_num == 1) {
3198 read_lock(&bond->lock);
3199 read_lock(&bond->curr_slave_lock);
3200 if (netif_carrier_ok(bond->dev))
3201 mii->val_out = BMSR_LSTATUS;
3203 read_unlock(&bond->curr_slave_lock);
3204 read_unlock(&bond->lock);
3208 case BOND_INFO_QUERY_OLD:
3209 case SIOCBONDINFOQUERY:
3210 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3212 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3215 res = bond_info_query(bond_dev, &k_binfo);
3217 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3221 case BOND_SLAVE_INFO_QUERY_OLD:
3222 case SIOCBONDSLAVEINFOQUERY:
3223 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3225 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3228 res = bond_slave_info_query(bond_dev, &k_sinfo);
3230 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3239 net = dev_net(bond_dev);
3241 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3244 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3246 pr_debug("slave_dev=%p:\n", slave_dev);
3251 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3253 case BOND_ENSLAVE_OLD:
3254 case SIOCBONDENSLAVE:
3255 res = bond_enslave(bond_dev, slave_dev);
3257 case BOND_RELEASE_OLD:
3258 case SIOCBONDRELEASE:
3259 res = bond_release(bond_dev, slave_dev);
3261 case BOND_SETHWADDR_OLD:
3262 case SIOCBONDSETHWADDR:
3263 bond_set_dev_addr(bond_dev, slave_dev);
3266 case BOND_CHANGE_ACTIVE_OLD:
3267 case SIOCBONDCHANGEACTIVE:
3268 bond_opt_initstr(&newval, slave_dev->name);
3269 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3278 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3280 struct bonding *bond = netdev_priv(bond_dev);
3282 if (change & IFF_PROMISC)
3283 bond_set_promiscuity(bond,
3284 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3286 if (change & IFF_ALLMULTI)
3287 bond_set_allmulti(bond,
3288 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3291 static void bond_set_rx_mode(struct net_device *bond_dev)
3293 struct bonding *bond = netdev_priv(bond_dev);
3294 struct list_head *iter;
3295 struct slave *slave;
3299 if (USES_PRIMARY(bond->params.mode)) {
3300 slave = rcu_dereference(bond->curr_active_slave);
3302 dev_uc_sync(slave->dev, bond_dev);
3303 dev_mc_sync(slave->dev, bond_dev);
3306 bond_for_each_slave_rcu(bond, slave, iter) {
3307 dev_uc_sync_multiple(slave->dev, bond_dev);
3308 dev_mc_sync_multiple(slave->dev, bond_dev);
3314 static int bond_neigh_init(struct neighbour *n)
3316 struct bonding *bond = netdev_priv(n->dev);
3317 const struct net_device_ops *slave_ops;
3318 struct neigh_parms parms;
3319 struct slave *slave;
3322 slave = bond_first_slave(bond);
3325 slave_ops = slave->dev->netdev_ops;
3326 if (!slave_ops->ndo_neigh_setup)
3329 parms.neigh_setup = NULL;
3330 parms.neigh_cleanup = NULL;
3331 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3336 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3337 * after the last slave has been detached. Assumes that all slaves
3338 * utilize the same neigh_cleanup (true at this writing as only user
3341 n->parms->neigh_cleanup = parms.neigh_cleanup;
3343 if (!parms.neigh_setup)
3346 return parms.neigh_setup(n);
3350 * The bonding ndo_neigh_setup is called at init time beofre any
3351 * slave exists. So we must declare proxy setup function which will
3352 * be used at run time to resolve the actual slave neigh param setup.
3354 * It's also called by master devices (such as vlans) to setup their
3355 * underlying devices. In that case - do nothing, we're already set up from
3358 static int bond_neigh_setup(struct net_device *dev,
3359 struct neigh_parms *parms)
3361 /* modify only our neigh_parms */
3362 if (parms->dev == dev)
3363 parms->neigh_setup = bond_neigh_init;
3369 * Change the MTU of all of a master's slaves to match the master
3371 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3373 struct bonding *bond = netdev_priv(bond_dev);
3374 struct slave *slave, *rollback_slave;
3375 struct list_head *iter;
3378 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3379 (bond_dev ? bond_dev->name : "None"), new_mtu);
3381 /* Can't hold bond->lock with bh disabled here since
3382 * some base drivers panic. On the other hand we can't
3383 * hold bond->lock without bh disabled because we'll
3384 * deadlock. The only solution is to rely on the fact
3385 * that we're under rtnl_lock here, and the slaves
3386 * list won't change. This doesn't solve the problem
3387 * of setting the slave's MTU while it is
3388 * transmitting, but the assumption is that the base
3389 * driver can handle that.
3391 * TODO: figure out a way to safely iterate the slaves
3392 * list, but without holding a lock around the actual
3393 * call to the base driver.
3396 bond_for_each_slave(bond, slave, iter) {
3397 pr_debug("s %p c_m %p\n",
3399 slave->dev->netdev_ops->ndo_change_mtu);
3401 res = dev_set_mtu(slave->dev, new_mtu);
3404 /* If we failed to set the slave's mtu to the new value
3405 * we must abort the operation even in ACTIVE_BACKUP
3406 * mode, because if we allow the backup slaves to have
3407 * different mtu values than the active slave we'll
3408 * need to change their mtu when doing a failover. That
3409 * means changing their mtu from timer context, which
3410 * is probably not a good idea.
3412 pr_debug("err %d %s\n", res, slave->dev->name);
3417 bond_dev->mtu = new_mtu;
3422 /* unwind from head to the slave that failed */
3423 bond_for_each_slave(bond, rollback_slave, iter) {
3426 if (rollback_slave == slave)
3429 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3431 pr_debug("unwind err %d dev %s\n",
3432 tmp_res, rollback_slave->dev->name);
3442 * Note that many devices must be down to change the HW address, and
3443 * downing the master releases all slaves. We can make bonds full of
3444 * bonding devices to test this, however.
3446 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3448 struct bonding *bond = netdev_priv(bond_dev);
3449 struct slave *slave, *rollback_slave;
3450 struct sockaddr *sa = addr, tmp_sa;
3451 struct list_head *iter;
3454 if (bond->params.mode == BOND_MODE_ALB)
3455 return bond_alb_set_mac_address(bond_dev, addr);
3458 pr_debug("bond=%p, name=%s\n",
3459 bond, bond_dev ? bond_dev->name : "None");
3461 /* If fail_over_mac is enabled, do nothing and return success.
3462 * Returning an error causes ifenslave to fail.
3464 if (bond->params.fail_over_mac &&
3465 bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3468 if (!is_valid_ether_addr(sa->sa_data))
3469 return -EADDRNOTAVAIL;
3471 /* Can't hold bond->lock with bh disabled here since
3472 * some base drivers panic. On the other hand we can't
3473 * hold bond->lock without bh disabled because we'll
3474 * deadlock. The only solution is to rely on the fact
3475 * that we're under rtnl_lock here, and the slaves
3476 * list won't change. This doesn't solve the problem
3477 * of setting the slave's hw address while it is
3478 * transmitting, but the assumption is that the base
3479 * driver can handle that.
3481 * TODO: figure out a way to safely iterate the slaves
3482 * list, but without holding a lock around the actual
3483 * call to the base driver.
3486 bond_for_each_slave(bond, slave, iter) {
3487 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3488 pr_debug("slave %p %s\n", slave, slave->dev->name);
3490 if (slave_ops->ndo_set_mac_address == NULL) {
3492 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3496 res = dev_set_mac_address(slave->dev, addr);
3498 /* TODO: consider downing the slave
3500 * User should expect communications
3501 * breakage anyway until ARP finish
3504 pr_debug("err %d %s\n", res, slave->dev->name);
3510 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3514 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3515 tmp_sa.sa_family = bond_dev->type;
3517 /* unwind from head to the slave that failed */
3518 bond_for_each_slave(bond, rollback_slave, iter) {
3521 if (rollback_slave == slave)
3524 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3526 pr_debug("unwind err %d dev %s\n",
3527 tmp_res, rollback_slave->dev->name);
3535 * bond_xmit_slave_id - transmit skb through slave with slave_id
3536 * @bond: bonding device that is transmitting
3537 * @skb: buffer to transmit
3538 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3540 * This function tries to transmit through slave with slave_id but in case
3541 * it fails, it tries to find the first available slave for transmission.
3542 * The skb is consumed in all cases, thus the function is void.
3544 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3546 struct list_head *iter;
3547 struct slave *slave;
3550 /* Here we start from the slave with slave_id */
3551 bond_for_each_slave_rcu(bond, slave, iter) {
3553 if (slave_can_tx(slave)) {
3554 bond_dev_queue_xmit(bond, skb, slave->dev);
3560 /* Here we start from the first slave up to slave_id */
3562 bond_for_each_slave_rcu(bond, slave, iter) {
3565 if (slave_can_tx(slave)) {
3566 bond_dev_queue_xmit(bond, skb, slave->dev);
3570 /* no slave that can tx has been found */
3575 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3576 * @bond: bonding device to use
3578 * Based on the value of the bonding device's packets_per_slave parameter
3579 * this function generates a slave id, which is usually used as the next
3580 * slave to transmit through.
3582 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3585 struct reciprocal_value reciprocal_packets_per_slave;
3586 int packets_per_slave = bond->params.packets_per_slave;
3588 switch (packets_per_slave) {
3590 slave_id = prandom_u32();
3593 slave_id = bond->rr_tx_counter;
3596 reciprocal_packets_per_slave =
3597 bond->params.reciprocal_packets_per_slave;
3598 slave_id = reciprocal_divide(bond->rr_tx_counter,
3599 reciprocal_packets_per_slave);
3602 bond->rr_tx_counter++;
3607 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3609 struct bonding *bond = netdev_priv(bond_dev);
3610 struct iphdr *iph = ip_hdr(skb);
3611 struct slave *slave;
3614 /* Start with the curr_active_slave that joined the bond as the
3615 * default for sending IGMP traffic. For failover purposes one
3616 * needs to maintain some consistency for the interface that will
3617 * send the join/membership reports. The curr_active_slave found
3618 * will send all of this type of traffic.
3620 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3621 slave = rcu_dereference(bond->curr_active_slave);
3622 if (slave && slave_can_tx(slave))
3623 bond_dev_queue_xmit(bond, skb, slave->dev);
3625 bond_xmit_slave_id(bond, skb, 0);
3627 slave_id = bond_rr_gen_slave_id(bond);
3628 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3631 return NETDEV_TX_OK;
3635 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3636 * the bond has a usable interface.
3638 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3640 struct bonding *bond = netdev_priv(bond_dev);
3641 struct slave *slave;
3643 slave = rcu_dereference(bond->curr_active_slave);
3645 bond_dev_queue_xmit(bond, skb, slave->dev);
3649 return NETDEV_TX_OK;
3652 /* In bond_xmit_xor() , we determine the output device by using a pre-
3653 * determined xmit_hash_policy(), If the selected device is not enabled,
3654 * find the next active slave.
3656 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3658 struct bonding *bond = netdev_priv(bond_dev);
3660 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3662 return NETDEV_TX_OK;
3665 /* in broadcast mode, we send everything to all usable interfaces. */
3666 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3668 struct bonding *bond = netdev_priv(bond_dev);
3669 struct slave *slave = NULL;
3670 struct list_head *iter;
3672 bond_for_each_slave_rcu(bond, slave, iter) {
3673 if (bond_is_last_slave(bond, slave))
3675 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3676 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3679 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3683 /* bond_dev_queue_xmit always returns 0 */
3684 bond_dev_queue_xmit(bond, skb2, slave->dev);
3687 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3688 bond_dev_queue_xmit(bond, skb, slave->dev);
3692 return NETDEV_TX_OK;
3695 /*------------------------- Device initialization ---------------------------*/
3698 * Lookup the slave that corresponds to a qid
3700 static inline int bond_slave_override(struct bonding *bond,
3701 struct sk_buff *skb)
3703 struct slave *slave = NULL;
3704 struct list_head *iter;
3706 if (!skb->queue_mapping)
3709 /* Find out if any slaves have the same mapping as this skb. */
3710 bond_for_each_slave_rcu(bond, slave, iter) {
3711 if (slave->queue_id == skb->queue_mapping) {
3712 if (slave_can_tx(slave)) {
3713 bond_dev_queue_xmit(bond, skb, slave->dev);
3716 /* If the slave isn't UP, use default transmit policy. */
3725 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3726 void *accel_priv, select_queue_fallback_t fallback)
3729 * This helper function exists to help dev_pick_tx get the correct
3730 * destination queue. Using a helper function skips a call to
3731 * skb_tx_hash and will put the skbs in the queue we expect on their
3732 * way down to the bonding driver.
3734 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3737 * Save the original txq to restore before passing to the driver
3739 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3741 if (unlikely(txq >= dev->real_num_tx_queues)) {
3743 txq -= dev->real_num_tx_queues;
3744 } while (txq >= dev->real_num_tx_queues);
3749 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3751 struct bonding *bond = netdev_priv(dev);
3753 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3754 if (!bond_slave_override(bond, skb))
3755 return NETDEV_TX_OK;
3758 switch (bond->params.mode) {
3759 case BOND_MODE_ROUNDROBIN:
3760 return bond_xmit_roundrobin(skb, dev);
3761 case BOND_MODE_ACTIVEBACKUP:
3762 return bond_xmit_activebackup(skb, dev);
3764 return bond_xmit_xor(skb, dev);
3765 case BOND_MODE_BROADCAST:
3766 return bond_xmit_broadcast(skb, dev);
3767 case BOND_MODE_8023AD:
3768 return bond_3ad_xmit_xor(skb, dev);
3771 return bond_alb_xmit(skb, dev);
3773 /* Should never happen, mode already checked */
3774 pr_err("%s: Error: Unknown bonding mode %d\n",
3775 dev->name, bond->params.mode);
3778 return NETDEV_TX_OK;
3782 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3784 struct bonding *bond = netdev_priv(dev);
3785 netdev_tx_t ret = NETDEV_TX_OK;
3788 * If we risk deadlock from transmitting this in the
3789 * netpoll path, tell netpoll to queue the frame for later tx
3791 if (is_netpoll_tx_blocked(dev))
3792 return NETDEV_TX_BUSY;
3795 if (bond_has_slaves(bond))
3796 ret = __bond_start_xmit(skb, dev);
3804 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3805 struct ethtool_cmd *ecmd)
3807 struct bonding *bond = netdev_priv(bond_dev);
3808 unsigned long speed = 0;
3809 struct list_head *iter;
3810 struct slave *slave;
3812 ecmd->duplex = DUPLEX_UNKNOWN;
3813 ecmd->port = PORT_OTHER;
3815 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3816 * do not need to check mode. Though link speed might not represent
3817 * the true receive or transmit bandwidth (not all modes are symmetric)
3818 * this is an accurate maximum.
3820 read_lock(&bond->lock);
3821 bond_for_each_slave(bond, slave, iter) {
3822 if (SLAVE_IS_OK(slave)) {
3823 if (slave->speed != SPEED_UNKNOWN)
3824 speed += slave->speed;
3825 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3826 slave->duplex != DUPLEX_UNKNOWN)
3827 ecmd->duplex = slave->duplex;
3830 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3831 read_unlock(&bond->lock);
3836 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3837 struct ethtool_drvinfo *drvinfo)
3839 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3840 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3841 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3845 static const struct ethtool_ops bond_ethtool_ops = {
3846 .get_drvinfo = bond_ethtool_get_drvinfo,
3847 .get_settings = bond_ethtool_get_settings,
3848 .get_link = ethtool_op_get_link,
3851 static const struct net_device_ops bond_netdev_ops = {
3852 .ndo_init = bond_init,
3853 .ndo_uninit = bond_uninit,
3854 .ndo_open = bond_open,
3855 .ndo_stop = bond_close,
3856 .ndo_start_xmit = bond_start_xmit,
3857 .ndo_select_queue = bond_select_queue,
3858 .ndo_get_stats64 = bond_get_stats,
3859 .ndo_do_ioctl = bond_do_ioctl,
3860 .ndo_change_rx_flags = bond_change_rx_flags,
3861 .ndo_set_rx_mode = bond_set_rx_mode,
3862 .ndo_change_mtu = bond_change_mtu,
3863 .ndo_set_mac_address = bond_set_mac_address,
3864 .ndo_neigh_setup = bond_neigh_setup,
3865 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3866 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3867 #ifdef CONFIG_NET_POLL_CONTROLLER
3868 .ndo_netpoll_setup = bond_netpoll_setup,
3869 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3870 .ndo_poll_controller = bond_poll_controller,
3872 .ndo_add_slave = bond_enslave,
3873 .ndo_del_slave = bond_release,
3874 .ndo_fix_features = bond_fix_features,
3877 static const struct device_type bond_type = {
3881 static void bond_destructor(struct net_device *bond_dev)
3883 struct bonding *bond = netdev_priv(bond_dev);
3885 destroy_workqueue(bond->wq);
3886 free_netdev(bond_dev);
3889 void bond_setup(struct net_device *bond_dev)
3891 struct bonding *bond = netdev_priv(bond_dev);
3893 /* initialize rwlocks */
3894 rwlock_init(&bond->lock);
3895 rwlock_init(&bond->curr_slave_lock);
3896 bond->params = bonding_defaults;
3898 /* Initialize pointers */
3899 bond->dev = bond_dev;
3901 /* Initialize the device entry points */
3902 ether_setup(bond_dev);
3903 bond_dev->netdev_ops = &bond_netdev_ops;
3904 bond_dev->ethtool_ops = &bond_ethtool_ops;
3906 bond_dev->destructor = bond_destructor;
3908 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3910 /* Initialize the device options */
3911 bond_dev->tx_queue_len = 0;
3912 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3913 bond_dev->priv_flags |= IFF_BONDING;
3914 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3916 /* At first, we block adding VLANs. That's the only way to
3917 * prevent problems that occur when adding VLANs over an
3918 * empty bond. The block will be removed once non-challenged
3919 * slaves are enslaved.
3921 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3923 /* don't acquire bond device's netif_tx_lock when
3925 bond_dev->features |= NETIF_F_LLTX;
3927 /* By default, we declare the bond to be fully
3928 * VLAN hardware accelerated capable. Special
3929 * care is taken in the various xmit functions
3930 * when there are slaves that are not hw accel
3934 /* Don't allow bond devices to change network namespaces. */
3935 bond_dev->features |= NETIF_F_NETNS_LOCAL;
3937 bond_dev->hw_features = BOND_VLAN_FEATURES |
3938 NETIF_F_HW_VLAN_CTAG_TX |
3939 NETIF_F_HW_VLAN_CTAG_RX |
3940 NETIF_F_HW_VLAN_CTAG_FILTER;
3942 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3943 bond_dev->features |= bond_dev->hw_features;
3947 * Destroy a bonding device.
3948 * Must be under rtnl_lock when this function is called.
3950 static void bond_uninit(struct net_device *bond_dev)
3952 struct bonding *bond = netdev_priv(bond_dev);
3953 struct list_head *iter;
3954 struct slave *slave;
3956 bond_netpoll_cleanup(bond_dev);
3958 /* Release the bonded slaves */
3959 bond_for_each_slave(bond, slave, iter)
3960 __bond_release_one(bond_dev, slave->dev, true);
3961 pr_info("%s: released all slaves\n", bond_dev->name);
3963 list_del(&bond->bond_list);
3965 bond_debug_unregister(bond);
3968 /*------------------------- Module initialization ---------------------------*/
3970 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3974 for (i = 0; tbl[i].modename; i++)
3975 if (mode == tbl[i].mode)
3981 static int bond_parm_tbl_lookup_name(const char *modename,
3982 const struct bond_parm_tbl *tbl)
3986 for (i = 0; tbl[i].modename; i++)
3987 if (strcmp(modename, tbl[i].modename) == 0)
3994 * Convert string input module parms. Accept either the
3995 * number of the mode or its string name. A bit complicated because
3996 * some mode names are substrings of other names, and calls from sysfs
3997 * may have whitespace in the name (trailing newlines, for example).
3999 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4002 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
4004 for (p = (char *)buf; *p; p++)
4005 if (!(isdigit(*p) || isspace(*p)))
4008 if (*p && sscanf(buf, "%20s", modestr) != 0)
4009 return bond_parm_tbl_lookup_name(modestr, tbl);
4010 else if (sscanf(buf, "%d", &modeint) != 0)
4011 return bond_parm_tbl_lookup(modeint, tbl);
4016 static int bond_check_params(struct bond_params *params)
4018 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4019 struct bond_opt_value newval, *valptr;
4020 int arp_all_targets_value;
4023 * Convert string parameters.
4026 bond_opt_initstr(&newval, mode);
4027 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4029 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4032 bond_mode = valptr->value;
4035 if (xmit_hash_policy) {
4036 if ((bond_mode != BOND_MODE_XOR) &&
4037 (bond_mode != BOND_MODE_8023AD)) {
4038 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4039 bond_mode_name(bond_mode));
4041 bond_opt_initstr(&newval, xmit_hash_policy);
4042 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4045 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4049 xmit_hashtype = valptr->value;
4054 if (bond_mode != BOND_MODE_8023AD) {
4055 pr_info("lacp_rate param is irrelevant in mode %s\n",
4056 bond_mode_name(bond_mode));
4058 bond_opt_initstr(&newval, lacp_rate);
4059 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4062 pr_err("Error: Invalid lacp rate \"%s\"\n",
4066 lacp_fast = valptr->value;
4071 bond_opt_initstr(&newval, lacp_rate);
4072 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4075 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4078 params->ad_select = valptr->value;
4079 if (bond_mode != BOND_MODE_8023AD)
4080 pr_warning("ad_select param only affects 802.3ad mode\n");
4082 params->ad_select = BOND_AD_STABLE;
4085 if (max_bonds < 0) {
4086 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4087 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4088 max_bonds = BOND_DEFAULT_MAX_BONDS;
4092 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4098 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4103 if (downdelay < 0) {
4104 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4105 downdelay, INT_MAX);
4109 if ((use_carrier != 0) && (use_carrier != 1)) {
4110 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4115 if (num_peer_notif < 0 || num_peer_notif > 255) {
4116 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4121 /* reset values for 802.3ad/TLB/ALB */
4122 if (BOND_NO_USES_ARP(bond_mode)) {
4124 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4125 pr_warning("Forcing miimon to 100msec\n");
4126 miimon = BOND_DEFAULT_MIIMON;
4130 if (tx_queues < 1 || tx_queues > 255) {
4131 pr_warning("Warning: tx_queues (%d) should be between "
4132 "1 and 255, resetting to %d\n",
4133 tx_queues, BOND_DEFAULT_TX_QUEUES);
4134 tx_queues = BOND_DEFAULT_TX_QUEUES;
4137 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4138 pr_warning("Warning: all_slaves_active module parameter (%d), "
4139 "not of valid value (0/1), so it was set to "
4140 "0\n", all_slaves_active);
4141 all_slaves_active = 0;
4144 if (resend_igmp < 0 || resend_igmp > 255) {
4145 pr_warning("Warning: resend_igmp (%d) should be between "
4146 "0 and 255, resetting to %d\n",
4147 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4148 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4151 bond_opt_initval(&newval, packets_per_slave);
4152 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4153 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4154 packets_per_slave, USHRT_MAX);
4155 packets_per_slave = 1;
4158 if (bond_mode == BOND_MODE_ALB) {
4159 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",
4164 if (updelay || downdelay) {
4165 /* just warn the user the up/down delay will have
4166 * no effect since miimon is zero...
4168 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4169 updelay, downdelay);
4172 /* don't allow arp monitoring */
4174 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4175 miimon, arp_interval);
4179 if ((updelay % miimon) != 0) {
4180 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4182 (updelay / miimon) * miimon);
4187 if ((downdelay % miimon) != 0) {
4188 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4190 (downdelay / miimon) * miimon);
4193 downdelay /= miimon;
4196 if (arp_interval < 0) {
4197 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to 0\n",
4198 arp_interval, INT_MAX);
4202 for (arp_ip_count = 0, i = 0;
4203 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4204 /* not complete check, but should be good enough to
4207 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4208 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4209 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4213 if (bond_get_targets_ip(arp_target, ip) == -1)
4214 arp_target[arp_ip_count++] = ip;
4216 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4221 if (arp_interval && !arp_ip_count) {
4222 /* don't allow arping if no arp_ip_target given... */
4223 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4229 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4230 pr_err("arp_validate only supported in active-backup mode\n");
4233 if (!arp_interval) {
4234 pr_err("arp_validate requires arp_interval\n");
4238 bond_opt_initstr(&newval, arp_validate);
4239 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4242 pr_err("Error: invalid arp_validate \"%s\"\n",
4246 arp_validate_value = valptr->value;
4248 arp_validate_value = 0;
4251 arp_all_targets_value = 0;
4252 if (arp_all_targets) {
4253 bond_opt_initstr(&newval, arp_all_targets);
4254 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4257 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4259 arp_all_targets_value = 0;
4261 arp_all_targets_value = valptr->value;
4266 pr_info("MII link monitoring set to %d ms\n", miimon);
4267 } else if (arp_interval) {
4268 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4269 arp_validate_value);
4270 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4271 arp_interval, valptr->string, arp_ip_count);
4273 for (i = 0; i < arp_ip_count; i++)
4274 pr_info(" %s", arp_ip_target[i]);
4278 } else if (max_bonds) {
4279 /* miimon and arp_interval not set, we need one so things
4280 * work as expected, see bonding.txt for details
4282 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");
4285 if (primary && !USES_PRIMARY(bond_mode)) {
4286 /* currently, using a primary only makes sense
4287 * in active backup, TLB or ALB modes
4289 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4290 primary, bond_mode_name(bond_mode));
4294 if (primary && primary_reselect) {
4295 bond_opt_initstr(&newval, primary_reselect);
4296 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4299 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4303 primary_reselect_value = valptr->value;
4305 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4308 if (fail_over_mac) {
4309 bond_opt_initstr(&newval, fail_over_mac);
4310 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4313 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4317 fail_over_mac_value = valptr->value;
4318 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4319 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4321 fail_over_mac_value = BOND_FOM_NONE;
4324 if (lp_interval == 0) {
4325 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4326 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4327 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4330 /* fill params struct with the proper values */
4331 params->mode = bond_mode;
4332 params->xmit_policy = xmit_hashtype;
4333 params->miimon = miimon;
4334 params->num_peer_notif = num_peer_notif;
4335 params->arp_interval = arp_interval;
4336 params->arp_validate = arp_validate_value;
4337 params->arp_all_targets = arp_all_targets_value;
4338 params->updelay = updelay;
4339 params->downdelay = downdelay;
4340 params->use_carrier = use_carrier;
4341 params->lacp_fast = lacp_fast;
4342 params->primary[0] = 0;
4343 params->primary_reselect = primary_reselect_value;
4344 params->fail_over_mac = fail_over_mac_value;
4345 params->tx_queues = tx_queues;
4346 params->all_slaves_active = all_slaves_active;
4347 params->resend_igmp = resend_igmp;
4348 params->min_links = min_links;
4349 params->lp_interval = lp_interval;
4350 params->packets_per_slave = packets_per_slave;
4351 if (packets_per_slave > 0) {
4352 params->reciprocal_packets_per_slave =
4353 reciprocal_value(packets_per_slave);
4355 /* reciprocal_packets_per_slave is unused if
4356 * packets_per_slave is 0 or 1, just initialize it
4358 params->reciprocal_packets_per_slave =
4359 (struct reciprocal_value) { 0 };
4363 strncpy(params->primary, primary, IFNAMSIZ);
4364 params->primary[IFNAMSIZ - 1] = 0;
4367 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4372 static struct lock_class_key bonding_netdev_xmit_lock_key;
4373 static struct lock_class_key bonding_netdev_addr_lock_key;
4374 static struct lock_class_key bonding_tx_busylock_key;
4376 static void bond_set_lockdep_class_one(struct net_device *dev,
4377 struct netdev_queue *txq,
4380 lockdep_set_class(&txq->_xmit_lock,
4381 &bonding_netdev_xmit_lock_key);
4384 static void bond_set_lockdep_class(struct net_device *dev)
4386 lockdep_set_class(&dev->addr_list_lock,
4387 &bonding_netdev_addr_lock_key);
4388 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4389 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4393 * Called from registration process
4395 static int bond_init(struct net_device *bond_dev)
4397 struct bonding *bond = netdev_priv(bond_dev);
4398 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4399 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4401 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4404 * Initialize locks that may be required during
4405 * en/deslave operations. All of the bond_open work
4406 * (of which this is part) should really be moved to
4407 * a phase prior to dev_open
4409 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4410 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4412 bond->wq = create_singlethread_workqueue(bond_dev->name);
4416 bond_set_lockdep_class(bond_dev);
4418 list_add_tail(&bond->bond_list, &bn->dev_list);
4420 bond_prepare_sysfs_group(bond);
4422 bond_debug_register(bond);
4424 /* Ensure valid dev_addr */
4425 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4426 bond_dev->addr_assign_type == NET_ADDR_PERM)
4427 eth_hw_addr_random(bond_dev);
4432 unsigned int bond_get_num_tx_queues(void)
4437 /* Create a new bond based on the specified name and bonding parameters.
4438 * If name is NULL, obtain a suitable "bond%d" name for us.
4439 * Caller must NOT hold rtnl_lock; we need to release it here before we
4440 * set up our sysfs entries.
4442 int bond_create(struct net *net, const char *name)
4444 struct net_device *bond_dev;
4449 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4450 name ? name : "bond%d",
4451 bond_setup, tx_queues);
4453 pr_err("%s: eek! can't alloc netdev!\n", name);
4458 dev_net_set(bond_dev, net);
4459 bond_dev->rtnl_link_ops = &bond_link_ops;
4461 res = register_netdevice(bond_dev);
4463 netif_carrier_off(bond_dev);
4467 bond_destructor(bond_dev);
4471 static int __net_init bond_net_init(struct net *net)
4473 struct bond_net *bn = net_generic(net, bond_net_id);
4476 INIT_LIST_HEAD(&bn->dev_list);
4478 bond_create_proc_dir(bn);
4479 bond_create_sysfs(bn);
4484 static void __net_exit bond_net_exit(struct net *net)
4486 struct bond_net *bn = net_generic(net, bond_net_id);
4487 struct bonding *bond, *tmp_bond;
4490 bond_destroy_sysfs(bn);
4491 bond_destroy_proc_dir(bn);
4493 /* Kill off any bonds created after unregistering bond rtnl ops */
4495 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4496 unregister_netdevice_queue(bond->dev, &list);
4497 unregister_netdevice_many(&list);
4501 static struct pernet_operations bond_net_ops = {
4502 .init = bond_net_init,
4503 .exit = bond_net_exit,
4505 .size = sizeof(struct bond_net),
4508 static int __init bonding_init(void)
4513 pr_info("%s", bond_version);
4515 res = bond_check_params(&bonding_defaults);
4519 res = register_pernet_subsys(&bond_net_ops);
4523 res = bond_netlink_init();
4527 bond_create_debugfs();
4529 for (i = 0; i < max_bonds; i++) {
4530 res = bond_create(&init_net, NULL);
4535 register_netdevice_notifier(&bond_netdev_notifier);
4539 bond_destroy_debugfs();
4540 bond_netlink_fini();
4542 unregister_pernet_subsys(&bond_net_ops);
4547 static void __exit bonding_exit(void)
4549 unregister_netdevice_notifier(&bond_netdev_notifier);
4551 bond_destroy_debugfs();
4553 bond_netlink_fini();
4554 unregister_pernet_subsys(&bond_net_ops);
4556 #ifdef CONFIG_NET_POLL_CONTROLLER
4558 * Make sure we don't have an imbalance on our netpoll blocking
4560 WARN_ON(atomic_read(&netpoll_block_tx));
4564 module_init(bonding_init);
4565 module_exit(bonding_exit);
4566 MODULE_LICENSE("GPL");
4567 MODULE_VERSION(DRV_VERSION);
4568 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4569 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");