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>
85 /*---------------------------- Module parameters ----------------------------*/
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV 0
89 #define BOND_LINK_ARP_INTERV 0
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon = BOND_LINK_MON_INTERV;
97 static int use_carrier = 1;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval = BOND_LINK_ARP_INTERV;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
120 "failover event (alias of num_unsol_na)");
121 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
123 "failover event (alias of num_grat_arp)");
124 module_param(miimon, int, 0);
125 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
126 module_param(updelay, int, 0);
127 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
128 module_param(downdelay, int, 0);
129 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
131 module_param(use_carrier, int, 0);
132 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
133 "0 for off, 1 for on (default)");
134 module_param(mode, charp, 0);
135 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
136 "1 for active-backup, 2 for balance-xor, "
137 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
138 "6 for balance-alb");
139 module_param(primary, charp, 0);
140 MODULE_PARM_DESC(primary, "Primary network device to use");
141 module_param(primary_reselect, charp, 0);
142 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
144 "0 for always (default), "
145 "1 for only if speed of primary is "
147 "2 for only on active slave "
149 module_param(lacp_rate, charp, 0);
150 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
151 "0 for slow, 1 for fast");
152 module_param(ad_select, charp, 0);
153 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
154 "0 for stable (default), 1 for bandwidth, "
156 module_param(min_links, int, 0);
157 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
159 module_param(xmit_hash_policy, charp, 0);
160 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
161 "0 for layer 2 (default), 1 for layer 3+4, "
163 module_param(arp_interval, int, 0);
164 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
165 module_param_array(arp_ip_target, charp, NULL, 0);
166 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
167 module_param(arp_validate, charp, 0);
168 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
169 "0 for none (default), 1 for active, "
170 "2 for backup, 3 for all");
171 module_param(arp_all_targets, charp, 0);
172 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
173 module_param(fail_over_mac, charp, 0);
174 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
175 "the same MAC; 0 for none (default), "
176 "1 for active, 2 for follow");
177 module_param(all_slaves_active, int, 0);
178 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
179 "by setting active flag for all slaves; "
180 "0 for never (default), 1 for always.");
181 module_param(resend_igmp, int, 0);
182 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
185 /*----------------------------- Global variables ----------------------------*/
187 #ifdef CONFIG_NET_POLL_CONTROLLER
188 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
191 int bond_net_id __read_mostly;
193 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
194 static int arp_ip_count;
195 static int bond_mode = BOND_MODE_ROUNDROBIN;
196 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
197 static int lacp_fast;
199 const struct bond_parm_tbl bond_lacp_tbl[] = {
200 { "slow", AD_LACP_SLOW},
201 { "fast", AD_LACP_FAST},
205 const struct bond_parm_tbl bond_mode_tbl[] = {
206 { "balance-rr", BOND_MODE_ROUNDROBIN},
207 { "active-backup", BOND_MODE_ACTIVEBACKUP},
208 { "balance-xor", BOND_MODE_XOR},
209 { "broadcast", BOND_MODE_BROADCAST},
210 { "802.3ad", BOND_MODE_8023AD},
211 { "balance-tlb", BOND_MODE_TLB},
212 { "balance-alb", BOND_MODE_ALB},
216 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
217 { "layer2", BOND_XMIT_POLICY_LAYER2},
218 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
219 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
223 const struct bond_parm_tbl arp_all_targets_tbl[] = {
224 { "any", BOND_ARP_TARGETS_ANY},
225 { "all", BOND_ARP_TARGETS_ALL},
229 const struct bond_parm_tbl arp_validate_tbl[] = {
230 { "none", BOND_ARP_VALIDATE_NONE},
231 { "active", BOND_ARP_VALIDATE_ACTIVE},
232 { "backup", BOND_ARP_VALIDATE_BACKUP},
233 { "all", BOND_ARP_VALIDATE_ALL},
237 const struct bond_parm_tbl fail_over_mac_tbl[] = {
238 { "none", BOND_FOM_NONE},
239 { "active", BOND_FOM_ACTIVE},
240 { "follow", BOND_FOM_FOLLOW},
244 const struct bond_parm_tbl pri_reselect_tbl[] = {
245 { "always", BOND_PRI_RESELECT_ALWAYS},
246 { "better", BOND_PRI_RESELECT_BETTER},
247 { "failure", BOND_PRI_RESELECT_FAILURE},
251 struct bond_parm_tbl ad_select_tbl[] = {
252 { "stable", BOND_AD_STABLE},
253 { "bandwidth", BOND_AD_BANDWIDTH},
254 { "count", BOND_AD_COUNT},
258 /*-------------------------- Forward declarations ---------------------------*/
260 static int bond_init(struct net_device *bond_dev);
261 static void bond_uninit(struct net_device *bond_dev);
263 /*---------------------------- General routines -----------------------------*/
265 const char *bond_mode_name(int mode)
267 static const char *names[] = {
268 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
269 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
270 [BOND_MODE_XOR] = "load balancing (xor)",
271 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
272 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
273 [BOND_MODE_TLB] = "transmit load balancing",
274 [BOND_MODE_ALB] = "adaptive load balancing",
277 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
283 /*---------------------------------- VLAN -----------------------------------*/
286 * bond_dev_queue_xmit - Prepare skb for xmit.
288 * @bond: bond device that got this skb for tx.
289 * @skb: hw accel VLAN tagged skb to transmit
290 * @slave_dev: slave that is supposed to xmit this skbuff
292 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
293 struct net_device *slave_dev)
295 skb->dev = slave_dev;
297 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
298 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
299 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
301 if (unlikely(netpoll_tx_running(bond->dev)))
302 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
310 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
311 * We don't protect the slave list iteration with a lock because:
312 * a. This operation is performed in IOCTL context,
313 * b. The operation is protected by the RTNL semaphore in the 8021q code,
314 * c. Holding a lock with BH disabled while directly calling a base driver
315 * entry point is generally a BAD idea.
317 * The design of synchronization/protection for this operation in the 8021q
318 * module is good for one or more VLAN devices over a single physical device
319 * and cannot be extended for a teaming solution like bonding, so there is a
320 * potential race condition here where a net device from the vlan group might
321 * be referenced (either by a base driver or the 8021q code) while it is being
322 * removed from the system. However, it turns out we're not making matters
323 * worse, and if it works for regular VLAN usage it will work here too.
327 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
328 * @bond_dev: bonding net device that got called
329 * @vid: vlan id being added
331 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
332 __be16 proto, u16 vid)
334 struct bonding *bond = netdev_priv(bond_dev);
335 struct slave *slave, *rollback_slave;
336 struct list_head *iter;
339 bond_for_each_slave(bond, slave, iter) {
340 res = vlan_vid_add(slave->dev, proto, vid);
348 /* unwind to the slave that failed */
349 bond_for_each_slave(bond, rollback_slave, iter) {
350 if (rollback_slave == slave)
353 vlan_vid_del(rollback_slave->dev, proto, vid);
360 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
361 * @bond_dev: bonding net device that got called
362 * @vid: vlan id being removed
364 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
365 __be16 proto, u16 vid)
367 struct bonding *bond = netdev_priv(bond_dev);
368 struct list_head *iter;
371 bond_for_each_slave(bond, slave, iter)
372 vlan_vid_del(slave->dev, proto, vid);
374 if (bond_is_lb(bond))
375 bond_alb_clear_vlan(bond, vid);
380 /*------------------------------- Link status -------------------------------*/
383 * Set the carrier state for the master according to the state of its
384 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
385 * do special 802.3ad magic.
387 * Returns zero if carrier state does not change, nonzero if it does.
389 static int bond_set_carrier(struct bonding *bond)
391 struct list_head *iter;
394 if (list_empty(&bond->slave_list))
397 if (bond->params.mode == BOND_MODE_8023AD)
398 return bond_3ad_set_carrier(bond);
400 bond_for_each_slave(bond, slave, iter) {
401 if (slave->link == BOND_LINK_UP) {
402 if (!netif_carrier_ok(bond->dev)) {
403 netif_carrier_on(bond->dev);
411 if (netif_carrier_ok(bond->dev)) {
412 netif_carrier_off(bond->dev);
419 * Get link speed and duplex from the slave's base driver
420 * using ethtool. If for some reason the call fails or the
421 * values are invalid, set speed and duplex to -1,
424 static void bond_update_speed_duplex(struct slave *slave)
426 struct net_device *slave_dev = slave->dev;
427 struct ethtool_cmd ecmd;
431 slave->speed = SPEED_UNKNOWN;
432 slave->duplex = DUPLEX_UNKNOWN;
434 res = __ethtool_get_settings(slave_dev, &ecmd);
438 slave_speed = ethtool_cmd_speed(&ecmd);
439 if (slave_speed == 0 || slave_speed == ((__u32) -1))
442 switch (ecmd.duplex) {
450 slave->speed = slave_speed;
451 slave->duplex = ecmd.duplex;
457 * if <dev> supports MII link status reporting, check its link status.
459 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
460 * depending upon the setting of the use_carrier parameter.
462 * Return either BMSR_LSTATUS, meaning that the link is up (or we
463 * can't tell and just pretend it is), or 0, meaning that the link is
466 * If reporting is non-zero, instead of faking link up, return -1 if
467 * both ETHTOOL and MII ioctls fail (meaning the device does not
468 * support them). If use_carrier is set, return whatever it says.
469 * It'd be nice if there was a good way to tell if a driver supports
470 * netif_carrier, but there really isn't.
472 static int bond_check_dev_link(struct bonding *bond,
473 struct net_device *slave_dev, int reporting)
475 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
476 int (*ioctl)(struct net_device *, struct ifreq *, int);
478 struct mii_ioctl_data *mii;
480 if (!reporting && !netif_running(slave_dev))
483 if (bond->params.use_carrier)
484 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
486 /* Try to get link status using Ethtool first. */
487 if (slave_dev->ethtool_ops->get_link)
488 return slave_dev->ethtool_ops->get_link(slave_dev) ?
491 /* Ethtool can't be used, fallback to MII ioctls. */
492 ioctl = slave_ops->ndo_do_ioctl;
494 /* TODO: set pointer to correct ioctl on a per team member */
495 /* bases to make this more efficient. that is, once */
496 /* we determine the correct ioctl, we will always */
497 /* call it and not the others for that team */
501 * We cannot assume that SIOCGMIIPHY will also read a
502 * register; not all network drivers (e.g., e100)
506 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
507 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
509 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
510 mii->reg_num = MII_BMSR;
511 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
512 return mii->val_out & BMSR_LSTATUS;
517 * If reporting, report that either there's no dev->do_ioctl,
518 * or both SIOCGMIIREG and get_link failed (meaning that we
519 * cannot report link status). If not reporting, pretend
522 return reporting ? -1 : BMSR_LSTATUS;
525 /*----------------------------- Multicast list ------------------------------*/
528 * Push the promiscuity flag down to appropriate slaves
530 static int bond_set_promiscuity(struct bonding *bond, int inc)
532 struct list_head *iter;
535 if (USES_PRIMARY(bond->params.mode)) {
536 /* write lock already acquired */
537 if (bond->curr_active_slave) {
538 err = dev_set_promiscuity(bond->curr_active_slave->dev,
544 bond_for_each_slave(bond, slave, iter) {
545 err = dev_set_promiscuity(slave->dev, inc);
554 * Push the allmulti flag down to all slaves
556 static int bond_set_allmulti(struct bonding *bond, int inc)
558 struct list_head *iter;
561 if (USES_PRIMARY(bond->params.mode)) {
562 /* write lock already acquired */
563 if (bond->curr_active_slave) {
564 err = dev_set_allmulti(bond->curr_active_slave->dev,
570 bond_for_each_slave(bond, slave, iter) {
571 err = dev_set_allmulti(slave->dev, inc);
580 * Retrieve the list of registered multicast addresses for the bonding
581 * device and retransmit an IGMP JOIN request to the current active
584 static void bond_resend_igmp_join_requests(struct bonding *bond)
586 if (!rtnl_trylock()) {
587 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
590 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
593 /* We use curr_slave_lock to protect against concurrent access to
594 * igmp_retrans from multiple running instances of this function and
595 * bond_change_active_slave
597 write_lock_bh(&bond->curr_slave_lock);
598 if (bond->igmp_retrans > 1) {
599 bond->igmp_retrans--;
600 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
602 write_unlock_bh(&bond->curr_slave_lock);
605 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
607 struct bonding *bond = container_of(work, struct bonding,
610 bond_resend_igmp_join_requests(bond);
613 /* Flush bond's hardware addresses from slave
615 static void bond_hw_addr_flush(struct net_device *bond_dev,
616 struct net_device *slave_dev)
618 struct bonding *bond = netdev_priv(bond_dev);
620 dev_uc_unsync(slave_dev, bond_dev);
621 dev_mc_unsync(slave_dev, bond_dev);
623 if (bond->params.mode == BOND_MODE_8023AD) {
624 /* del lacpdu mc addr from mc list */
625 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
627 dev_mc_del(slave_dev, lacpdu_multicast);
631 /*--------------------------- Active slave change ---------------------------*/
633 /* Update the hardware address list and promisc/allmulti for the new and
634 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
635 * slaves up date at all times; only the USES_PRIMARY modes need to call
636 * this function to swap these settings during a failover.
638 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
639 struct slave *old_active)
644 if (bond->dev->flags & IFF_PROMISC)
645 dev_set_promiscuity(old_active->dev, -1);
647 if (bond->dev->flags & IFF_ALLMULTI)
648 dev_set_allmulti(old_active->dev, -1);
650 bond_hw_addr_flush(bond->dev, old_active->dev);
654 /* FIXME: Signal errors upstream. */
655 if (bond->dev->flags & IFF_PROMISC)
656 dev_set_promiscuity(new_active->dev, 1);
658 if (bond->dev->flags & IFF_ALLMULTI)
659 dev_set_allmulti(new_active->dev, 1);
661 netif_addr_lock_bh(bond->dev);
662 dev_uc_sync(new_active->dev, bond->dev);
663 dev_mc_sync(new_active->dev, bond->dev);
664 netif_addr_unlock_bh(bond->dev);
669 * bond_set_dev_addr - clone slave's address to bond
670 * @bond_dev: bond net device
671 * @slave_dev: slave net device
673 * Should be called with RTNL held.
675 static void bond_set_dev_addr(struct net_device *bond_dev,
676 struct net_device *slave_dev)
678 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
679 bond_dev, slave_dev, slave_dev->addr_len);
680 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
681 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
682 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
686 * bond_do_fail_over_mac
688 * Perform special MAC address swapping for fail_over_mac settings
690 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
692 static void bond_do_fail_over_mac(struct bonding *bond,
693 struct slave *new_active,
694 struct slave *old_active)
695 __releases(&bond->curr_slave_lock)
696 __releases(&bond->lock)
697 __acquires(&bond->lock)
698 __acquires(&bond->curr_slave_lock)
700 u8 tmp_mac[ETH_ALEN];
701 struct sockaddr saddr;
704 switch (bond->params.fail_over_mac) {
705 case BOND_FOM_ACTIVE:
707 write_unlock_bh(&bond->curr_slave_lock);
708 read_unlock(&bond->lock);
709 bond_set_dev_addr(bond->dev, new_active->dev);
710 read_lock(&bond->lock);
711 write_lock_bh(&bond->curr_slave_lock);
714 case BOND_FOM_FOLLOW:
716 * if new_active && old_active, swap them
717 * if just old_active, do nothing (going to no active slave)
718 * if just new_active, set new_active to bond's MAC
723 write_unlock_bh(&bond->curr_slave_lock);
724 read_unlock(&bond->lock);
727 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
728 memcpy(saddr.sa_data, old_active->dev->dev_addr,
730 saddr.sa_family = new_active->dev->type;
732 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
733 saddr.sa_family = bond->dev->type;
736 rv = dev_set_mac_address(new_active->dev, &saddr);
738 pr_err("%s: Error %d setting MAC of slave %s\n",
739 bond->dev->name, -rv, new_active->dev->name);
746 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
747 saddr.sa_family = old_active->dev->type;
749 rv = dev_set_mac_address(old_active->dev, &saddr);
751 pr_err("%s: Error %d setting MAC of slave %s\n",
752 bond->dev->name, -rv, new_active->dev->name);
754 read_lock(&bond->lock);
755 write_lock_bh(&bond->curr_slave_lock);
758 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
759 bond->dev->name, bond->params.fail_over_mac);
765 static bool bond_should_change_active(struct bonding *bond)
767 struct slave *prim = bond->primary_slave;
768 struct slave *curr = bond->curr_active_slave;
770 if (!prim || !curr || curr->link != BOND_LINK_UP)
772 if (bond->force_primary) {
773 bond->force_primary = false;
776 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
777 (prim->speed < curr->speed ||
778 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
780 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
786 * find_best_interface - select the best available slave to be the active one
787 * @bond: our bonding struct
789 static struct slave *bond_find_best_slave(struct bonding *bond)
791 struct slave *slave, *bestslave = NULL;
792 struct list_head *iter;
793 int mintime = bond->params.updelay;
795 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
796 bond_should_change_active(bond))
797 return bond->primary_slave;
799 bond_for_each_slave(bond, slave, iter) {
800 if (slave->link == BOND_LINK_UP)
802 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
803 slave->delay < mintime) {
804 mintime = slave->delay;
812 static bool bond_should_notify_peers(struct bonding *bond)
814 struct slave *slave = bond->curr_active_slave;
816 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
817 bond->dev->name, slave ? slave->dev->name : "NULL");
819 if (!slave || !bond->send_peer_notif ||
820 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
827 * change_active_interface - change the active slave into the specified one
828 * @bond: our bonding struct
829 * @new: the new slave to make the active one
831 * Set the new slave to the bond's settings and unset them on the old
833 * Setting include flags, mc-list, promiscuity, allmulti, etc.
835 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
836 * because it is apparently the best available slave we have, even though its
837 * updelay hasn't timed out yet.
839 * If new_active is not NULL, caller must hold bond->lock for read and
840 * curr_slave_lock for write_bh.
842 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
844 struct slave *old_active = bond->curr_active_slave;
846 if (old_active == new_active)
850 new_active->jiffies = jiffies;
852 if (new_active->link == BOND_LINK_BACK) {
853 if (USES_PRIMARY(bond->params.mode)) {
854 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
855 bond->dev->name, new_active->dev->name,
856 (bond->params.updelay - new_active->delay) * bond->params.miimon);
859 new_active->delay = 0;
860 new_active->link = BOND_LINK_UP;
862 if (bond->params.mode == BOND_MODE_8023AD)
863 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
865 if (bond_is_lb(bond))
866 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
868 if (USES_PRIMARY(bond->params.mode)) {
869 pr_info("%s: making interface %s the new active one.\n",
870 bond->dev->name, new_active->dev->name);
875 if (USES_PRIMARY(bond->params.mode))
876 bond_hw_addr_swap(bond, new_active, old_active);
878 if (bond_is_lb(bond)) {
879 bond_alb_handle_active_change(bond, new_active);
881 bond_set_slave_inactive_flags(old_active);
883 bond_set_slave_active_flags(new_active);
885 rcu_assign_pointer(bond->curr_active_slave, new_active);
888 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
890 bond_set_slave_inactive_flags(old_active);
893 bool should_notify_peers = false;
895 bond_set_slave_active_flags(new_active);
897 if (bond->params.fail_over_mac)
898 bond_do_fail_over_mac(bond, new_active,
901 if (netif_running(bond->dev)) {
902 bond->send_peer_notif =
903 bond->params.num_peer_notif;
904 should_notify_peers =
905 bond_should_notify_peers(bond);
908 write_unlock_bh(&bond->curr_slave_lock);
909 read_unlock(&bond->lock);
911 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
912 if (should_notify_peers)
913 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
916 read_lock(&bond->lock);
917 write_lock_bh(&bond->curr_slave_lock);
921 /* resend IGMP joins since active slave has changed or
922 * all were sent on curr_active_slave.
923 * resend only if bond is brought up with the affected
924 * bonding modes and the retransmission is enabled */
925 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
926 ((USES_PRIMARY(bond->params.mode) && new_active) ||
927 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
928 bond->igmp_retrans = bond->params.resend_igmp;
929 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
934 * bond_select_active_slave - select a new active slave, if needed
935 * @bond: our bonding struct
937 * This functions should be called when one of the following occurs:
938 * - The old curr_active_slave has been released or lost its link.
939 * - The primary_slave has got its link back.
940 * - A slave has got its link back and there's no old curr_active_slave.
942 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
944 void bond_select_active_slave(struct bonding *bond)
946 struct slave *best_slave;
949 best_slave = bond_find_best_slave(bond);
950 if (best_slave != bond->curr_active_slave) {
951 bond_change_active_slave(bond, best_slave);
952 rv = bond_set_carrier(bond);
956 if (netif_carrier_ok(bond->dev)) {
957 pr_info("%s: first active interface up!\n",
960 pr_info("%s: now running without any active interface !\n",
966 /*--------------------------- slave list handling ---------------------------*/
969 * This function attaches the slave to the end of list.
971 * bond->lock held for writing by caller.
973 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
975 list_add_tail_rcu(&new_slave->list, &bond->slave_list);
980 * This function detaches the slave from the list.
981 * WARNING: no check is made to verify if the slave effectively
983 * Nothing is freed on return, structures are just unchained.
984 * If any slave pointer in bond was pointing to <slave>,
985 * it should be changed by the calling function.
987 * bond->lock held for writing by caller.
989 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
991 list_del_rcu(&slave->list);
995 #ifdef CONFIG_NET_POLL_CONTROLLER
996 static inline int slave_enable_netpoll(struct slave *slave)
1001 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1006 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1015 static inline void slave_disable_netpoll(struct slave *slave)
1017 struct netpoll *np = slave->np;
1023 __netpoll_free_async(np);
1025 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1027 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1029 if (!slave_dev->netdev_ops->ndo_poll_controller)
1034 static void bond_poll_controller(struct net_device *bond_dev)
1038 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 struct bonding *bond = netdev_priv(bond_dev);
1041 struct list_head *iter;
1042 struct slave *slave;
1044 bond_for_each_slave(bond, slave, iter)
1045 if (IS_UP(slave->dev))
1046 slave_disable_netpoll(slave);
1049 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1051 struct bonding *bond = netdev_priv(dev);
1052 struct list_head *iter;
1053 struct slave *slave;
1056 bond_for_each_slave(bond, slave, iter) {
1057 err = slave_enable_netpoll(slave);
1059 bond_netpoll_cleanup(dev);
1066 static inline int slave_enable_netpoll(struct slave *slave)
1070 static inline void slave_disable_netpoll(struct slave *slave)
1073 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1078 /*---------------------------------- IOCTL ----------------------------------*/
1080 static netdev_features_t bond_fix_features(struct net_device *dev,
1081 netdev_features_t features)
1083 struct bonding *bond = netdev_priv(dev);
1084 struct list_head *iter;
1085 netdev_features_t mask;
1086 struct slave *slave;
1088 if (list_empty(&bond->slave_list)) {
1089 /* Disable adding VLANs to empty bond. But why? --mq */
1090 features |= NETIF_F_VLAN_CHALLENGED;
1095 features &= ~NETIF_F_ONE_FOR_ALL;
1096 features |= NETIF_F_ALL_FOR_ALL;
1098 bond_for_each_slave(bond, slave, iter) {
1099 features = netdev_increment_features(features,
1100 slave->dev->features,
1103 features = netdev_add_tso_features(features, mask);
1108 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1109 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1110 NETIF_F_HIGHDMA | NETIF_F_LRO)
1112 static void bond_compute_features(struct bonding *bond)
1114 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1115 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1116 struct net_device *bond_dev = bond->dev;
1117 struct list_head *iter;
1118 struct slave *slave;
1119 unsigned short max_hard_header_len = ETH_HLEN;
1120 unsigned int gso_max_size = GSO_MAX_SIZE;
1121 u16 gso_max_segs = GSO_MAX_SEGS;
1123 if (list_empty(&bond->slave_list))
1126 bond_for_each_slave(bond, slave, iter) {
1127 vlan_features = netdev_increment_features(vlan_features,
1128 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1130 dst_release_flag &= slave->dev->priv_flags;
1131 if (slave->dev->hard_header_len > max_hard_header_len)
1132 max_hard_header_len = slave->dev->hard_header_len;
1134 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1135 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1139 bond_dev->vlan_features = vlan_features;
1140 bond_dev->hard_header_len = max_hard_header_len;
1141 bond_dev->gso_max_segs = gso_max_segs;
1142 netif_set_gso_max_size(bond_dev, gso_max_size);
1144 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1145 bond_dev->priv_flags = flags | dst_release_flag;
1147 netdev_change_features(bond_dev);
1150 static void bond_setup_by_slave(struct net_device *bond_dev,
1151 struct net_device *slave_dev)
1153 bond_dev->header_ops = slave_dev->header_ops;
1155 bond_dev->type = slave_dev->type;
1156 bond_dev->hard_header_len = slave_dev->hard_header_len;
1157 bond_dev->addr_len = slave_dev->addr_len;
1159 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1160 slave_dev->addr_len);
1163 /* On bonding slaves other than the currently active slave, suppress
1164 * duplicates except for alb non-mcast/bcast.
1166 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1167 struct slave *slave,
1168 struct bonding *bond)
1170 if (bond_is_slave_inactive(slave)) {
1171 if (bond->params.mode == BOND_MODE_ALB &&
1172 skb->pkt_type != PACKET_BROADCAST &&
1173 skb->pkt_type != PACKET_MULTICAST)
1180 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1182 struct sk_buff *skb = *pskb;
1183 struct slave *slave;
1184 struct bonding *bond;
1185 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1187 int ret = RX_HANDLER_ANOTHER;
1189 skb = skb_share_check(skb, GFP_ATOMIC);
1191 return RX_HANDLER_CONSUMED;
1195 slave = bond_slave_get_rcu(skb->dev);
1198 if (bond->params.arp_interval)
1199 slave->dev->last_rx = jiffies;
1201 recv_probe = ACCESS_ONCE(bond->recv_probe);
1203 ret = recv_probe(skb, bond, slave);
1204 if (ret == RX_HANDLER_CONSUMED) {
1210 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1211 return RX_HANDLER_EXACT;
1214 skb->dev = bond->dev;
1216 if (bond->params.mode == BOND_MODE_ALB &&
1217 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1218 skb->pkt_type == PACKET_HOST) {
1220 if (unlikely(skb_cow_head(skb,
1221 skb->data - skb_mac_header(skb)))) {
1223 return RX_HANDLER_CONSUMED;
1225 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1231 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1232 struct net_device *slave_dev,
1233 struct slave *slave)
1237 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1240 slave_dev->flags |= IFF_SLAVE;
1241 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1245 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1246 struct net_device *slave_dev)
1248 netdev_upper_dev_unlink(slave_dev, bond_dev);
1249 slave_dev->flags &= ~IFF_SLAVE;
1250 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1253 /* enslave device <slave> to bond device <master> */
1254 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1256 struct bonding *bond = netdev_priv(bond_dev);
1257 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1258 struct slave *new_slave = NULL;
1259 struct sockaddr addr;
1263 if (!bond->params.use_carrier &&
1264 slave_dev->ethtool_ops->get_link == NULL &&
1265 slave_ops->ndo_do_ioctl == NULL) {
1266 pr_warning("%s: Warning: no link monitoring support for %s\n",
1267 bond_dev->name, slave_dev->name);
1270 /* already enslaved */
1271 if (slave_dev->flags & IFF_SLAVE) {
1272 pr_debug("Error, Device was already enslaved\n");
1276 /* vlan challenged mutual exclusion */
1277 /* no need to lock since we're protected by rtnl_lock */
1278 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1279 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1280 if (vlan_uses_dev(bond_dev)) {
1281 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1282 bond_dev->name, slave_dev->name, bond_dev->name);
1285 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1286 bond_dev->name, slave_dev->name,
1287 slave_dev->name, bond_dev->name);
1290 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1294 * Old ifenslave binaries are no longer supported. These can
1295 * be identified with moderate accuracy by the state of the slave:
1296 * the current ifenslave will set the interface down prior to
1297 * enslaving it; the old ifenslave will not.
1299 if ((slave_dev->flags & IFF_UP)) {
1300 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1303 goto err_undo_flags;
1306 /* set bonding device ether type by slave - bonding netdevices are
1307 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1308 * there is a need to override some of the type dependent attribs/funcs.
1310 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1311 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1313 if (list_empty(&bond->slave_list)) {
1314 if (bond_dev->type != slave_dev->type) {
1315 pr_debug("%s: change device type from %d to %d\n",
1317 bond_dev->type, slave_dev->type);
1319 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1321 res = notifier_to_errno(res);
1323 pr_err("%s: refused to change device type\n",
1326 goto err_undo_flags;
1329 /* Flush unicast and multicast addresses */
1330 dev_uc_flush(bond_dev);
1331 dev_mc_flush(bond_dev);
1333 if (slave_dev->type != ARPHRD_ETHER)
1334 bond_setup_by_slave(bond_dev, slave_dev);
1336 ether_setup(bond_dev);
1337 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1340 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1343 } else if (bond_dev->type != slave_dev->type) {
1344 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1346 slave_dev->type, bond_dev->type);
1348 goto err_undo_flags;
1351 if (slave_ops->ndo_set_mac_address == NULL) {
1352 if (list_empty(&bond->slave_list)) {
1353 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1355 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1356 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1357 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",
1360 goto err_undo_flags;
1364 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1366 /* If this is the first slave, then we need to set the master's hardware
1367 * address to be the same as the slave's. */
1368 if (list_empty(&bond->slave_list) &&
1369 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1370 bond_set_dev_addr(bond->dev, slave_dev);
1372 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1375 goto err_undo_flags;
1377 INIT_LIST_HEAD(&new_slave->list);
1379 * Set the new_slave's queue_id to be zero. Queue ID mapping
1380 * is set via sysfs or module option if desired.
1382 new_slave->queue_id = 0;
1384 /* Save slave's original mtu and then set it to match the bond */
1385 new_slave->original_mtu = slave_dev->mtu;
1386 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1388 pr_debug("Error %d calling dev_set_mtu\n", res);
1393 * Save slave's original ("permanent") mac address for modes
1394 * that need it, and for restoring it upon release, and then
1395 * set it to the master's address
1397 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1399 if (!bond->params.fail_over_mac) {
1401 * Set slave to master's mac address. The application already
1402 * set the master's mac address to that of the first slave
1404 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1405 addr.sa_family = slave_dev->type;
1406 res = dev_set_mac_address(slave_dev, &addr);
1408 pr_debug("Error %d calling set_mac_address\n", res);
1409 goto err_restore_mtu;
1413 /* open the slave since the application closed it */
1414 res = dev_open(slave_dev);
1416 pr_debug("Opening slave %s failed\n", slave_dev->name);
1417 goto err_restore_mac;
1420 new_slave->bond = bond;
1421 new_slave->dev = slave_dev;
1422 slave_dev->priv_flags |= IFF_BONDING;
1424 if (bond_is_lb(bond)) {
1425 /* bond_alb_init_slave() must be called before all other stages since
1426 * it might fail and we do not want to have to undo everything
1428 res = bond_alb_init_slave(bond, new_slave);
1433 /* If the mode USES_PRIMARY, then the following is handled by
1434 * bond_change_active_slave().
1436 if (!USES_PRIMARY(bond->params.mode)) {
1437 /* set promiscuity level to new slave */
1438 if (bond_dev->flags & IFF_PROMISC) {
1439 res = dev_set_promiscuity(slave_dev, 1);
1444 /* set allmulti level to new slave */
1445 if (bond_dev->flags & IFF_ALLMULTI) {
1446 res = dev_set_allmulti(slave_dev, 1);
1451 netif_addr_lock_bh(bond_dev);
1453 dev_mc_sync_multiple(slave_dev, bond_dev);
1454 dev_uc_sync_multiple(slave_dev, bond_dev);
1456 netif_addr_unlock_bh(bond_dev);
1459 if (bond->params.mode == BOND_MODE_8023AD) {
1460 /* add lacpdu mc addr to mc list */
1461 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1463 dev_mc_add(slave_dev, lacpdu_multicast);
1466 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1468 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1469 bond_dev->name, slave_dev->name);
1473 write_lock_bh(&bond->lock);
1475 bond_attach_slave(bond, new_slave);
1477 new_slave->delay = 0;
1478 new_slave->link_failure_count = 0;
1480 write_unlock_bh(&bond->lock);
1482 bond_compute_features(bond);
1484 bond_update_speed_duplex(new_slave);
1486 read_lock(&bond->lock);
1488 new_slave->last_arp_rx = jiffies -
1489 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1490 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1491 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1493 if (bond->params.miimon && !bond->params.use_carrier) {
1494 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1496 if ((link_reporting == -1) && !bond->params.arp_interval) {
1498 * miimon is set but a bonded network driver
1499 * does not support ETHTOOL/MII and
1500 * arp_interval is not set. Note: if
1501 * use_carrier is enabled, we will never go
1502 * here (because netif_carrier is always
1503 * supported); thus, we don't need to change
1504 * the messages for netif_carrier.
1506 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",
1507 bond_dev->name, slave_dev->name);
1508 } else if (link_reporting == -1) {
1509 /* unable get link status using mii/ethtool */
1510 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",
1511 bond_dev->name, slave_dev->name);
1515 /* check for initial state */
1516 if (bond->params.miimon) {
1517 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1518 if (bond->params.updelay) {
1519 new_slave->link = BOND_LINK_BACK;
1520 new_slave->delay = bond->params.updelay;
1522 new_slave->link = BOND_LINK_UP;
1525 new_slave->link = BOND_LINK_DOWN;
1527 } else if (bond->params.arp_interval) {
1528 new_slave->link = (netif_carrier_ok(slave_dev) ?
1529 BOND_LINK_UP : BOND_LINK_DOWN);
1531 new_slave->link = BOND_LINK_UP;
1534 if (new_slave->link != BOND_LINK_DOWN)
1535 new_slave->jiffies = jiffies;
1536 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1537 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1538 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1540 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1541 /* if there is a primary slave, remember it */
1542 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1543 bond->primary_slave = new_slave;
1544 bond->force_primary = true;
1548 write_lock_bh(&bond->curr_slave_lock);
1550 switch (bond->params.mode) {
1551 case BOND_MODE_ACTIVEBACKUP:
1552 bond_set_slave_inactive_flags(new_slave);
1553 bond_select_active_slave(bond);
1555 case BOND_MODE_8023AD:
1556 /* in 802.3ad mode, the internal mechanism
1557 * will activate the slaves in the selected
1560 bond_set_slave_inactive_flags(new_slave);
1561 /* if this is the first slave */
1562 if (bond_first_slave(bond) == new_slave) {
1563 SLAVE_AD_INFO(new_slave).id = 1;
1564 /* Initialize AD with the number of times that the AD timer is called in 1 second
1565 * can be called only after the mac address of the bond is set
1567 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1569 struct slave *prev_slave;
1571 prev_slave = bond_prev_slave(bond, new_slave);
1572 SLAVE_AD_INFO(new_slave).id =
1573 SLAVE_AD_INFO(prev_slave).id + 1;
1576 bond_3ad_bind_slave(new_slave);
1580 bond_set_active_slave(new_slave);
1581 bond_set_slave_inactive_flags(new_slave);
1582 bond_select_active_slave(bond);
1585 pr_debug("This slave is always active in trunk mode\n");
1587 /* always active in trunk mode */
1588 bond_set_active_slave(new_slave);
1590 /* In trunking mode there is little meaning to curr_active_slave
1591 * anyway (it holds no special properties of the bond device),
1592 * so we can change it without calling change_active_interface()
1594 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1595 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1598 } /* switch(bond_mode) */
1600 write_unlock_bh(&bond->curr_slave_lock);
1602 bond_set_carrier(bond);
1604 #ifdef CONFIG_NET_POLL_CONTROLLER
1605 slave_dev->npinfo = bond->dev->npinfo;
1606 if (slave_dev->npinfo) {
1607 if (slave_enable_netpoll(new_slave)) {
1608 read_unlock(&bond->lock);
1609 pr_info("Error, %s: master_dev is using netpoll, "
1610 "but new slave device does not support netpoll.\n",
1618 read_unlock(&bond->lock);
1620 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1624 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1627 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1628 goto err_dest_symlinks;
1631 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1633 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1634 goto err_unregister;
1638 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1639 bond_dev->name, slave_dev->name,
1640 bond_is_active_slave(new_slave) ? "n active" : " backup",
1641 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1643 /* enslave is successful */
1646 /* Undo stages on error */
1648 netdev_rx_handler_unregister(slave_dev);
1651 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1654 if (!USES_PRIMARY(bond->params.mode))
1655 bond_hw_addr_flush(bond_dev, slave_dev);
1657 vlan_vids_del_by_dev(slave_dev, bond_dev);
1658 write_lock_bh(&bond->lock);
1659 bond_detach_slave(bond, new_slave);
1660 if (bond->primary_slave == new_slave)
1661 bond->primary_slave = NULL;
1662 if (bond->curr_active_slave == new_slave) {
1663 bond_change_active_slave(bond, NULL);
1664 write_unlock_bh(&bond->lock);
1665 read_lock(&bond->lock);
1666 write_lock_bh(&bond->curr_slave_lock);
1667 bond_select_active_slave(bond);
1668 write_unlock_bh(&bond->curr_slave_lock);
1669 read_unlock(&bond->lock);
1671 write_unlock_bh(&bond->lock);
1673 slave_disable_netpoll(new_slave);
1676 slave_dev->priv_flags &= ~IFF_BONDING;
1677 dev_close(slave_dev);
1680 if (!bond->params.fail_over_mac) {
1681 /* XXX TODO - fom follow mode needs to change master's
1682 * MAC if this slave's MAC is in use by the bond, or at
1683 * least print a warning.
1685 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1686 addr.sa_family = slave_dev->type;
1687 dev_set_mac_address(slave_dev, &addr);
1691 dev_set_mtu(slave_dev, new_slave->original_mtu);
1697 bond_compute_features(bond);
1698 /* Enslave of first slave has failed and we need to fix master's mac */
1699 if (list_empty(&bond->slave_list) &&
1700 ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1701 eth_hw_addr_random(bond_dev);
1707 * Try to release the slave device <slave> from the bond device <master>
1708 * It is legal to access curr_active_slave without a lock because all the function
1709 * is write-locked. If "all" is true it means that the function is being called
1710 * while destroying a bond interface and all slaves are being released.
1712 * The rules for slave state should be:
1713 * for Active/Backup:
1714 * Active stays on all backups go down
1715 * for Bonded connections:
1716 * The first up interface should be left on and all others downed.
1718 static int __bond_release_one(struct net_device *bond_dev,
1719 struct net_device *slave_dev,
1722 struct bonding *bond = netdev_priv(bond_dev);
1723 struct slave *slave, *oldcurrent;
1724 struct sockaddr addr;
1725 netdev_features_t old_features = bond_dev->features;
1727 /* slave is not a slave or master is not master of this slave */
1728 if (!(slave_dev->flags & IFF_SLAVE) ||
1729 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1730 pr_err("%s: Error: cannot release %s.\n",
1731 bond_dev->name, slave_dev->name);
1736 write_lock_bh(&bond->lock);
1738 slave = bond_get_slave_by_dev(bond, slave_dev);
1740 /* not a slave of this bond */
1741 pr_info("%s: %s not enslaved\n",
1742 bond_dev->name, slave_dev->name);
1743 write_unlock_bh(&bond->lock);
1744 unblock_netpoll_tx();
1748 write_unlock_bh(&bond->lock);
1750 bond_upper_dev_unlink(bond_dev, slave_dev);
1751 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1752 * for this slave anymore.
1754 netdev_rx_handler_unregister(slave_dev);
1755 write_lock_bh(&bond->lock);
1757 /* Inform AD package of unbinding of slave. */
1758 if (bond->params.mode == BOND_MODE_8023AD) {
1759 /* must be called before the slave is
1760 * detached from the list
1762 bond_3ad_unbind_slave(slave);
1765 pr_info("%s: releasing %s interface %s\n",
1767 bond_is_active_slave(slave) ? "active" : "backup",
1770 oldcurrent = bond->curr_active_slave;
1772 bond->current_arp_slave = NULL;
1774 /* release the slave from its bond */
1775 bond_detach_slave(bond, slave);
1777 if (!all && !bond->params.fail_over_mac) {
1778 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1779 !list_empty(&bond->slave_list))
1780 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",
1781 bond_dev->name, slave_dev->name,
1783 bond_dev->name, slave_dev->name);
1786 if (bond->primary_slave == slave)
1787 bond->primary_slave = NULL;
1789 if (oldcurrent == slave)
1790 bond_change_active_slave(bond, NULL);
1792 if (bond_is_lb(bond)) {
1793 /* Must be called only after the slave has been
1794 * detached from the list and the curr_active_slave
1795 * has been cleared (if our_slave == old_current),
1796 * but before a new active slave is selected.
1798 write_unlock_bh(&bond->lock);
1799 bond_alb_deinit_slave(bond, slave);
1800 write_lock_bh(&bond->lock);
1804 rcu_assign_pointer(bond->curr_active_slave, NULL);
1805 } else if (oldcurrent == slave) {
1807 * Note that we hold RTNL over this sequence, so there
1808 * is no concern that another slave add/remove event
1811 write_unlock_bh(&bond->lock);
1812 read_lock(&bond->lock);
1813 write_lock_bh(&bond->curr_slave_lock);
1815 bond_select_active_slave(bond);
1817 write_unlock_bh(&bond->curr_slave_lock);
1818 read_unlock(&bond->lock);
1819 write_lock_bh(&bond->lock);
1822 if (list_empty(&bond->slave_list)) {
1823 bond_set_carrier(bond);
1824 eth_hw_addr_random(bond_dev);
1826 if (vlan_uses_dev(bond_dev)) {
1827 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1828 bond_dev->name, bond_dev->name);
1829 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1834 write_unlock_bh(&bond->lock);
1835 unblock_netpoll_tx();
1838 if (list_empty(&bond->slave_list)) {
1839 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1840 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1843 bond_compute_features(bond);
1844 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1845 (old_features & NETIF_F_VLAN_CHALLENGED))
1846 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1847 bond_dev->name, slave_dev->name, bond_dev->name);
1849 /* must do this from outside any spinlocks */
1850 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1852 vlan_vids_del_by_dev(slave_dev, bond_dev);
1854 /* If the mode USES_PRIMARY, then this cases was handled above by
1855 * bond_change_active_slave(..., NULL)
1857 if (!USES_PRIMARY(bond->params.mode)) {
1858 /* unset promiscuity level from slave */
1859 if (bond_dev->flags & IFF_PROMISC)
1860 dev_set_promiscuity(slave_dev, -1);
1862 /* unset allmulti level from slave */
1863 if (bond_dev->flags & IFF_ALLMULTI)
1864 dev_set_allmulti(slave_dev, -1);
1866 bond_hw_addr_flush(bond_dev, slave_dev);
1869 slave_disable_netpoll(slave);
1871 /* close slave before restoring its mac address */
1872 dev_close(slave_dev);
1874 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1875 /* restore original ("permanent") mac address */
1876 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1877 addr.sa_family = slave_dev->type;
1878 dev_set_mac_address(slave_dev, &addr);
1881 dev_set_mtu(slave_dev, slave->original_mtu);
1883 slave_dev->priv_flags &= ~IFF_BONDING;
1887 return 0; /* deletion OK */
1890 /* A wrapper used because of ndo_del_link */
1891 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1893 return __bond_release_one(bond_dev, slave_dev, false);
1897 * First release a slave and then destroy the bond if no more slaves are left.
1898 * Must be under rtnl_lock when this function is called.
1900 static int bond_release_and_destroy(struct net_device *bond_dev,
1901 struct net_device *slave_dev)
1903 struct bonding *bond = netdev_priv(bond_dev);
1906 ret = bond_release(bond_dev, slave_dev);
1907 if (ret == 0 && list_empty(&bond->slave_list)) {
1908 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1909 pr_info("%s: destroying bond %s.\n",
1910 bond_dev->name, bond_dev->name);
1911 unregister_netdevice(bond_dev);
1917 * This function changes the active slave to slave <slave_dev>.
1918 * It returns -EINVAL in the following cases.
1919 * - <slave_dev> is not found in the list.
1920 * - There is not active slave now.
1921 * - <slave_dev> is already active.
1922 * - The link state of <slave_dev> is not BOND_LINK_UP.
1923 * - <slave_dev> is not running.
1924 * In these cases, this function does nothing.
1925 * In the other cases, current_slave pointer is changed and 0 is returned.
1927 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1929 struct bonding *bond = netdev_priv(bond_dev);
1930 struct slave *old_active = NULL;
1931 struct slave *new_active = NULL;
1934 if (!USES_PRIMARY(bond->params.mode))
1937 /* Verify that bond_dev is indeed the master of slave_dev */
1938 if (!(slave_dev->flags & IFF_SLAVE) ||
1939 !netdev_has_upper_dev(slave_dev, bond_dev))
1942 read_lock(&bond->lock);
1944 old_active = bond->curr_active_slave;
1945 new_active = bond_get_slave_by_dev(bond, slave_dev);
1947 * Changing to the current active: do nothing; return success.
1949 if (new_active && new_active == old_active) {
1950 read_unlock(&bond->lock);
1956 new_active->link == BOND_LINK_UP &&
1957 IS_UP(new_active->dev)) {
1959 write_lock_bh(&bond->curr_slave_lock);
1960 bond_change_active_slave(bond, new_active);
1961 write_unlock_bh(&bond->curr_slave_lock);
1962 unblock_netpoll_tx();
1966 read_unlock(&bond->lock);
1971 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1973 struct bonding *bond = netdev_priv(bond_dev);
1975 info->bond_mode = bond->params.mode;
1976 info->miimon = bond->params.miimon;
1978 read_lock(&bond->lock);
1979 info->num_slaves = bond->slave_cnt;
1980 read_unlock(&bond->lock);
1985 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1987 struct bonding *bond = netdev_priv(bond_dev);
1988 struct list_head *iter;
1989 int i = 0, res = -ENODEV;
1990 struct slave *slave;
1992 read_lock(&bond->lock);
1993 bond_for_each_slave(bond, slave, iter) {
1994 if (i++ == (int)info->slave_id) {
1996 strcpy(info->slave_name, slave->dev->name);
1997 info->link = slave->link;
1998 info->state = bond_slave_state(slave);
1999 info->link_failure_count = slave->link_failure_count;
2003 read_unlock(&bond->lock);
2008 /*-------------------------------- Monitoring -------------------------------*/
2011 static int bond_miimon_inspect(struct bonding *bond)
2013 int link_state, commit = 0;
2014 struct list_head *iter;
2015 struct slave *slave;
2016 bool ignore_updelay;
2018 ignore_updelay = !bond->curr_active_slave ? true : false;
2020 bond_for_each_slave(bond, slave, iter) {
2021 slave->new_link = BOND_LINK_NOCHANGE;
2023 link_state = bond_check_dev_link(bond, slave->dev, 0);
2025 switch (slave->link) {
2030 slave->link = BOND_LINK_FAIL;
2031 slave->delay = bond->params.downdelay;
2033 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2035 (bond->params.mode ==
2036 BOND_MODE_ACTIVEBACKUP) ?
2037 (bond_is_active_slave(slave) ?
2038 "active " : "backup ") : "",
2040 bond->params.downdelay * bond->params.miimon);
2043 case BOND_LINK_FAIL:
2046 * recovered before downdelay expired
2048 slave->link = BOND_LINK_UP;
2049 slave->jiffies = jiffies;
2050 pr_info("%s: link status up again after %d ms for interface %s.\n",
2052 (bond->params.downdelay - slave->delay) *
2053 bond->params.miimon,
2058 if (slave->delay <= 0) {
2059 slave->new_link = BOND_LINK_DOWN;
2067 case BOND_LINK_DOWN:
2071 slave->link = BOND_LINK_BACK;
2072 slave->delay = bond->params.updelay;
2075 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2076 bond->dev->name, slave->dev->name,
2077 ignore_updelay ? 0 :
2078 bond->params.updelay *
2079 bond->params.miimon);
2082 case BOND_LINK_BACK:
2084 slave->link = BOND_LINK_DOWN;
2085 pr_info("%s: link status down again after %d ms for interface %s.\n",
2087 (bond->params.updelay - slave->delay) *
2088 bond->params.miimon,
2097 if (slave->delay <= 0) {
2098 slave->new_link = BOND_LINK_UP;
2100 ignore_updelay = false;
2112 static void bond_miimon_commit(struct bonding *bond)
2114 struct list_head *iter;
2115 struct slave *slave;
2117 bond_for_each_slave(bond, slave, iter) {
2118 switch (slave->new_link) {
2119 case BOND_LINK_NOCHANGE:
2123 slave->link = BOND_LINK_UP;
2124 slave->jiffies = jiffies;
2126 if (bond->params.mode == BOND_MODE_8023AD) {
2127 /* prevent it from being the active one */
2128 bond_set_backup_slave(slave);
2129 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2130 /* make it immediately active */
2131 bond_set_active_slave(slave);
2132 } else if (slave != bond->primary_slave) {
2133 /* prevent it from being the active one */
2134 bond_set_backup_slave(slave);
2137 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2138 bond->dev->name, slave->dev->name,
2139 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2140 slave->duplex ? "full" : "half");
2142 /* notify ad that the link status has changed */
2143 if (bond->params.mode == BOND_MODE_8023AD)
2144 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2146 if (bond_is_lb(bond))
2147 bond_alb_handle_link_change(bond, slave,
2150 if (!bond->curr_active_slave ||
2151 (slave == bond->primary_slave))
2156 case BOND_LINK_DOWN:
2157 if (slave->link_failure_count < UINT_MAX)
2158 slave->link_failure_count++;
2160 slave->link = BOND_LINK_DOWN;
2162 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2163 bond->params.mode == BOND_MODE_8023AD)
2164 bond_set_slave_inactive_flags(slave);
2166 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2167 bond->dev->name, slave->dev->name);
2169 if (bond->params.mode == BOND_MODE_8023AD)
2170 bond_3ad_handle_link_change(slave,
2173 if (bond_is_lb(bond))
2174 bond_alb_handle_link_change(bond, slave,
2177 if (slave == bond->curr_active_slave)
2183 pr_err("%s: invalid new link %d on slave %s\n",
2184 bond->dev->name, slave->new_link,
2186 slave->new_link = BOND_LINK_NOCHANGE;
2194 write_lock_bh(&bond->curr_slave_lock);
2195 bond_select_active_slave(bond);
2196 write_unlock_bh(&bond->curr_slave_lock);
2197 unblock_netpoll_tx();
2200 bond_set_carrier(bond);
2206 * Really a wrapper that splits the mii monitor into two phases: an
2207 * inspection, then (if inspection indicates something needs to be done)
2208 * an acquisition of appropriate locks followed by a commit phase to
2209 * implement whatever link state changes are indicated.
2211 void bond_mii_monitor(struct work_struct *work)
2213 struct bonding *bond = container_of(work, struct bonding,
2215 bool should_notify_peers = false;
2216 unsigned long delay;
2218 read_lock(&bond->lock);
2220 delay = msecs_to_jiffies(bond->params.miimon);
2222 if (list_empty(&bond->slave_list))
2225 should_notify_peers = bond_should_notify_peers(bond);
2227 if (bond_miimon_inspect(bond)) {
2228 read_unlock(&bond->lock);
2230 /* Race avoidance with bond_close cancel of workqueue */
2231 if (!rtnl_trylock()) {
2232 read_lock(&bond->lock);
2234 should_notify_peers = false;
2238 read_lock(&bond->lock);
2240 bond_miimon_commit(bond);
2242 read_unlock(&bond->lock);
2243 rtnl_unlock(); /* might sleep, hold no other locks */
2244 read_lock(&bond->lock);
2248 if (bond->params.miimon)
2249 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2251 read_unlock(&bond->lock);
2253 if (should_notify_peers) {
2254 if (!rtnl_trylock())
2256 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2261 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2263 struct net_device *upper;
2264 struct list_head *iter;
2267 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2271 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2272 if (ip == bond_confirm_addr(upper, 0, ip)) {
2283 * We go to the (large) trouble of VLAN tagging ARP frames because
2284 * switches in VLAN mode (especially if ports are configured as
2285 * "native" to a VLAN) might not pass non-tagged frames.
2287 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2289 struct sk_buff *skb;
2291 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2292 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2294 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2295 NULL, slave_dev->dev_addr, NULL);
2298 pr_err("ARP packet allocation failed\n");
2302 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2304 pr_err("failed to insert VLAN tag\n");
2312 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2314 struct net_device *upper, *vlan_upper;
2315 struct list_head *iter, *vlan_iter;
2317 __be32 *targets = bond->params.arp_targets, addr;
2320 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2321 pr_debug("basa: target %pI4\n", &targets[i]);
2323 /* Find out through which dev should the packet go */
2324 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2327 pr_debug("%s: no route to arp_ip_target %pI4\n",
2328 bond->dev->name, &targets[i]);
2334 /* bond device itself */
2335 if (rt->dst.dev == bond->dev)
2339 /* first we search only for vlan devices. for every vlan
2340 * found we verify its upper dev list, searching for the
2341 * rt->dst.dev. If found we save the tag of the vlan and
2342 * proceed to send the packet.
2346 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2348 if (!is_vlan_dev(vlan_upper))
2350 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2352 if (upper == rt->dst.dev) {
2353 vlan_id = vlan_dev_vlan_id(vlan_upper);
2360 /* if the device we're looking for is not on top of any of
2361 * our upper vlans, then just search for any dev that
2362 * matches, and in case it's a vlan - save the id
2364 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2365 if (upper == rt->dst.dev) {
2366 /* if it's a vlan - get its VID */
2367 if (is_vlan_dev(upper))
2368 vlan_id = vlan_dev_vlan_id(upper);
2376 /* Not our device - skip */
2377 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2378 bond->dev->name, &targets[i],
2379 rt->dst.dev ? rt->dst.dev->name : "NULL");
2385 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2387 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2392 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2396 if (!sip || !bond_has_this_ip(bond, tip)) {
2397 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2401 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2403 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2406 slave->last_arp_rx = jiffies;
2407 slave->target_last_arp_rx[i] = jiffies;
2410 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2411 struct slave *slave)
2413 struct arphdr *arp = (struct arphdr *)skb->data;
2414 unsigned char *arp_ptr;
2418 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2419 return RX_HANDLER_ANOTHER;
2421 read_lock(&bond->lock);
2423 if (!slave_do_arp_validate(bond, slave))
2426 alen = arp_hdr_len(bond->dev);
2428 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2429 bond->dev->name, skb->dev->name);
2431 if (alen > skb_headlen(skb)) {
2432 arp = kmalloc(alen, GFP_ATOMIC);
2435 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2439 if (arp->ar_hln != bond->dev->addr_len ||
2440 skb->pkt_type == PACKET_OTHERHOST ||
2441 skb->pkt_type == PACKET_LOOPBACK ||
2442 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2443 arp->ar_pro != htons(ETH_P_IP) ||
2447 arp_ptr = (unsigned char *)(arp + 1);
2448 arp_ptr += bond->dev->addr_len;
2449 memcpy(&sip, arp_ptr, 4);
2450 arp_ptr += 4 + bond->dev->addr_len;
2451 memcpy(&tip, arp_ptr, 4);
2453 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2454 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2455 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2459 * Backup slaves won't see the ARP reply, but do come through
2460 * here for each ARP probe (so we swap the sip/tip to validate
2461 * the probe). In a "redundant switch, common router" type of
2462 * configuration, the ARP probe will (hopefully) travel from
2463 * the active, through one switch, the router, then the other
2464 * switch before reaching the backup.
2466 * We 'trust' the arp requests if there is an active slave and
2467 * it received valid arp reply(s) after it became active. This
2468 * is done to avoid endless looping when we can't reach the
2469 * arp_ip_target and fool ourselves with our own arp requests.
2471 if (bond_is_active_slave(slave))
2472 bond_validate_arp(bond, slave, sip, tip);
2473 else if (bond->curr_active_slave &&
2474 time_after(slave_last_rx(bond, bond->curr_active_slave),
2475 bond->curr_active_slave->jiffies))
2476 bond_validate_arp(bond, slave, tip, sip);
2479 read_unlock(&bond->lock);
2480 if (arp != (struct arphdr *)skb->data)
2482 return RX_HANDLER_ANOTHER;
2485 /* function to verify if we're in the arp_interval timeslice, returns true if
2486 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2487 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2489 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2492 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2494 return time_in_range(jiffies,
2495 last_act - delta_in_ticks,
2496 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2500 * this function is called regularly to monitor each slave's link
2501 * ensuring that traffic is being sent and received when arp monitoring
2502 * is used in load-balancing mode. if the adapter has been dormant, then an
2503 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2504 * arp monitoring in active backup mode.
2506 void bond_loadbalance_arp_mon(struct work_struct *work)
2508 struct bonding *bond = container_of(work, struct bonding,
2510 struct slave *slave, *oldcurrent;
2511 struct list_head *iter;
2512 int do_failover = 0;
2514 read_lock(&bond->lock);
2516 if (list_empty(&bond->slave_list))
2519 oldcurrent = bond->curr_active_slave;
2520 /* see if any of the previous devices are up now (i.e. they have
2521 * xmt and rcv traffic). the curr_active_slave does not come into
2522 * the picture unless it is null. also, slave->jiffies is not needed
2523 * here because we send an arp on each slave and give a slave as
2524 * long as it needs to get the tx/rx within the delta.
2525 * TODO: what about up/down delay in arp mode? it wasn't here before
2528 bond_for_each_slave(bond, slave, iter) {
2529 unsigned long trans_start = dev_trans_start(slave->dev);
2531 if (slave->link != BOND_LINK_UP) {
2532 if (bond_time_in_interval(bond, trans_start, 1) &&
2533 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2535 slave->link = BOND_LINK_UP;
2536 bond_set_active_slave(slave);
2538 /* primary_slave has no meaning in round-robin
2539 * mode. the window of a slave being up and
2540 * curr_active_slave being null after enslaving
2544 pr_info("%s: link status definitely up for interface %s, ",
2549 pr_info("%s: interface %s is now up\n",
2555 /* slave->link == BOND_LINK_UP */
2557 /* not all switches will respond to an arp request
2558 * when the source ip is 0, so don't take the link down
2559 * if we don't know our ip yet
2561 if (!bond_time_in_interval(bond, trans_start, 2) ||
2562 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2564 slave->link = BOND_LINK_DOWN;
2565 bond_set_backup_slave(slave);
2567 if (slave->link_failure_count < UINT_MAX)
2568 slave->link_failure_count++;
2570 pr_info("%s: interface %s is now down.\n",
2574 if (slave == oldcurrent)
2579 /* note: if switch is in round-robin mode, all links
2580 * must tx arp to ensure all links rx an arp - otherwise
2581 * links may oscillate or not come up at all; if switch is
2582 * in something like xor mode, there is nothing we can
2583 * do - all replies will be rx'ed on same link causing slaves
2584 * to be unstable during low/no traffic periods
2586 if (IS_UP(slave->dev))
2587 bond_arp_send_all(bond, slave);
2592 write_lock_bh(&bond->curr_slave_lock);
2594 bond_select_active_slave(bond);
2596 write_unlock_bh(&bond->curr_slave_lock);
2597 unblock_netpoll_tx();
2601 if (bond->params.arp_interval)
2602 queue_delayed_work(bond->wq, &bond->arp_work,
2603 msecs_to_jiffies(bond->params.arp_interval));
2605 read_unlock(&bond->lock);
2609 * Called to inspect slaves for active-backup mode ARP monitor link state
2610 * changes. Sets new_link in slaves to specify what action should take
2611 * place for the slave. Returns 0 if no changes are found, >0 if changes
2612 * to link states must be committed.
2614 * Called with bond->lock held for read.
2616 static int bond_ab_arp_inspect(struct bonding *bond)
2618 unsigned long trans_start, last_rx;
2619 struct list_head *iter;
2620 struct slave *slave;
2623 bond_for_each_slave(bond, slave, iter) {
2624 slave->new_link = BOND_LINK_NOCHANGE;
2625 last_rx = slave_last_rx(bond, slave);
2627 if (slave->link != BOND_LINK_UP) {
2628 if (bond_time_in_interval(bond, last_rx, 1)) {
2629 slave->new_link = BOND_LINK_UP;
2636 * Give slaves 2*delta after being enslaved or made
2637 * active. This avoids bouncing, as the last receive
2638 * times need a full ARP monitor cycle to be updated.
2640 if (bond_time_in_interval(bond, slave->jiffies, 2))
2644 * Backup slave is down if:
2645 * - No current_arp_slave AND
2646 * - more than 3*delta since last receive AND
2647 * - the bond has an IP address
2649 * Note: a non-null current_arp_slave indicates
2650 * the curr_active_slave went down and we are
2651 * searching for a new one; under this condition
2652 * we only take the curr_active_slave down - this
2653 * gives each slave a chance to tx/rx traffic
2654 * before being taken out
2656 if (!bond_is_active_slave(slave) &&
2657 !bond->current_arp_slave &&
2658 !bond_time_in_interval(bond, last_rx, 3)) {
2659 slave->new_link = BOND_LINK_DOWN;
2664 * Active slave is down if:
2665 * - more than 2*delta since transmitting OR
2666 * - (more than 2*delta since receive AND
2667 * the bond has an IP address)
2669 trans_start = dev_trans_start(slave->dev);
2670 if (bond_is_active_slave(slave) &&
2671 (!bond_time_in_interval(bond, trans_start, 2) ||
2672 !bond_time_in_interval(bond, last_rx, 2))) {
2673 slave->new_link = BOND_LINK_DOWN;
2682 * Called to commit link state changes noted by inspection step of
2683 * active-backup mode ARP monitor.
2685 * Called with RTNL and bond->lock for read.
2687 static void bond_ab_arp_commit(struct bonding *bond)
2689 unsigned long trans_start;
2690 struct list_head *iter;
2691 struct slave *slave;
2693 bond_for_each_slave(bond, slave, iter) {
2694 switch (slave->new_link) {
2695 case BOND_LINK_NOCHANGE:
2699 trans_start = dev_trans_start(slave->dev);
2700 if (bond->curr_active_slave != slave ||
2701 (!bond->curr_active_slave &&
2702 bond_time_in_interval(bond, trans_start, 1))) {
2703 slave->link = BOND_LINK_UP;
2704 if (bond->current_arp_slave) {
2705 bond_set_slave_inactive_flags(
2706 bond->current_arp_slave);
2707 bond->current_arp_slave = NULL;
2710 pr_info("%s: link status definitely up for interface %s.\n",
2711 bond->dev->name, slave->dev->name);
2713 if (!bond->curr_active_slave ||
2714 (slave == bond->primary_slave))
2721 case BOND_LINK_DOWN:
2722 if (slave->link_failure_count < UINT_MAX)
2723 slave->link_failure_count++;
2725 slave->link = BOND_LINK_DOWN;
2726 bond_set_slave_inactive_flags(slave);
2728 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2729 bond->dev->name, slave->dev->name);
2731 if (slave == bond->curr_active_slave) {
2732 bond->current_arp_slave = NULL;
2739 pr_err("%s: impossible: new_link %d on slave %s\n",
2740 bond->dev->name, slave->new_link,
2748 write_lock_bh(&bond->curr_slave_lock);
2749 bond_select_active_slave(bond);
2750 write_unlock_bh(&bond->curr_slave_lock);
2751 unblock_netpoll_tx();
2754 bond_set_carrier(bond);
2758 * Send ARP probes for active-backup mode ARP monitor.
2760 * Called with bond->lock held for read.
2762 static void bond_ab_arp_probe(struct bonding *bond)
2764 struct slave *slave, *before = NULL, *new_slave = NULL;
2765 struct list_head *iter;
2768 read_lock(&bond->curr_slave_lock);
2770 if (bond->current_arp_slave && bond->curr_active_slave)
2771 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2772 bond->current_arp_slave->dev->name,
2773 bond->curr_active_slave->dev->name);
2775 if (bond->curr_active_slave) {
2776 bond_arp_send_all(bond, bond->curr_active_slave);
2777 read_unlock(&bond->curr_slave_lock);
2781 read_unlock(&bond->curr_slave_lock);
2783 /* if we don't have a curr_active_slave, search for the next available
2784 * backup slave from the current_arp_slave and make it the candidate
2785 * for becoming the curr_active_slave
2788 if (!bond->current_arp_slave) {
2789 bond->current_arp_slave = bond_first_slave(bond);
2790 if (!bond->current_arp_slave)
2794 bond_set_slave_inactive_flags(bond->current_arp_slave);
2796 bond_for_each_slave(bond, slave, iter) {
2797 if (!found && !before && IS_UP(slave->dev))
2800 if (found && !new_slave && IS_UP(slave->dev))
2802 /* if the link state is up at this point, we
2803 * mark it down - this can happen if we have
2804 * simultaneous link failures and
2805 * reselect_active_interface doesn't make this
2806 * one the current slave so it is still marked
2807 * up when it is actually down
2809 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2810 slave->link = BOND_LINK_DOWN;
2811 if (slave->link_failure_count < UINT_MAX)
2812 slave->link_failure_count++;
2814 bond_set_slave_inactive_flags(slave);
2816 pr_info("%s: backup interface %s is now down.\n",
2817 bond->dev->name, slave->dev->name);
2819 if (slave == bond->current_arp_slave)
2823 if (!new_slave && before)
2829 new_slave->link = BOND_LINK_BACK;
2830 bond_set_slave_active_flags(new_slave);
2831 bond_arp_send_all(bond, new_slave);
2832 new_slave->jiffies = jiffies;
2833 bond->current_arp_slave = new_slave;
2837 void bond_activebackup_arp_mon(struct work_struct *work)
2839 struct bonding *bond = container_of(work, struct bonding,
2841 bool should_notify_peers = false;
2844 read_lock(&bond->lock);
2846 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2848 if (list_empty(&bond->slave_list))
2851 should_notify_peers = bond_should_notify_peers(bond);
2853 if (bond_ab_arp_inspect(bond)) {
2854 read_unlock(&bond->lock);
2856 /* Race avoidance with bond_close flush of workqueue */
2857 if (!rtnl_trylock()) {
2858 read_lock(&bond->lock);
2860 should_notify_peers = false;
2864 read_lock(&bond->lock);
2866 bond_ab_arp_commit(bond);
2868 read_unlock(&bond->lock);
2870 read_lock(&bond->lock);
2873 bond_ab_arp_probe(bond);
2876 if (bond->params.arp_interval)
2877 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2879 read_unlock(&bond->lock);
2881 if (should_notify_peers) {
2882 if (!rtnl_trylock())
2884 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2889 /*-------------------------- netdev event handling --------------------------*/
2892 * Change device name
2894 static int bond_event_changename(struct bonding *bond)
2896 bond_remove_proc_entry(bond);
2897 bond_create_proc_entry(bond);
2899 bond_debug_reregister(bond);
2904 static int bond_master_netdev_event(unsigned long event,
2905 struct net_device *bond_dev)
2907 struct bonding *event_bond = netdev_priv(bond_dev);
2910 case NETDEV_CHANGENAME:
2911 return bond_event_changename(event_bond);
2912 case NETDEV_UNREGISTER:
2913 bond_remove_proc_entry(event_bond);
2915 case NETDEV_REGISTER:
2916 bond_create_proc_entry(event_bond);
2918 case NETDEV_NOTIFY_PEERS:
2919 if (event_bond->send_peer_notif)
2920 event_bond->send_peer_notif--;
2929 static int bond_slave_netdev_event(unsigned long event,
2930 struct net_device *slave_dev)
2932 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2933 struct bonding *bond;
2934 struct net_device *bond_dev;
2938 /* A netdev event can be generated while enslaving a device
2939 * before netdev_rx_handler_register is called in which case
2940 * slave will be NULL
2944 bond_dev = slave->bond->dev;
2948 case NETDEV_UNREGISTER:
2949 if (bond_dev->type != ARPHRD_ETHER)
2950 bond_release_and_destroy(bond_dev, slave_dev);
2952 bond_release(bond_dev, slave_dev);
2956 old_speed = slave->speed;
2957 old_duplex = slave->duplex;
2959 bond_update_speed_duplex(slave);
2961 if (bond->params.mode == BOND_MODE_8023AD) {
2962 if (old_speed != slave->speed)
2963 bond_3ad_adapter_speed_changed(slave);
2964 if (old_duplex != slave->duplex)
2965 bond_3ad_adapter_duplex_changed(slave);
2970 * ... Or is it this?
2973 case NETDEV_CHANGEMTU:
2975 * TODO: Should slaves be allowed to
2976 * independently alter their MTU? For
2977 * an active-backup bond, slaves need
2978 * not be the same type of device, so
2979 * MTUs may vary. For other modes,
2980 * slaves arguably should have the
2981 * same MTUs. To do this, we'd need to
2982 * take over the slave's change_mtu
2983 * function for the duration of their
2987 case NETDEV_CHANGENAME:
2989 * TODO: handle changing the primary's name
2992 case NETDEV_FEAT_CHANGE:
2993 bond_compute_features(bond);
2995 case NETDEV_RESEND_IGMP:
2996 /* Propagate to master device */
2997 call_netdevice_notifiers(event, slave->bond->dev);
3007 * bond_netdev_event: handle netdev notifier chain events.
3009 * This function receives events for the netdev chain. The caller (an
3010 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3011 * locks for us to safely manipulate the slave devices (RTNL lock,
3014 static int bond_netdev_event(struct notifier_block *this,
3015 unsigned long event, void *ptr)
3017 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3019 pr_debug("event_dev: %s, event: %lx\n",
3020 event_dev ? event_dev->name : "None",
3023 if (!(event_dev->priv_flags & IFF_BONDING))
3026 if (event_dev->flags & IFF_MASTER) {
3027 pr_debug("IFF_MASTER\n");
3028 return bond_master_netdev_event(event, event_dev);
3031 if (event_dev->flags & IFF_SLAVE) {
3032 pr_debug("IFF_SLAVE\n");
3033 return bond_slave_netdev_event(event, event_dev);
3039 static struct notifier_block bond_netdev_notifier = {
3040 .notifier_call = bond_netdev_event,
3043 /*---------------------------- Hashing Policies -----------------------------*/
3046 * Hash for the output device based upon layer 2 data
3048 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3050 struct ethhdr *data = (struct ethhdr *)skb->data;
3052 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3053 return (data->h_dest[5] ^ data->h_source[5]) % count;
3059 * Hash for the output device based upon layer 2 and layer 3 data. If
3060 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3062 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3064 const struct ethhdr *data;
3065 const struct iphdr *iph;
3066 const struct ipv6hdr *ipv6h;
3068 const __be32 *s, *d;
3070 if (skb->protocol == htons(ETH_P_IP) &&
3071 pskb_network_may_pull(skb, sizeof(*iph))) {
3073 data = (struct ethhdr *)skb->data;
3074 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3075 (data->h_dest[5] ^ data->h_source[5])) % count;
3076 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3077 pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3078 ipv6h = ipv6_hdr(skb);
3079 data = (struct ethhdr *)skb->data;
3080 s = &ipv6h->saddr.s6_addr32[0];
3081 d = &ipv6h->daddr.s6_addr32[0];
3082 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3083 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3084 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3087 return bond_xmit_hash_policy_l2(skb, count);
3091 * Hash for the output device based upon layer 3 and layer 4 data. If
3092 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3093 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3095 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3098 const struct iphdr *iph;
3099 const struct ipv6hdr *ipv6h;
3100 const __be32 *s, *d;
3101 const __be16 *l4 = NULL;
3103 int noff = skb_network_offset(skb);
3106 if (skb->protocol == htons(ETH_P_IP) &&
3107 pskb_may_pull(skb, noff + sizeof(*iph))) {
3109 poff = proto_ports_offset(iph->protocol);
3111 if (!ip_is_fragment(iph) && poff >= 0) {
3112 l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3115 layer4_xor = ntohs(l4[0] ^ l4[1]);
3117 return (layer4_xor ^
3118 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3119 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3120 pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3121 ipv6h = ipv6_hdr(skb);
3122 poff = proto_ports_offset(ipv6h->nexthdr);
3124 l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3127 layer4_xor = ntohs(l4[0] ^ l4[1]);
3129 s = &ipv6h->saddr.s6_addr32[0];
3130 d = &ipv6h->daddr.s6_addr32[0];
3131 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3132 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3134 return layer4_xor % count;
3137 return bond_xmit_hash_policy_l2(skb, count);
3140 /*-------------------------- Device entry points ----------------------------*/
3142 static void bond_work_init_all(struct bonding *bond)
3144 INIT_DELAYED_WORK(&bond->mcast_work,
3145 bond_resend_igmp_join_requests_delayed);
3146 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3147 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3148 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3149 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3151 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3152 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3155 static void bond_work_cancel_all(struct bonding *bond)
3157 cancel_delayed_work_sync(&bond->mii_work);
3158 cancel_delayed_work_sync(&bond->arp_work);
3159 cancel_delayed_work_sync(&bond->alb_work);
3160 cancel_delayed_work_sync(&bond->ad_work);
3161 cancel_delayed_work_sync(&bond->mcast_work);
3164 static int bond_open(struct net_device *bond_dev)
3166 struct bonding *bond = netdev_priv(bond_dev);
3167 struct list_head *iter;
3168 struct slave *slave;
3170 /* reset slave->backup and slave->inactive */
3171 read_lock(&bond->lock);
3172 if (!list_empty(&bond->slave_list)) {
3173 read_lock(&bond->curr_slave_lock);
3174 bond_for_each_slave(bond, slave, iter) {
3175 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3176 && (slave != bond->curr_active_slave)) {
3177 bond_set_slave_inactive_flags(slave);
3179 bond_set_slave_active_flags(slave);
3182 read_unlock(&bond->curr_slave_lock);
3184 read_unlock(&bond->lock);
3186 bond_work_init_all(bond);
3188 if (bond_is_lb(bond)) {
3189 /* bond_alb_initialize must be called before the timer
3192 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3194 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3197 if (bond->params.miimon) /* link check interval, in milliseconds. */
3198 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3200 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3201 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3202 if (bond->params.arp_validate)
3203 bond->recv_probe = bond_arp_rcv;
3206 if (bond->params.mode == BOND_MODE_8023AD) {
3207 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3208 /* register to receive LACPDUs */
3209 bond->recv_probe = bond_3ad_lacpdu_recv;
3210 bond_3ad_initiate_agg_selection(bond, 1);
3216 static int bond_close(struct net_device *bond_dev)
3218 struct bonding *bond = netdev_priv(bond_dev);
3220 bond_work_cancel_all(bond);
3221 bond->send_peer_notif = 0;
3222 if (bond_is_lb(bond))
3223 bond_alb_deinitialize(bond);
3224 bond->recv_probe = NULL;
3229 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3230 struct rtnl_link_stats64 *stats)
3232 struct bonding *bond = netdev_priv(bond_dev);
3233 struct rtnl_link_stats64 temp;
3234 struct list_head *iter;
3235 struct slave *slave;
3237 memset(stats, 0, sizeof(*stats));
3239 read_lock_bh(&bond->lock);
3240 bond_for_each_slave(bond, slave, iter) {
3241 const struct rtnl_link_stats64 *sstats =
3242 dev_get_stats(slave->dev, &temp);
3244 stats->rx_packets += sstats->rx_packets;
3245 stats->rx_bytes += sstats->rx_bytes;
3246 stats->rx_errors += sstats->rx_errors;
3247 stats->rx_dropped += sstats->rx_dropped;
3249 stats->tx_packets += sstats->tx_packets;
3250 stats->tx_bytes += sstats->tx_bytes;
3251 stats->tx_errors += sstats->tx_errors;
3252 stats->tx_dropped += sstats->tx_dropped;
3254 stats->multicast += sstats->multicast;
3255 stats->collisions += sstats->collisions;
3257 stats->rx_length_errors += sstats->rx_length_errors;
3258 stats->rx_over_errors += sstats->rx_over_errors;
3259 stats->rx_crc_errors += sstats->rx_crc_errors;
3260 stats->rx_frame_errors += sstats->rx_frame_errors;
3261 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3262 stats->rx_missed_errors += sstats->rx_missed_errors;
3264 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3265 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3266 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3267 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3268 stats->tx_window_errors += sstats->tx_window_errors;
3270 read_unlock_bh(&bond->lock);
3275 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3277 struct net_device *slave_dev = NULL;
3278 struct ifbond k_binfo;
3279 struct ifbond __user *u_binfo = NULL;
3280 struct ifslave k_sinfo;
3281 struct ifslave __user *u_sinfo = NULL;
3282 struct mii_ioctl_data *mii = NULL;
3286 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3298 * We do this again just in case we were called by SIOCGMIIREG
3299 * instead of SIOCGMIIPHY.
3306 if (mii->reg_num == 1) {
3307 struct bonding *bond = netdev_priv(bond_dev);
3309 read_lock(&bond->lock);
3310 read_lock(&bond->curr_slave_lock);
3311 if (netif_carrier_ok(bond->dev))
3312 mii->val_out = BMSR_LSTATUS;
3314 read_unlock(&bond->curr_slave_lock);
3315 read_unlock(&bond->lock);
3319 case BOND_INFO_QUERY_OLD:
3320 case SIOCBONDINFOQUERY:
3321 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3323 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3326 res = bond_info_query(bond_dev, &k_binfo);
3328 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3332 case BOND_SLAVE_INFO_QUERY_OLD:
3333 case SIOCBONDSLAVEINFOQUERY:
3334 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3336 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3339 res = bond_slave_info_query(bond_dev, &k_sinfo);
3341 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3350 net = dev_net(bond_dev);
3352 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3355 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3357 pr_debug("slave_dev=%p:\n", slave_dev);
3362 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3364 case BOND_ENSLAVE_OLD:
3365 case SIOCBONDENSLAVE:
3366 res = bond_enslave(bond_dev, slave_dev);
3368 case BOND_RELEASE_OLD:
3369 case SIOCBONDRELEASE:
3370 res = bond_release(bond_dev, slave_dev);
3372 case BOND_SETHWADDR_OLD:
3373 case SIOCBONDSETHWADDR:
3374 bond_set_dev_addr(bond_dev, slave_dev);
3377 case BOND_CHANGE_ACTIVE_OLD:
3378 case SIOCBONDCHANGEACTIVE:
3379 res = bond_ioctl_change_active(bond_dev, slave_dev);
3391 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3393 struct bonding *bond = netdev_priv(bond_dev);
3395 if (change & IFF_PROMISC)
3396 bond_set_promiscuity(bond,
3397 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3399 if (change & IFF_ALLMULTI)
3400 bond_set_allmulti(bond,
3401 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3404 static void bond_set_rx_mode(struct net_device *bond_dev)
3406 struct bonding *bond = netdev_priv(bond_dev);
3407 struct list_head *iter;
3408 struct slave *slave;
3412 if (USES_PRIMARY(bond->params.mode)) {
3413 slave = rtnl_dereference(bond->curr_active_slave);
3415 dev_uc_sync(slave->dev, bond_dev);
3416 dev_mc_sync(slave->dev, bond_dev);
3419 bond_for_each_slave(bond, slave, iter) {
3420 dev_uc_sync_multiple(slave->dev, bond_dev);
3421 dev_mc_sync_multiple(slave->dev, bond_dev);
3426 static int bond_neigh_init(struct neighbour *n)
3428 struct bonding *bond = netdev_priv(n->dev);
3429 const struct net_device_ops *slave_ops;
3430 struct neigh_parms parms;
3431 struct slave *slave;
3434 slave = bond_first_slave(bond);
3437 slave_ops = slave->dev->netdev_ops;
3438 if (!slave_ops->ndo_neigh_setup)
3441 parms.neigh_setup = NULL;
3442 parms.neigh_cleanup = NULL;
3443 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3448 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3449 * after the last slave has been detached. Assumes that all slaves
3450 * utilize the same neigh_cleanup (true at this writing as only user
3453 n->parms->neigh_cleanup = parms.neigh_cleanup;
3455 if (!parms.neigh_setup)
3458 return parms.neigh_setup(n);
3462 * The bonding ndo_neigh_setup is called at init time beofre any
3463 * slave exists. So we must declare proxy setup function which will
3464 * be used at run time to resolve the actual slave neigh param setup.
3466 * It's also called by master devices (such as vlans) to setup their
3467 * underlying devices. In that case - do nothing, we're already set up from
3470 static int bond_neigh_setup(struct net_device *dev,
3471 struct neigh_parms *parms)
3473 /* modify only our neigh_parms */
3474 if (parms->dev == dev)
3475 parms->neigh_setup = bond_neigh_init;
3481 * Change the MTU of all of a master's slaves to match the master
3483 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3485 struct bonding *bond = netdev_priv(bond_dev);
3486 struct slave *slave, *rollback_slave;
3487 struct list_head *iter;
3490 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3491 (bond_dev ? bond_dev->name : "None"), new_mtu);
3493 /* Can't hold bond->lock with bh disabled here since
3494 * some base drivers panic. On the other hand we can't
3495 * hold bond->lock without bh disabled because we'll
3496 * deadlock. The only solution is to rely on the fact
3497 * that we're under rtnl_lock here, and the slaves
3498 * list won't change. This doesn't solve the problem
3499 * of setting the slave's MTU while it is
3500 * transmitting, but the assumption is that the base
3501 * driver can handle that.
3503 * TODO: figure out a way to safely iterate the slaves
3504 * list, but without holding a lock around the actual
3505 * call to the base driver.
3508 bond_for_each_slave(bond, slave, iter) {
3509 pr_debug("s %p s->p %p c_m %p\n",
3511 bond_prev_slave(bond, slave),
3512 slave->dev->netdev_ops->ndo_change_mtu);
3514 res = dev_set_mtu(slave->dev, new_mtu);
3517 /* If we failed to set the slave's mtu to the new value
3518 * we must abort the operation even in ACTIVE_BACKUP
3519 * mode, because if we allow the backup slaves to have
3520 * different mtu values than the active slave we'll
3521 * need to change their mtu when doing a failover. That
3522 * means changing their mtu from timer context, which
3523 * is probably not a good idea.
3525 pr_debug("err %d %s\n", res, slave->dev->name);
3530 bond_dev->mtu = new_mtu;
3535 /* unwind from head to the slave that failed */
3536 bond_for_each_slave(bond, rollback_slave, iter) {
3539 if (rollback_slave == slave)
3542 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3544 pr_debug("unwind err %d dev %s\n",
3545 tmp_res, rollback_slave->dev->name);
3555 * Note that many devices must be down to change the HW address, and
3556 * downing the master releases all slaves. We can make bonds full of
3557 * bonding devices to test this, however.
3559 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3561 struct bonding *bond = netdev_priv(bond_dev);
3562 struct slave *slave, *rollback_slave;
3563 struct sockaddr *sa = addr, tmp_sa;
3564 struct list_head *iter;
3567 if (bond->params.mode == BOND_MODE_ALB)
3568 return bond_alb_set_mac_address(bond_dev, addr);
3571 pr_debug("bond=%p, name=%s\n",
3572 bond, bond_dev ? bond_dev->name : "None");
3574 /* If fail_over_mac is enabled, do nothing and return success.
3575 * Returning an error causes ifenslave to fail.
3577 if (bond->params.fail_over_mac)
3580 if (!is_valid_ether_addr(sa->sa_data))
3581 return -EADDRNOTAVAIL;
3583 /* Can't hold bond->lock with bh disabled here since
3584 * some base drivers panic. On the other hand we can't
3585 * hold bond->lock without bh disabled because we'll
3586 * deadlock. The only solution is to rely on the fact
3587 * that we're under rtnl_lock here, and the slaves
3588 * list won't change. This doesn't solve the problem
3589 * of setting the slave's hw address while it is
3590 * transmitting, but the assumption is that the base
3591 * driver can handle that.
3593 * TODO: figure out a way to safely iterate the slaves
3594 * list, but without holding a lock around the actual
3595 * call to the base driver.
3598 bond_for_each_slave(bond, slave, iter) {
3599 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3600 pr_debug("slave %p %s\n", slave, slave->dev->name);
3602 if (slave_ops->ndo_set_mac_address == NULL) {
3604 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3608 res = dev_set_mac_address(slave->dev, addr);
3610 /* TODO: consider downing the slave
3612 * User should expect communications
3613 * breakage anyway until ARP finish
3616 pr_debug("err %d %s\n", res, slave->dev->name);
3622 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3626 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3627 tmp_sa.sa_family = bond_dev->type;
3629 /* unwind from head to the slave that failed */
3630 bond_for_each_slave(bond, rollback_slave, iter) {
3633 if (rollback_slave == slave)
3636 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3638 pr_debug("unwind err %d dev %s\n",
3639 tmp_res, rollback_slave->dev->name);
3647 * bond_xmit_slave_id - transmit skb through slave with slave_id
3648 * @bond: bonding device that is transmitting
3649 * @skb: buffer to transmit
3650 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3652 * This function tries to transmit through slave with slave_id but in case
3653 * it fails, it tries to find the first available slave for transmission.
3654 * The skb is consumed in all cases, thus the function is void.
3656 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3658 struct list_head *iter;
3659 struct slave *slave;
3662 /* Here we start from the slave with slave_id */
3663 bond_for_each_slave_rcu(bond, slave, iter) {
3665 if (slave_can_tx(slave)) {
3666 bond_dev_queue_xmit(bond, skb, slave->dev);
3672 /* Here we start from the first slave up to slave_id */
3674 bond_for_each_slave_rcu(bond, slave, iter) {
3677 if (slave_can_tx(slave)) {
3678 bond_dev_queue_xmit(bond, skb, slave->dev);
3682 /* no slave that can tx has been found */
3686 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3688 struct bonding *bond = netdev_priv(bond_dev);
3689 struct iphdr *iph = ip_hdr(skb);
3690 struct slave *slave;
3693 * Start with the curr_active_slave that joined the bond as the
3694 * default for sending IGMP traffic. For failover purposes one
3695 * needs to maintain some consistency for the interface that will
3696 * send the join/membership reports. The curr_active_slave found
3697 * will send all of this type of traffic.
3699 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3700 slave = rcu_dereference(bond->curr_active_slave);
3701 if (slave && slave_can_tx(slave))
3702 bond_dev_queue_xmit(bond, skb, slave->dev);
3704 bond_xmit_slave_id(bond, skb, 0);
3706 bond_xmit_slave_id(bond, skb,
3707 bond->rr_tx_counter++ % bond->slave_cnt);
3710 return NETDEV_TX_OK;
3714 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3715 * the bond has a usable interface.
3717 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3719 struct bonding *bond = netdev_priv(bond_dev);
3720 struct slave *slave;
3722 slave = rcu_dereference(bond->curr_active_slave);
3724 bond_dev_queue_xmit(bond, skb, slave->dev);
3728 return NETDEV_TX_OK;
3732 * In bond_xmit_xor() , we determine the output device by using a pre-
3733 * determined xmit_hash_policy(), If the selected device is not enabled,
3734 * find the next active slave.
3736 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3738 struct bonding *bond = netdev_priv(bond_dev);
3740 bond_xmit_slave_id(bond, skb,
3741 bond->xmit_hash_policy(skb, bond->slave_cnt));
3743 return NETDEV_TX_OK;
3746 /* in broadcast mode, we send everything to all usable interfaces. */
3747 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3749 struct bonding *bond = netdev_priv(bond_dev);
3750 struct slave *slave = NULL;
3751 struct list_head *iter;
3753 bond_for_each_slave_rcu(bond, slave, iter) {
3754 if (bond_is_last_slave(bond, slave))
3756 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3757 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3760 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3764 /* bond_dev_queue_xmit always returns 0 */
3765 bond_dev_queue_xmit(bond, skb2, slave->dev);
3768 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3769 bond_dev_queue_xmit(bond, skb, slave->dev);
3773 return NETDEV_TX_OK;
3776 /*------------------------- Device initialization ---------------------------*/
3778 static void bond_set_xmit_hash_policy(struct bonding *bond)
3780 switch (bond->params.xmit_policy) {
3781 case BOND_XMIT_POLICY_LAYER23:
3782 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3784 case BOND_XMIT_POLICY_LAYER34:
3785 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3787 case BOND_XMIT_POLICY_LAYER2:
3789 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3795 * Lookup the slave that corresponds to a qid
3797 static inline int bond_slave_override(struct bonding *bond,
3798 struct sk_buff *skb)
3800 struct slave *slave = NULL;
3801 struct slave *check_slave;
3802 struct list_head *iter;
3805 if (!skb->queue_mapping)
3808 /* Find out if any slaves have the same mapping as this skb. */
3809 bond_for_each_slave_rcu(bond, check_slave, iter) {
3810 if (check_slave->queue_id == skb->queue_mapping) {
3811 slave = check_slave;
3816 /* If the slave isn't UP, use default transmit policy. */
3817 if (slave && slave->queue_id && IS_UP(slave->dev) &&
3818 (slave->link == BOND_LINK_UP)) {
3819 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3826 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3829 * This helper function exists to help dev_pick_tx get the correct
3830 * destination queue. Using a helper function skips a call to
3831 * skb_tx_hash and will put the skbs in the queue we expect on their
3832 * way down to the bonding driver.
3834 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3837 * Save the original txq to restore before passing to the driver
3839 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3841 if (unlikely(txq >= dev->real_num_tx_queues)) {
3843 txq -= dev->real_num_tx_queues;
3844 } while (txq >= dev->real_num_tx_queues);
3849 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3851 struct bonding *bond = netdev_priv(dev);
3853 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3854 if (!bond_slave_override(bond, skb))
3855 return NETDEV_TX_OK;
3858 switch (bond->params.mode) {
3859 case BOND_MODE_ROUNDROBIN:
3860 return bond_xmit_roundrobin(skb, dev);
3861 case BOND_MODE_ACTIVEBACKUP:
3862 return bond_xmit_activebackup(skb, dev);
3864 return bond_xmit_xor(skb, dev);
3865 case BOND_MODE_BROADCAST:
3866 return bond_xmit_broadcast(skb, dev);
3867 case BOND_MODE_8023AD:
3868 return bond_3ad_xmit_xor(skb, dev);
3871 return bond_alb_xmit(skb, dev);
3873 /* Should never happen, mode already checked */
3874 pr_err("%s: Error: Unknown bonding mode %d\n",
3875 dev->name, bond->params.mode);
3878 return NETDEV_TX_OK;
3882 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3884 struct bonding *bond = netdev_priv(dev);
3885 netdev_tx_t ret = NETDEV_TX_OK;
3888 * If we risk deadlock from transmitting this in the
3889 * netpoll path, tell netpoll to queue the frame for later tx
3891 if (is_netpoll_tx_blocked(dev))
3892 return NETDEV_TX_BUSY;
3895 if (!list_empty(&bond->slave_list))
3896 ret = __bond_start_xmit(skb, dev);
3905 * set bond mode specific net device operations
3907 void bond_set_mode_ops(struct bonding *bond, int mode)
3909 struct net_device *bond_dev = bond->dev;
3912 case BOND_MODE_ROUNDROBIN:
3914 case BOND_MODE_ACTIVEBACKUP:
3917 bond_set_xmit_hash_policy(bond);
3919 case BOND_MODE_BROADCAST:
3921 case BOND_MODE_8023AD:
3922 bond_set_xmit_hash_policy(bond);
3929 /* Should never happen, mode already checked */
3930 pr_err("%s: Error: Unknown bonding mode %d\n",
3931 bond_dev->name, mode);
3936 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3937 struct ethtool_cmd *ecmd)
3939 struct bonding *bond = netdev_priv(bond_dev);
3940 unsigned long speed = 0;
3941 struct list_head *iter;
3942 struct slave *slave;
3944 ecmd->duplex = DUPLEX_UNKNOWN;
3945 ecmd->port = PORT_OTHER;
3947 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3948 * do not need to check mode. Though link speed might not represent
3949 * the true receive or transmit bandwidth (not all modes are symmetric)
3950 * this is an accurate maximum.
3952 read_lock(&bond->lock);
3953 bond_for_each_slave(bond, slave, iter) {
3954 if (SLAVE_IS_OK(slave)) {
3955 if (slave->speed != SPEED_UNKNOWN)
3956 speed += slave->speed;
3957 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3958 slave->duplex != DUPLEX_UNKNOWN)
3959 ecmd->duplex = slave->duplex;
3962 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3963 read_unlock(&bond->lock);
3968 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3969 struct ethtool_drvinfo *drvinfo)
3971 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3972 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3973 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3977 static const struct ethtool_ops bond_ethtool_ops = {
3978 .get_drvinfo = bond_ethtool_get_drvinfo,
3979 .get_settings = bond_ethtool_get_settings,
3980 .get_link = ethtool_op_get_link,
3983 static const struct net_device_ops bond_netdev_ops = {
3984 .ndo_init = bond_init,
3985 .ndo_uninit = bond_uninit,
3986 .ndo_open = bond_open,
3987 .ndo_stop = bond_close,
3988 .ndo_start_xmit = bond_start_xmit,
3989 .ndo_select_queue = bond_select_queue,
3990 .ndo_get_stats64 = bond_get_stats,
3991 .ndo_do_ioctl = bond_do_ioctl,
3992 .ndo_change_rx_flags = bond_change_rx_flags,
3993 .ndo_set_rx_mode = bond_set_rx_mode,
3994 .ndo_change_mtu = bond_change_mtu,
3995 .ndo_set_mac_address = bond_set_mac_address,
3996 .ndo_neigh_setup = bond_neigh_setup,
3997 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3998 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3999 #ifdef CONFIG_NET_POLL_CONTROLLER
4000 .ndo_netpoll_setup = bond_netpoll_setup,
4001 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4002 .ndo_poll_controller = bond_poll_controller,
4004 .ndo_add_slave = bond_enslave,
4005 .ndo_del_slave = bond_release,
4006 .ndo_fix_features = bond_fix_features,
4009 static const struct device_type bond_type = {
4013 static void bond_destructor(struct net_device *bond_dev)
4015 struct bonding *bond = netdev_priv(bond_dev);
4017 destroy_workqueue(bond->wq);
4018 free_netdev(bond_dev);
4021 static void bond_setup(struct net_device *bond_dev)
4023 struct bonding *bond = netdev_priv(bond_dev);
4025 /* initialize rwlocks */
4026 rwlock_init(&bond->lock);
4027 rwlock_init(&bond->curr_slave_lock);
4028 INIT_LIST_HEAD(&bond->slave_list);
4029 bond->params = bonding_defaults;
4031 /* Initialize pointers */
4032 bond->dev = bond_dev;
4034 /* Initialize the device entry points */
4035 ether_setup(bond_dev);
4036 bond_dev->netdev_ops = &bond_netdev_ops;
4037 bond_dev->ethtool_ops = &bond_ethtool_ops;
4038 bond_set_mode_ops(bond, bond->params.mode);
4040 bond_dev->destructor = bond_destructor;
4042 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4044 /* Initialize the device options */
4045 bond_dev->tx_queue_len = 0;
4046 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4047 bond_dev->priv_flags |= IFF_BONDING;
4048 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4050 /* At first, we block adding VLANs. That's the only way to
4051 * prevent problems that occur when adding VLANs over an
4052 * empty bond. The block will be removed once non-challenged
4053 * slaves are enslaved.
4055 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4057 /* don't acquire bond device's netif_tx_lock when
4059 bond_dev->features |= NETIF_F_LLTX;
4061 /* By default, we declare the bond to be fully
4062 * VLAN hardware accelerated capable. Special
4063 * care is taken in the various xmit functions
4064 * when there are slaves that are not hw accel
4068 bond_dev->hw_features = BOND_VLAN_FEATURES |
4069 NETIF_F_HW_VLAN_CTAG_TX |
4070 NETIF_F_HW_VLAN_CTAG_RX |
4071 NETIF_F_HW_VLAN_CTAG_FILTER;
4073 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4074 bond_dev->features |= bond_dev->hw_features;
4078 * Destroy a bonding device.
4079 * Must be under rtnl_lock when this function is called.
4081 static void bond_uninit(struct net_device *bond_dev)
4083 struct bonding *bond = netdev_priv(bond_dev);
4084 struct list_head *iter;
4085 struct slave *slave;
4087 bond_netpoll_cleanup(bond_dev);
4089 /* Release the bonded slaves */
4090 bond_for_each_slave(bond, slave, iter)
4091 __bond_release_one(bond_dev, slave->dev, true);
4092 pr_info("%s: released all slaves\n", bond_dev->name);
4094 list_del(&bond->bond_list);
4096 bond_debug_unregister(bond);
4099 /*------------------------- Module initialization ---------------------------*/
4102 * Convert string input module parms. Accept either the
4103 * number of the mode or its string name. A bit complicated because
4104 * some mode names are substrings of other names, and calls from sysfs
4105 * may have whitespace in the name (trailing newlines, for example).
4107 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4109 int modeint = -1, i, rv;
4110 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4112 for (p = (char *)buf; *p; p++)
4113 if (!(isdigit(*p) || isspace(*p)))
4117 rv = sscanf(buf, "%20s", modestr);
4119 rv = sscanf(buf, "%d", &modeint);
4124 for (i = 0; tbl[i].modename; i++) {
4125 if (modeint == tbl[i].mode)
4127 if (strcmp(modestr, tbl[i].modename) == 0)
4134 static int bond_check_params(struct bond_params *params)
4136 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4137 int arp_all_targets_value;
4140 * Convert string parameters.
4143 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4144 if (bond_mode == -1) {
4145 pr_err("Error: Invalid bonding mode \"%s\"\n",
4146 mode == NULL ? "NULL" : mode);
4151 if (xmit_hash_policy) {
4152 if ((bond_mode != BOND_MODE_XOR) &&
4153 (bond_mode != BOND_MODE_8023AD)) {
4154 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4155 bond_mode_name(bond_mode));
4157 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4159 if (xmit_hashtype == -1) {
4160 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4161 xmit_hash_policy == NULL ? "NULL" :
4169 if (bond_mode != BOND_MODE_8023AD) {
4170 pr_info("lacp_rate param is irrelevant in mode %s\n",
4171 bond_mode_name(bond_mode));
4173 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4174 if (lacp_fast == -1) {
4175 pr_err("Error: Invalid lacp rate \"%s\"\n",
4176 lacp_rate == NULL ? "NULL" : lacp_rate);
4183 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4184 if (params->ad_select == -1) {
4185 pr_err("Error: Invalid ad_select \"%s\"\n",
4186 ad_select == NULL ? "NULL" : ad_select);
4190 if (bond_mode != BOND_MODE_8023AD) {
4191 pr_warning("ad_select param only affects 802.3ad mode\n");
4194 params->ad_select = BOND_AD_STABLE;
4197 if (max_bonds < 0) {
4198 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4199 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4200 max_bonds = BOND_DEFAULT_MAX_BONDS;
4204 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4205 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4206 miimon = BOND_LINK_MON_INTERV;
4210 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4215 if (downdelay < 0) {
4216 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4217 downdelay, INT_MAX);
4221 if ((use_carrier != 0) && (use_carrier != 1)) {
4222 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4227 if (num_peer_notif < 0 || num_peer_notif > 255) {
4228 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4233 /* reset values for 802.3ad */
4234 if (bond_mode == BOND_MODE_8023AD) {
4236 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");
4237 pr_warning("Forcing miimon to 100msec\n");
4242 if (tx_queues < 1 || tx_queues > 255) {
4243 pr_warning("Warning: tx_queues (%d) should be between "
4244 "1 and 255, resetting to %d\n",
4245 tx_queues, BOND_DEFAULT_TX_QUEUES);
4246 tx_queues = BOND_DEFAULT_TX_QUEUES;
4249 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4250 pr_warning("Warning: all_slaves_active module parameter (%d), "
4251 "not of valid value (0/1), so it was set to "
4252 "0\n", all_slaves_active);
4253 all_slaves_active = 0;
4256 if (resend_igmp < 0 || resend_igmp > 255) {
4257 pr_warning("Warning: resend_igmp (%d) should be between "
4258 "0 and 255, resetting to %d\n",
4259 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4260 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4263 /* reset values for TLB/ALB */
4264 if ((bond_mode == BOND_MODE_TLB) ||
4265 (bond_mode == BOND_MODE_ALB)) {
4267 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4268 pr_warning("Forcing miimon to 100msec\n");
4273 if (bond_mode == BOND_MODE_ALB) {
4274 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",
4279 if (updelay || downdelay) {
4280 /* just warn the user the up/down delay will have
4281 * no effect since miimon is zero...
4283 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",
4284 updelay, downdelay);
4287 /* don't allow arp monitoring */
4289 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4290 miimon, arp_interval);
4294 if ((updelay % miimon) != 0) {
4295 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4297 (updelay / miimon) * miimon);
4302 if ((downdelay % miimon) != 0) {
4303 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4305 (downdelay / miimon) * miimon);
4308 downdelay /= miimon;
4311 if (arp_interval < 0) {
4312 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4313 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4314 arp_interval = BOND_LINK_ARP_INTERV;
4317 for (arp_ip_count = 0, i = 0;
4318 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4319 /* not complete check, but should be good enough to
4321 __be32 ip = in_aton(arp_ip_target[i]);
4322 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4323 ip == htonl(INADDR_BROADCAST)) {
4324 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4328 if (bond_get_targets_ip(arp_target, ip) == -1)
4329 arp_target[arp_ip_count++] = ip;
4331 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4336 if (arp_interval && !arp_ip_count) {
4337 /* don't allow arping if no arp_ip_target given... */
4338 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4344 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4345 pr_err("arp_validate only supported in active-backup mode\n");
4348 if (!arp_interval) {
4349 pr_err("arp_validate requires arp_interval\n");
4353 arp_validate_value = bond_parse_parm(arp_validate,
4355 if (arp_validate_value == -1) {
4356 pr_err("Error: invalid arp_validate \"%s\"\n",
4357 arp_validate == NULL ? "NULL" : arp_validate);
4361 arp_validate_value = 0;
4363 arp_all_targets_value = 0;
4364 if (arp_all_targets) {
4365 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4366 arp_all_targets_tbl);
4368 if (arp_all_targets_value == -1) {
4369 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4371 arp_all_targets_value = 0;
4376 pr_info("MII link monitoring set to %d ms\n", miimon);
4377 } else if (arp_interval) {
4378 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4380 arp_validate_tbl[arp_validate_value].modename,
4383 for (i = 0; i < arp_ip_count; i++)
4384 pr_info(" %s", arp_ip_target[i]);
4388 } else if (max_bonds) {
4389 /* miimon and arp_interval not set, we need one so things
4390 * work as expected, see bonding.txt for details
4392 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");
4395 if (primary && !USES_PRIMARY(bond_mode)) {
4396 /* currently, using a primary only makes sense
4397 * in active backup, TLB or ALB modes
4399 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4400 primary, bond_mode_name(bond_mode));
4404 if (primary && primary_reselect) {
4405 primary_reselect_value = bond_parse_parm(primary_reselect,
4407 if (primary_reselect_value == -1) {
4408 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4410 NULL ? "NULL" : primary_reselect);
4414 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4417 if (fail_over_mac) {
4418 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4420 if (fail_over_mac_value == -1) {
4421 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4422 arp_validate == NULL ? "NULL" : arp_validate);
4426 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4427 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4429 fail_over_mac_value = BOND_FOM_NONE;
4432 /* fill params struct with the proper values */
4433 params->mode = bond_mode;
4434 params->xmit_policy = xmit_hashtype;
4435 params->miimon = miimon;
4436 params->num_peer_notif = num_peer_notif;
4437 params->arp_interval = arp_interval;
4438 params->arp_validate = arp_validate_value;
4439 params->arp_all_targets = arp_all_targets_value;
4440 params->updelay = updelay;
4441 params->downdelay = downdelay;
4442 params->use_carrier = use_carrier;
4443 params->lacp_fast = lacp_fast;
4444 params->primary[0] = 0;
4445 params->primary_reselect = primary_reselect_value;
4446 params->fail_over_mac = fail_over_mac_value;
4447 params->tx_queues = tx_queues;
4448 params->all_slaves_active = all_slaves_active;
4449 params->resend_igmp = resend_igmp;
4450 params->min_links = min_links;
4451 params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4454 strncpy(params->primary, primary, IFNAMSIZ);
4455 params->primary[IFNAMSIZ - 1] = 0;
4458 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4463 static struct lock_class_key bonding_netdev_xmit_lock_key;
4464 static struct lock_class_key bonding_netdev_addr_lock_key;
4465 static struct lock_class_key bonding_tx_busylock_key;
4467 static void bond_set_lockdep_class_one(struct net_device *dev,
4468 struct netdev_queue *txq,
4471 lockdep_set_class(&txq->_xmit_lock,
4472 &bonding_netdev_xmit_lock_key);
4475 static void bond_set_lockdep_class(struct net_device *dev)
4477 lockdep_set_class(&dev->addr_list_lock,
4478 &bonding_netdev_addr_lock_key);
4479 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4480 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4484 * Called from registration process
4486 static int bond_init(struct net_device *bond_dev)
4488 struct bonding *bond = netdev_priv(bond_dev);
4489 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4490 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4492 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4495 * Initialize locks that may be required during
4496 * en/deslave operations. All of the bond_open work
4497 * (of which this is part) should really be moved to
4498 * a phase prior to dev_open
4500 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4501 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4503 bond->wq = create_singlethread_workqueue(bond_dev->name);
4507 bond_set_lockdep_class(bond_dev);
4509 list_add_tail(&bond->bond_list, &bn->dev_list);
4511 bond_prepare_sysfs_group(bond);
4513 bond_debug_register(bond);
4515 /* Ensure valid dev_addr */
4516 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4517 bond_dev->addr_assign_type == NET_ADDR_PERM)
4518 eth_hw_addr_random(bond_dev);
4523 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4525 if (tb[IFLA_ADDRESS]) {
4526 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4528 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4529 return -EADDRNOTAVAIL;
4534 static unsigned int bond_get_num_tx_queues(void)
4539 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4541 .priv_size = sizeof(struct bonding),
4542 .setup = bond_setup,
4543 .validate = bond_validate,
4544 .get_num_tx_queues = bond_get_num_tx_queues,
4545 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4549 /* Create a new bond based on the specified name and bonding parameters.
4550 * If name is NULL, obtain a suitable "bond%d" name for us.
4551 * Caller must NOT hold rtnl_lock; we need to release it here before we
4552 * set up our sysfs entries.
4554 int bond_create(struct net *net, const char *name)
4556 struct net_device *bond_dev;
4561 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4562 name ? name : "bond%d",
4563 bond_setup, tx_queues);
4565 pr_err("%s: eek! can't alloc netdev!\n", name);
4570 dev_net_set(bond_dev, net);
4571 bond_dev->rtnl_link_ops = &bond_link_ops;
4573 res = register_netdevice(bond_dev);
4575 netif_carrier_off(bond_dev);
4579 bond_destructor(bond_dev);
4583 static int __net_init bond_net_init(struct net *net)
4585 struct bond_net *bn = net_generic(net, bond_net_id);
4588 INIT_LIST_HEAD(&bn->dev_list);
4590 bond_create_proc_dir(bn);
4591 bond_create_sysfs(bn);
4596 static void __net_exit bond_net_exit(struct net *net)
4598 struct bond_net *bn = net_generic(net, bond_net_id);
4599 struct bonding *bond, *tmp_bond;
4602 bond_destroy_sysfs(bn);
4603 bond_destroy_proc_dir(bn);
4605 /* Kill off any bonds created after unregistering bond rtnl ops */
4607 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4608 unregister_netdevice_queue(bond->dev, &list);
4609 unregister_netdevice_many(&list);
4613 static struct pernet_operations bond_net_ops = {
4614 .init = bond_net_init,
4615 .exit = bond_net_exit,
4617 .size = sizeof(struct bond_net),
4620 static int __init bonding_init(void)
4625 pr_info("%s", bond_version);
4627 res = bond_check_params(&bonding_defaults);
4631 res = register_pernet_subsys(&bond_net_ops);
4635 res = rtnl_link_register(&bond_link_ops);
4639 bond_create_debugfs();
4641 for (i = 0; i < max_bonds; i++) {
4642 res = bond_create(&init_net, NULL);
4647 register_netdevice_notifier(&bond_netdev_notifier);
4651 rtnl_link_unregister(&bond_link_ops);
4653 unregister_pernet_subsys(&bond_net_ops);
4658 static void __exit bonding_exit(void)
4660 unregister_netdevice_notifier(&bond_netdev_notifier);
4662 bond_destroy_debugfs();
4664 rtnl_link_unregister(&bond_link_ops);
4665 unregister_pernet_subsys(&bond_net_ops);
4667 #ifdef CONFIG_NET_POLL_CONTROLLER
4669 * Make sure we don't have an imbalance on our netpoll blocking
4671 WARN_ON(atomic_read(&netpoll_block_tx));
4675 module_init(bonding_init);
4676 module_exit(bonding_exit);
4677 MODULE_LICENSE("GPL");
4678 MODULE_VERSION(DRV_VERSION);
4679 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4680 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4681 MODULE_ALIAS_RTNL_LINK("bond");